20140957 Ver 2_Attachment 6_Wilson Report_20170818

Attachment 6

Evaluating Market Need for the Atlantic Coast Pipeline
James F. Wilson, Wilson Energy Economics
Prepared for the Southern Environmental Law Center
April 3, 2017
I. INTRODUCTION
1. On September 18, 2015, Atlantic Coast Pipeline, LLC ("Atlantic") filed an
abbreviated application for a Certificate of Public Convenience and Necessity to
construct the Atlantic Coast Pipeline ("ACP") in Federal Energy Regulatory Commission
("Commission") Docket No. CP15-554 ("Application"). The Application states (p. 33),
"Atlantic's precedent agreements, for nearly all of the Project capacity, demonstrate the
long-term market need for the Project from major electric utilities and local distribution
companies in Virginia and North Carolina." The Application further states (pp. 6-7) that
an estimated 79.2 percent of the natural gas transported by the ACP would be used as a
fuel for existing and proposed power generation facilities, and shows (Exhibit 1) that 93%
of the contracted ACP capacity is under executed precedent agreements with affiliates
of the project's sponsors.' In particular, 70% of the ACP capacity under executed
precedent agreements would serve Dominion Virginia Power ("DVP"), Duke Energy
Carolinas, LLC ("Duke Energy Carolinas"), and Duke Energy Progress, LLC ("Duke Energy
Progress").
2. This report first discusses the Commission's policy for reviewing pipeline
applications and evaluating the market need for a project, focusing on the Commission's
historical reliance on precedent agreements without "looking behind" those
agreements. The report discusses whether this practice remains appropriate for a new
class of projects before the Commission at this time, of which ACP is one example, that
' Application, pp. 3-4 and Exhibit I.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 1 of 23

are primarily intended to provide gas supply to new gas-fired electric generation, but
are largely supported by regulated LDCs and EDCs affiliated with the project sponsors.
3. The report then reviews the DVP, Duke Energy Carolinas, and Duke
Energy Progress forecasts of future electric loads that motivated entering into contracts
for ACP capacity. DVP, Duke Energy Carolinas, and Duke Energy Progress all filed
Integrated Resource Plans ("IRPs") in 2016.2 Pursuant to earlier assignments in these
state proceedings, James Wilson reviewed the three utilities' forecasts of peak loads and
capacity requirements .3 This report draws on this earlier work, which is attached
hereto.
II. SUMMARY AND CONCLUSIONS
4. In reviewing applications for authorization to build new natural gas
pipelines, the Commission is charged with identifying whether there is need for a
project, and Commission policy calls for considering all evidence submitted. However,
under a long-standing practice, the Commission generally considers long-term contracts
for the pipeline's firm capacity as evidence of market need, even if the contracts are
with affiliates of the pipeline's sponsors. Affiliated customers typically have either been
at risk for realizing the value of the capacity (e.g., producers or marketers), or have been
regulated LDCs with incremental needs to serve firm customers. Accordingly, the
2 Dominion Virginia Power's and Dominion North Carolina Power's Report of its Integrated Resource Plan
("DVP 2016 IRP") filed April 29, 2016 in Virginia State Corporation Commission Case No. PUE-2016-00049;
Duke Energy Carolinas North Carolina Integrated Resource Plan (Biennial Report) ("Duke Energy Carolinas
2016 IRP") and Duke Energy Progress North Carolina Integrated Resource Plan (Biennial Report) ("Duke
Energy Progress 2016 IRP"), both filed on September 1, 2016 in North Carolina Utilities Commission
Docket No. E-100 Sub 147.
3 Wilson, James F., Direct Testimony on behalf of Environmental Respondents ("DVP Testimony"), filed in
Virginia State Corporation Commission Case No. PUE-2016-00049 on August 17, 2016; Wilson, James F.,
Review and Evaluation of the Peak Load Forecasts for the Duke Energy Carolinas and Duke Energy
Progress 2016 Integrated Resource Plans ("Duke Load Forecast Report") and Review and Evaluation of the
Reserve Margin Determinations for the Duke Energy Carolinas and Duke Energy Progress 2016 Integrated
Resource Plans ("Duke Reserve Margin Report"), Attachments A and B, respectively, to the Comments of
Southern Alliance for Clean Energy, Natural Resources Defense Council and the Sierra Club, filed in North
Carolina Utilities Commission Docket No. E-100 Sub 147 on February 17, 2017; attached hereto.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 2 of 23

Commission has generally not "looked behind" contracts and separately evaluated the
details of the contracts, or whether demand and supply conditions indicate a need for
the new pipeline.
5. At the present time, a new group of natural gas pipeline proposals is
before the Commission whose characteristics and circumstances are unlike those of any
project the Commission has considered and approved in the past. These projects are
large, and largely intended to provide incremental gas supply for planned gas-fired
power plants that are not yet under construction. However, the projects are not
primarily supported by merchant generators, or by at -risk marketers, or by gas
producers. Instead, this new group of proposed pipelines is relying upon precedent
agreements with local gas or electric distribution companies ("LDCs", "EDCs"), either
directly or through affiliates of the LDC or EDC. These regulated utilities, who in many
instances are affiliated with the pipeline projects' owners, would seek to recover the
cost of the long-term pipeline commitments from their captive ratepayers.
6. At the present time, the future need for incremental gas supply for new
gas-fired electric generation is highly uncertain, due to weak or non-existent electric
load growth, the uncertain pace of coal and nuclear plant retirements, and the
increasing penetration of wind, solar and other renewable resources, among other
factors. The potential need for a particular new pipeline is further complicated by the
growing list of competing pipeline projects before the Commission at this time (of which
many provide incremental takeaway capacity from the Marcellus/Utica region in various
directions). Year after year, electric load forecasts are adjusted downward and new
pipeline projects are put forward. However, LDCs and EDCs may not be re-evaluating
their planned commitments to their affiliates' pipeline projects, to determine whether
the capacity is still considered needed, or whether other, more economical alternatives
have become available.
7. Where LDCs and EDCs have (directly or indirectly) entered into precedent
agreements with (often affiliated) pipeline developers, there generally would be some
form of state -level review of the contracts, and cost recovery could ultimately be
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 3 of 23

disallowed if the cost of the pipeline commitment is found imprudent. However, the
state -level review is typically very cursory, and may take into account considerations
independent of the need for the project, such as potential benefits of the project due to
its impact on regional natural gas and electricity prices. Consequently, even with some
form of state -level approval, LDC or EDC commitments to a pipeline project may not
indicate need for the project. Even worse, the state -level approval may reflect a
willingness to subsidize the pipeline project at ratepayers' expense, in pursuit of the
potential benefits to the state of suppressing natural gas and electric prices below the
levels that would otherwise have obtained.
8. Accordingly, under these circumstances a pipeline project's precedent
agreements with LDCs or EDCs, even if the contracts have been approved at the state
level, may not be a reliable indicator of the market need for a project, which in any case
is highly uncertain at this time due to weak electric load growth. To remain true to its
well-established and successful policy for reviewing and approving pipeline projects, the
Commission will need to set aside its long-standing practice of not looking behind the
firm, binding contracts that support a pipeline project. This is necessary at this time to
ensure there is market need for a pipeline, and also to verify that a project is not being
supported largely as a subsidy in order to suppress prices. Such a subsidized project
would depress the value of existing and proposed competing pipeline capacity, and this
could have a negative impact on the robust market for pipeline expansions that has
developed under the Commission's current policies.
9. With regard to ACP, the precedent agreements were signed in 2014. At
the time, DVP was forecasting robust growth in electric demand (14 percent increase
from 2017 to 2025). However, the trend in DVP's electric load has been flat over the
past decade, with the only growth coming from data centers. Other forecasts, such as
the more recent forecast for the Dominion zone prepared by PJM Interconnection, LLC
("PJM"), anticipate much slower load growth -- only 3.5% to 2025. While new and
expanded data centers are likely to provide growth in this zone, at least some of the
companies building data centers intend to supply them with renewable sources of
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 4 of 23

electricity. As discussed further below, the electric demand DVP was forecasting for
2020 now appears unlikely to be reached for at least a decade. Consequently, a re-
evaluation of DVP's need for the ACP capacity would show that any such need is many
years away at best.
10. Duke Energy Carolinas has also reduced its load forecast sharply, and
even before reducing the forecast, had no plans to build new gas-fired generation
before 2022. If Duke Energy Carolinas were to re-evaluate its commitment to ACP, it
would likely find that the commitment is not needed at this time, it is unclear when such
capacity might be needed, and it is also unknown whether better options might be
available at such time as incremental pipeline capacity does become needed.
11. Similarly, Duke Energy Progress has also lowered its load forecast since
2014, at which time it was not planning to build additional gas-fired generation until
2021. If Duke Energy Progress were to re-evaluate its commitment to ACP, like DVP
and Duke Energy Carolinas, it would likely find that the commitment is not needed at
this time, it is unclear when such capacity might be needed, and it is also unknown
whether better options might be available at such time as incremental pipeline capacity
does become needed.
12. The remainder of this report is organized as follows. The next section
discusses the Commission's policy for evaluating the market need for pipeline proposals,
and whether the historical practice of not "looking behind" precedent agreements
remains appropriate for projects that are largely supported by LDCs and EDCs affiliated
with a project's sponsors. The final three sections review the past and updated electric
load forecasts for DVP, Duke Energy Carolinas, and Duke Energy Progress, and whether
these forecasts suggest a need for the ACP capacity.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 5 of 23

III. EVALUATING MARKET NEED FOR NATURAL GAS PIPELINE PROJECTS:
COMMISSION PRACTICES UNDER CHANGING CIRCUMSTANCES
13. The Commission's policies for reviewing pipeline projects evolved
through the 1980s and 1990s, culminating in the 1999 Certificate Policy Statement.4
The Certificate Policy Statement established criteria for determining whether there is a
need for a project and whether it will serve the public interest. The Commission
balances the public benefits against potential adverse consequences, and gives
appropriate consideration to, among other things, enhancement of competitive
alternatives, the possibility of overbuilding, unnecessary disruption of the environment,
and unneeded exercise of eminent domains A threshold requirement is that the
pipeline and its customers must be prepared to financially support the project without
relying on subsidization from the pipeline's existing customers.6
14. To evaluate the need for a project, "the Commission considers all
evidence submitted reflecting on the need for the project, including, but not limited to,
precedent agreements, demand projections, potential cost savings to consumers, or a
comparison of projected demand with the amount of capacity currently serving the
market."' However, the Commission has generally found that long-term commitments
serve as significant evidence of demand for a project.$
15. On a number of occasions, intervenors have challenged a pipeline's
showing of need when it relied upon long-term commitments with affiliates of the
pipeline. The Certificate Policy Statement acknowledged that "[a] project that has
precedent agreements with multiple new customers may present a greater indication of
4 Certification of New Interstate Natural Gas Pipeline Facilities, 88 FERC $ 61,227 (1999), order on
clarification, 90 FERC ¶ 61,128, order on clarification, 92 FERC ¶ 61,094 (2000) ("Certificate Policy
Statement").
6 See, for instance, Transcontinental Gas Pipe Line Company, LLC, 158 FERC ¶ 61,125 (2017) at P 20.
61d at P 21.
Id at P 28.
8 Id at P 28.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 6 of 23

need than a project with only a precedent agreement with an affiliate.i9 However, the
Commission generally has given "equal weight" to contracts with affiliates and non -
affiliates, and has not "look[ed] behind" contracts to determine whether customer
commitments represent genuine growth of market demand,10 as long as the
commitments are long-term and binding, and there is no evidence of "self-dealing" to
support the need for the project, 11 or evidence of a violation of the Commission's rules
regarding affiliate relationships.12
16. Precedent agreements typically commit the buyer to enter into a firm
contract subject to the satisfaction of various conditions precedent, such as receipt of
various regulatory authorizations. In granting certificate authorizations, the Commission
typically requires execution of firm contracts for volumes and service terms equivalent
to those in its precedent agreements prior to the commencement of construction. 13
17. In most certificate proceedings, the issue of contracts with affiliates does
not arise, because there are no such contracts or they represent only a small fraction of
the pipeline's commitments. Where the issue has been raised, the affiliated entities
have often been natural gas producers or marketers. The Commission has recognized
that such entities would generally be at risk to realize value and recover the cost of their
firm pipeline commitments in a competitive market, and, therefore, contracts with
producers or marketers, even if affiliated with the pipeline, would indicate market
need .14
9 Certificate Policy Statement at pp. 25-26.
10 See, for instance, Paiute Pipeline Company, 151 FERC ¶ 61,132 (2015) at P 32.
11 Constitution Pipeline Company, LLC, 149 FERC ¶ 61,199 (2014) at P 28.
12 Transcontinental Gas Pipe Line Company, LLC, 141 FERC ¶ 61,091 (2012) at P 21.
13 See, for instance, Transcontinental Gas Pipe Line Company, LLC, 158 FERC ¶ 61,125 (2017) at P 174;
Eastern Shore Natural Gas Company, 132 FERC ¶ 61,204 (2010) P 32.
14 See, for instance, Millenium Pipeline Company, L.P., 100 FERC ¶ 61,277 (2002) at P 57; Constitution
Pipeline Company, LLC, 149 FERC ¶ 61,199 (2014) at P 28; Transcontinental Gas Pipe Line Company, LLC,
141 FERC ¶ 61,091 (2012) at P 21.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 7 of 23

18. In other instances, the contracts were with affiliated LDCs. However,
LDCs' incremental pipeline capacity needs are typically small and associated with firm
customer demand, and, therefore, are indicative of need .15
19. In one unusual instance, the Commission found that an agreement with
an affiliate was not reliable evidence of market need. The affiliate had been formed and
the agreement signed just days before the application was filed, apparently to avoid
dismissal of the application .16 Notwithstanding this exception, the Commission has
been consistent in its policy of considering long-term, binding agreements as indicative
of market need, even if the agreements are with affiliated entities, and not looking
behind the contracts or requiring additional analysis of market demand.
20. However, recently a few substantial pipeline projects have been
proposed that are primarily supported by affiliates and that have become controversial.
These new pipeline proposals differ from the projects the Commission has approved in
the past despite contractual support by affiliated entities: they are large; the capacity is
primarily intended to supply new gas-fired power plants that are not yet under
construction; and much of the contractual support is by affiliated, regulated LDCs
and/or EDCs.
21. The underlying market need for the incremental gas supply these
pipelines would provide is reasonably considered highly uncertain at this time. Electric
load growth has been very weak over the past decade, and forecasts have consistently
been revised downward. In addition, the need for new gas-fired power plants also
depends upon the uncertain timing of coal plant retirements and the rate of growth of
renewable resources such as wind and solar, among other factors. And in some cases,
there are competing pipeline projects that may have stronger market support.
22. As noted above, the Commission would generally not be concerned by a
high degree of uncertainty about the need for a pipeline if it was supported by contracts
is See, for instance, Eastern Shore Natural Gas Company, 132 FERC ¶ 61,204 (2010).
16 Independence Pipeline Company, 89 FERC ¶ 61,283 (1999) at p. 51.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 8 of 23

with entities who will be at risk to realize value and recover the cost of their firm
pipeline commitments in a competitive market. However, some of these new projects
are primarily supported by entities who likely would not be at risk, but would instead
attempt to recover much or all of the cost of the commitments from captive ratepayers.
This might not cause serious concern about the need for a project, if there was evidence
that the commitment, despite the uncertainties, had been judged reasonable and
prudent from the ratepayers' standpoint, for instance by a state regulatory body. In
various instances such review has not occurred, or was cursory, so it remains unclear
whether a detailed and objective evaluation of the need has been performed.
23. In addition, with the passage of time, load forecasts are revised (recently,
nearly always downward), and other, competing pipelines are approved and begin
construction. The need for a pipeline project might change. But LDCs and EDCs might
be reticent to pull support for their affiliate's project, even if the case for its need
weakens, or if more economical alternatives are available. While LDCs and EDCs face
some risk that the state regulator might disallow the costs of a long-term pipeline
commitment if deemed imprudent, there is generally a high hurdle to take such action
when the regulator has earlier given some form of approval to the commitment.
24. In other instances, a state regulator may have supported a precedent
agreement, despite the uncertainties and risks to ratepayers, taking into consideration
studies showing benefits to the state resulting from the impact of the pipeline project
on natural gas and electricity prices. 17 That is, the regulator may be supporting the
project, and the contracts that put ratepayers at risk for the value of the pipeline
capacity, taking into account alleged price suppression that would benefit other energy
17 See, for instance, ICF International, The Economic Impacts of the Atlantic Coast Pipeline, prepared for
Dominion Transmission, Inc., February 9, 2015 (estimating the benefits of the pipeline due to lowering of
natural gas and electricity prices in Virginia and North Carolina, among other alleged benefits); ICF
Resources, LLC, Impact of the NEXUS Pipeline on Michigan Energy Markets, submitted to DTE Electric
November 2015 (estimating the benefits of the pipeline due to lowering of Michigan natural gas and
electric prices, among other alleged benefits); ICF International, Access Northeast Project - Reliability
Benefits and Energy Cost Savings to New England, prepared for Eversource Energy and Spectra Energy,
February 18, 2015 (estimating the benefits of additional pipeline capacity to New England due to lowering
of natural gas and electricity prices, among other alleged benefits).
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 9 of 23

consumers in the state. In such instances the state support is arguably a subsidy to the
pipeline project, and an exercise of buyer market power to suppress prices. This raises
additional concern about the impact on other pipeline projects in the near term. In
addition, the Commission should be concerned that entry by subsidized pipeline
projects that suppress prices could have a dampening effect on future market-driven
pipeline expansions, undermining the Commission's policies and ultimately harming
consumers.
25. The following are examples of pipeline projects currently or recently
before the Commission that are primarily intended to supply future electric generation,
and that are largely supported not by at -risk electric generators, but by affiliated LDCs
and EDCs:
a. NEXUS Gas Transmission (application filed November 20, 2015 in FERC
Docket No. CP16-22): 1,500,000 Dth/d from Ohio to Michigan; largely
supported by Michigan and Ontario LDCs and EDCs affiliated with the
project sponsors, and primarily needed to serve future gas-fired
generation.
b. Access Northeast Expansion (application filed November 3, 2015 in FERC
Docket No. PF16-1, currently on hold): 925,000 Dth/d from Pennsylvania
to New England; was to be contracted to EDCs affiliated with the project
sponsors, with intent to release the capacity to merchant generators.
c. Northeast Energy Direct (application filed November 20, 2015 in FERC
Docket No. CP16-21, later withdrawn): 1,200,000 Dth/d from
Pennsylvania to New England; the downstream segment was largely
supported by LDCs with intent to release the capacity to merchant
generators.
d. and Atlantic Coast Pipeline, the subject of this report.
26. The Commission has approved numerous projects in recent years, of
which most will provide incremental takeaway capacity from the Marcellus/Utica
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 10 of 23

production region and, in contrast to the projects listed above, were generally
supported by at -risk entities: producers, marketers and mid -stream companies.18
Projects largely intended to serve future needs for gas-fired electric generation, but
mainly supported by LDCs and EDCs, attempt to work around the fact that merchant
electric generators and generation project developers typically will not commit to firm
gas supply, but instead rely upon short-term natural gas markets. This is usually more
economical for electric generation than committing to firm pipeline capacity.
27. This discussion suggests that in instances where a pipeline is largely
intended to serve electric generation, but much of its support is from LDCs and EDCs
affiliated with the project's sponsors, it is appropriate for the Commission to take a
closer look at the actual need for a pipeline, consistent with the Commission's policy to
"consider all evidence submitted reflecting on the need for the project." To the extent a
propose pipeline is supported by contracts with downstream affiliates of the sponsors,
the Commission should consider the extent of the state -level review of the
commitments to firm pipeline capacity by LDCs and EDCs, and the current, updated
electric and natural gas demand and supply circumstances. There is some evidence the
Commission is beginning to "look behind" the contracts in such instances.19 Where
there has been very little state -level examination of the need for a pipeline and for
specific utility commitments, the Commission should give additional attention to
evidence presented regarding downstream electric and natural gas demands. Where
the state -level support may be partly based upon potential energy price suppression
benefits, the Commission should further consider whether the support represents a
subsidy of the project by ratepayers. If so, the Commission might further scrutinize the
18 See Federal Energy Regulatory Commission, Approved Major Pipeline Projects (2009 -Present), updated
regularly and available at https://www.ferc.gov/industries/gas/indus-act/pipelines/approved-
projects.asp.
19 See, for instance, Memorandum dated October 25, 2016 summarizing discussions between FERC staff
and the New Jersey Division of Rate Counsel regarding the state -level process for review of the need for
gas capacity, affiliate relationships, and other matters, filed October 26, 2016 in FERC Docket No. CF15-
558.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 11 of 23

showing of project need, and/or condition any approval of the project on provisions to
mitigate inappropriate price impacts.
28. The remaining three sections of this report review some of the evidence
of market need for ACP based on the IRP filings of DVP, Duke Energy Carolinas and Duke
Energy Progress. With regard to DVP, the Virginia State Corporation Commission
approved its 2016 IRP filing, but only as a "planning document", noting that its approval
"does not in any way create the slightest presumption that resource options contained
in the approved IRP will be approved in a future certificate of public and convenience
and necessity, rate adjustment clause, fuel factor or other type of proceeding governed
by different statutes ."20 With regard to Duke Energy Carolinas and Duke Energy
Progress, both companies received approval for their precedent agreements with ACP in
2014,21 but neither company mentioned ACP in its 2016 IRP filing. The companies did
submit testimony in the IRP docket on February 16, 2017, discussing natural gas issues
and the precedent agreements with ACP .22
IV. DVP'S LOAD FORECASTS AND CAPACITY NEEDS
29. To evaluate a long-term load forecast, it is useful to begin with a review
of recent trends in load. Actual peak loads will tend to vary substantially from year to
year, reflecting the presence or absence of the type of extremely hot weather than can
cause the highest summer peak loads. Consequently, it is useful to review recent
"weather -normalized" peak loads (also called "weather -adjusted" peak loads; these
terms are used synonymously in this report). Weather -normalized peak loads are
20 Virginia State Corporation Commission Case No. PUE-2016-00049, Final Order, December 14, 2016, pp.
2-3, acronyms and citations omitted; available at
http://www.scc.virginia.gov/docketsearch/DOCS/3c`/`23vOl!.PDF
21 North Carolina Utilities Commission Docket Nos. E-2 Sub 1052 and E-7 Sub 1062, Order Accepting
Affiliate Agreements, Allowing Payment Thereunder, and Granting Limited Waiver of Code of Conduct,
October 29, 2014.
22 North Carolina Utilities Commission Docket No. E-100 Sub 147, Duke Energy Carolinas, LLC and Duke
Energy Progress, LLC's Testimony on Natural Gas Issues, February 17, 2017, available at
http://starwl.ncuc.net/NCUC/ViewFile.aspx?ld=5edd7611-95bl-4822-917f-e4O970de9f72
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 12 of 23

estimates of what the peak load would have been in a historical period had the peak
occurred on a day with the typical peak -causing weather. Weather -normalized
historical peak loads remove the impact of weather variability and reveal the underlying
peak load trend due to other factors such as economic and demographic trends,
changes in industry and end-use technologies, and energy efficiency.
30. DVP does not calculate weather -normalized historical peaks. However,
PJM Interconnection, LLC ("PJM"), the Regional Transmission Organization that manages
the wholesale power markets for all or part of 13 states (including Virginia and North
Carolina) and the District of Columbia, prepares load forecasts and calculates weather -
normalized peaks for the Dominion transmission zone. DVP represents about 87%, and
other load -serving entities the remaining 13%, of the Dominion zone's peak loads.
31. Figure 1 shows recent Dominion zone weather -normalized peak loads, as
calculated by PJM. The weather -normalized peak loads have been quite flat over the
past decade; the 2007 and 2015 values are nearly identical. Even in the post -recession
period (from about 2010 to present) when economic growth returned, peak loads have
been flat on a weather -normalized basis.
32. Figure 1 also shows PJM's January 2017 forecast of peak loads for the
Dominion zone. PJM's forecast anticipates growth in the Dominion zone peak load
despite the recent trends. This may reflect expectations of somewhat stronger
economic growth than has occurred in recent years. In light of the recent trend, this
forecast appears somewhat optimistic, but within a reasonable range.
33. In recent years a number of data centers have been constructed in
Northern Virginia and have become a major new source of electric demand in the
Dominion zone. Additional construction and expansion of data centers is expected in
this area over the coming years. Accordingly, both DVP and PJM have adjusted their
recent Dominion zone forecasts applying estimates of future data center loads. For this
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 13 of 23

purpose both DVP and PJM rely on a study and forecast by Quanta Technology.23 Using
this information on the data center loads, Figure 2 shows the weather -normalized
history and PJM forecast for the total Dominion zone load, and also for all loads other
than the data center loads.
Figure 1: PJM's 2017 Dominion Zone Peak Load Forecast
(unrestricted, non -coincident summer peak loads; IVIW)
24,000
23,000 —4—PJM DOM Zone peak forecast (Jan. 2017)
22,000
—i—Weather-normalized peak
21,000
20,000
19,000
18,000
17,000
Sources: PJM; PJM 2017 Load Forecast Report and supporting data.
16,000
2005 2007 2009 2011 2013 2015 2017 2019 2021 2023 2025 2027
34. Figure 2 reveals that the Dominion zone trend over the past decade is
actually down, for all loads other than the data centers. PJM's forecast anticipates very
little growth in Dominion zone peak loads in the coming years for all consumers outside
the data center segment. Of course, the future growth in data centers and data center
electric loads are highly uncertain, especially beyond the first year or two, as it is
unknown where and when additional construction of data centers may occur. It should
23 PJM, Item 8: Dominion Load Forecast Adjustment, PJM Load Analysis Subcommittee meeting October
19, 2017, available at http://www.pjm.com/—/media/committees-
groups/subcommittees/las/20161019/20161019-item-08-dominion-load-forecast-adjustment.ashx.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 14 of 23

also be noted that some of the companies that build data centers have committed to
meeting 100% of their energy needs from renewable sources .24
24,000
23,000
22,000
21,000
20,000
19,000
18,000
17,000
16,000 '
2005
Figure 2: PJM 2017 Dominion Zone Forecast w/o Data Centers
(unrestricted, non -coincident summer peak loads; MW)
+PJM DOM Zone peak forecast (Jan. 201
—*— PJM DOM Zane 2017 net of data center:
—*—Weather -normalized peak
—+— W/N peaks net of data centers
Sources: PJM; PJM 2017 Load Forecast Report and suppo
2007 2009 2011 2013 2015 2017 2019 2021 2023 2025 2027
35. Figure 3 adds DVP's peak load forecast, from its 2016 IRP. This forecast,
prepared in September 2015, appears extremely high. For 2021, it exceeds PJM's
forecast (which also appears somewhat optimistic compared to recent trends), by over
2,000 MW. DVP's forecast is too high, primarily because it uses an econometric
approach based on a thirty-year historical period, which fails to capture recent trends (a
detailed critique is included in my DVP Testimony, pp. 7-22).
24 For example, Google has announced that it now purchases 100% renewable energy to match
consumption for its global operations including data centers and offices; see Google Data Centers —
Renewable Energy, available at https://www.google.com/about/datacenters/renewable/index.htmi.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 15 of 23

24,000
23,000
22,000
21,000
20,000
19,000
18,000
17,000
Figure 3: DVP's Dominion Zone Peak Load Forecast
(unrestricted, non -coincident summer peak loads; MW)
t DVP 2016 IRP DOM zone peak forecast
—0— DVP 2016 IRP forecast net of data centers
PJM DOM Zone peak forecast (Jan. 2017)
—*— PJM DOM Zone 2017 net of data centers
—r--Weather-normalized peak
+— W/N peaks net of data centers
rxr
�r
dr
r
♦..
Sources: DVP 2016 IRP; PJM; PJM 2017 Load Forecast Report and supporting data.
16,000
2005 2007 2009 2011 2013 2015 2017 2019 2021 2023 2025 2027
36. To determine generating capacity requirements, DVP includes a 12.46%
reserve margin over its peak load forecast (roughly based on PJM's approach to
determining capacity requirements, and taking into account that PJM assigns capacity
needs based on coincident peak loads). Therefore, the gap between the capacity
requirements as reflected in the DVP 2016 IRP, and requirements based on PJM's more
recent forecast, is even larger.
37. This discussion has focused on peak load forecasts; while fuel needs will
depend more on energy (GWHs) than peaks (MWs). However, while energy drives fuel
consumption, the need for new gas-fired capacity would generally be based upon
capacity not energy needs. In any case, energy trends tend to follow peak load trends.
Figure 4 compares PJM's and DVP's Dominion zone energy forecasts, showing that, as
with peaks, DVP's forecast is much higher. Based on PJM's forecast, incremental
capacity needs would be put off for many years.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 16 of 23

38. In its 2016 IRP, DVP evaluated multiple resource scenarios, but most
required additional gas-fired capacity beyond plants already under construction only in
2022 (under one scenario, 2021; DVP 2016 IRP, Figure 1.3.1, p. 5, and Figure 1.4.1, 2016
Studied Plans, p. 13).
116%
114%
112%
110%
108%
106%
104%
Figure 4: PJM and DVP Energy Growth Forecasts
(2017=100)
113.9%
fDVP's forecast (DVP LSE)
112.4%
—*—PJM Forecast (Dominion zone)
110.7
1%
06.9%
08.9%
103.2% "
103.3% 103.5%
102/0 102.6%
01.5402.5% 102.5% 102.3% 102.3%
100%
Sources: DVP 20161RP; PJM; PJM 2017 Load Forecast Report and supporting data.
98%
2016 2017
2018 2019 2020 2021 2022 2023 2024 2025 2026 1
39. While the DVP 2016 IRP mentions ACP, it also suggests that DVP has
already secured firm transportation for its new gas-fired Greenville County Power
Station that is under construction, and that ACP would primarily provide flexibility and
reliability (DVP 2016 IRP, p. 87). Consequently, with an updated load forecast much
closer to PJM's, DVP would not appear to need any ACP capacity in the coming years.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 17 of 23

V. DUKE ENERGY CAROLINAS' LOAD FORECASTS AND CAPACITY NEEDS
40. Like DVP, Duke Energy Carolinas anticipated robust electric load growth
in 2014, when it entered into the precedent agreement with ACP.25 However, by 2016
its expectations had been substantially lowered. Wilson Energy Economics' Duke Load
Forecast Report found that Duke Energy Carolinas' summer peak load forecast reflected
growth in excess of trend, but appeared to fall within a reasonable range, while its
winter peak load forecast appeared somewhat high (p. 13). Figure 5 shows Duke Energy
Carolinas' 2014 and 2016 forecasts (with energy efficiency programs), along with its
weather adjusted historical peaks and a trend line based on the weather adjusted peaks.
22,000
21,500
21,000
20,500
20,000
19,500
19,000
18,500
18,000
17,500
17,000
16,500
Figure 5: Duke Energy Carolinas Summer Peaks, Historical and Forecast
X2014 IRP Forecast
1-2016 IRP Forecast
+ Historical peaks (weather adj
---- Linear (2009-2016)
16,000
2008 2010 2012 2014 2016 2018 2020 2022 2024
25 Duke Energy Carolinas, Integrated Resource Plan (Annual Report), September 1, 2014, filed in North
Carolina Utilities Commission Docket No. E-100 Sub 141.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 18 of 23

41. In 2014, Duke Energy Carolinas was anticipating summer peak loads over
20,000 MW in 2020. In its more recent IRP (2016), it is only expecting such loads in
2023; and recent trends suggest such loads would not be reached until after 2024. The
2016 forecast with energy efficiency for 2020 is 949 MW lower than the 2014 forecast.
42. The Duke Energy Carolinas 2016 IRP recommended an increase in its
reserve margin from 14.5% to 15%, based upon recommendations in a resource
adequacy study prepared by Astrape Consulting26 ("Duke Energy Carolinas RA Study";
"RA Study"). This recommendation largely resulted from concerns about winter
resource adequacy. In recent years, extremely cold winter days have resulted in very
high electric demand in the morning hours on the Duke Energy Carolinas and Duke
Energy Progress systems. These load spikes have been very brief; loads in other hours
on the same day, and on other days, were much lower. However, the Wilson Energy
Economics Duke Reserve Margin Report concluded that the RA Study did not support
the recommended increase in the reserve margin, due to flaws in the study that
improperly inflate the planning reserve margin:
a. The RA Study relied upon regressions to represent the impact of extreme
cold on load levels that greatly overstate the impact; more accurate
regressions more focused on colder temperatures suggest a much more
moderate impact of extreme cold on load.
b. The RA Study included multiple years of economic load forecast
uncertainty, but failed to take into account how capacity commitments
can be expanded in response to unexpectedly strong load growth.
c. The RA Study also overstated the probability of unexpectedly strong load
growth; the probability distribution used was not supported by the data
it was based upon.
26 Astrape Consulting, Duke Energy Carolinas (DEC) 2016 Resource Adequacy Study, August 30, 2016.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 19 of 23

d. Finally, the RA Study assumed that the load forecast values were mean
(or expected) values; but the forecast values were actually substantially
greater than mean values, due to wholesale commitments that were
added to the forecast and that are very infrequently called upon.
43. ACP reported that Duke Energy Carolinas has no plans to add new gas-
fired generation before 2022.Z' The Duke Energy Carolinas 2016 IRP identifies a need
for incremental generating capacity only in 2022/2023 (p. 38 Chart 8-A). Consequently,
it appears Duke Energy Carolinas does not need ACP capacity at this time, and will not
need additional pipeline capacity for at least several more years. If Duke Energy
Carolinas were to re-evaluate its commitment to ACP, it would likely find that the
commitment is not needed at this time, it is unclear when such capacity might be
needed, and it is also unknown whether better options might be available at such time
as incremental pipeline capacity does become needed.
VI. DUKE ENERGY PROGRESS' LOAD FORECASTS AND CAPACITY NEEDS
44. Like DVP and Duke Energy Carolinas, Duke Energy Progress has lowered
its summer peak load forecast since 2014, when it entered into the precedent
agreement with AC p.28 This is shown in Figure 6, along with the weather adjusted
historical peaks (which were available only for a few past years) and trend. The 2016
forecast is 380 MW lower than the 2014 forecast, and the trend is lower still.
45. While historically Duke Energy Progress has been a summer peaking
system, based on the recent "polar vortex" experience Duke Energy Progress now
predicts winter peaks slightly higher than summer peaks. However, this is based on the
potential for extremely cold days with very high loads; such infrequent loads do not
27 Atlantic Coast Pipeline, LLC and Dominion Transmission, Inc., Docket Nos. CP -15-554 & CP15-555,
Response to Data Request dated November 23, 1016, Question No. 3.
28 Duke Energy Progress, Integrated Resource Plan (Annual Report), September 1, 2014, filed in North
Carolina Utilities Commission Docket No. E-100 Sub 141.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 20 of 23

drive new plant construction or energy needs. Energy needs, and associated fuel needs,
remain greater in summer than winter.
Figure 6: Duke Energy Progress Summer Peaks, Historical and Forecast
15,500
--0-2014 IRP Forecast
15,000 12016 IRP Forecast
—Historical peaks (weather adjusted)
14,500 _ _ - Linear (Historical peaks (weather adjusted))
14,000
13,500
13,000 -
12,500
12,000
11,500
2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
46. As did Duke Energy Carolinas, Duke Energy Progress recommended an
increase in its reserve margin from 14.5% to 15% in its 2016 IRP, based upon a resource
adequacy study prepared by Astrape Consulting ("Duke Energy Carolinas RA Study").29
The Wilson Energy Economics Duke Reserve Margin Report concluded that the study did
not support the recommended increase in the reserve margin, due to same flaws
discussed above with regard to the similar study for Duke Energy Carolinas.
29 Astrape Consulting, Duke Energy Carolinas (DEC) 2016 Resource Adequacy Study, August 30, 2016.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 21 of 23

47. ACP reports that Duke Energy Progress, like Duke Energy Carolinas, has
no plans to add new gas-fired generation before 2022.30 The Duke Energy Progress
2016 IRP identified a need for incremental generating capacity only in 2021/22 (p. 39
Chart 8-A).
48. If Duke Energy Progress were to re-evaluate its commitment to ACP, like
DVP and Duke Energy Carolinas, it would likely find that the commitment is not needed
at this time, it is unclear when such capacity might be needed, and it is also unknown
whether better options might be available at such time as incremental pipeline capacity
does become needed.
30 Atlantic Coast Pipeline, LLC and Dominion Transmission, Inc., Docket No CP -15-554 &CP15-555,
Response to Data Request dated November 23, 1016, Questions No. 3.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 22 of 23

APPENDIX: QUALIFICATIONS OF JAMES F. WILSON
James F. Wilson is an economist and independent consultant doing business as
Wilson Energy Economics, with a business address of 4800 Hampden Lane Suite 200,
Bethesda, Maryland 20814. Mr. Wilson has over 30 years of consulting experience,
primarily in the electric power and natural gas industries. Many of his consulting
assignments have pertained to the economic and policy issues arising from the interplay
of competition and regulation in these industries, including restructuring policies,
market design, market analysis and market power. Other recent engagements have
involved resource adequacy and capacity markets, contract litigation and damages,
forecasting and market evaluation, pipeline rate cases and evaluating allegations of
market manipulation. Mr. Wilson's experience and qualifications are further detailed in
his CV, available at www.wilsonenec.com.
Wilson Report on Evidence of Need for the Atlantic Coast Pipeline Page 23 of 23

Southern 201 West Main Street, Suite 14
Charlottesville, VA 22902-5065
Environmental 434-977-4090
Fax [phone]
_Y Law Center SouthernEnvironment.org
August 17, 2016
VIA ELECTRONIC FILING
Mr. Joel H. Peck, Clerk
c/o Document Control Center
State Corporation Commission
Tyler Building — First Floor
1300 East Main Street
Richmond, Virginia 23219
RE: Application of Virginia Electric and Power Company in re: Virginia
Electric and Power Company's Integrated Resource Plan filing
pursuant to Va. Code § 56-597 et seq
Case No. PUE-2016-00049
Dear Mr. Peck:
Attached for filing in the above -referenced matter is the Direct Testimony of James F.
Wilson, which is being submitted on behalf of Appalachian Voices, the Chesapeake Climate
Action Network, and the Natural Resources Defense Council (collectively, "Environmental
Respondents"). Included with this testimony is Mr. Wilson's one-page summary and five
exhibits. This filing is being completed electronically, pursuant to the Commission's Electronic
Document Filing system.
If you should have any questions regarding this filing, please contact me at (434) 977 -
.M
Regards,
William C. Cleveland
cc: Parties on Service List
Commission Staff
Charlottesville • Chapel Hill • Atlanta • Asheville • Birmingham • Charleston • Nashville • Richmond • Washington, DC
100% recycles' paper

COMMONWEALTH OF VIRGINIA
STATE CORPORATION COMMISSION
APPLICATION OF VIRGINIA ELECTRIC
AND POWER COMPANY
In Reference Virginia Electric and Power )
Company's Integrated Resource Plan filing )
pursuant to Va. Code § 56-597 et seq. )
Case No. PUE-2016-00049
Summary of Direct Testimony of
James F. Wilson
On Behalf of
Environmental Respondents
August 17, 2015

1 Summary of the Testimony of James F. Wilson
2 My testimony evaluates the Company's peak load forecast and the calculation of the
3 Total Resource Requirements used in the 2016 Plan.
4 I first review past trends in peak loads in the Dominion zone, and show that over a decade
5 the trend has been down for all loads other than data centers. With regard to the Company's
6 peak load forecast, I conclude the following: (1) the Company's peak load forecast is very likely
7 to be too high, primarily because it fails to capture trends toward slower peak load growth and
8 the increasing efficiency of electricity use during peak periods; (2) PJM's peak load forecast,
9 based on a methodology that has several recent enhancements, is likely to be more accurate than
10 the Company's, although it is still very conservative and likely to over -forecast future peak
I I loads; (3) the Company's calculations understate the peak loads of other Load Serving Entities in
12 the DOM zone, which are growing faster than the Company's, resulting in overstating the peaks
13 for the Dominion Load Serving Entity.
14 With regard to the Company's calculation of Total Resource Requirement, I find that
15 while the Company attempts to follow PJM's approach, its approach is different, and the values
16 used for some parameters are not correct.
17 Finally, I offer several recommendations with regard to load forecasting and the
18 calculation of Total Resource Requirements for future Integrated Resource Plans.

COMMONWEALTH OF VIRGINIA
STATE CORPORATION COMMISSION
APPLICATION OF VIRGINIA ELECTRIC
AND POWER COMPANY
In Reference Virginia Electric and Power )
Company's Integrated Resource Plan filing )
pursuant to Va. Code § 56-597 et seq. )
Case No. PUE-2016-00049
Direct Testimony of
James F. Wilson
On Behalf of
Environmental Respondents
August 17, 2015

CONTENTS
Page
I. Introduction and Qualifications.............................................................................................1
II. Summary and Recommendations..........................................................................................2
111. Dom Zone Peak Loads: Recent Trends and Current Forecasts.............................................5
IV. Peak Load Forecasting: The Company's, and PJM's, Methodologies................................1 1
V. Load Forecast Adjustments to Reflect Anticipated Data Center Loads..............................17
VI. Reserve Margin and Total Resource Requirement Calculations.........................................22
VII. Conclusions and Recommendations....................................................................................29
Wilson Direct Testimony Page 1

1 L INTRODUCTION AND QUALIFICATIONS
2 Q 1: Please state your name, position and business address.
3 A: My name is James F. Wilson. I am an economist and independent consultant doing
4 business as Wilson Energy Economics. My business address is 4800 Hampden Lane
5 Suite 200, Bethesda, Maryland 20814
6 Q 2: On whose behalf are you testifying in this proceeding?
7 A: I am testifying on behalf of the Environmental Respondents: Natural Resources Defense
8 Council, Appalachian Voices, and the Chesapeake Climate Action Network.
9 Q 3: Please describe your experience and qualifications.
10 A: I have over thirty years of consulting experience, primarily in the electric power and
11 natural gas industries. Many of my assignments have pertained to the economic and
12 policy issues arising from the interplay of competition and regulation in these industries,
13 including restructuring policies, market design, market analysis and market power. Other
14 recent engagements have involved resource adequacy and capacity markets, contract
15 litigation and damages, forecasting and market evaluation, pipeline rate cases and
16 evaluating allegations of market manipulation. I also spent five years in Russia in the
17 early 1990s advising on the reform, restructuring, and development of the Russian
18 electricity and natural gas industries for the World Bank and other clients.
19 With respect to the load forecasting and capacity requirements issues I will address in this
20 testimony, I have been actively involved in these issues in the PJM Interconnection,
21 L.L.C. ("PJM") region for many years, participating in PJM stakeholder processes,
22 performing and presenting analysis of these issues, and submitting affidavits in various
23 regulatory proceedings.
Wilson Direct Testimony Page 1

I I have submitted affidavits and presented testimony in proceedings of the FERC, state
2 regulatory agencies, and U.S. district court. I hold a B.A. in Mathematics from Oberlin
3 College and an M.S. in Engineering -Economic Systems from Stanford University. My
4 curriculum vitae, summarizing my experience and listing past testimony, is attached as
5 Attachment JFW-1.
6 Q 4: Have you previously submitted testimony in a Virginia State Corporation
7 Commission ("Commission") proceeding?
8 A: Yes. I submitted direct testimony on behalf of Commission staff in Case No. PUE-2009-
9 00043 (Application of PATH Allegheny Virginia Transmission Corporation for
10 Certificates of Public Convenience and Necessity).
11 Q 5: What is the scope and purpose of your testimony in this case?
12 A: This proceeding involves the 2016 Integrated Resource Plan ("2016 Plan") for Dominion
13 Virginia Power ("Dominion" or the "Company"). My assignment was to evaluate the
14 forecasts of peak loads and Total Resource Requirements included in the 2016 Plan and
15 provide any recommendations.
16
17 Il. SUMMARY AND RECOMMENDATIONS
18 Q 6: How are the Company's forecasts of peak loads and Total Resource Requirements
19 used in the 2016 Plan?
20 A: The Total Resource Requirements ("TRK') are the Company's estimates of the amount
21 of capacity that will be assigned to the Company by PJM for purposes of allocating
22 capacity costs. The TRRs represent the Company's estimates of its customers' future
23 generating capacity needs, and the 2016 Plan describes how the Company plans to meet
Wilson Direct Testimony Page 2

I these needs through owned and contracted resources. TRRs are calculated as the forecast
2 peak load for the Dominion Load -Serving Entity ("DOM LSE") plus a reserve margin.
3 Q 7: Please summarize your conclusions with respect to the peak load forecast used by
4 the Company for the 2016 Plan.
5 A: Peak loads in the DOM Zone, and especially for the DOM LSE, have been quite flat over
6 the past decade, and the primary source of growth has been data centers.
7 I conclude that the Company's peak load forecasts are very likely to be too high,
8 primarily because they fail to capture trends toward slower peak load growth and the
9 increasing efficiency of electricity use during peak periods. Both the Company and PJM
10 adjust their forecasts upward to reflect anticipated strong growth in data center loads.
11 I conclude that PJM's peak load forecast, while still very conservative and likely to over -
12 forecast future peak loads, is likely to be more accurate than the Company's due to two
13 key differences in the methodologies. In addition, the Company's calculations overstate
14 the DOM LSE peak loads due to understating the peak loads of other LSEs in the DOM
15 Zone, which are growing faster than the Company's.
16 Q 8: Please summarize your conclusions with respect to the Company's reserve margin
17 and TRR calculations.
18 A: The Company attempts to follow PJM's approach for its reserve margin and total
19 resource requirement calculations, but its approach is different, and some of the values
20 used were not accurate. I have estimated the TRRs based on PJM's forecast, a revised
21 estimate of Other LSE loads, and corrected reserve margin parameters.
Wilson Direct Testimony Page 3

I Q 9: Please present your revised peak load forecast and TRR values.
2 A: Table 1 presents the results.
Table 1: Load Forecast and Total Resource Requirements (MW)
2020 12021 12022 2023 2024 2025
DOM Zone Peak Load Forecast
2016 Plan
21,847
22,263
22,546
22,792 23,260 23,566
Based on PJM
20,700
20,871
21,059
21,235 21,452 21,664
Difference
-1,147
-1,392
-1,487
-1,557 -1,808 -1,902
Other LSEs' Peaks
2016 Plan
2,722
2,773 2,808
25840 2,898 2,936
Revised
2,883
2,956 3,005
3,048 3,129 3,182
Difference
161
183 197
208 231 246
DOM LSE Adjusted Peak Load Forecast
2016 Plan 18,891 19,257 19,509 19,724 20,132 20,399
Revised 17,583 17,683 17,825 17,958 18,092 18,250
Difference -1,308 -1,574 -1,684 -1,766 -2,040 -2,149
DOM LSE Total Resource Requirement
2016 Plan
21,245
21,657
21,940
22,181
22,640
22,941
Revised
19,761
19,894
20,025
20,215
20,355
20,505
Difference
-1,484
-1,763
-1,915
-1,966
-2,285
-2,436
3 Q 10: Do you have recommendations with regard to peak load forecasting and TRR
4 calculations for the purposes of Integrated Resource Plans?
5 A: Yes I do. I have included several such recommendations in the last section of my
6 testimony.
7 Q 11: How is the remainder of your testimony organized?
s A: The next section reviews recent trends in peak loads in the DOM Zone, and presents the
9 Company's and PJM's forecasts. Section IV discusses load forecasting methodologies
10 and explains why PJM's approach is likely to be more accurate than the Company's.
11 Section V discusses the adjustments both the Company and PJM made to their forecasts
12 to represent anticipated strong growth in data center loads. Section VI reviews the
13 Company's calculation of TRRs and identifies some issues in that regard. The final
Wilson Direct Testimony Page 4

I section presents recommendations with respect to load forecasts and TRR calculations for
2 future Integrated Resource Plans.
3
4 III. DOM ZONE PEAK LOADS: RECENT TRENDS AND CURRENT FORECASTS
5 Q 12: Please present the recent peak loads in the DOM transmission zone.
6 A: Figure JFW-Al presents the actual DOM Zone peak loads since 2003. These are the
7 "unrestricted" peak loads, where any demand response or demand-side management by
8 PJM or the Company that may have occurred during the peak hour has been added back.
9 Q 13: Do these peak load values exhibit any clear trend?
10 A: No. These actual peak loads reflect the actual weather that occurred each summer; the
11 peaks will tend to be high in years in which a very extreme period of hot weather
12 occurred (leading to high electricity use for air conditioning), and they will tend to be low
13 in years in which the most extreme weather that occurred during the summer was
14 relatively mild. Because the actual peak loads reflect changeable weather, their pattern
15 over relatively short periods of time may not reflect any trend, or may even be
16 misleading, suggesting a trend that does not in fact exist.
17 Q 14: Is there a way to reveal the trends in past peak loads?
is A: Yes. To discern trends in peak loads, it is very helpful to calculate "weather -normalized"
19 historical peaks. Weather -normalized historical peak loads are estimates of what the
20 peak loads would have been in past years had the weather, at the time of the summer peak
21 load, been the typical (very hot) weather that tends to occur at the time of the annual peak
'All Figures included in Attachment JFW-1.
Wilson Direct Testimony Page 5

I load. This removes the year-to-year variability due to weather in the historical peak
2 loads. With the weather variability removed, the more stable trends in peak loads due to
3 forces such as economic and demographic growth and changing electrical equipment
4 stocks are revealed.
5 Q 15: How do weather -normalized historical peak loads compare to forecast peak loads?
6 A: Weather -normalized historical peaks, and forecast peaks, are essentially the same thing.
7 A forecast peak load is generally intended to be a median (or "50-50") peak; that is, the
8 peak load level that has an equal chance of being exceeded, or not being exceeded, in the
9 future year. The weather -normalized historical peak is exactly the same concept — it is
10 the peak load level in the historical year that had a 50-50 chance of being exceeded.
11 Put another way, the weather -normalized historical peak load is exactly the peak load that
12 past and current peak load forecasting efforts attempt to determine. And, accordingly, we
13 would expect that a peak load forecast would generally be consistent with the trend
14 reflected in past weather -normalized peaks. This is why I consider historical weather -
15 normalized peak loads extremely useful in understanding likely trends in future peak
16 loads.
17 Q 16: Does the Company prepare weather -normalized historical peaks?
18 A: No .2 However, PJM does prepare weather -normalized historical peaks.
2 Response to Data Request ER -3-48, included here in Attachment JFW-3.
Wilson Direct Testimony Page 6

1 Q 17: How are weather -normalized peak loads calculated?
2 A: Various approaches can be used to estimate historical weather -normalized peak loads.
3 PJM has refined its methodology from time to time, and in 2015 evaluated multiple
4 approaches3 and then updated its methodology.
5 Q 18: Please present and discuss the recent trends in weather -normalized peak loads for
6 the DOM Zone.
7 A: Figure JFW-B presents PJM's weather -normalized peaks for the DOM Zone. The
8 weather -normalized peak loads have been quite flat over the past decade; the 2007 and
9 2015 values are nearly identical. Even in the post -recession period (from about 2010 to
10 present), peak loads have been flat.
11 Q 19: The 2016 Plan notes recent growth in data center loads (p. 24). What is the trend in
12 the DOM Zone peak load for all loads other than data centers?
13 A: Figure JFW-B also shows the weather -normalized peak loads net of data center peak
14 loads. The trend in the Company's peak load for all customers other than the data centers
15 is actually down over the past decade: the weather -normalized peak for 2015 was lower
16 than for 2006. Focusing only on the post -recession period, the trend has also been down.
17 Q 20: Now please present the Company's peak load forecast for the DOM Zone.
18 Figure JFW-C presents the Company's forecast that was relied upon for the 2016 Plan
19 (App. 2G). The Company's forecast suggests robust growth in peak loads, starting right
20 in 2016. Figure JFW-C also shows the Company's forecast net of the forecast of data
3 PJM, Weather Normalization of Peak Load, Load Analysis Subcommittee meeting September 2, 2015, Item 3,
available at http://www.pjm.com/—/media/committees-groups/subcommittees/las/20150902/20150902-item-03-
weather-normalization. ashx.
Wilson Direct Testimony Page 7

I center peak loads. For all loads other than data centers, the forecast also suggests rapid
2 growth in peak loads.
3 Q 21: Is the Company's peak load forecast consistent with recent trends?
4 A: No. While peak loads have been flat, or even down when data centers are separated, the
5 Company's forecast suggests strong growth. The Company's forecasts deviate sharply
6 from the recent trends, and this raises questions and concerns about the potential accuracy
7 of this forecast.
8 Q 22: What are the primary drivers of peak load growth under the Company's forecasting
9 methodology?
10 A: The Company's econometric approach relies upon various economic and demographic
11 forecasts as independent variables that drive future peak load growth. These are
12 summarized in Figure 2.2.6 in the 2016 Plan, and include trends in the number of
13 customers and households, per capita income, and employment. However, as the
14 Company acknowledges (2016 Plan p. 24), a "key driver" is the forecast of the Virginia
15 economy.
16 Q 23: Is the Company's forecast of robust growth in peak load explained by the economic
17 and demographic forecasts?
18 No; the trends in these independent variables have been rather steady recently, and they
19 are expected to continue to show steady increases over the forecast period, as suggested
20 by Figure 2.2.6. These forecasts do not explain the sharp deviation from trend reflected
21 in the Company's peak load forecast. Figure JFW-D shows DOM Zone economic and
Wilson Direct Testimony Page 8

1 demographic variables in the form of a composite index used by PJM in forecasting the
2 DOM Zone .4 This index combines six economic -demographic variables (households,
3 population, personal income, non -manufacturing employment, U.S. gross domestic
4 product, and state or metropolitan product). These are the same or similar economic -
5 demographic variables used by the Company in its forecasting, and sourced from the
6 same vendor (Moody's economy.com).
7 The figure shows that while DOM Zone peak loads were flat or declining over the past
8 decade, the economic -demographic variable continued to climb. The figure further
9 shows that while the economic -demographic variable is expected to continue to rise in
10 future years, it would not appear to explain the sharp increase in the Company's peak
1 I load forecast.
12 Q 24: How can peak loads have remained flat or declined while the economic and
13 demographic drivers were increasing?
14 A: Peak loads can be flat or declining while economic and demographic measures rise due to
15 increasing efficiency in the use of electricity, as I will discuss later in this testimony.
16 Q 25: If the economic and demographic forecasts do not point to robust growth in peak
17 loads, why does the Company's peak load forecast rise so sharply?
18 A: One reason the Company's forecast suggests robust peak load growth is that the
19 Company's forecasting methodology, rather than capturing and reflecting the recent trend
a PJM, 2016 Economic Variable Data, available at hLtp://www.pjm.com/—/medig/
ashx.
5 PJM, PJM Manual 19: Load Forecasting and Analysis, Revision: 31 Effective Date: 06/01/2016, p. 18, available
at http://www.pjm.com/—/media/documents/manuals/ml9.ashx.
Wilson Direct Testimony Page 9

1 in peak load growth, instead reaches back and includes thirty years of historical data to
2 estimate the parameters for its econometric model.
3 Many years ago, the DOM Zone, and other regions of the country, did indeed experience
4 much faster peak load growth. However, more recently, there has been a trend of
5 slowing peak load growth, both in absolute terms, and in relation to economic and
6 demographic growth; and this trend has continued or even strengthened in recent years.
7 Including the long -ago history in the Company's forecasting leads the model to discount
8 the trend over the past decade, and place weight on the higher rates of peak load growth
9 seen ten to thirty years ago.
10 Q 26: Now please present PJM's forecast of peak loads for the DOM Zone.
11 A: Figure JFW-E adds PJM's DOM Zone forecast from its 2016 Load Forecast Report?
12 (published in January of 2016). The Figure also shows PJM's latest forecast, based on its
13 mid -year update in July 20168 (the mid -year update provides an updated coincident peak
14 forecast, and only for 2016 through 2019; the series shown here is estimated based on the
15 ratio of non -coincident to coincident peaks from the 2016 Load Forecast Report).
16 PJM's forecasts are considerably lower than the Company's forecast. PJM's most recent
17 forecast is over 1,100 MW lower than the Company's for 2020, growing to more than a
18 1,500 MW difference by 2023. However, PJM's forecasts still appear to be very
6 2016 Plan p. 24.
7 PJM, PJM Load Forecast Report January 2016, available at
hllp://www.p j m. com/—/media/documents/reports/2016-load-report. ashx
8 PJM, Load Forecast Update — July 2016, available at http://www.pjm.com/—/media/planning/res-adeq/load-
forecast/pjm-load-forecast-update July-2016.ashx
Wilson Direct Testimony Page 10

I conservative and more likely to over- than under -state future peak loads, because they
2 anticipate peak load growth much faster than recent trends suggest.
3 Q 27: Why are PJM's forecasts so much lower than the Company's?
4 A: This reflects differences in the forecasting methodologies, as I will discuss in the next
5 section of this testimony.
6
7 IV. PEAK LOAD FORECASTING: THE COMPANY'S, AND PJM'S, METHODOLOGIES
s Q 28: What topic will you address in this section of your testimony?
9 A: I will explain why traditional econometric approaches to forecasting future peak loads,
10 which both the Company and PJM have used for many years, have resulted in chronic
11 over -forecasting in recent years. I will further explain that PJM's new methodology,
12 following various enhancements designed and implemented in 2015, is still likely to be
13 conservative and to over -forecast peak loads, but is likely to be more accurate than the
14 Company's methodology, which is unchanged over many years and similar to PJM's
15 approach before the recent enhancements.
16 Q 29: Please describe the Company's approach to forecasting peak loads.
17 A: The Company uses an econometric regression model that takes some inputs from a
18 separate model of sales by customer class.9 The regression model forecasts peak loads
19 based on various economic and demographic independent variables (shown in Figure
9 Virginia Electric and Power Company, Peak Demand and Energy Sales Forecast Model Documentation,
Attachment ER Set 2-14(a).
Wilson Direct Testimony Page 11

1 2.2.6 and Appendix 2K; forecasts from September 2015), as noted above. The
2 methodology is described in the 2016 Plan at pp. 19-20.
3 Q 30: Please describe PJM's approach to forecasting peak loads for the DOM Zone and
4 compare it to the Company's approach.
5 A: PJM also uses an econometric approach based on similar economic and demographic
6 forecasts. While there are numerous differences between the Company's and PJM's
7 econometric models (of which some are described in the 2016 Plan at p. 24), two are
8 likely the most important factors leading to the different results:
9 1. PJM uses an 18 -year historical period for estimating the model, while the Company
10 uses 30 years. As a result, PJM's forecast will reflect recent trends to a somewhat
11 greater extent.
12 2. PJM's methodology has recently been enhanced to better capture trends in energy
13 efficiency (discussed further below).
14 Q 31: Please summarize the recent performance of the Company's peak load forecasts.
15 A: In recent years, both the Company's and PJM's forecasts have been too high. For
16 example, in the Company's 2010 Plan, the utility Base Forecast summer peak load for
17 2015 was 19,247 MW (Appendix 2H, line 1.A.1 a). In the 2013 Plan, the same value was
18 17,695 MW, and in the 2015 Plan, the same value was lowered again, to 17,475 MW —
19 despite the recent increases in data center loads. PJM's forecasts of the Dominion zone
20 (and of all or most other PJM zones) have also been too high over recent years.
Wilson Direct Testimony Page 12

I Q 32: Why have these forecasts been too high?
2 A: A key reason for over -forecasting has been inaccuracy in the underlying economic
3 forecasts. Econometric forecasting approaches rely on forecasts of economic conditions
4 (for the DOM Zone, primarily forecasts of growth in the Virginia economy) as the
5 primary driver of growth in future peak loads. These forecasts have proven to be overly
6 optimistic, as growth in the Virginia economy, and in the U.S. and world economies more
7 broadly, has been slower than expected, both during the recession that began in around
8 2008, and also in the post -recession period.
9 For example, the 2016 Plan anticipates the Virginia economy will grow at a compound
10 annual rate of 2.09% over the coming fifteen years (Figure 2.2.6, p. 25). However, as
11 recently as 2013 for the 2013 Plan, the anticipated annual growth rate was 2.4% (2013
12 Plan Figure 2.2.4, p. 23).
13 Q 33: Please elaborate on how economic forecasts are the primary driver of peak load
14 growth in econometric forecasting models.
15 A: While these models include other independent variables, the primary driver is generally
16 economic growth. Econometric approaches assess how peak loads have risen with
17 economic growth in the past, and then assume a similar relationship will hold in the
18 future.
19 For example, if the economy in a zone grew by, say, 40% over the past thirty years, while
20 peak loads grew by 30%, this suggests peak loads grow at about 75% of the rate of
21 economic growth (that is, an elasticity of peak load growth to economic growth of 0.75,
22 30%/40%). So if the economic forecast suggested economic growth would be 20% over
23 the coming ten years, the econometric approach would anticipate peak load growth of
Wilson Direct Testimony Page 13

1 roughly 15% over that period (20% x 0.75). While econometric approaches are more
2 complex than this, this is the fundamental structure.
3 Note that this simple example also illustrates the importance of the chosen historical
4 period. While the example suggests an elasticity of 0.75 over the past thirty years,
5 perhaps the elasticity has been 0.5 over the past ten years (say, 10% economic growth
6 and 5% peak load growth). Then the forecasting approach would suggest 10% peak load
7 growth (20% x 0.5) rather than 15% over the coming ten years.
8 Q 34: Please summarize the recent economic forecasts used by the Company and PJM for
9 forecasting the DOM Zone.
10 A: Both the Company and PJM have used economic forecasts provided by Moody's. Both
11 during the recession that began in around 2008, and for a few years following the
12 recession, Moody's expected a very robust recovery in the U.S. economy. However, this
13 did not occur; instead, post -recession growth has been modest. More recently, the
14 Moody's forecasts have anticipated continued, modest economic growth going forward.
15 Q 35: If economic growth, and forecasts of future economic growth, have stabilized in
16 recent years, does this mean the current econometric forecasts should be more
17 accurate?
18 A: This should reduce the impact of the main cause of recent over -forecasting. However,
19 the inaccurate economic forecasts have not been the only cause of over -forecasting. As
20 shown in Figure JFW-D, peak loads have remained flat or fallen while the economic -
21 demographic measures have continued to rise. PJM has recently identified that this is
22 due to the increasing efficiency of electricity use, which has not been captured by the
23 econometric forecasting approaches.
Wilson Direct Testimony Page 14

I Q 36: Please elaborate on how PJM staff came to the conclusion that the increasing
2 efficiency of electricity use was not being captured by PJM's peak load forecasting
3 methodology.
4 A: In the first few years following the recession, PJM staff believed the over -forecasting was
5 due to the inaccurate economic forecasts, and that, removing this source of error, their
6 forecasting approach was accurate. However, in around 2014, PJM staff determined that
7 the economic forecasts no longer explained the forecast error, and began an internal effort
8 to determine causes and design solutions. In March 2015, PJM staff initiated a process to
9 discuss the problem and its proposed solutions with PJM stakeholders through the PJM
10 Load Analysis Subcommittee ("LAS").
11 Q 37: Was PJM able to identify enhancements to its forecasting approach that will
12 improve its accuracy?
13 A: Yes. PJM evaluated a number of potential enhancements, and identified a few
14 enhancements that can be expected to improve accuracy. PJM evaluated the accuracy of
15 its forecasting with the recommended enhancements using various historical periods
16 across the many zones that it forecasts, and showed that the proposed enhancements
17 improve accuracy. See, for instance, slides 46 to 54 of PJM's presentation to LAS on
18 September 2, 2015.10
10 PJM, Updates To Load Forecast Methodology, Load Analysis Subcommittee September 2, 2015, available at
http://www.pj m. com/—/media/committees-groups/subcommittees/las/20150902/20150902-item-04-forecast-
update.ashx
Wilson Direct Testimony Page 15

I Q 38: What was the result of this work by PJM staff and the LAS?
2 A: PJM's recommended enhancements to its load forecasting methodology were endorsed
3 by the PJM Markets and Reliability Committee at its November 19, 2015 meeting with
4 no objections. The enhancements were reflected in the forecast documented in the 2016
5 PJM Load Forecast Report (January 2016).
6 Q 39: Please describe the main enhancements PJM made to its forecasting methodology in
7 2015.
8 A: The two most important changes were as follows:
9 1. New independent variables to capture past regional trends and forward-looking
10 forecasts of equipment and appliance efficiency and penetration. These variables are
11 prepared by Itron, Inc. based on U.S. Energy Information Administration ("EIA")
12 data.
13 2. Improvements to the use of weather splines, to more accurately represent the
14 relationship between weather and loads during periods of extreme weather and high
15 loads.
16 Q 40: With these enhancements, do you expect PJM's forecasts will no longer consistently
17 over -forecast future peak loads?
18 A: This is possible. However, while I believe these enhancements will improve the accuracy
19 of PJM's forecasts and reduce the over -forecasting, PJM's methodology likely will
20 continue to over -forecast peak loads for the PJM RTO and most of its zones, including
21 the DOM Zone. PJM continues to use historical data back to 1998 to estimate its model,
22 which, as noted above, will result in the model being slow to reflect more recent trends of
23 slowing peak load growth, which may have additional causes.
Wilson Direct Testimony Page 16

I Q 41: Please summarize your testimony with regard to the Company's and PJM's peak
2 load forecasting methodologies.
3 A: The Company and PJM employ similar methodologies. However, PJM has enhanced its
4 approach to better capture trends in equipment penetration and efficiency, which PJM has
5 shown will improve the accuracy of its forecasts. PJM's methodology also uses a shorter
6 historical period, which will better reflect recent trends. I conclude that PJM's peak load
7 forecasts are likely to be more accurate than the Company's.
K
9 V. LOAD FORECAST ADJUSTMENTS TO REFLECT ANTICIPATED DATA CENTER LOADS
to Q 42: You mentioned that both the Company and PJM adjusted their forecasts to take
11 into account anticipated strong growth in data center loads. Please explain the
12 rationale for such adjustments.
13 A: In recent years a number of data centers have been constructed in Northern Virginia and
14 have been a major source of load growth in the DOM Zone. Additional construction and
15 expansion of data centers is expected in this area over the coming years. The Company
16 and PJM share a concern that their econometric forecasting approaches will fail to
17 accurately forecast the growth in such loads, because the trend is fairly recent. PJM has
18 been adjusting its forecast for the DOM Zone since its 2014 Load Forecast Report,
19 prepared in 2013 and published in January of 2014.
Wilson Direct Testimony Page 17

I Q 43: How did the Company and PJM determine the adjustments to apply for data center
2 load growth?
3 A: The Company and PJM both relied upon the same Quanta Technology report prepared in
4 2015 ("QuantaReport")." Previous adjustments were based on a similar analysis by
5 Quanta Technology in 2013.12 The 2015 Quanta Report provides forecasts of future data
6 center peak loads along with estimates of the amount of the load growth that is captured
7 by the type of econometric forecasting methods used by both the Company and PJM.
8 The Quanta Report then recommends adjusting the DOM Zone forecasts by the amount
9 of the anticipated increase in data center load that is not captured by the econometric
10 forecasting approaches. Both the Company and PJM applied such an adjustment based
1 I on the Quanta Report.
12 Q 44: How certain is this forecast growth in data center loads?
13 A: The growth is highly uncertain; it could be considerably different from the forecast in
14 either direction. The Quanta Report notes (p. 13) that data center owners are "deliberately
15 optimistic in giving the utility completion dates and future loads", because they want no
16 utility -side constraints on when they can get the power they need.
17 While it may be very likely that there will be strong growth in electric demand for data
18 centers in North America, it is highly uncertain exactly when and where that growth will
19 occur. For example, a recent report by Jones Lang LaSalle Inc. ("JLL") noted anticipated
11 Quanta Technology, Dominion Northern Virginia Load Forecast Dominion Virginia Power, Oct. 23, 2015,
Attachment ER Set 2-28(a).
12 Quanta Technology, Dominion Northern Virginia Load Forecast Dominion Virginia Power, Oct. 17, 2013,
Attachment ER Set 3-46(a), included here in Attachment JFW-4.
Wilson Direct Testimony Page 18

growth in data centers, but discussed sixteen locations across North America, in addition
2 to Northern Virginia, where data centers are being constructed. 13
3 And a recent report by Lawrence Berkeley National Laboratory suggests that increasing
4 energy efficiency at data centers will result in little additional growth in their electricity
5 demands at the national level in the coming years, despite strong growth in the demand
6 for their services: 14
7 The combination of these efficiency trends has resulted in a relatively steady U.S data
8 center electricity demand over the past 5 years, with little growth expected for the
9 remainder of this decade. It is important to note that this near constant electricity demand
10 across the decade is occurring while simultaneously meeting a drastic increase in demand
11 for data center services; data center electricity use would be significantly higher without
12 these energy efficiency improvements.
13
14 Q 45: How much of the data center load growth does the Company assume is already
15 captured by the Company's econometric forecasting approach?
16 A: This is found by comparing the total forecast data center peak load growth (Table 4-5 of
17 the Quanta Report) to the recommended adjustment (Table 4-6). This reveals that the
18 Company has used the Quanta Report's estimate that very little of the growth is captured
19 by the forecasting approach. Specifically, of 1,903 MW of data center peak load growth
13 Jones Lang LaSalle Inc., Data Center Outlook— Strong Demand, Smart Growth, 2016, available at
http://www.us. ill.com/united-states/en-us/research/7319/us-north-america-data-center-outlook-2016-j ll.
14 U.S. Department of Energy, Ernest Orlando Lawrence Berkeley National Laboratory, United States Data Center
Energy Usage Report, June 2016 (LBNL-1005775), p. ES -2, available at http:Heta.lbl.gov/sites/all/files/lbnl-
1005775ev2.pdf.
Wilson Direct Testimony Page 19

I expected from 2015 to 2025, the Quanta Report estimates that only 136 MW (7.1%) of
2 this growth is captured by the Company's econometric approach. The Quanta Report
3 therefore recommends that the remainder (1,767 MW) should be added to the Company's
4 forecast, and the Company has done so.
5 Q 46: You stated that PJM also adjusted its forecast for data center loads, based on the
6 same Quanta Technology report. Please describe how PJM adjusted its forecast.
7 A: PJM also adjusted its forecast; the values used are in Table B-9 of the 2016 Load
8 Forecast Report. PJM used a somewhat smaller adjustment than the Company (for
9 example, 730 MW for 2020, compared to 1,062 MW used by the Company), having
10 concluded that "Quanta's projections are designed to capture the entire data center
11 industry but are significantly higher than can be supported by Dominion's own detailed
12 projection. ,15 The work by Quanta Technology resulted in multiple scenarios of future
13 data center loads, and PJM selected a more modest scenario.
14 Q 47: Does the Quanta Report's forecast include data centers that are not served by DOM
15 LSE?
16 A: Yes; and many of the data centers are not served by DOM LSE. The Quanta Report
17 forecasts all data centers served through Dominion's transmission system (p. 13), that is,
18 in the DOM Zone:
19 In particular, a significant amount of new data center load has
20 developed and much more has been announced in the service
15 PJM, Load Forecast Adjustment Guidelines — Dominion, December 10, 2015, p. 2, available at
http://www.pj m. com/—/media/committees-groups/subcommittees/las/20151210/20151210-item-05-load-forecast-
adi ustment-dominion. ashx.
Wilson Direct Testimony Page 20

I territory of Northern Virginia Electric Company [sic] (NOVEC),
2 which provides retail electric service to some parts of northern
3 Virginia where Dominion Virginia Power does not. However,
4 Dominion Virginia Power provides power to the NOVEC's
5 distribution system through its transmission system, so NOVEC
6 loads directly and entirely impact loading on the Dominion
7 Virginia Power system and thus need to be included in any forecast
8 used for transmission planning.
9 In its report on the data center adjustment, PJM noted that of the nine "Category II" data
10 centers that account for 600 MW of the forecasted growth to 2020, four are located in the
11 Northern Virginia Electric Cooperative sub -area and, accordingly, are not served by
12 DOM LSE.
13 Q 48: Did the Quanta Report or the Company identify the data center growth that will be
14 served by DOM LSE?
15 A: No. The Company's estimate of "Other LSE" peak loads is discussed in the next section
16 of my testimony.
17 Q 49: What do you conclude with regard to these adjustments for data center loads?
18 A: An adjustment for data center load growth is reasonable in light of the substantial growth
19 in such loads in the DOM Zone in recent years, which is projected to continue and may
20 increase. The Company and PJM have applied somewhat different adjustments, and the
21 amount of data center load captured in their econometric analyses is also likely different;
22 PJM's methodology, with a shorter historical period, likely captures more of the
Wilson Direct Testimony Page 21

1 historical trend. Given the uncertainty about future data center loads, I believe PJM's
2 more modest adjustment for data center loads is a reasonable and conservative approach.
3 Q 50: What is your conclusion and recommendation with respect to the peak load forecast
4 for the 2016 Plan?
5 A: I conclude that the Company's forecast is very likely to be inaccurate and to overstate
6 future peak loads, and that PJM's forecast is very likely to be more accurate.
7
8 VI. RESERVE MARGIN AND TOTAL RESOURCE REQUIREMENT CALCULATIONS
9 Q 51: Please describe how the Company calculated its Total Resource Requirement
10 ("TRR").
11 A: The annual TRR values shown in Figure 4.2.2.1 were apparently calculated as follows
12 (references are to 2016 Plan appendices):
13 1. The starting point was the Company's forecast summer peak load for the DOM Zone
14 (Appendix 2G).
15 2. From the forecast DOM Zone summer peaks, the forecast peak loads of "other LSEs"
16 were deducted (2016 Plan p. 20) to determine the DOM LSE peak load "base
17 forecast" shown in Appendix 2I line 1 a.
18 3. The DOM LSE "base forecast" was adjusted for conservation and efficiency
19 (Appendix 2I, line 2) and for a "peak adjustment" (Appendix 2I, line 5) to determine
20 the DOM LSE adjusted peak load shown at Appendix 2I line 6.
21 4. The reserve margin was determined using an "effective reserve margin" combining
22 two components:
Wilson Direct Testimony Page 22

1 (1) a "coincidence factor", to estimate the DOM LSE PJM RTO -coincident peak load
2 based on the non -coincident peak load. As described on page 64, this is done
3 because PJM capacity cost is allocated based on coincident peak load. The
4 coincidence factor used was the average of the ratio of coincident to non -
5 coincident DOM Zone peaks as reflected in PJM's forecasts over 2016-2019
6 (0.9653).
7 (2) PJM's recommended installed reserve margin for 2019 (16.5%). The resulting
8 effective reserve margin was 0.1246 (0.9653 x 1.165 — 1).
9 5. The effective reserve margin (0.1246) was multiplied by the DOM LSE adjusted peak
10 load (step 3 above) to determine the reserve margin in MW shown in Appendix 2J
11 line 1 a.
12 6. The reserve margin was then added to the DOM LSE adjusted peak load to determine
13 the Total Resource Requirement.
14 Q 52: Does PJM follow these steps in determining capacity obligations?
15 A: No. PJM begins with its forecast of coincident peaks (Table BIO in its load forecast
16 reports), and determines capacity obligations on an "unforced" capacity ("UCAP") basis
17 by applying its Forecast Pool Requirement ("FPR") to the coincident peaks. 16 The
18 installed reserve margin (16.5%) is not used to determine capacity obligations, which are
19 expressed in UCAP terms.
16 PJM, Planning Period Parameters for the 2019-2020 Base Residual Auction, tab 2019-2020 Parameters (showing
that the Reliability Requirement is calculated based on the FPR, and the installed capacity reserve margin is used
only in the calculations of the shape of the VRR curve), available at http://www.pjm.com/—/media/markets-
ops/rpm/rpm-auction-info/2019-2020-bra-planning-parameters. ashx.
Wilson Direct Testimony Page 23

I Q 53: How were the forecast peak loads of the Other LSEs in the DOM Zone (your Step 2)
2 determined?
3 A: The Company estimated the Other LSE peak loads using a regression of past DOM LSE
4 peaks compared to DOM Zone peaks. The details were provided in response to a data
5 request. 17 The regression used peaks for DOM Zone and DOM LSE over 2006 through
6 2015. The Company then assumed DOM LSE over the coming years would represent the
7 same average fraction of DOM Zone peak load as it had over the historical period. The
8 implied Other LSE peak load is the difference between the DOM Zone peak and the
9 DOM LSE peak based on this method.
10 Q 54: Is this an accurate way to forecast the DOM LSE and Other LSE peaks?
] I A: No. The historical data shows that the peak loads of the Other LSEs in the DOM Zone
12 are generally rising faster than DOM LSE peak loads, and represent an increasing
13 fraction of the DOM Zone peak over time. This likely reflects, perhaps among other
14 factors, the strong growth in data center loads served by other LSEs in the DOM Zone.
15 The regression used by the Company to determine the DOM LSE peak loads captures this
16 trend. However, instead of using the regression to estimate the growing share of Other
17 LSE peak loads during the forecast period, the Company ignored the trend, and simply
18 applied the historical average throughout the forecast period.
17 Data Request Attachment ER -3-43(b).
Wilson Direct Testimony Page 24

I Q 55: What was the "peak adjustment" that was applied to the DOM LSE forecast (your
2 step 3)?
3 A: The "peak adjustment" reflects behind -the -meter generation (55 MW) and the results of
4 RPM auctions for 2016-2019.18 The forecast is essentially increased based on the amount
5 of excess capacity cleared in RPM auctions.
6 Q 56: Please comment on the peak adjustments.
7 A: I have not reviewed and have no comment on the behind -the -meter estimate. As to the
8 adjustment of the peak based on RPM auctions, this does not make sense (as I will
9 explain later in my testimony) and I have not included it in my estimates of TRRs.
10 Q 57: Are there other issues with the Company's approach to calculating TRRs?
11 A: Yes, there are a few other issues that have small impacts.
12 1. The Company used a single coincidence factor (averaged over 2016-2019) for all
13 years. However, PJM forecasts coincident and non -coincident peaks by year, and the
14 coincidence factor varies from year to year.
15 2. The Company used a single reserve margin (16.5%) for all years. However, the PJM
16 study that identified this reserve margin recommended values by year through 2025,
17 and the values vary over time (they are lower for 2022 — 2025).19
'g Response to Data Request ER -3-35, included here in Attachment JFW-5.
19 PJM, 2015 Reserve Requirement Study, Table I-2 p. 14, available at http://www.pim.com/—/media/plannin /r�es-
adeq/2015 -pj m -re s erve-re quirement-study. ashx.
Wilson Direct Testimony Page 25

I Q 58: Have you calculated the DOM LSE Total Resource Requirement correcting the
2 various issues you have identified?
3 A: Yes I have. While I would prefer to simply apply PJM's approach, some of the data that
4 would be required is not available or available only on a non -coincident peak basis. So
5 my calculations follow the Company's approach, with the various alternative forecasts
6 and corrections. My estimates of the DOM LSE TRR reflect the following differences
7 from the Company's estimates:
8 1. I used PJM's latest DOM Zone coincident peak load forecast (July 2016), extended
9 based on the ratio of coincident to non -coincident peaks in the January 2016 forecast.
10 2. I re -estimated Other LSE peak loads using the Company's regression for this purpose.
11 3. I removed the peak adjustment for RPM results.
12 4. I used PJM's annual installed reserve margin values.
13 The results of the calculation were shown above in Table 1.
14
15 Q 59: The 2016 Plan states (p. 63) that the Company, as a PJM member and signatory to
16 PJM's Reliability Assurance Agreement ("RAA"), is obligated to own or procure
17 sufficient capacity to maintain overall system reliability. Is it correct that the RAA
18 obligates the Company to own or procure capacity?
19 A: No. PJM acquires commitments to provide the capacity needed for resource adequacy
20 through its Reliability Pricing Model ("RPM") capacity construct. The RAA assigns
21 capacity responsibility for the purpose of allocating RPM costs to zones and to LSEs.
22 However, the RAA does not obligate any party to own or procure capacity; its references
Wilson Direct Testimony Page 26

1 to "capacity obligations" ultimately have to do with cost allocation.20 Indeed, many
2 LSEs in PJM do not own capacity or have capacity under contract.
3 Q 60: The 2016 Plan (p. 65) also identifies an "upper bound reserve margin", and states
4 that the Company may be required to meet this reserve margin in the future. Is this
5 correct?
6 A: No. Again, PJM does not require acquisition of capacity or any particular reserve
7 margin. The relevant calculations are only for purposes of cost allocation.
8 The Company calculates this higher reserve margin noting that RPM has often resulted in
9 total capacity commitments in excess of reliability targets. But this is merely a result of
10 the sloped RPM capacity demand ("VRR") curve used in the RPM auctions. The sloped
11 VRR curve ensures that when capacity is relatively scarce and costly, RPM's auctions
12 will result in a relatively low amount of committed capacity and high capacity prices; and
13 when capacity is relatively abundant and low cost (as it has been in recent years), RPM
14 will result in a total amount of committed capacity in excess of resource adequacy
15 targets, and relatively low capacity prices. This approach sends a price signal about the
16 need for capacity.
17 Q 61: Would it be prudent for the Company to plan for the higher reserve margins that
is often result from the RPM auctions?
19 A: No, that would not be prudent, and it would make no sense. When RPM results in excess
20 committed capacity, this occurs at a relatively low capacity price, signaling that capacity
21 is abundant and incremental capacity is not needed. Under such circumstances, while the
20 Response to Data Request ER -3-37.
Wilson Direct Testimony Page 27

1 nominal amount of capacity to be allocated to zones and LSEs is higher, the total capacity
2 cost to be allocated is actually much lower. To the extent market participants expect
3 RPM to result in excess capacity at low cost, it would make more sense for market
4 participants to react to such a situation of abundance by planning relatively less, not
5 more, capacity.
6 Q 62: Please explain how, when RPM clears excess capacity, the total capacity cost is
7 actually lower.
8 A: Consider the following example, using the parameters from the most recent RPM base
9 residual auction. If RPM cleared at the target reliability requirement, the clearing price
10 would be $434.46/MW-day and the total market cost would be $25 billion. If instead, as
11 actually occurred, RPM clears a large excess at $100/MW-day, the total market cost
12 would be closer to $6 billion (ignoring higher prices in some zones). Thus, when RPM
13 clears excess capacity, it results in less, not more capacity cost allocated to Dominion and
14 other LSEs.
15 Q 63: The 2016 Plan also suggests (pp. 65-66) that it may be prudent to plan for a higher
16 reserve margin at this time, because potential retirements cause uncertainty. Do
17 you agree?
18 A: No. While there have been many retirements in PJM, they have occurred in an
19 environment of low prices, excess capacity, and robust entry by new gas-fired generation.
20 Demand response, wind, and solar resources are also increasing, as costs decline.
21 The retirements reflect older, mainly coal-fired plants that have become uneconomic and
22 essentially been pushed into retirement by new, lower cost resources. Retirements are
23 not the driver, they are an outcome of the entry of low-cost resources.
Wilson Direct Testimony Page 28

I VII. CONCLUSIONS AND RECOMMENDATIONS
2 Q 64: Please summarize your conclusions with regard to the peak load forecast and Total
3 Resource Requirement values used in the 2016 Plan.
4 A: I conclude that the peak load values are too high, and the Other LSE peaks are
5 understated. More accurate estimates of peak loads, based on PJM's forecasts, and of
6 TRR values are shown in Table 1 above.
7 Q 65: Do you have other recommendations with regard to the load forecasts used in
8 Integrated Resource Plans?
9 A: Yes. With regard to the peak load forecast, I recommend that the Commission consider
10 requiring the following of the Company, in future plans:
11 1. To use the econometric forecasting models to forecast future loads exclusive of data
12 centers loads (by removing historical data center loads from the database, and
13 applying the models to forecast all other loads), in addition to preparing a separate
14 forecast for data center loads, as recommended by Quanta Technology in its 2013
15 report (p. 33).
16 2. To present recent weather -normalized peak loads for the Dominion zone (either
17 prepared by the Company, or PJM's estimates), and to discuss recent trends in
18 weather -normalized peak loads.
19 3. To present load forecasts determined using 20- and 10 -year historical estimation
20 periods, in addition to the longer period currently used, and to provide a rationale for
21 the choice of historical period.
Wilson Direct Testimony Page 29

1 4. To consider enhancing the methodology along the lines of the enhancements
2 implemented by PJM in 2015 (in particular, to capture trends in energy efficiency),
3 and, to the extent similar enhancements are not adopted, to explain why not.
4 Q 66: Do you have other recommendations with regard to the calculation of TRRs used in
5 Integrated Resource Plans?
6 A: Yes. With regard to the calculation of TRRs, I recommend that the Commission consider
7 requiring the following of the Company, in future plans:
8 1. To provide an explicit forecast of the peak loads of Other LSEs in the DOM Zone,
9 with a discussion of recent trends and how the forecast was prepared.
10 2. To use PJM's Forecast Pool Requirement ("FPR") values, applied to a forecast of
11 coincident peak loads, to determine the TRR in unforced capacity terms, consistent
12 with how PJM allocates capacity cost, and to present TRRs in installed capacity
13 terms, if needed, by applying a DOM LSE resource average forced outage rate, again
14 consistent with PJM's approach.
15 Q 67: Does this complete your testimony?
16 A: Yes it does.
Wilson Direct Testimony Page 30

Att. JFW- I

Att. JFW-1
Prefiled Direct Testimony of
James F. Wilson
Figures JFW-A to JFW-E

Att. JFW-1
Figure JFW-A: DOM Zone Actual Peak Loads
(unrestricted, non -coincident summer peak loads; MW)
23,000
22,000
21,000
20,000
4
19,000
18,000
17,000
Source: PJM.
16,000
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Att. JFW-1
Figure JFW-B: DOM Zone Weather -Normalized Peak Loads
(unrestricted, non -coincident summer peak loads; MW)
23,000
+Actual peak
22,000
-4- Weather -normalized peak
21,000 }WAN peaks net of data centers
20,000
19,000
18,000
17,000
Sources: PJM; Data Request ER Set 3-46(c).
16,000
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Att. JFW-1
Figure JFW-C: DOM Zone Peak Load Forecast from the 2016 Plan
(unrestricted, non -coincident summer peak loads; MW)
23,000
(2016 Plan DOM zone peak forecast
22,000
+2016 Plan forecast net of data centers
-4—Weather-normalized peak
21,000 —*--W/N peaks net of data centers
20,000
19,000
18,000
17,000
Sources: 2016 Plan App. 2G; PJM; Quanta Report.
16,000
2003 2005 2007 2009 2011 2013 2015 2017 2019 2021 2023 2025

23,000
22,000
21,000
20,000
19,000
18,000
Figure JFW-D: DOM Zone Peak Load and Economic Forecasts
(unrestricted, non -coincident summer peak loads; MW)
—4-2016 Plan DOM Zone peak forecast
X2016 Plan forecast net of data centers
♦Weather -normalized peak
—+—W/N peaks net of data centers
---PJM DOM Zone Economic Variable
"M-R�
aw
Att. JFW-1
1.6
1.4
1.2
1.0
M•
17,000 0.6
Sources: 2016 Plan App. 2G; PJM; Quanta Report.
16,000 0.4
2003 2005 2007 2009 2011 2013 2015 2017 2019 2021 2023 2025

23,000
22,000
21,000
20,000
19,000
17,000
Att. JFW-1
Figure JFW-E: PJM's DOM Zone Peak Load Forecasts
(unrestricted, non -coincident summer peak loads; MW)
16,000
2003 2005 2007 2009 2011 2013 2015 2017
UM
1.6
1.4
1.2
1.0
M•
0.4
2019 2021 2023 2025

Att. JFW-2

Att. JFW-2
James F. Wilson
Principal, Wilson Energy Economics
4800 Hampden Lane Suite 200
Bethesda, Maryland 20814 USA
Phone: [phone]
Cell: [phone]
Fax: [phone]
Email: [e-mail]
www.wilsonenec.com
SUMMARY
James F. Wilson is an economist with 30 years of consulting experience, primarily in the electric power
and natural gas industries. Many of his assignments have pertained to the economic and policy issues
arising from the interplay of competition and regulation in these industries, including restructuring policies,
market design, market analysis and market power. Other recent engagements have involved resource
adequacy and capacity markets, contract litigation and damages, forecasting and market evaluation,
pipeline rate cases and evaluating allegations of market manipulation. Mr. Wilson has been involved in
electricity restructuring and wholesale market design for over twenty years in California, PJM, New
England, Russia and other regions. He also spent five years in Russia in the early 1990s advising on the
reform, restructuring and development of the Russian electricity and natural gas industries.
Mr. Wilson has submitted affidavits and testified in Federal Energy Regulatory Commission and state
regulatory proceedings. His papers have appeared in the Energy Journal, Electricity Journal, Public
Utilities Fortnightly and other publications, and he often presents at industry conferences.
Prior to founding Wilson Energy Economics, Mr. Wilson was a Principal at LECG, LLC. He has also
worked for ICF Resources, Decision Focus Inc., and as an independent consultant.
EDUCATION
MS, Engineering -Economic Systems, Stanford University, 1982
BA, Mathematics, Oberlin College, 1977
RECENT ENGAGEMENTS
• Various consulting assignments on wholesale electric capacity market design issues in PJM, New
England, the Midwest, Texas, and California.
• Cost -benefit analysis of a new natural gas pipeline.
• Evaluation of the impacts of demand response on electric generation capacity mix and emissions.
• Panelist on a FERC technical conference on capacity markets.
• Affidavit on the potential for market power over natural gas storage.
• Executive briefing on wind integration and linkages to short-term and longer-term resource
adequacy approaches.
• Affidavit on the impact of a centralized capacity market on the potential benefits of participation in
a Regional Transmission Organization (RTO).
• Participated in a panel teleseminar on resource adequacy policy and modeling.
• Affidavit on opt -out rules for centralized capacity markets.
• Affidavits on minimum offer price rules for RTO centralized capacity markets.
• Evaluated electric utility avoided cost in a tax dispute.
• Advised on pricing approaches for RTO backstop short-term capacity procurement.

Att. JFW-2
• Affidavit evaluating the potential impact on reliability of demand response products limited in the
number or duration of calls.
• Evaluated changing patterns of natural gas production and pipeline flows, developed approaches
for pipeline tolls and cost recovery.
• Evaluated an electricity peak load forecasting methodology and forecast; evaluated regional
transmission needs for resource adequacy.
• Participated on a panel teleseminar on natural gas price forecasting.
• Affidavit evaluating a shortage pricing mechanism and recommending changes.
• Testimony in support of proposed changes to a forward capacity market mechanism.
• Reviewed and critiqued an analysis of the economic impacts of restrictions on oil and gas
development.
• Advised on the development of metrics for evaluating the performance of Regional Transmission
Organizations and their markets.
• Prepared affidavit on the efficiency benefits of excess capacity sales in readjustment auctions for
installed capacity.
• Prepared affidavit on the potential impacts of long lead time and multiple uncertainties on clearing
prices in an auction for standard offer electric generation service.
EARLIER PROFESSIONAL EXPERIENCE
LECG, LCC, Washington, DC 1998-2009.
Principal
• Reviewed and commented on an analysis of the target installed capacity reserve margin for the
Mid Atlantic region; recommended improvements to the analysis and assumptions.
• Evaluated an electric generating capacity mechanism and the price levels to support adequate
capacity; recommended changes to improve efficiency.
• Analyzed and critiqued the methodology and assumptions used in preparation of a long run
electricity peak load forecast.
• Evaluated results of an electric generating capacity incentive mechanism and critiqued the
mechanism's design; prepared a detailed report. Evaluated the impacts of the mechanism's flaws
on prices and costs and prepared testimony in support of a formal complaint.
• Analyzed impacts and potential damages of natural gas migration from a storage field.
• Evaluated allegations of manipulation of natural gas prices and assessed the potential impacts of
natural gas trading strategies.
• Prepared affidavit evaluating a pipeline's application for market-based rates for interruptible
transportation and the potential for market power.
• Prepared testimony on natural gas industry contracting practices and damages in a contract
dispute.
• Prepared affidavits on design issues for an electric generating capacity mechanism for an eastern
US regional transmission organization; participated in extensive settlement discussions.
• Prepared testimony on the appropriateness of zonal rates for a natural gas pipeline.
• Evaluated market power issues raised by a possible gas -electric merger.
• Prepared testimony on whether rates for a pipeline extension should be rolled -in or incremental
under Federal Energy Regulatory Commission ("FERC") policy.
• Prepared an expert report on damages in a natural gas contract dispute.
• Prepared testimony regarding the incentive impacts of a ratemaking method for natural gas
pipelines.
• Prepared testimony evaluating natural gas procurement incentive mechanisms.
• Analyzed the need for and value of additional natural gas storage in the southwestern US.
• Evaluated market issues in the restructured Russian electric power market, including the need to
introduce financial transmission rights, and policies for evaluating mergers.

Att. JFW-2
• Affidavit on market conditions in western US natural gas markets and the potential for a new
merchant gas storage facility to exercise market power.
• Testimony on the advantages of a system of firm, tradable natural gas transmission and storage
rights, and the performance of a market structure based on such policies.
• Testimony on the potential benefits of new independent natural gas storage and policies for
providing transmission access to storage users.
• Testimony on the causes of California natural gas price increases during 2000-2001 and the
possible exercise of market power to raise natural gas prices at the California border.
• Advised a major US utility with regard to the Federal Energy Regulatory Commission's proposed
Standard Market Design and its potential impacts on the company.
• Reviewed and critiqued draft legislation and detailed market rules for reforming the Russian
electricity industry, for a major investor in the sector.
• Analyzed the causes of high prices in California wholesale electric markets during 2000 and
developed recommendations, including alternatives for price mitigation. Testimony on price
mitigation measures.
• Summarized and critiqued wholesale and retail restructuring and competition policies for electric
power and natural gas in select US states, for a Pacific Rim government contemplating energy
reforms.
• Presented testimony regarding divestiture of hydroelectric generation assets, potential market
power issues, and mitigation approaches to the California Public Utilities Commission.
• Reviewed the reasonableness of an electric utility's wholesale power purchases and sales in a
restructured power market during a period of high prices.
• Presented an expert report on failure to perform and liquidated damages in a natural gas contract
dispute.
• Presented a workshop on Market Monitoring to a group of electric utilities in the process of
forming an RTO.
• Authored a report on the screening approaches used by market monitors for assessing exercise
of market power, material impacts of conduct, and workable competition.
• Developed recommendations for mitigating locational market power, as part of a package of
congestion management reforms.
• Provided analysis in support of a transmission owner involved in a contract dispute with
generators providing services related to local grid reliability.
• Authored a report on the role of regional transmission organizations in market monitoring.
• Prepared market power analyses in support of electric generators' applications to FERC for
market-based rates for energy and ancillary services.
• Analyzed western electricity markets and the potential market power of a large producer under
various asset acquisition or divestiture strategies.
• Testified before a state commission regarding the potential benefits of retail electric competition
and issues that must be addressed to implement it.
• Prepared a market power analysis in support of an acquisition of generating capacity in the New
England market.
• Advised a California utility regarding reform strategies for the California natural gas industry,
addressing market power issues and policy options for providing system balancing services.
ICF RESOURCES, INC., Fairfax, VA, 1997-1998.
Project Manager
• Reviewed, critiqued and submitted testimony on a New Jersey electric utility's restructuring
proposal, as part of a management audit for the state regulatory commission.
• Assisted a group of US utilities in developing a proposal to form a regional Independent System
Operator (ISO).
• Researched and reported on the emergence of Independent System Operators and their role in
reliability, for the Department of Energy.

Att. JFW-2
• Provided analytical support to the Secretary of Energy's Task Force on Electric System Reliability
on various topics, including ISOs. Wrote white papers on the potential role of markets in ensuring
reliability.
• Recommended near-term strategies for addressing the potential stranded costs of non-utility
generator contracts for an eastern utility; analyzed and evaluated the potential benefits of various
contract modifications, including buyout and buydown options; designed a reverse auction
approach to stimulating competition in the renegotiation process.
• Designed an auction process for divestiture of a Northeastern electric utility's generation assets
and entitlements (power purchase agreements).
• Participated in several projects involving analysis of regional power markets and valuation of
existing or proposed generation assets.
IRIS MARKET ENVIRONMENT PROJECT, 1994-1996.
Proiect Director. Moscow. Russia
Established and led a policy analysis group advising the Russian Federal Energy Commission and
Ministry of Economy on economic policies for the electric power, natural gas, oil pipeline,
telecommunications, and rail transport industries (the Program on Natural Monopolies, a project of the
IRIS Center of the University of Maryland Department of Economics, funded by USAID):
• Advised on industry reforms and the establishment of federal regulatory institutions.
• Advised the Russian Federal Energy Commission on electricity restructuring, development of a
competitive wholesale market for electric power, tariff improvements, and other issues of electric
power and natural gas industry reform.
• Developed policy conditions for the IMF's $10 billion Extended Funding Facility.
• Performed industry diagnostic analyses with detailed policy recommendations for electric power
(1994), natural gas, rail transport and telecommunications (1995), oil transport (1996).
Independent Consultant stationed in Moscow, Russia, 1991-1996
Projects for the WORLD BANK, 1992-1996:
• Bank Strategy for the Russian Electricity Sector. Developed a policy paper outlining current
industry problems and necessary policies, and recommending World Bank strategy.
• Russian Electric Power Industry Restructuring. Participated in work to develop recommendations
to the Russian Government on electric power industry restructuring.
• Russian Electric Power Sector Update. Led project to review developments in sector
restructuring, regulation, demand, supply, tariffs, and investment.
• Russian Coal Industry Restructuring. Analyzed Russian and export coal markets and developed
forecasts of future demand for Russian coal.
• World Bank/IEA Electricity Options Study for the G-7. Analyzed mid- and long-term electric power
demand and efficiency prospects and developed forecasts.
• Russian Energy Pricing and Taxation. Developed recommendations for liberalizing energy
markets, eliminating subsidies and restructuring tariffs for all energy resources.
Other consulting assignments in Russia, 1991-1994:
• Advised on projects pertaining to Russian energy policy and the transition to a market economy in
the energy industries, for the Institute for Energy Research of the Russian Academy of Sciences.
• Presented seminars on the structure, economics, planning, and regulation of the energy and
electric power industries in the US, for various Russian clients.

Att. JFW-2
DECISION FOCUS INC., Mountain View, CA, 1983-1992
Senior Associate, 1985-1992.
• For the Electric Power Research Institute, led projects to develop decision -analytic methodologies
and models for evaluating long term fuel and electric power contracting and procurement
strategies. Applied the methodologies and models in numerous case studies, and presented
several workshops and training sessions on the approaches.
• Analyzed long-term and short-term natural gas supply decisions for a large California gas
distribution company following gas industry unbundling and restructuring.
• Analyzed long term coal and rail alternatives for a midwest electric utility, including alternative
coal supply regions, suppliers and contract structures; spot/contract mix; rail arrangements;
power purchases; conversion to gas.
• Evaluated bulk power purchase alternatives and strategies for a New Jersey electric utility.
• Performed a financial and economic analysis of a proposed hydroelectric project.
• For a natural gas pipeline company serving the Northeastern US, forecasted long-term natural
gas supply and transportation volumes. Developed a forecasting system for staff use.
• Analyzed potential benefits of diversification of suppliers for a natural gas pipeline company.
• Evaluated uranium contracting strategies for an electric utility.
• Analyzed telecommunications services markets under deregulation, developed and implemented
a pricing strategy model. Evaluated potential responses of residential and business customers to
changes in the client's and competitors' telecommunications services and prices.
• Analyzed coal contract terms and supplier diversification strategies for an eastern electric utility.
• Analyzed oil and natural gas contracting strategies for an electric utility.
TESTIMONY AND AFFIDAVITS
In the Matter of the Application of DTE Electric Company for Authority to Implement a Power Supply
Cost Recovery Plan in its Rate Schedules for 2016 Metered Jurisdictional Sales of Electricity,
Michigan Public Service Commission Case No. U-17920, Direct Testimony on behalf of Michigan
Environmental Council and the Sierra Club, March 14, 2016.
In the Matter of the Application Seeking Approval of Ohio Power Company's Proposal to Enter into
an Affiliate Power Purchase Agreement for Inclusion in the Power Purchase Agreement Rider, Public
Utilities Commission of Ohio Case No. 14-1693-EL-RDR: Direct Testimony on Behalf of the Office of
the Ohio Consumers' Counsel, September 11, 2015; deposition, September 30, 2015; supplemental
deposition, October 16, 2015; testimony at hearings, October 21, 2015; supplemental testimony
December 28, 2015; second supplemental deposition, December 30, 2015; testimony at hearings
January 8, 2016.
Indicated Market Participants v. PJM Interconnection, L.L.C., FERC Docket No. EL15-88, Affidavit on
behalf of the Joint Consumer Representatives and Interested State Commissions, August 17, 2015.
ISO New England Inc. and New England Power Pool Participants Committee, FERC Docket No.
ER15-2208, Testimony on Behalf of the New England States Committee on Electricity, August 5,
2015.
Joint Consumer Representatives v. PJM Interconnection, L.L.C., FERC Docket No. EL15-83,
Affidavit in Support of the Motion to Intervene and Comments of the Public Power Association of
New Jersey, July 20, 2015.
In the Matter of the Tariff Revisions Filed by ENSTAR Natural Gas Company, a Division of SEMCO
Energy, Inc., Regulatory Commission of Alaska Case No. U-14-111, Testimony on Behalf of
Matanuska Electric Association, Inc., May 13, 2015.
In the Matter of the Application of Ohio Edison Company, The Cleveland Electric Illuminating
Company and The Toledo Edison Company for Authority to Provide for a Standard Service Offer
Pursuant to R.C. 4928.143 in the Form of an Electric Security Plan, Public Utilities Commission of
Ohio Case No. 14-1297-EL-SSO: Direct Testimony on Behalf of the Office of the Ohio Consumers'

Att. JFW-2
Counsel and Northeast Ohio Public Energy Council, December 22, 2014; deposition, February 10,
2015; supplemental testimony May 11, 2015; second deposition May 26, 2015; testimony at
hearings, October 2, 2015; second supplemental testimony December 30, 2015; third deposition
January 8, 2016; testimony at hearings January 19, 2016; rehearing direct testimony June 22, 2016;
fourth deposition July 5, 2016; testimony at hearings July 14, 2016.
PJM Interconnection, L.L.C., FERC Docket No. ER14-2940 (RPM Triennial Review), Affidavit in
Support of the Protest of the PJM Load Group, October 16, 2014.
In the Matter of the Application of Duke Energy Ohio for Authority to Establish a Standard Service
Offer in the Form of an Electric Security Plan, Public Utilities Commission of Ohio Case No. 14-841-
EL-SSO: Direct Testimony on Behalf of the Office of the Ohio Consumers' Counsel, September 26,
2014; deposition, October 6, 2014; testimony at hearings, November 5, 2014.
In the Matter of the Application of Ohio Power Company for Authority to Establish a Standard Service
Offer in the Form of an Electric Security Plan, Public Utilities Commission of Ohio Case No. 13-2385-
EL-SSO: Direct Testimony on Behalf of the Office of the Ohio Consumers' Counsel, May 6, 2014;
deposition, May 29, 2014; testimony at hearings, June 16, 2014.
PJM Interconnection, L.L.C., FERC Docket No. ER14-504 (Clearing of Demand Response in RPM),
Affidavit in Support of the Protest of the Joint Consumer Advocates and Public Interest
Organizations, December 20, 2013.
New England Power Generators Association, Inc. v. ISO New England Inc., FERC Docket No. EL14-
7, Testimony in Support of the Protest of the New England States Committee on Electricity,
November 27, 2013.
Midwest Independent Transmission System Operator, Inc., FERC Docket No. ER11-4081, Affidavit
In Support of Brief of the Midwest TDUs, October 11, 2013.
ANR Storage Company, FERC Docket No. RP12-479, Prepared Answering Testimony on behalf of
the Joint Intervenor Group, April 2, 2013; Prepared Cross -answering Testimony, May 15, 2013;
testimony at hearings, September 4, 2013.
In the Matter of the Application of The Dayton Power and Light Company for Approval of its Market
Rate Offer, Public Utilities Commission of Ohio Case No. 12-426-EL-SSO: Direct Testimony on
Behalf of the Office of the Ohio Consumers' Counsel, March 5, 2013; deposition, March 11, 2013.
PJM Interconnection, L.L.C., FERC Docket No. ER13-535 (Minimum Offer Price Rule), Affidavit in
Support of the Protest and Comments of the Joint Consumer Advocates, December 28, 2012.
In the Matter of the Application of Ohio Edison Company, et al for Authority to Provide for a Standard
Service Offer in the Form of an Electric Security Plan, Public Utilities Commission of Ohio Case No.
12-1230-EL-SSO: Direct Testimony on Behalf of the Office of the Ohio Consumers' Counsel, May
21, 2012; deposition, May 30, 2012; testimony at hearings, June 5, 2012.
PJM Interconnection, L.L.C., FERC Docket No. ER12-513, Affidavit in Support of Protest of the Joint
Consumer Advocates and Demand Response Supporters (changes to RPM), December 22, 2011.
People of the State of Illinois ex rel. Leon A. Greenblatt, III v Commonwealth Edison Company,
Circuit Court of Cook County, Illinois, deposition, September 22, 2011; interrogatory, Feb. 22, 2011.
In the Matter of the Application of Union Electric Company for Authority to Continue the Transfer of
Functional Control of Its Transmission System to the Midwest Independent Transmission System
Operator, Inc., Missouri PSC Case No. EO -2011-0128, Testimony in hearings, February 9, 2012;
Rebuttal Testimony and Response to Commission Questions On Behalf Of The Missouri Joint
Municipal Electric Utility Commission, September 14, 2011.
PJM Interconnection, L.L.C., and PJM Power Providers Group v. PJM Interconnection, L.L.C., FERC
Docket Nos. ER11-2875 and EL11-20 (Minimum Offer Price Rule), Affidavit in Support of Protest of
New Jersey Division of Rate Counsel, March 4, 2011, and Affidavit in Support of Request for
Rehearing and for Expedited Consideration of New Jersey Division of Rate Counsel, May 12, 2011.
PJM Interconnection, L.L.C., FERC Docket No. ER11-2288 (Demand response "saturation" issue),
Affidavit in Support of Protest and Comments of the Joint Consumer Advocates, December 23, 2010.

Att. JFW-2
North American Electric Reliability Corporation, FERC Docket No. RM10-10, Comments on
Proposed Reliability Standard BAL-502-RFC-02: Planning Resource Adequacy Analysis,
Assessment and Documentation, December 23, 2010.
In the Matter of the Reliability Pricing Model and the 2013/2014 Delivery Year Base Residual Auction
Results, Maryland Public Service Commission Administrative Docket PC22, Comments and
Responses to Questions On Behalf of Southern Maryland Electric Cooperative, October 15, 2010.
PJM Interconnection, L.L.C., FERC Docket No. ER09-1063-004 (PJM compliance filing on pricing
during operating reserve shortages): Affidavit In Support of Comments and Protest of the
Pennsylvania Public Utility Commission, July 30, 2010.
ISO New England, Inc. and New England Power Pool, FERC Docket No. ER10-787-000 on Forward
Capacity Market Revisions: Direct Testimony On Behalf Of The Connecticut Department of Public
Utility Control, March 30, 2010; Direct Testimony in Support of First Brief of the Joint Filing
Supporters, July 1, 2010; Supplemental Testimony in Support of Second Brief of the Joint Filing
Supporters, September 1, 2010.
PJM Interconnection, L.L.C., FERC Docket No. ER09-412-006: Affidavit In Support of Protest of
Indicated Consumer Interests, January 19, 2010.
In the Matter of the Application of Ohio Edison Company, et al for Approval of a Market Rate Offer to
Conduct a Competitive Bidding Process for Standard Service Offer Electric Generation Supply,
Public Utilities Commission of Ohio Case No. 09-906-EL-SSO: Direct Testimony on Behalf of the
Office of the Ohio Consumers' Counsel, December 7, 2009; deposition, December 10, 2009,
testimony at hearings, December 22, 2009.
Application of PATH Allegheny Virginia Transmission Corporation for Certificates of Public
Convenience and Necessity to Construct Facilities: 765 kV Transmission Line through Loudon,
Frederick and Clarke Counties, Virginia State Corporation Commission Case No. PUE-2009-00043:
Direct Testimony on Behalf of Commission Staff, December 8, 2009.
PJM Interconnection, L.L.C., FERC Docket No. ER09-412-000: Affidavit On Proposed Changes to
the Reliability Pricing Model On Behalf Of RPM Load Group, January 9, 2009; Reply Affidavit,
January 26, 2009.
PJM Interconnection, L.L.C., FERC Docket No. ER09-412-000: Affidavit In Support of the Protest
Regarding Load Forecast To Be Used in May 2009 RPM Auction, January 9, 2009.
Maryland Public Service Commission et al v. PJM Interconnection, L.L.C., FERC Docket No. EL08-
67-000: Affidavit in Support Complaint of the RPM Buyers, May 30, 2008; Supplemental Affidavit,
July 28, 2008.
PJM Interconnection, L.L.C., FERC Docket No. ER08-516: Affidavit On PJM's Proposed Change To
RPM Parameters On Behalf Of RPM Buyers, March 6, 2008.
PJM Interconnection, L.L.C., Reliability Pricing Model Compliance Filing, FERC Docket Nos. ER05-
1410 and EL05-148: Affidavit Addressing RPM Compliance Filing Issues on Behalf of the Public
Power Association of New Jersey, October 15, 2007.
TXU Energy Retail Company LP v. Leprino Foods Company, Inc., US District Court for the Northern
District of California, Case No. C01-20289: Testimony at trial, November 15-29, 2006; Deposition,
April 7, 2006; Expert Report on Behalf of Leprino Foods Company, March 10, 2006.
Gas Transmission Northwest Corporation, Federal Energy Regulation Commission Docket No.
RP06-407: Reply Affidavit, October 26, 2006; Affidavit on Behalf of the Canadian Association of
Petroleum Producers, October 18, 2006.
PJM Interconnection, L.L.C., Reliability Pricing Model, FERC Docket Nos. ER05-1410 and EL05-
148: Supplemental Affidavit on Technical Conference Issues, June 22, 2006; Supplemental Affidavit
Addressing Paper Hearing Topics, June 2, 2006; Affidavit on Behalf of the Public Power Association
of New Jersey, October 19, 2005.

Att. JFW-2
Maritimes & Northeast Pipeline, L.L.C., FERC Docket No. RP04-360-000: Prepared Cross
Answering Testimony, March 11, 2005; Prepared Direct and Answering Testimony on Behalf of Firm
Shipper Group, February 11, 2005.
Dynegy Marketing and Trade v. Multiut Corporation, US District Court of the Northern District of
Illinois, Case. No. 02 C 7446: Deposition, September 1, 2005; Expert Report in response to
Defendant's counterclaims, March 21, 2005; Expert Report on damages, October 15, 2004.
Application of Pacific Gas and Electric Company, California Public Utilities Commission proceeding
A.04-03-021: Prepared Testimony, Policy for Throughput -Based Backbone Rates, on behalf of
Pacific Gas and Electric Company, May 21, 2004.
Gas Market Activities, California Public Utilities Commission Order Instituting Investigation 1.02-11-
040: Testimony at hearings, July, 2004; Prepared Testimony, Comparison of Incentives Under Gas
Procurement Incentive Mechanisms, on behalf of Pacific Gas and Electric Company, December 10,
2003.
Application of Red Lake Gas Storage, L.P., FERC Docket No. CP02-420, Affidavit in support of
application for market-based rates for a proposed merchant gas storage facility, March 3, 2003.
Application of Pacific Gas and Electric Company, California Public Utilities Commission proceeding
A.01-10-011: Testimony at hearings, April 1-2, 2003; Rebuttal Testimony, March 24, 2003; Prepared
Testimony, Performance of the Gas Accord Market Structure, on behalf of Pacific Gas and Electric
Company, January 13, 2003.
Application of Wild Goose Storage, Inc., California Public Utilities Commission proceeding A.01-06-
029: Testimony at hearings, November, 2001; Prepared testimony regarding policies for backbone
expansion and tolls, and potential ratepayer benefits of new storage, on behalf of Pacific Gas and
Electric Company, October 24, 2001.
Public Utilities Commission of the State of California v. EI Paso Natural Gas Co., FERC Docket No.
RP00-241: Testimony at hearings, May -June, 2001; Prepared Testimony on behalf of Pacific Gas
and Electric Company, May 8, 2001.
Application of Pacific Gas and Electric Company, California Public Utilities Commission proceeding
A.99-09-053: Prepared testimony regarding market power consequences of divestiture of
hydroelectric assets, December 5, 2000.
San Diego Gas & Electric Company, et al, FERC Docket No. EL00-95: Prepared testimony regarding
proposed price mitigation measures on behalf of Pacific Gas and Electric Company, November 22,
2000.
Application of Harbor Cogeneration Company, FERC Docket No. ER99-1248: Affidavit in support of
application for market-based rates for energy, capacity and ancillary services, December 1998.
Application of and Complaint of Residential Electric, Incorporated vs. Public Service Company of
New Mexico, New Mexico Public Utility Commission Case Nos. 2867 and 2868: Testimony at
hearings, November, 1998; Direct Testimony on behalf of Public Service Company of New Mexico
on retail access issues, November, 1998.
Management audit of Public Service Electric and Gas' restructuring proposal for the New Jersey
Board of Public Utilities: Prepared testimony on reliability and basic generation service, March 1998.
PUBLISHED ARTICLES
Forward Capacity Market CONEfusion, Electricity Journal Vol. 23 Issue 9, November 2010.
Reconsidering Resource Adequacy (Part 2): Capacity Planning for the Smart Grid, Public Utilities
Fortnightly, May 2010.
Reconsidering Resource Adequacy (Part 1): Has the One -Day -in -Ten -Years Criterion Outlived Its
Usefulness? Public Utilities Fortnightly, April 2010.

Att. JFW-2
A Hard Look at Incentive Mechanisms for Natural Gas Procurement, with K. Costello, National
Regulatory Research Institute Report No. 06-15, November 2006.
Natural Gas Procurement: A Hard Look at Incentive Mechanisms, with K. Costello, Public Utilities
Fortnightly, February 2006, p. 42.
After the Gas Bubble: An Economic Evaluation of the Recent National Petroleum Council Study, with
K. Costello and H. Huntington, Energy Journal Vol. 26 No. 2 (2005).
High Natural Gas Prices in California 2000-2001: Causes and Lessons, Journal of Industry,
Competition and Trade, vol. 2:1/2, November 2002.
Restructuring the Electric Power Industry: Past Problems, Future Directions, Natural Resources and
Environment, ABA Section of Environment, Energy and Resources, Volume 16 No. 4, Spring, 2002.
Scarcity, Market Power, Price Spikes, and Price Caps, Electricity Journal, November, 2000.
The New York ISO's Market Power Screens, Thresholds, and Mitigation: Why It Is Not A Model For
Other Market Monitors, Electricity Journal, August/September 2000.
ISOs: A Grid -by -Grid Comparison, Public Utilities Fortnightly, January 1, 1998.
Economic Policy in the Natural Monopoly Industries in Russia: History and Prospects (with V.
Capelik), Voprosi Ekonomiki, November 1995.
Meeting Russia's Electric Power Needs: Uncertainty, Risk and Economic Reform, Financial and
Business News, April 1993.
Russian Energy Policy through the Eyes of an American Economist, Energeticheskoye Stroitelstvo,
December 1992, p 2.
Fuel Contracting Under Uncertainty, with R. B. Fancher and H. A. Mueller, IEEE Transactions on
Power Systems, February, 1986, p. 26-33.
OTHER ARTICLES, REPORTS AND PRESENTATIONS
Panel: What is the PJM Load Forecast, Organization of PJM States, Inc. Annual Meeting, October
12, 2015.
PJM's "Capacity Performance" Tariff Changes: Estimated Impact on the Cost of Capacity, prepared
for the American Public Power Association, October, 2015.
Panel: Capacity Performance (and Incentive) Reform, EUCI Conference on Capacity Markets:
Gauging Their Real Impact on Resource Development & Reliability, August 15, 2015.
Panel on Load Forecasting, Organization of PJM States, Inc. Spring Strategy Meeting, April 13,
2015.
Panelist for Session 2: Balancing Bulk Power System and Distribution System Reliability in the
Eastern Interconnection, Meeting of the Eastern Interconnection States' Planning Council, December
11, 2014.
Panel: Impact of PJM Capacity Performance Proposal on Demand Response, Mid -Atlantic
Distributed Resources Initiative (MADRI) Working Group Meeting #36, December 9, 2014.
Panel: Applying the Lessons Learned from Extreme Weather Events — What Changes Are Needed
In PJM Markets and Obligations? Infocast PJM Market Summit, October 28, 2014.
Panel on RPM: What Changes Are Proposed This Year? Organization of PJM States, Inc. 1011
Annual Meeting, Chicago Illinois, October 13-14, 2014.
Panel on centralized capacity market design going forward, Centralized Capacity Markets in
Regional Transmission Organizations and Independent System Operators, Docket No. AD13-7,
September 25, 2013; post -conference comments, January 8, 2014.
Economics of Planning for Resource Adequacy, NARUC Summer Meetings, Denver, Colorado, July
21, 2013.

Att. JFW-2
The Increasing Need for Flexible Resources: Considerations for Forward Procurement, EUCI
Conference on Fast and Flexi-Ramp Resources, Chicago, Illinois, April 23-24, 2013.
Panel on RPM Issues: Long Term Vision and Recommendations for Now, Organization of PJM
States, Inc. Spring Strategy Meeting, April 3, 2013.
Comments On: The Economic Ramifications of Resource Adequacy Whitepaper, peer review of
whitepaper prepared for EISPC and NARUC, March 24, 2013.
Resource Adequacy: Criteria, Constructs, Emerging Issues, Coal Finance 2013, Institute for Policy
Integrity, NYU School of Law, March 19, 2013.
Panel Discussion — Alternative Models and Best Practices in Other Regions, Long -Term Resource
Adequacy Summit, California Public Utilities Commission and California ISO, San Francisco,
California, February 26, 2013.
Fundamental Capacity Market Design Choices: How Far Forward? How Locational? EUCI Capacity
Markets Conference, October 3, 2012.
One Day in Ten Years? Economics of Resource Adequacy, Mid-America Regulatory Conference
Annual Meeting, June 12, 2012.
Reliability and Economics: Separate Realities? Harvard Electricity Policy Group Sixty -Fifth Plenary
Session, December 1, 2011.
National Regulatory Research Institute Teleseminar: The Economics of Resource Adequacy
Planning: Should Reserve Margins Be About More Than Keeping the Lights On?, panelist,
September 15, 2011.
Improving RTO -Operated Wholesale Electricity Markets: Recommendations for Market Reforms,
American Public Power Association Symposium, panelist, January 13, 2011.
Shortage Pricing Issues, panelist, Organization of PJM States, Inc. Sixth Annual Meeting, October 8,
2010.
National Regulatory Research Institute Teleseminar: Forecasting Natural Gas Prices, panelist, July
28, 2010.
Comments on the NARUC-Initiated Report: Analysis of the Social, Economic and Environmental
Effects of Maintaining Oil and Gas Exploration Moratoria On and Beneath Federal Lands (February
15, 2010) submitted to NARUC on June 22, 2010.
Forward Capacity Market CONEfusion, Advanced Workshop in Regulation and Competition, 29th
Annual Eastern Conference of the Center for Research in Regulated Industries, Rutgers University,
May 21, 2010.
One Day in Ten Years? Resource Adequacy for the Smart Grid, revised draft November 2009.
Approaches to Local Resource Adequacy, presented at Electric Utility Consultants' Smart Capacity
Markets Conference, November 9, 2009.
One Day in Ten Years? Resource Adequacy for the Smarter Grid, Advanced Workshop in
Regulation and Competition, 28th Annual Eastern Conference of the Center for Research in
Regulated Industries, Rutgers University, May 15, 2009.
Resource Adequacy in Restructured Electricity Markets: Initial Results of PJM's Reliability Pricing
Model (RPM), Advanced Workshop in Regulation and Competition, 27th Annual Eastern Conference
of the Center for Research in Regulated Industries, Rutgers University, May 15, 2008.
Statement at Federal Energy Regulatory Commission technical conference, Capacity Markets in
Regions with Organized Electric Markets, Docket No. AD08-4-000, May 7, 2008.
Raising the Stakes on Capacity Incentives: PJM's Reliability Pricing Model (RPM), presentation at
the University of California Energy Institute's 13th Annual POWER Research Conference, Berkeley,
California, March 21, 2008.

Att. JFW-2
Raising the Stakes on Capacity Incentives: PJM's Reliability Pricing Model (RPM), report prepared
for the American Public Power Association, March 14, 2008.
Comments on GTN's Request for Market -Based Rates for Interruptible Transportation, presentation
at technical conference in Federal Energy Regulatory Commission Docket No. RP06-407,
September 26-27, 2006 on behalf of Canadian Association of Petroleum Producers.
Comments on Policies to Encourage Natural Gas Infrastructure, and Supplemental Comments on
Market -Based Rates Policy For New Natural Gas Storage, State of the Natural Gas Industry
Conference, Federal Energy Regulatory Commission Docket No. AD05-14, October 12 and 26,
2005.
After the Gas Bubble: A Critique of the Modeling and Policy Evaluation Contained in the National
Petroleum Council's 2003 Natural Gas Study, with K. Costello and H. Huntington, presented at the
24th Annual North American Conference of the USAEE/IAEE, July 2004.
Comments on the Pipeline Capacity Reserve Concept, State of the Natural Gas Industry
Conference, Federal Energy Regulatory Commission Docket No. PL04-17, October 21, 2004.
Southwest Natural Gas Market and the Need for Storage, Federal Energy Regulatory Commission's
Southwestern Gas Storage Technical Conference, docket AD03-11, August 2003.
Assessing Market Power in Power Markets: the "Pivotal Supplier" Approach and Variants, presented
at Electric Utility Consultants' Ancillary Services Conference, November 1, 2001.
Scarcity and Price Mitigation in Western Power Markets, presented at Electric Utility Consultants'
conference: What To Expect In Western Power Markets This Summer, May 1-2, 2001.
Market Power: Definition, Detection, Mitigation, pre -conference workshop, with Scott Harvey,
January 24, 2001.
Market Monitoring in the U.S.: Evolution and Current Issues, presented at the Association of Power
Exchanges' APEx 2000 Conference, October 25, 2000.
Ancillary Services and Market Power, presented at the Electric Utility Consultants' Ancillary Services
Conference (New Business Opportunities in Competitive Ancillary Services Markets), Sept. 14, 2000.
Market Monitoring Workshop, presented to RTO West Market Monitoring Work Group, June 2000.
Screens and Thresholds Used In Market Monitoring, presented at the Conference on RTOs and
Market Monitoring, Edison Electric Institute and Energy Daily, May 19, 2000.
The Regional Transmission Organization's Role in Market Monitoring, report for the Edison Electric
Institute attached to their comments on the FERC's NOPR on RTOs, August, 1999.
The Independent System Operator's Mission and Role in Reliability, presented at the Electric Utility
Consultants' Conference on ISOs and Transmission Pricing, March 1998.
Independent System Operators and Their Role in Maintaining Reliability in a Restructured Electric
Power Industry, ICF Resources for the U. S. Department of Energy, 1997.
Rail Transport in the Russian Federation, Diagnostic Analysis and Policy Recommendations, with V.
Capelik and others, IRIS Market Environment Project, 1995.
Telecommunications in the Russian Federation: Diagnostic Analysis and Policy Recommendations,
with E. Whitlock and V. Capelik, IRIS Market Environment Project, 1995.
Russian Natural Gas Industry: Diagnostic Analysis and Policy Recommendations, with I. Sorokin and
V. Eskin, IRIS Market Environment Project, 1995.
Russian Electric Power Industry: Diagnostic Analysis and Policy Recommendations, with I. Sorokin,
IRIS Market Environment Project, 1995.

PROFESSIONAL ASSOCIATIONS
United States Association for Energy Economics
Natural Gas Roundtable
Energy Bar Association
Att. JFW-2
July 2016

Att. JFW-3

Att. JFW-3
Vir'nia Electric and Power Company
Case No. PUE-2016-00049
Environmental Respondents
Third Set
The following response to Question No. 48 of the Third Set of Interrogatories and Requests for
Production of Documents propounded by the Environmental Respondents received on August 2,
2016 has been prepared under my supervision.
Karim §iamer
Lead Economist,
Load Research. and Forecast
Dominion Resources Services, Inco
Question No. 48
Reference response to ER 2-24:
a) Provide the simulated weather normalized energy for all historical years for which
it is available.
b) Has the Company ever estimated a weather -normalized peak load for any prior
year? If so, provide the Company's most recent estimates of past weather -
normalized peals loads.
Response:
(a) -(b) The weather normalized energy referred to in the Company's response to Question No. 24
of the Environmental Respondents Second Set was for forecast periods only. The Company does
not weather normalize peak loads.
DOM-2016VAI RP -000257

Att. JFW-4

Att. JFW-4
Attachment ER Set 3-46(a
Page 1 of 3:
QUANTA
TECHNOLOGY
Dominion Northern Virginia Load Forecast
Dominion Virginia Power
10/17/13
Prepared for: Dominion Virginia Power
Prepared by: Quanta Technology, LLC
Contact: Lee Willis, P.E.
lwilliskquanta-technology
Srijib Mukherjee, Ph.D., P.E.
smukherjee °,quanta -technology com
***** Dominion Load Forecast**** Confidential/Proprietary
33
Page 0 of

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 2 of 3:
TECHNOLOGY
Table of Contents
1 - Executive Summary and Introduction 2
2 - Data Centers and Data Center Growth in Northern Virginia 4
3 - Conclusions and Recommendations 21
4 — Findings, Conclusions, and Recommendations 29
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page I of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 3 of 3:
• •
-ATECHN
1 - Executive Summary and Introduction
This report describes a project to analyze and forecast the impact of the data center industry on
the electric peak load growth in the Northern Virginia area of Dominion Virginia Power's service
territory in the period 2016 — 2023, and into the possibility that load forecasts being used for
transmission planning for that period underestimate the impact that that industry's continued
growth will have on peak load growth. This work was performed by Quanta Technology under
contract to Dominion Virginia Power. The work was performed by a project team with extensive
experience in electric utility load forecasting including data center analysis and forecasting for
FPL, SCE, PGE, and Duke Power as well as many other domestic and some foreign utilities and
governments.
Data center peak load in Northern Virginia has grown from next to nothing to 351 MW (2012) in
only eight years. New data centers currently under construction, and those announced for
construction over the next two years by operators with credible track records, leave little doubt
the rapid pace of growth will continue through 2015. The focus of this report is therefore on the
time period beginning in 2016.
Data centers have extremely high load densities — an order of magnitude higher than other
industries in comparable -size buildings. This means they are of considerable importance to
electric utilities and the planning of utility power systems. Data centers also have a number of
other characteristics — an extremely low number of employees per site, relatively little interaction
with local and surrounding businesses and services, and a product shipped in and out
electronically — that make them almost invisible in the county -by -county or state -by -state
econometric/demographic trend datasets provided by services like Moody's Econometrics,
Woods and Poole and some state governments. These datasets provide the foundations for the
load forecasting models most widely used in the power industry including the PJM methodology
that provides the basis for the transmission planning forecasts used by Dominion Virginia Power.
For this reason some utilities that have seen substantial data center growth in recent years,
including Dominion Virginia Power, are concerned that the peak load forecasts they use for
system planning may not have correctly projected the expected continuation of aggressive data
center growth in their service territories.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 2 of 33

Att. JFW-4
Attachment ER Set 3-46(a)
QUANTA Page 4 of 33
TECHNOLOGY
The Quanta Technology project team analyzed the data center industry's recent growth in
Northern Virginia and forecast its most likely course of future growth in the 2016 - 2023 time
period. The work concluded that data center growth in Northern Virginia can be expected to
continue to be extremely strong for at least another decade, and that very likely the PJM -forecast
driven peak load projection used by Dominion Virginia Power's transmission planners
underestimates the expected levels of peak load for Northern Virginia by year 2023 by nearly 900
MW.
This report provides background information on the data center industry and data center loads and
load growth characteristics, describes the growth of that industry in Northern Virginia, and
explains the project team's forecast of continuing data center growth in that region (Section 2).
Section 3 then reports on examinations of the transmission planning forecast used by Dominion
Virginia Power, and a forecast methodology test, that led the forecast team to conclude that
slightly more than 60% of data center peak load growth expected by 2023 is not included in the
forecast currently being used. Finally, Section 4 gives the project team's overall conclusions and
recommendations, including year -by -year peak load estimates of that missing margin of data
center growth that can be added to forecasts used for transmission planning or other similar
purposes by Dominion Virginia Power and PJM to correct for this under -forecasting.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 3 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 5 of 3:
TECHNOLOGY
2 - Data Centers and Data Center Growth in Northern Virginia
Data centers, also called data warehouses or server farms or warehouses, are industrial -scale
facilities built to serve the internet, cloud computing, and business and institutional sectors of the
economy as storage and transportation hubs for electronic data and data transfer. In only the last
dozen years, this industry has gone from next to nothing as regards the number of large, purpose-
built industrial -scale facilities to a major industry that, among other things, represents a very
noticeable portion of the total electric peak demand of some utilities in the US.
2.1 Development and Growth of the Data Center Industry
"Data centers" in some form have existed as long as there has been a computing industry and
certainly as long as there has been an internet. However, prior to the beginning of this century,
most "data centers" were small closet-, room- or multi -room server facilities in office,
institutional, or government buildings that had been fitted out to accommodate the needs of and
owned and operated by that business or institution.
Beginning roughly at the beginning of this century, industrial -scale data centers — entire buildings
devoted to nothing but this function — began to be built at sites throughout the US and the rest of
the world to serve the growing public and government demand for internet services, cloud, and
portable computing. Many of these large, industrial -scale data centers were owned and operated
by very large commercial or industrial companies or government owners for their own use. Many
others were "co -location" centers which rented data center capacity in either a wholesale or retail
manner to users who wanted it. It is these large, industrial -scale data centers that are the focus of
the work covered by this report. Throughout the rest of this report, the term data center will refer
only to an entire building dedicated solely to data -center service, whether serving exclusively just
the needs of the government or the company that owns it, or providing co -location services.
Figure 1 shows the growth of data centers in the US developed by the project team from industry
data. The nearly three -to -one discrepancy in total number among sources is because some
industry analysts refer to an address with multiple data -center buildings as a single data center,
while others count each building as a separate data center. 1
1 For example, the data center industry watch group datacentermap.com counts Latisys-Ashburn's multiple
buildings at addresses from 20147 to 21635 Red Run Drive in Ashburn, VA, as a single data center.
Dominion Virginia Power records indicate four separate data -center buildings with four separately -tracked
services provided at just the 21635 address alone.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 4 of 33

Att. JFW-4
Attachment ER Set 3-46(a)
QUANTA Page 6 of 33
TECHNOLOGY
8000
N
i
N 7000
v aJ
U *'
t6
2 6000
m �
o �
N 5000
in M
� 4000
r
N C
7 '
r-
0 3000
O i
CJ CL 2000
-0 p
1000
Z
2001 2005 2010 2012
Year
Cisco Datacentermap.com Quanta Technology
Figure 1: Count of industrial -scale data centers operating in the US as developed from three
sources. Datacentermap.com counts all buildings at an address or adjacent addresses as a single
"center." Data developed from Cisco data as well as Quanta Technology's study counts
individual service locations. All three show an exponential trend of growth with 3X growth over
the period 2007 — 2012.
Data Centers. From the outside industrial -scale data centers data centers most often look like
warehouses or large distribution centers — low, wide and long windowless buildings built with
efficiency and low cost in mind and with little regard for esthetics.2 Inside they are filled with
tightly packed floor -to -ceiling racks of servers, memory and digital switching units — equipment
needed to fulfill their mission. They have extraordinarily large chiller (air conditioning) systems
compared to other buildings of comparable size, because their dense electronics create truly
prodigious amounts of heat that must be removed lest that equipment overheat to the point that it
2 When necessary data centers can and have been built as multi -story buildings, as a very few have been in
southern Florida, Europe, and South America. However, the economics of data center design, particularly
the accommodation of the most recent modular server -rack designs and the arrangement of cooling systems
— a significant factor in their design - favor a low building with a large roof area. The industry is usually
successful at finding low-cost industrial park sites that permit it to build single -story data centers.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 5 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 7 of 3:
TECHNOLOGY
does not function well, or is ruined. Chiller and ancillary systems account for over half the load
at many facilities.
Very High and Very High Density Loads. As a result of both the high density of electronic
equipment and the large chiller units needed, data centers have among the highest load densities
of any type of industry. Load densities can be as high as 100 watts per square foot of building
space, or roughly ten times the maximums seen in many other industries. Furthermore, data
centers do not require extensive parking lots or set -backs like most other commercial and
industrial buildings, so the relative load density on a per -address basis compared to other types of
load is higher still. A modestly -sized data center located in a site that would meet the needs of a
typical suburban grocery store might have a peak demand up to 40 times that of the grocery store.
The largest data centers, those the size of a Big -Box retail chain distribution center, may have a
peak demand exceeding 50 MW.
Very High Load Factor. Like many other heavy industrial activities, data centers run at nearly
full capacity almost every hour of every day. As a result their electric demand varies little from
hour to hour or day to day. Load factor - the ratio of average hourly use to peak hourly usage - is
roughly twice that of many other loads: roughly 97% as opposed to around 54% — 40% for many
other types of commercial and industrial electric load.
Geographic Clustering and Modular -Unit Growth Characteristics. Although the data center
industry has in the most part been able to fmd relatively inexpensive land on which to build the
vast majority of centers built to date, it does have very specific needs as to location. For reasons
analogous to why power generation owners want to locate their plants near major transmission
lines, data centers owners want to locate their facilities near major data-com transmission
corridors, where buried fiber, etc., provides the highest levels of bandwidth and internet access.
Proximity to Data Center Demand. Although the internet is everywhere and user demand can
to a great extent be accommodated by a server regardless of location, there are definite technical
and business reasons why data center owners want to locate data centers relatively near areas of
high demand for their service. Part of the reason is that doing so reduces data-com tolling
charges and potential problems in much the same way proximity of generation to major load
centers reduces the likelihood of high transmission fees and congestion problems. In addition
some large commercial and institutional users have technical issues related to the timing and
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 6 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 8 of 3:
TECHNOLOGY
synchronization of their computing needs that constrain where they need to add capacity to best
serve their needs.
Secure or confidential Sites. In Quanta Technology's experience, and as determined by an
examination of data center industry website databases, roughly 20% of data center owners
consider their data center and its operation confidential to the extent that they limit the
information they disclose about the site, its purpose, and characteristics. Some are not even listed
in data -center industry inventories of sites, and for many others various industry websites will list
the location only by county, not town, street, or street number. Utilities providing service are
usually provided the peak demand and service date that is expected for first use, but are often not
informed about operating schedules that might vary demand, or expansion plans down the road.
Data Center Owned and Operated Generation. A large portion of data centers nationwide — a
substantial majority — have their own generation and fuel storage on site. This generation is
capable of sustaining the data center's operation through major utility outages of even a day or
more, although perhaps at some diminished output. Owners often operate this generation when
there is no utility outage, usually infrequently, but for reasons and according to schedules that fit
their business and operating needs, and for which they are not forthcoming with communications.
These schedules do not necessarily correspond with utility peak load periods of peak pricing
periods and can appear "random" or at least unpredictable to the utility serving the load.
Weather Sensitivity of Data Center Loads. Like some other industrial loads, data centers
normally show only the slightest amount of weather sensitivity in their demand for power. The
electronic equipment's demand does not vary as a function of temperature, and the vast majority
of their cooling needs, even in the peak of summer's high-temperature periods, are due to the heat
created by the data center's electronic equipment, not ambient temperatures. Therefore high
temperatures make less impact on heat -removal needs as they vary. During winter, data center
loads vary insignificantly as a function of temperature: the cooling systems job is to get heat out
of the building: colder outside temperatures help the chiller system to be a bit more efficient but
load hardly varies regardless. Given other factors, in most cases Quanta Technology does not try
to adjust data center loads for temperature and does not recommend this be done. Given this and
the unpredictability of schedules and self -generation operation, the peak demand observed at a
site regardless of time (e.g., on -peak, off peak) should be used as the expected coincident peak
load for planning purposes.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 7 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 9 of 3:
TECHNOLOGY
Growth on Initial Load. Individual data centers typically takes about two years to scope, plan,
design, build and commission. They initially come on line at less than their full design load,
being brought on line when only a portion of their equipment is in place and working. Over a
period of months thereafter — typically 18 to 30 — they will work up to full load, after which no
further growth at that building will occur. Figure 2 shows the typical "maturation curve" of peak
demand for new data centers.
c
E
v
0
CU
a
Year
Figure 2: The amount of load a utility can expect as a function of the number of years after the
data center owner announces it will build a 1 MW data center. This curve was developed from
Dominion Virginia Power information on 41 data centers in operation four or more years in its
service territory as well as about fifty other data centers in other areas of the country Quanta
Technology has worked on in projects for other utilities. Expected load two years after
announced construction is almost exactly the 1 MW the average developer promises for that time,
but demand continues to rise to roughly twice that amount as the center "matures" over the
subsequent three years.
Continued growth of data center capacity, when needed by the owners, is nearly always
accommodated through addition of additional data center buildings at an address or at sites
nearby, and generally not by expanding the capacity of an existing building. In this manner, too,
the data center industry is analogous to the power industry. Just as a utility may increase the
capacity of a substation over its lifetime by adding one, two, three, and eventually four power
transformers to that site, a "data center" may initially be built with a single building at an address
and additional buildings added as business dictates their need. This raises the aforementioned
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 8 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 10 of 3:
TECHNOLOGY
confusing nomenclature issue that must be worked through when dealing with articles and
historical accounts of the industry's growth: what does the term "data center" mean? Is it a single
building or a facility with multiple buildings at one site owned by one business? Usage varies
within the data center industry.
Fewer but Larger Future Data Centers. The data center industry reports that the number of
new data centers being built is decreasing slightly as new centers are designed and built larger —
roughly twice as large in physical size, in capability, and in electric load per site as only five to
ten years ago. Thus, counts of the number of new data centers being built is not growing as fast as
the capacity added (or electric load added).
2.2 Historical Data Center Industry Growth in Northern Virginia
Over the past decade, Dominion Virginia Power has seen an extremely high rate of growth of the
data center industry in Northern Virginia. The industry has gone from nothing to 130 data center
buildings at 80 separate addresses comprising 351 MW of peak load (Figure 3 and Table 4).
Nationally as well as globally, this high rate of growth has been driven by a continuously growing
business and government demand for internet, cloud computing and data services, as well as by
the public's increasing appetite for data-com and internet services.
Data center growth in Northern Virginia has been quite high for several reasons. First, the
Washington DC area creates much more than typical local demands for data center services. The
US Federal government is arguably the largest single consumer of data and data-com services in
the world, and its needs are growing. Many businesses aligned either directly or indirectly with
the US government, such as Lockheed Martin, have located major plants and business hubs in
Northern Virginia. They, too, have substantially above-average data and data-com needs.
Second, major data-com transportation trunks pass through parts of Fairfax and Loudoun counties
in Northern Virginia, an area nicknamed the Virginia "Fiber Alley" by the data-com industry.
As mentioned earlier, data centers prefer to locate adjacent to major data transportation corridors.
Land here is plentiful, easy to build on, relatively inexpensive, and close to one of the major data-
com corridors in the US. For this reason most of the data center growth in and around the
Washington DC metroplex has occurred there. Overall, data center activity in Northern Virginia
has qualitatively mirrored the growth of the data center industry while quantitatively slightly
outpacing it (3.1 times rather than right at 3.04 times for the industry as a whole, from 2007 to
2012 (Figure 3).
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 9 of 33

Att. JFW-4
Attachment ER Set 3-46(7)
QUANTA Page 11 of 33
TECHNOLOGY
1.00
0
4
0
0 n
N
2r -'
.E N .75
o
c v
L N
v
Y E .50-
0 L
Z O
Z
C �
ro �
C
.25
� L
3
}J
U
0
2001 2005 2010 2012
Year
Cisco Datacentermap.com Quanta Technology Dominion
Figure 3: Twelve year trends of data -center industry growth given in Figure 1, along with
Dominion Virginia Power's recorded peak load for the data centers in Northern Virginia, all
normalized to 2012 peak = 1.00. Dominion Virginia Power records show data centers identified
as a specific load sub -category only back to 2005, but a few probably existed prior to that time
and have been estimated by the project team to complete the load history shown here.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 10 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 12 of 3:
TECHNOLOGY
2.3 Will Data Center Growth Continue to Be Strong in Northern Virginia?
The historical trend of data center growth in northern Virginia is quite strong, but it is also a trend
of short duration, increasing the risk that extrapolation of recent trends alone could lead to an
inaccurate prediction of future load growth. The project team took this (extrapolation) along with
another approach it considers to be effective to create different scenarios of expected future data
center peak load growth in Northern Virginia.
Growth Over the Next Few Years Is Being Built Now
Dominion Virginia Power provided the project team with a table of information on all data
centers it services, or has been contacted about service for, from year 2005 to the present. It lists
138 separate data center buildings at about 88 separate addresses or locations, 131 and 80
respectively being in Northern Virginia's "Fiber Alley." Twenty-five of those had no load in
2012 or projected through at least the first part of 2013: those are in the process of being built or
are planned: their owners have provided that information to Dominion Virginia Power in order to
arrange service. Another 72 are on-line but have not completed their expected maturation cycle
(Figure 2) and have a load that can be expected to grow. In aggregate all current and announced
data centers represent an expected 673 MW — roughly twice the 2012 peak demand level, the vast
bulk of which growth can expected to develop and mature in the period 2013 — 2015. For these
reasons, short term growth is almost certain to be quite strong and the project therefore focused
on studying the potential for growth beyond 2015.
2.4.2 Ten -Year Forecast Based on Extrapolation of Data Center Peak Load History
It is highly likely that as time passes other data centers will be announced and this overall process
of data center growth will continue into the future: the discussion the data center industry given in
Sections 2.1 — 2.3 provides no indication that growth nationally or in Northern Virginia will cease
any time soon. The blue line in Figure 4 shows an extrapolation of the historical peak load on
data centers in Northern Virginia. To produce this extrapolation, the project team fitted a second
order polynomial to the historical peak load data for 2005 — 2012 and extrapolated it through
2023.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 11 of 33

Att. JFW-4
Attachment ER Set 3-46(7
QUANTA Page 13 of 33
TECHNOLOGY
2500
2000
1500
E Extrapolation
a)
in Data history
c
1000
c
Q
Y
f6
aa' 500
0 - - - - -
2005 2012 2023
Year
Figure 4: Extrapolation of data center growth in Northern Virginia through 2023 based on
extrapolation using a quadratic polynomial fitted to the 2005 — 2012 peak load history.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 12 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 14 of 3:
TECHNOLOGY
2.4.3 Forecasts Based on the Data Center
Industry's Forecasts of Its Own Growth
The project team's other, and preferred, way to project peak load, particularly for industry -
specific load categories such as data centers, is to tie that growth to a credible, previously forecast
economic or demographic variable that is a direct or driving factor in causing that electric load
growth. The forecast given here is based on using the data center industry's own projections of its
future growth to help forecast its growth in Northern Virginia. As was shown in Figure 3,
throughout the history of this industry in Northern Virginia its growth has followed the industry
trend, if slightly outpacing it. The forecast developed here assumes that will continue into the
foreseeable future, and that the data center industry is capable of foreseeing its own future well.
The data center industry spends considerable effort monitoring and studying its market in order to
project future demand for its services, and in planning how it will accommodate that growth.
When considered in its largest interpretation — of not just companies that own and operate data
centers but as also including companies that supply or use the services, it includes a number of
the largest companies in the world, many with excellent track records of running their business in
growing industries well over long periods of time, such as Microsoft and Cisco. These
companies know and understand that industry and produce and share their forecasts of expected
growth with the public for a variety of mostly selfish reasons.3 The forecasts seem credible both
because of their track record and because many are using these forecasts as guidelines for their
investment of billions of dollars in new and future data centers.
The project team had a good familiarity with data centers, their load characteristics, and growth
of that load category from prior data center projects going back to 2003, but devoted a good deal
of research into updating its knowledge of current industry practices, particularly of data center
activity in Northern Virginia and of that industry's expectations for its own growth over the next
decade. Tables 2 and 3 gives the sources of information on data center industry growth trends
nationally and in Northern Virginia the project team gathered. Most of these projections and
3 Publically-held companies publish their forecasts at conference and on their websites to inform potential
investors of their continued viability. Without doubt a great deal of thought goes into their decisions about
what information they share and if and how they "spin" it to the public. However, there are regulations
about misleading investors, and the companies' reputations as well as the credibility of future statements
they make are at stake, so they have some incentive to produce the best forecasts they can for public
consumption. Beyond that, many projections available are by "neutral parties" (or as close as one can get) —
non -owner, industry observers such as datacentermap.com.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 13 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 15 of 3:
TECHNOLOGY
Table 1: Information Sources Behind Data Center Growth Scenarios For This Forecast.
Source
Descripton
Obtained from ...
Dominion Data
Extrapolation of data center load growth trend
from Dominion's historical data, 2005 - 2012.
Dominion Virginia Power
"PROJECTING ANNUAL NEW DATACENTER
http://www.datacenterdynamics.com
Microsoft
CONSTRUCTION MARKETSIZE"- industraytrade
/focus/archive/2011/03/sizing-data-
show presentation by Microsoft
center -construction -market
http://newsroom.cisco.com/press-
release-co ntent?articleId=888280
Cisco
Cisco's VNI Forecast Projects the Internet Will Be
Four Times as Large in Four Years
http://www.theatlantic.com/technoI
ogy/archive/2012/05/the-future-
The Atlantic
The Future Growth of the Internet in One Chart
and One Table
growth-of-the-internet-in-one-chart-
and-one-graph/2578111
Kliener Perkins
Caulfied Byer
Internet Trends
http://www.slideshare.net/kleinerpe
rkins/kpcb-internet-trends-2013
Digital Trends
http://www.digitaltrends.com/web/r
eport-web-traffic-to-quadruple-by-
(Industry Association
Web Traffic to Quadruple by 2016
2016
Group)
DataCenter
Various White papers and articles particularly their
http://www.datacenterknowledge.co
Knowledge.com
Industry Perspectives page
m/
DataCenter
dynamics.com
Various articles and white papers on their site in
addition to the Microsoft presentations and others
from sponsored conferences.
http://www.datacenterdynamics.com
Datacentermap.com
Statistics, maps, and reference material on the
location and types of data centers built throughout
http://www.datacentermap.com/
the US and many other countries.
Datacenter
A source much like that listed above, and used to
mapping.com
confirm and ugment that site's data, and vice
http://www.datacentermapping.com/
versa.
Yet another data center map/inventory site and a
http://www.datacent
er9.com/
long-time favorite of Quanta's project team.
Slightly better organized maps and serach engines
but not as up to date an inventory list of as some
http://www.datacenter9.com/
other sites.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 14 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 16 of 3:
TECHNOLOGY
Table 2: Sources of information obtained in interviews or discussions
Person/Company
Information Provided
Northern Virginia Electric services areas that intermingle with Dominion's service in
Northern Virginia. Mr. Bisson confirmed that data center growth has been aggresive over
the last five to seven years and is expected to continue and that many data center owners
Bob Bisson, Vice
are secretive about plans. NOVEC's experience is that anew data center grows its load
President, Northern
steadily over the first two years it is installed but then grows no further. Growth over the
Virginia Electric Co.
long run occurs due to increases in number rather than continued growth of existing data
centers (this corrsponds to Dominion's experience). Weather sensitivity in the noticeable
if not nearly as significant for other loads in summer but not in the winter.
Buddy Rizer, Economic Development Director Loudoun County: Loudoun County's
population grew almost 84% from 2000to 2010. It is expected to grow another 10to 20%
Buddy Rizer,
by 2014. In the past decade Loudoun County has created more than 46,000jobs. Loudoun
Economic
county has been the "value" alternative to Fairfax county. More than 50% of the world's
Development
Internet traffic passes through Loudoun County. This county has seen data center growth
Director Loudoun
to be more than 180% since 2000. Data centers in the county typically occupy about 40
County
buildings and more than 4.3 million square feet of land. Such growth is considered strong
and very likely to continue into the foreseeable future.
The forthcoming slowdown the industry speaks of has to be kept in context - the growth
the last five to seven years has been close to unmanageable. Planning and justification
by owners is done on a three-year cycle. Equipment technoogy changes but "obsolete
Norm Cann, Data
technology" (servers and systems not of the very latest chip design, etc., will be left in
Center Manager,
place at a working center for at least seven years before renewal, maybe longer: changing
Verizon
out the servers, etc., for newer may not be worth the trouble because the higher density
would require more and redesigned cooling. Best is to leave the site as was built until it
no longer is a viable business contributor, then shut it down or completely rebuild. This
would be more than ten and perhaps up to twenty years.
Data centers are justified by owners on the basis of 18 -month payback - they will not
build if break-even is not around that timeframe. "Long-term" justification is based on
Sarah Corbrett, Data
nothing past seven years (a period linked to business tax deductions and credits) and
Center Architect,
requires that the incremental profit projected after five to seven years have at least a 30%
Norbson Smith and
net margin. An increasing number of sites request design with on-site generation
Co.
intended to more than just backup power- the owner intending to use that generation
during some periods to reduce energy cost where permitted and advantageous.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 15 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 17 of 3:
TECHNOLOGY
studies date from analysis gathered and analyzed from late 2011 to late 2012. These data sources
lead to the different scenarios of data center growth in Northern Virginia to be discussed below.
Salient characteristics of data center load and load growth developed from this information are:
• Growth rate of completed data centers dropped temporarily during the recession from
2008 to 2012 due to economic slowdowns in data-com markets and plans by internet
providers. A return to economic growth means a return to higher growth through
2014 as these centers are completed as well as new centers announced.
• There is plenty of available, suitable land to continue to build data centers on in
Northern Virginia's Loudoun and Fairfax counties, and those county governments are
disposed to encourage that growth (and the resulting tax revenue). Land prices are
rising slowly but are reported by the country governments to be "very competitive"
with other areas in and around Washington DC.
• Data center load growth, not just in Northern Virginia but in all utility service areas
Quanta Technology has seen, is mostly due to the increase in number of centers, not
their individual loads.
o As noted earlier, after first going into operation, a new data center facility
will usually take about 18 to 30 months to "mature" - to come up to its full,
final peak load level. After that the site will operate at roughly at that peak
load level and its load will not grow. This industry -wide characteristic
mirrors the experience the project team observed in both Northern Virginia
and its prior projects (Figure 3).
o As was discussed in Sections 2.1 and 2.3, newer data centers are on
average designed and built to be roughly twice as large in physical size,
in capability, and in electric load those built earlier in the industry's
history. Counts of the number of new data centers being built are not
growing nearly as fast as the capacity added.
• The economics of facility design and operation dictate that there is both a strong
business incentive for the company to locate near data-com trunk routes, such as that
running through Northern Virginia, and to cluster more than one site near one
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 16 of 33

Att. JFW-4
Attachment ER Set 3-46(a)
QUANTA Page 18 of 33
TECHNOLOGY
another. This was known but was a point repeated in every interview and every
industry overview the project team had.
• With no notable exceptions, all published projections of data center industry activity
forecast a healthy growth of demand for their service, although there are considerable
differences in the details and projected long-term growth rates. However, almost all
studies and published expectations the project team could find agreed on several key
points:
o Industry growth in the US will slow slightly as compared to the last three
to five years even if it continues at record rates overseas: internet and
data center development in the US is about five years ahead of the rest of
the world. .
o This anticipated "slowdown" must be kept in perspective: while opinions
as to the degree of slowdown differ, they average to a simple rule of
thumb: whereas data center numbers and load in North America tripled
in the six year period 2006-2012 they will probably double or grow
slightly more than that over next six: industry rate of growth will drop by
10% to 40% depending on forecast.
o This excepted slowing in data center industry construction is not
attributed primarily to an expected slowdown of growth of demand for
internet and data -center related services (which is still very strong). It is
mostly due to measureable increases in data center efficiency: the
industry has had nearly a decade to work, adapt, and put into place
lessons learned earlier in its history, and as scheduling and operational
software becomes more efficient. Simply put, the average new data
center can provide more service than it could five years ago.
o Although not obvious in the time period shown in the diagrams, where
they talk about long-term prospects all scenarios acknowledge the
inevitable stabilization and cessation of high growth, as would
correspond to the classic "S" curve characteristic of developing
industries: growth rate gradually falls until it reaches near -zero at some
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 17 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 19 of 3:
TECHNOLOGY
future time perhaps years or even decades in the future.4 These different
sources and projections all see this happening, and if not explicitly
discussing that possibility, are all consistent with a view that it would
occur far into the future and not in the next ten years. They differ as to
how fast they see that tapering of load rate occurring and therefore the
ultimate level of build out the industry reaches.
Four scenarios of possible data center industry growth were developed to represent the range of
the data center industry's views of its future: that the industry's growth would not slow, or that it
would slow by 15%, by 30%, or 45%. These trends in growth rate reduction - none, 15%, 30%,
and 45% - were then applied to the extrapolation of Northern Virginia data center growth
developed above (the blue line in Figure 4) by simply reducing its growth rates by those amounts
each year. The "no reduction" scenario yielded the same curve as the extrapolation did , the three
additional forecasts of Northern Virginia data center growth correspond to three perspectives of
the data center industry's view of its future trends, imposed on the slightly higher than industry -
average growth in Northern Virginia as seen over the last eight years.
All four scenario trends include a "recession bounce -back" effect that has been accounted for in
the extrapolation. As mentioned earlier the recession caused delays in completion of some
already -started data centers (growth in 2008 — 2012 was about 10-15 MW lower each year than
originally announced). A return to healthier economic growth in 2012 and beyond means
4 This S-curve is and has been a fixture in forecasting of any new technology or industry for decades, and is
the overall qualitative trend in almost all emerging industries, although in cases where the industry has a
broad market, as for the internet and data centers, the S-curve may be decades long (the railroad industry is
one of the longest on record since the dawn of the industrial era, taking about 15 decades). See
Technological Forecasting by Joseph P. Martino, McGraw Hill, third edition 1992.
The S-curve is also an almost universal fixture in observed and forecast load growth for utilities, although
again the full S curve shape may take decades to develop for particularly large areas or industries (see
Spatial Electric Load Forecasting — 2nd Edition, by H. Lee Willis, Marcel Dekker, 2002, particularly
Chapter 7).
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 18 of 33

Att. JFW-4
Attachment ER Set 3-46(7)
QUANTA Page 20 of 33
TECHNOLOGY
Table 3: Four Scenarios of Data Center Growth
Scenario Description
1 High: continuation of historical trend (blue line from Figure 4)
2 Slight reduction corresponding to 15% drop in industry growth
3 Mean industry expectation — 30% drop from historical trends
4 Significant drop — a 45% reduction in historical trends
Table 4: Historical and Forecast Data Center Loads in Northern Virginia.
With Just Forecast Scenario
2005 86
2006 115
2007 145
2008 180
2009 206
2010 271
2011 316
2012 351 351 351 351 351 351
2013 474 492 470 449 428
2014 528 582 544 507 472
2015 575 682 623 568 516
2016 790 707 631 561
2017 908 797 697 607
2018 1034 891 765 654
2019 1170 991 835 701
2020 1314 1095 907 748
2021 1468 1203 982 796
2022 1630 1317 1058 845
2023 1802 1434 1136 894
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 19 of 33
Year
History
Completion
1
2
3
4

2500
2000
1500
ra
M
E
0 1000
QUANTA
TECHNOLOGY
L11
1❑
2005 2012 2023
Year
Att. JFW-4
Attachment ER Set 3-46(a)
Page 21 of 33
Scenarios
1
2
3
4
Figure 5: Total Northern Virginia forecast data center peak load for the four scenarios of data
center industry growth.
completion of these deferred data centers has accelerated even as more are announced. This is
included in Dominion's data for the period through 2015; however, the extrapolation (blue line
show in Figure 4) is based on the long-term trend and deliberately did not include the effects of
this recent recession -end effect in its model fitting.
These four projected peak load scenarios are listed in Table 4 and plotted in Figure 5. All four
trends start with the same conditions and peak data center load in 2012. Differences in the
amount of new data center load are small in the first few forecast years but grow significantly
over time.
These scenarios, their interpretation and use will be discussed further in Section 4: Findings,
Conclusions, and Recommendations.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 20 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 22 of 3:
TECHNOLOGY
3 Is Data Center Growth Adequately Included
in Load Forecasts Being Used By Dominion?
3.1 Examination of the Load Forecast Used for Transmission Planning: 2016 - 2023
Table 5 shows the historical and forecast peak loads for Northern Virginia that Dominion
Virginia Power currently uses in the planning of its transmission system. These figures are based
on the PJM load forecasts for three sub -regions of the Dominion Virginia Power service territory:
Alexandria -Arlington, Fairfax, and Woodbridge. Together these constitute the loads typically
used for "Northern Virginia" planning. Over the 24 -year period 1988 to 2012, the region's peak
electric load grew by an average of 2.9% per year, and over the preceding decade (2002-2012) at
3%. It is projected to grow at 2.5% through 2023, the roughly .4% - .5% reduction from long -
historical averages being typical of what the project team has seen throughout most of the nation
due to energy efficiency and changes in demographics and styles of living and business activity
(These will be discussed further is section 3.2). This rate of growth is higher for Virginia and the
nation as a whole and reflects a robust local economy and proximity to Washington DC.
Table 6 shows the projected year-to-year increases in peak load in this Northern Virginia
Transmission Planning region for the years 2016 and 2023 — the amount of load growth the
transmission planning forecast expects in Northern Virginia in those years. Also shown are
similar year -2016 and year -2023 increases for data center peak demand from Scenario 2, and the
percent of total Northern Virginia load growth that represents.
If the data center forecast values (Scenario 2, Table 4) are included in the transmission planning
forecast values (Table 5) then data center growth constitutes 71% of the region's total peak load
growth in year 2023. That means that all the other load in Northern Virginia is growing only
about a third of its historical long-term average growth rate. This seems unrealistic, given that
projections for growth of population, business activity, and net productivity for the region as done
by Moody's Econometrics and similar firms and projected tax revenues and community needs as
projected by the governments of Loudoun, Fairfax, and Arlington Counties as well as the
independent City of Alexandria, all show no decrease from historical levels of growth. For
example, econometric -demographic data provided by Dominion Virginia Power and gathered by
the project team indicate that during the 2016 - 2023 period population and expected county and
city tax revenues in Northern Virginia are both expected to increase by about 2% per year. These
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 21 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 23 of 3:
TECHNOLOGY
growth rates are broadly similar to the growth in those numbers seen in the past and lead to the
conclusion that non -data center growth will continue to be very similar to past rates through 2023.
This discussion will return to the comparison of load forecast magnitudes in Tables 4, 5, and 6
later in this section, after looking at the way econometric factors like those projected by Moody's
and country governments effect the growth of load and are used in electric utility load forecasts.
Table 5: Peak loads Used for Transmission Planning of the Northern
Virginia area in the Dominion Virginia Power Transmission System
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 22 of 33
YEAR
S
WE
TO TAL
AREAA10 S_AREA11
S AREA12
1988
1,349
1,540
918
3,807
1989
1,361
1,560
942
3,863
1990
1,297
1,472
915
3,685
1991
1,429
1,632
981
4,041
1992
1,439
1,643
993
4,075
1993
1,408
1,608
986
4,001
1994
1,511
1,726
1,054
4,291
1995
1,519
1,741
1,062
4,321
1996
1,321
1,727
891
3,939
1997
1,528
2,026
1,008
4,562
1998
1,554
2,028
1,037
4,618
1999
1,629
2,218
1,174
5,022
2000
1,524
2,078
1,086
4,688
2001
1,635
2,396
1,212
5,244
2002
1,714
2,464
1,221
5,399
2003
1,624
2,459
1,241
5,323
2004
1,547
2,379
1,217
5,143
2005
1,750
2,771
1,546
6,068
2006
1,842
2,922
1,603
6,367
2007
1,829
2,989
1,661
6,480
2008
1,787
3,015
1,596
6,398
2009
1,720
2,853
1,519
6,093
2010
1,788
3,057
1,663
6,508
2011
1,912
3,283
1,830
7,026
2012
1,835
3,237
1,857
6,928
2013
1,833
3,303
1,957
7,093
2014
1,842
3,419
2,029
7,290
2015
1,870
3,553
2,111
7,534
2016
1,892
3,671
2,184
7,747
2017
1,905
3,774
2,247
7,926
2018
1,913
3,868
2,304
8,085
2019
1,922
3,968
2,365
8,255
2020
1,932
4,071
2,427
8,430
2021
1,938
4,169
2,486
8,593
2022
1,942
4,265
2,544
8,752
2023
1,948
4,367
2,605
8,920
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 22 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 24 of 3:
TECHNOLOGY
Table 6: Growth (MW) in Peak Load in the Transmission Forecast,
Compared to that Projected Data Center Scenario 2: 2016 - 2023
Time Transmission Data Centers % of Tran. Forecast
Period Forecast - MW Forecast - MW Data Centers Represent
2015-2016
213
84
39%
2022-2023
169
118
70%
2016 —2023
1,173
726
62%
3.2 Commonly Used Load Forecast Methods Often
Fail to Pick Up on Data Center Growth Trends
Data centers have a number of characteristics that make them nearly "invisible" to the most
commonly -used forecasting methods in the electric utility industry. Those forecast methods
regress weather -corrected annual or monthly electric demand histories against a set of similarly
annual or monthly econometric and demographic trends such as population, employment counts,
gross product, total payroll or other similar statistical measures of population and economic
activity. These data trends are often provided on a country -by -county basis so that, when
necessary, study and forecasts of areas within a utility, such as for just Northern Virginia, can be
carried out, rather than only forecast at a utility -systems or state -by -state basis.
These load forecasting methods are well -proven and for the most part do a more than acceptable
job in tying electric load growth to economic factors that are then forecast: Companies like
Moody's Econometrics, Woods and Poole, and others, along with some state governments,
provide projections of those same econometric and demographic statistics, again on a year -by -
year or monthly for several decades into the future. The numerical model fitted to the utility's
load and econometric -demographic history is applied to these trends to determine estimates of
future electric demand. That raw forecast may be adjusted by some utilities or ISOs to account
for expected changes due to energy efficiency, DER programs, or other energy-related factors,
but those econometric -demographic driven trends form the basis for the forecast and determine its
overall character
Data centers are "invisible" to the type of forecasting method described above because they do
not register (show up) in these econometric -demographic statistics in any way even close to
proportional to their extremely high loads, as compared to other industries and businesses:
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 23 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 25 of 3:
TECHNOLOGY
o While data centers have very high loads, they create almost no jobs. A few data
centers are completely automatic, but most have the barest minimum staff. often just
one person on site in any one shift: four or five employees perhaps. By contrast a
building or land parcel of comparable size in more other industries would have 50 to
500 employees, or ten to one hundred times as many. Factor in the difference in load
density — typically at least another order of magnitude, and on an employees -per -MW
of load added basis, data centers typically register about three orders of magnitude
lower than other industries in a region. With respect to both employment count and
payroll totals, their contribution to a region is basically within the range considered
noise in the data and regression fitting of many models.
o Data centers do little or no business with nearby businesses, as is typical of the vast
majority of commerce and industry. A grocery or big box store, a distribution center,
a new office building or a manufacturing plant built on a comparable site would not
only create jobs at its location, but also buy supplies and services locally, further
adding in a very noticeable way to the local economy and thus employment and
business activity statistics as measured for that area. By contrast data centers do
almost no such local business beyond buying electricity from the local utility.
o Finally, any "value-added" data centers create - any profitability or gain they create
for their owners is shipped in and out over the internet without monitoring in the way
manufactured goods and retail and commercial services are tracked. The value may
show up on the corporate books of the company that owns the data center (or not) but
that is often in another region or state, not on the local area's tally of "gross product,"
etc.
Therefore, from the standpoint of the econometric variables used in many of the power industry's
electric utility load forecast methods such as additions to employment, increases directly and
indirectly in local business activities and payrolls, the total gross product on a country or state or
utility service area basis, and business -to -business activity, etc., data centers largely don't exist.
The effect of a significant data center load growth history, as has occurred in Northern Virginia
on an electric load forecast method using these econometric -demographic data types will vary
depending on the exact type of method, the amount of data center load, and the timing of that data
center trend. The project team's past experience with many such forecasting tools at many
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 24 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 26 of 3:
TECHNOLOGY
utilities leads it to believe that, given enough scale and enough time — many data centers
developing over a period of time that represents most if not all of the historical period over which
the regression fit and analysis of growth is done — many methods would "figure it out" and adjust
so that they would, in fact, forecast some if not most of the impact of data center growth on the
growth of electric load.
But this is not the case here. The typical historical period for forecast model fitting is 25 years.
Data centers in Northern Virginia have existed as an identifiable subcategory only since 2005,
and they have represented a significant load — more than 2% of the regions peak - only since
2007. In many if not most of the econometric forecast methods used in the industry, a recent data
center trend like this will cause a slight increase in fitting residuals over the final few years of the
forecast period, with only a slight impact on the forecasted trend which has been fit to all the load
over the much longer historical period
Test of A Typical Econometric -Demographic based Forecast Method
The project team has developed a econometric -demographic peak load forecast method that it has
used in load forecast projects and studies it has done over the past seven years, a tool much like
many others used for system peak load forecasting throughout the power industry, if having
advantages as to regional and spatial growth projection that make it particularly adaptable to
T&D load forecasting. This method has given good results for Quanta Technology and forms the
basis for forecast tools the project team has helped put in place at a number of US utilities,
including PG&E, Duke, Nashville Electric, KCPL and MG&E.
Quanta Technology's forecast model works with econometric -demographic historical and
forecast trends as provided by Moody's Econometrics, Woods and Poole, or the US Census, by
regressing peak load growth adjusted for weather and other factors against those variables, in
total or in customer classes as the case may be, and with leads and lags in time analyzed, and
using other modeling features common to such models.
This forecast model was run in two cases in order to determine how well such a model "sees"
data center load growth. First, it was calibrated on (fitted to) weather -corrected historical peak
load, population and economic data for Northern Virginia, with data center loads included. The
project team made an effort to calibrate their model in this case so that its forecast results closely
resembled the Dominion Virginia Power transmission -planning forecast, itself the sum of three
regions of the PJM forecast, by picking econometric -demographic variables that when fitted, gave
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 25 of 33

Att. JFW-4
Attachment ER Set 3-46(7
QUANTA Page 27 of 33
TECHNOLOGY
a nearly identical forecast trend. This was done because the project team's intention was not to
create an independent forecast as much as "reverse engineer" an econometric -demographic
forecast model that would most likely respond not just like most industry models would, but like
PJM's forecast tool would. This fitted model was then used to produce the "with" forecast shown
in Table 7 and Figure 6. Average mismatch with the PJM forecast is .19% with the worst
deviation being .80% in the final year. This forecast model was then re -fit (allowed to readjust its
parameter coefficients) to those same economic data trends but with peak load data in which the
data center peak load histories had been subtracted. This case produced the "Without" forecast
shown in Table 7 and Figure 6.
Table 7: Dominion's Ten -Year Transmission Planning
Forecast Compared to Forecasts Produced With and
without Inclusion of Data Centers in the Historical Load
Data
Dominion Quanta Technology
Year Tranm. Team's Forecast
Planning
Forecast W, Data Ctrs Without
2013 7,093 7,090 6,722
2014 7,290 7,253 6,868
2015 7,534 7,516 7,114
2016 7,747 7,728 7,309
2017 7,926 7,942 7,506
2018 8,085 8,099 7,646
2019 8,255 8,257 7,787
2020 8,430 8,415 7,928
2021 8,593 8,591 8,087
2022 8,752 8,773 8,252
2023 8,920 8,850 8,312
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 26 of 33

1o,D
• err
6,000
C
E
4,04,'3
W
0
� QQUANTA
TECHNOLOGY
2013 2018 2023
Year
Att. JFW-4
Attachment ER Set 3-46(a)
Page 28 of 33
Forecasts
Transrn. P ann"ng
QT - with
QT -without
Figure 6: Three forecasts of Northern Virginia area loads. Dominion's transmission planning
and the project team's "With" forecast are nearly identical. The "Without" forecast is lower by
between 350 and 540 MW.
Analysis and Interpretation of the Differences
• Dominion's Transmission Planning forecast projects an average compounded annual
growth rate over the 2013-2023 period of 2.32% annual peak load growth for a 2023
projected peak demand of 8,920 MW.
• With the data center loads included, the project team's forecast method projects an
average compounded annual growth rate over the 2013-2023 period of 2.24% annual
peak load growth for a 2023 projected peak demand of 8,850 MW.
• Without the data center loads included, the project team's forecast method projects an
average compounded annual growth rate over the 2013-2023 period of 2.14% annual
peak load growth for a 2023 projected peak demand of 8,312 MW.
Table 8 summarizes this test, which was basically a base -line comparison of existing forecast
methods to illustrate the response of the existing forecast to the historical data center trend for
Northern Virginia. The difference between the project team's "With" and "Without" forecasts is
the the response of the model to the data center peak load history. project team's interpretation of
that difference. The largest portion of the difference between the two forecasts is due to the fact
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 27 of 33

Att. JFW-4
Attachment ER Set 3-46(7
QUANTA Page 29 of 33
TECHNOLOGY
that the "With" forecast starts out with 351 MW of additional data center peak load in 2012,
which it carries through to 2023, growing it to 445 MW. In addition, the addition load — 351 MW
in 2012 but lesser amounts going back a number of years, means the model saw a higher recent
historical load growth rate leading up to the forecast period, which caused it to slightly ramp up
its projected overall growth rate (i.e., that it applies to all load) for 2013 to 2023, from an average
2.14% in the "Without" case to 2.24% in the "With" case, adding 93 MW of growth to the
"other," non -data center load. The total difference is 538 MW more load in the "With" forecast
for 2023 than the "Without."
The 538 MW of data center load in 2023 is about 350 MW lower than the lowest scenario's
projected data center load (Table 4) and over 1,200 MW less than the high scenario's projection
for 2023.
Table 8: Project Team's Assessment of the "Response" of An Econometric Model to Data
Center Load History
Quanta Technology
1
2013 7,090 6,722 368 356 12
2014 7,253 6,868 385 365 20
2015 7,516 7,114 402 378 24
2016 7,728 7,309 419 388 31
2017 7,942 7,506 436 399 37
2018 8,099 7,646 453 407 46
2019 8,257 7,787 470 415 55
2020 8,415 7,928 487 423 64
2021 8,591 8,087 504 432 72
2022 8,773 8,252 521 441 80
2023 8,850 8,312 538 445 93
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 28 of 33
Difference: Growth of Higher
Original
Year
Team's Forecast
With minus
Without
g
351MW Data
Center Load
Growth of
Non -Data -
Center Load
w. Data Ctrs
Without

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 30 of 3:
TECHNOLOGY
4: Findings, Conclusions and Recommendations
4.1 Data Center Industry Growth in Northern Virginia
is Very Likely to Be Extremely High Through 2023
All evidence suggests that data center growth in Northern Virginia will continue to be very robust
for at least the next decade. The four scenarios of data -center industry growth developed for
Northern Virginia in Section 2 differ over a two -to -one range as regards the expected amount of
2023 peak data -center load, but even the lowest shows data center peak load in 2023 at nearly
three times 2012 levels.
Scenario 1, the highest growth scenario, represents a continuation of trends seen over the past
decade. Any proposal not to use that scenario as a forecast should have sound reasons for
rejecting the historical precedent. One reason to do so is that the data center industry itself
expects a slight slowdown in growth. On the other hand, the factors that have caused Northern
Virginia's data center growth to slightly outpace that industry's growth nationwide appear to be
as strong, and relevant, as ever. Virginia's "Fiber Alley" is one of the major data-com corridors
in the nation. Land there suitable for data center construction is abundant and competitive in
price. Local governments support continued development on this industry in their areas. The
Washington DC metroplex is home to some one of the largest consumers of data center services
in the world so demand for locally -located data centers will continue to be very strong.
Scenarios 2, 3 and 4 represent the diverse range of opinion that members of the data center
industry have of their own industry's future and the slowdown in growth that is expected. Very
few in that industry expect no slowdown at all, just as none see a slowdown beyond a reduction in
growth rate of 50%. But within that range, opinions differ as to how much slowdown there may
be. One option would be to simply pick the mean of these four scenarios (a 22.5% reduction
scenario). However, the project team puts significant weight on Northern Virginia's historical
trend of slightly outpacing the industry in growth, and to the above-average advantages the area
has for data centers and continued development. It also considers the fact that the short-term
trend (that developed by Dominion from existing and announced construction) best matches the
15% scenario as an indicator that it may be more likely than the others. Thus, on balance, the
project team recommends Scenario 2, the 15% -reduction scenario, for planning purposes.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 29 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 31 of 3:
TECHNOLOGY
4.2 The Load Forecast Dominion Virginia Power Is Using for
Transmission Planning Includes Less Than Half of the
Most Likely Expected Data Center Load Growth
Section 3 provided an examination of the transmission planning forecast numbers themselves in
comparison to the expected data center growth (Section 3.1), which indicated Dominion's
transmission forecast does not include the data center trends discussed above. Simply put, the
trend of data center loads is just not adequately included in the forecasted totals. The reason is
that the commonly accepted econometric -demographic load forecast method used for load
forecast throughout the electric utility industry, similar to that used by PJM, is partially "blind"
when it comes to explaining and forecasting load growth driven by industries like data centers,
because data centers do not register in the most meaningful ways in the econometric data trends
upon which these forecasting methods work. Section 3.2 outlined how the project team tested
one such forecast model to determine how it would respond to a data center history such as has
occurred in Northern Virginia. That test indicated it would forecast slightly less than 38% of the
growth expected from a detailed assessment of the industry growth itself and not include about
62% of what should realistically be expected in its load forecasts. This test looked at only one
method, but the project team thinks that method is broadly indicative of the behavior that can be
expected from other methods of the same type, including that used by PJM.
Loads for All of Dominion Virginia Power, Not Just Northern Virginia, Are Affected, Too.
This focus of this project has been on the Northern Virginia area of the Dominion Virginia Power
service territory, where 131 of the 138 data centers the company is dealing with are located.
These other seven data centers are scattered throughout the other planning regions of Dominion
Virginia Power's service territory. Project resources did not permit a detailed assessment of other
operating regions' growth. However, there is no reason to think that the situation as regards
under -forecasting and continued strong growth in those regions would be materially different than
for Northern Virginia.
4.3 Forecasts for Transmission Planning Should Be Adjusted Upward
For the reasons outlined above, the project team recommends that Dominion Virginia Power
adjust the peak loads it uses for transmission planning in Northern Virginia, and for Virginia
overall, by adding the amounts shown in Table 9 to the values it uses for transmission planning in
the 2016 — 2023 timeframe. These amounts were determined by subtracting the amount of data
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 30 of 33

Att. JFW-4
Attachment ER Set 3-46(7
QUANTA Page 32 of 33
TECHNOLOGY
center load the tests in Section 3.2 determined would be included in an econometric forecast
(Table 8 column marked "Difference With minus Without") from the projected data center loads
for the four scenarios developed in Section 2 (Table 4). Again, Scenario 2 is recommended.
Adjustments shown in Table 9 for Virginia overall (lower part of table) include the adjustments
for Northern Virginia (upper part). For example, the 944 MW listed in scenario 2 for the state as
a whole for 2023 includes the 896 MW listed for Northern Virginia. Northern Virginia peak
loads for that year would be adjusted upward by 896 MW, while the 48 MW difference between
that and the 944 MW for the state as a whole should be assigned to other regions in the state. The
project team recommends Dominion Virginia Power do further study to determine into which
other planning regions that load should be added.
Table 9: Amounts that Should Be Added to the Dominion Virginia Power Transmission
Planning Forecast to Represent Each Scenario of Data Center Growth - MW
Amount of Additional Amount that Should Be Added to Northern Virginia Peak Loads
Year Data Center Scenario 1 Scenario 2 Mean Scenario 3 Scenario 4
Load Probably Extraolated Recommended- Medium Lowest
in Forecast trend 15% reduction 22% red 30% reduction 45% reduction
2016 419 371 288 253 212 142
2017 436 472 361 316 261 171
2018 453 581 438 383 312 201
2019 470 700 521 454 365 231
2020 487 827 608 529 420 261
2021 504 964 699 608 478 292
2022 521 1109 796 691 537 324
2023 538 1264 896 778 598 356
Amount of Additional Amount that Should Be Added to Dominion Virginia Power Overall
Data Center Scenario 1 Scenario 2 Mean Scenario 3 Scenario 4
Year Load Probably Extraolated Recommended- Medium Lowest
in Forecasto 22% red o 0
trend 15 /o reduction 30 /o reduction 45 /o reduction
2016 441 391 304 267 224 150
2017 459 497 380 333 275 180
2018 477 612 462 403 329 211
2019 495 737 548 478 385 243
2020 513 871 640 557 443 275
2021 531 1015 737 641 503 308
2022 549 1168 838 728 565 341
2023 567 1331 944 820 630 375
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 31 of 33

Att. JFW-4
Attachment ER Set 3-46(a
QUANTA Page 33 of 3:
TECHNOLOGY
Recommended Method for Producing Forecasts in the Future
Section 3.2 discussed why the commonly accepted method of forecasting load growth based on
standard econometric and demographic data trends will not fully track and forecast data center
growth. To avoid this situation in future forecasts, the project team recommends the following
revised method that still utilizes that standard widely -accepted load forecast method for the bulk
of the forecasting:
1) Remove the data center load from the historical load, leaving a demand history of all
loads except data centers.
2) Perform weather adjustment and any other normalizations of that demand history and set
up, calibrate, and fit the forecast model to the demand history of all load except data
centers, to produce a forecast of all load except data centers.
3) Perform a separate forecast of data center loads that does not use an econometric -
demographic method, but instead develops the forecast using a method like that used in
Section 2 to produce the four scenarios of growth shown in Table 4.
4) Add the two forecasts together to produce the overall forecast of all demands.
Analysis and Forecast of Data Center Growth in Northern Virginia Oct. 17, 2013 Confidential/Proprietary
Page 32 of 33

Att. JFW-5

Att. JFW-5
Virginia Electric and Power Company
Case No. PUE-2016-00049
Erivirontmental Respondents
Third Set
The following response to Question No. 35 of the Third Set of Interrogatories and Requests for
Production of Documents propounded by the Environmental Respondents received on August 2,
2016 has been prepared under my supervision,
G .�.
Ashwani Vaswani
Manager, Transmission Planning Integration
Dominion Resources Services, Inc.
Question No. 35
Reference Appendix 21 of the IRP
a) Please explain line 5, "Peak Adjustment.
Response,
The Peak Adjustment line includes the results of the RPM auction (in years 2017 — 2019), as well
as 55 MW of behind -the -meter ("BTM") generation, not included in the load forecast.
11101E■tWN1-UTMrNZ1111WA y

CERTIFICATE OF SERVICE
I hereby certify that the following have been served with a true and accurate copy of the
foregoing via first-class mail, postage pre -paid:
Michael D. Thomas, Esquire
Matt Roussy, Esquire
K.B. Clowers, IV, Esquire
Office of General Counsel
STATE CORPORATION COMMISSION
P.O. Box 1197
Richmond, VA 23218
Vishwa B. Link, Esquire
Jennifer D. Valaika, Esquire
MCGUIRE WOODS, LLP
Gateway Plaza
800 East Canal Street
Richmond, VA 23219
Donald J. Sipe, Esquire
PRETI, FLAHERTY, BELIVEAU & PACHIOS, LLP
45 Memorial Circle
P.O. Box 1058
Augusta, ME 04332-1058
Brian R. Greene, Esquire
Eric J. Wallace, Esquire
Bruce H. Burcat, Esquire
GREENEHURLOCKER, PLC
1807 Libbie Avenue, Suite 102
Richmond, VA 23226
Louis R. Monacell, Esquire
Edward L. Petrini, Esquire
James G. Ritter, Esquire
CHRISTIAN & BARTON, LLP
909 East Main Street, Suite 1200
Richmond, VA 23219-3095
William T. Reisinger, Esquire
GREENEHURLoCKER, PLC
1807 Libbie Avenue, Suite 102
Richmond, VA 23226
Lisa S. Booth, Esquire
Charlotte P. McAfee, Esquire
DOMINION RESOURCES SERVICES, INC.
Law Department
120 Tredegar Street
Richmond, VA 23219
C. Meade Browder, Jr., Esquire
C. Mitch Burton, Jr., Esquire
Kiva Pierce, Esquire
Division of Consumer Counsel
OFFICE OF THE ATTORNEY GENERAL
202 North Ninth Street
Richmond, VA 23219
Irene A. Kowalczyk, Esquire
WESTROCK COMPANY
7 Penn Plaza, Suite 1606
New York, NY 10001
Peter W. Brown, Esquire
Nathan R. Fennessy, Esquire
PRETI, FLAHERTY, BELIVEAU & PACHIOS, LLP
P.O. Box 1318
Concord, NH 03302-1318
Joseph M. Lovett, Esquire
Evan D. Johns, Esquire
APPALACHIAN MOUNTAIN ADVOCATES
415 Seventh Street Northeast
Charlottesville, VA 22902
Bobbi Jo Alexis, Esquire
Culpeper County Attorney
306 N. Main Street
Culpeper, VA 22701

William C. Cleveland
Southern Environmental Law Center
DATED: August 17, 2016

Attachment A
PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
Review and Evaluation of the Peak Load Forecasts for the
Duke Energy Carolinas and Duke Energy Progress 2016 Integrated Resource Plans
James F. Wilson, Wilson Energy Economics
Prepared on behalf of Natural Resources Defense Council, Southern Alliance for Clean
Energy, and the Sierra Club
February 17, 2017
I. INTRODUCTION
1. Duke Energy Progress, LLC ("DEP") and Duke Energy Carolinas, LLC
("DEC") filed their 2016 Integrated Resource Plans ("2016 IRP") on September 1, 2016 in
Docket No. E-100 Sub 147. The peak load forecasts and reserve margins serve as the
basis for each utility's determination of the total generating capacity required over the
IRP planning horizon. This report evaluates the peak load forecasts used in the 2016
IRPs. The determination of the reserve margins used in the 2016 IRPs is the subject of a
separate Wilson Energy Economics report ("Reserve Margin Report").
II. PEAK LOAD FORECASTING METHODOLOGY
2. The 2016 DEC and DEP IRPs include peak load and energy forecasts for
the respective service territories over the 2017 to 2031 time period. The forecasts
encompass residential, commercial and industrial retail customers and also wholesale
customer loads. The companies use econometric models to forecast the residential,
commercial and industrial customer classes separately. The forecasts rely upon
economic and demographic projections from Moody's Analytics, and projections of
appliance efficiencies and saturations from Itron, based on U.S. Energy Information
Administration data.
3. In recent years peak load growth has been weak across most of the
country, and forecasts are generally being revised downward. The weak growth is partly
due to the economic downturn and weak recovery over the past decade. However,
there are also trends toward more efficient use of electricity, and a weaker relationship
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 1 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
between economic growth and load growth. For example, PJM Interconnection, LLC,
which prepares peak load forecasts for all or part of 13 states (including North Carolina)
and the District of Columbia, has been continually enhancing its load forecasting
methodology, and its most recent forecasts have again been sharply reduced.'
III. COMPARISONS OF PEAK LOAD FORECASTS TO RECENT TRENDS
4. This evaluation begins with a review of the actual and weather -adjusted
summer and winter peak load trends. Actual peak loads will tend to vary substantially
from year to year, reflecting the presence or absence of the type of extremely hot or
cold weather than can cause the highest summer or winter peak loads, respectively.
Weather -adjusted peak loads are estimates of what the peak load would have been in a
historical period had the peak occurred on a day with the typical peak -causing weather.
Weather -adjusted historical peak loads remove the impact of weather variability, and
reveal the underlying peak load trend due to other factors such as economic and
demographic trends, changes in industry and end-use technologies, and energy
efficiency.
5. Peak load forecasts are generally considered median or 50-50 forecasts
(meaning that the forecasters consider the actual future peak load to be equally likely to
exceed, or to fall short of, the forecast value) or mean forecasts (averages or expected
values of the future peaks). Similarly, a weather -adjusted historical peak is generally
also considered a median value for the past period (a value that, based on actual
weather, had a 50-50 chance of being exceeded) or perhaps a mean value. Thus, the
weather -adjusted actual peaks correspond to what the peak load forecasts are
attempting to estimate. Peak load forecasts can be compared to the trends in weather -
adjusted historical peaks, and we should expect the forecasts to be generally consistent
with those trends. Therefore, a comparison of a peak load forecast to the
1 PJM, PJM Load Forecast Report, January 2017 (noting at p. 2 that the forecast for 2020 has
been reduced two percent compared to the prior forecast prepared one year earlier).
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 2 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
corresponding weather -adjusted peak load trend is a useful starting point for evaluating
the reasonableness of a forecast.
6. However, for a meaningful comparison of trends in weather -adjusted
peaks to forecast peaks, both sets of data must represent roughly the same underlying
loads. For example, the obligations under a large new wholesale contract would not be
reflected in the historical trends but might be included in the forecast, and this would
cause the forecast to appear high compared to the historical trend unless the impact of
the contract is taken into account.
7. In addition, anticipated shifts in the underlying economic, demographic
or technological trends that drive peak load growth could cause the peak forecast to
deviate from the historical trend. For example, economic growth is a driver of peak load
growth, and if economic growth is expected to be stronger over the forecast period than
it has been in recent years, this could cause the forecast to increase more rapidly than
the trend, other things equal. Similarly, improvements in energy efficiency affect peak
load growth, so changing trends in this regard can also cause a forecast to deviate from
past trends.
8. DEC and DEP provided summer and winter peak load forecasts (response
to Public Staff data request 1-7) both with and without the load -reducing impacts of
future energy efficiency program implementation (discussed in DEC 2016 IRP, pp. 9-10).
This report's comparisons will be based on the peak load forecasts without the forecast
impacts of the future implementation of these programs, whose additional impacts are
in any case rather small in the first years of the forecast. It should be noted that the
forecasts without the future energy efficiency implementations will still reflect the
future impacts of prior energy efficiency implementation. DEC and DEP also provided
historical actual and weather -adjusted peak loads (responses to SACE DEP 2-9, SACE DEC
2-9, Public Staff DEC 1-3, DEC 1-4, DEP 1-3, DEP 1-4).
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 3 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
IV. PEAK LOAD FORECASTS COMPARED TO RECENT TRENDS: DEC
9. Figure JFW-1 compares recent DEC actual and weather -adjusted
historical summer peak loads to the forecast peaks. The figure also shows a trend line
based on the weather -adjusted peaks over the 2009 to 2016 period, during which
period they exhibited a quite steady trend.
21,W0
20,500
20,000
19,500
19,000
18,500
18,000
17,500
17,000
16,500
16,000
15,500
15,000
Figure JFW-1: DEC Summer Peaks, Historical and Forecast
Sources: Data Requests NC Public Staff 1-3, 1-4, 1-6 and SAC 2-9
45 -0 ,y0 ,y0 -0 y0 'e"'e" ,y0 y ,t0 T ,y0 ,10 T ,y0 "0 0 ,PO
10. The peak load forecast reflects some changes in DEC's wholesale sales
contracts over time, documented in DEC 2016 IRP, Table H-1. The cumulative total
change in such contracts is , so the composition of the peak load forecast
in that year is quit the historical trend. Figure JFW-1 suggests that the
forecast is roughly 200 MW, or about 1%, in excess of the trend in 2019, and that after
2019 the forecast peaks grow at a somewhat faster rate.
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 4 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
11. However, in a recent supplemental response to a data request, DEC also
states that the forecast reflects 540 MW of a backstand agreement with North Carolina
Electric Membership Corporation ("NCEMC"), and that this would not be reflected in the
historical values except when the Catawba Nuclear Station was not in operation
(responses and follow-up responses to Data Requests SACE 3-2 and 3-3)
If indeed this 540
MW is included in the forecast and is not reflected in the historical data, it is appropriate
to adjust the forecast downward by this amount for comparison purposes, in which case
the forecast would lie somewhat below the trend line for most of the forecast period.
12. Further, if this 540 MW obligation, which apparently is unlikely to be
called upon, was included in the forecast, the forecast can no longer be considered a
median or mean forecast. This has implications for the use of the forecast in the
resource adequacy study, as discussed in the Reserve Margin Report.
13. In addition, DEC's peak load forecasts are based on economic forecasts
that anticipate faster growth in the North Carolina economy over the coming years than
was seen in the past decade. Figure JFW-2 shows the North Carolina Gross State
Product ("NC GSP") forecast used to prepare the load forecasts, from Moody's Analytics.
As shown in the figure, the North Carolina economy grew at about a 1.4% per year rate
during 2009 to 2015, while the forecast rate of growth is 2.9% per year. This would help
to explain the somewhat faster rate of growth in the peak load forecast compared to
trend.
14. The load forecasts also use historical and projected appliance saturation
and penetration data from Itron and the U.S. Energy Information Administration (DEC
2016 IRP, p. 16). While I have not performed a detailed review of this data, some key
indicators, such as central air conditioner efficiency, suggest a slowing rate of
improvement in energy efficiency over the coming years compared to recent trends.
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 5 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
Figure JFW-2: North Carolina Real Gross State Product ($ mil.)
$750,000
MUM(]
i.
$650,000 I
I
2016 to 2030 (forecast): 2.95%/year
$600,000
I
I
I
$550,000 I
I
I
$500,000 I
2005 to 2015 (historical): 1.43%/year
2009 to 2015 (historical): 1.3$%/year
$450,000
1
I
I
$400,000
1
v
n
$350,000
Source: Response to SALE 2-2. Data is from Moody's Analytics; July values are used.
$300,000
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 2022 2024 2026 2028 2030
15. Overall, DEC's summer peak load forecast appears with a reasonable
range, especially if it is correct that the 540 MW NCEMC backstand agreement is
included in the forecast but not reflected in the historical data.
16. Turning now to DEC's winter peak load forecast, Figure JFW-3 presents
the comparison to recent actual and weather -adjusted peaks. Figure JFW-3 also shows
the very high actual winter peak loads seen in 2014 and 2015, which occurred under
conditions of very extreme cold. As noted in the discussion of summer peaks, the
forecast reflects changes in wholesale contracts documented in DEC 2016 IRP Table H-1,
and the cumulative change is , so that year is perhaps the best to focus on
for comparison purposes. The winter peak forecast is roughly 700 MW in excess of
trend in 2019. The 540 MW NCEMC backstand agreement, noted above, would explain
most of that discrepancy, if indeed it is included in the forecast but not reflected in the
historical data. Anticipated stronger economic growth would also help to explain the
discrepancy.
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 6 of 15

21,000
20,500
20,000
19,500
19,000
18,s00
18,000
17,500
17,000
16,500
16,000
15,500
15,000
PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
Figure JFW-3. DEC Winter Peaks, Historical and Forecast
winter Peak: Actual
—winter Peak: weather Adjusted
—Forecast winter Peak no Fe
--Linear (winter Peak: weather Adjusted)
�Ources: uaia Kequesis 14L rumic ]Lair 1-3, 1-4, 1-h ana JAIL L-7
o� o§� t$o yti �-" �,,�+a 1. b ,�� �� ti� ti° titi titi ti3 tia
y0 y0 do Lq tip tiQ y0 ,LO IV ,te+ ti4 ti0 yC? y0 y0 ,LO L� L©
17. Figure JFW-4 shows the hourly load patterns on the two highest winter
peak load days of 2014 and of 2015, based on the hourly data provided in response to
data request SACE 2-82. The lowest temperatures on these days (under 10 degrees) had
not occurred for nearly twenty years, since 1996 (based on the hourly temperature data
provided in the response to SACE 2-16). Note also that the highest loads on these days
occurred for only a very brief period around 8 AM. Changes in end-use technologies
may be affecting these brief, extreme winter peak loads under extreme cold conditions.
However, DEC states that it has not performed any formal analysis to determine which
end uses are contributing to these load spikes on extremely cold winter mornings
(response to Data Request SACE 2-11).
2 It should be noted that there were some quite substantial differences between the summer
and winter actual peak loads for 2014 to 2016 as reported in the responses to data requests
Public Staff 1-3 and 1-4 and the hourly historical data provided in response to SACE 2-8. Figures
JFW-4 and JFW-5 rely upon the hourly data.
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 7 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
Figure JFW-4: DEC Hourly Leads - Recent Extreme Winter Peak Days
20,000
19,000
18,wo
17,000
16,000
15,000
14,000
01/07/2014
13,000
—01/30/2014
12,000
—01/08/2015
11,000 —02/20/2015
10,000
9,000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour Ending
18. Note also that the extreme winter loads occurred for a brief period of
time, and only on a few extremely cold winter days. Figure JFW-5 shows the daily peaks
on the highest winter load days, expressed as a percent of the annual weather -adjusted
peak, for the top six peak winter days in each of the past six years. As could be
expected, three years had peaks above the weather -adjusted peak and three had peaks
below that level. However, even in the "polar vortex" year (2014), the fourth highest
daily peak that winter was far below the highest peak, and also below that year's
weather -adjusted peak. So while there may be indications that winter peak loads can
exhibit brief, extraordinary spikes on days with extremely low temperatures that are
rarely seen, this should not result in a substantial increase in a peak load forecast that is
intended to be a median or mean peak load forecast.
19. Overall, the DEC winter peak forecast seems somewhat high compared to
the trend in the weather -adjusted peaks, even if it is appropriate to adjust it by 540 MW
for the NCEMC backstand agreement. While the extreme winter peak loads in 2014 and
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 8 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
Figure JFW-5: DEC Winter Daily Peak Load Durations
(actual peaks expressed as percents of weather -adjusted peaks)
110%
—2011 +2012
105%
-*-2013 +2014
--*--2015 -4- 2016
s
m 100%
a
a
m
a
95%
t
m
a
J
85%
80%
2, 3 4 5 6
Days
2015 suggest that changes in end uses may be contributing to higher winter peak loads
under extreme cold, this may be a phenomenon that only has large impacts under very
extreme conditions that occur very infrequently. It is not clear that under the very cold,
but less extreme temperatures typical of the annual winter peak day in most years that
such extreme loads should be expected. And again, if the 540 MW for the backstand
agreement has been added, this is no longer a median or mean forecast.
20. Figure JFW-6 shows the DEC winter and summer peak load history and
forecast on the same graph. There has been a steady differential between the weather -
adjusted summer and winter peaks during recent years, averaging 750 MW over 2009 to
2016, and averaging 683 MW over 2014 to 2016. The forecast breaks from this pattern,
again suggesting that the winter peak forecast is high.
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 9 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
Figure JFW-6: DEC Summer and Winter Peaks, Historical and Forecast
21,000
20,500
20,000
r
19,500
19,000
18,500
1s,0oo -�
17,500
—Summer Peak: Weather Adjusted
17,000
blunter Peak: Weather Adjusted
16,500 Forecast Summer Peak no EE
16,000 -4-Forecast Winter Peak no EE
15,500
Sources: Data Requests NC Public Staff 1-3,1-4,1-6 and SALE 2-9
15,000
4° Q1 d`b 0°� ti� titi titi �3 ,yA yh ti0 til 11b 1C) ,ti0 by y^L 'L'' ryN
V. PEAK LOAD FORECASTS COMPARED TO RECENT TRENDS: DEP
21. Figures JFW-7, JFW-8 and JFW-9 compare recent DEP actual and weather -
adjusted historical summer peak loads to the forecast peaks. However, in contrast to
the DEC data, very little historical data was provided for DEP, making it difficult to
discern past trends. Figure JFW-7 shows a trend line based on the weather -adjusted
peaks over the 2013 to 2016 period, during which period they exhibited a rather steady
trend. As for DEC, there were changes in the DEP wholesale sales contracts
documented in DEP 2016 IRP, Table H-1. The cumulative total change is -
and , so the peak load forecast in 2020 is quite
comparable to the historical trend. Figure JFW-7 suggests that the forecast is roughly
200 MW in excess of the trend in 2020. After 2020 the forecast peaks grow at a faster
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 10 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
rate than the trend, which may be explained by the anticipated faster economic growth
discussed earlier.
Figure JFW-7: DEP Summer Peaks, Historical and Forecast
16000
15500
15000
14500
14000
13500
13000-w – –
12500 ~
—�—Summer Peak: Actual
12000
—Summer Peak: weather Adjusted
11500 —Forecast Summer Peak no EE
11000 - - - Linear (Summer Peak: weather Adjusted)
10500
Sources: Data Requests NC Public Staff 1-3, 1-4, 1-6 and SACE 2-9
10000
2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
22. Figure JFW-8 presents DEP's winter peak forecast and the available
historical data. The forecast is rather consistent with a trend line based on the 2013 and
2014 winter peaks. The actual peaks were considerably higher in the winters of 2015
and 2016, due to the extreme cold noted above. The weather -adjusted peak values that
were provided for 2015 and 2016 were also considerably higher than the values for
2013 and 2014, which may reflect that the methodology for determining weather -
adjusted peaks does not fully compensate for weather impacts, and the weather -
adjusted values for 2015 and 2016 may be overstated. In responses to data requests
(SACE 3-2 and 3-3d) DEP stated that the 2016 actual results were not available when the
forecast was prepared, and also suggested that its forecasting model may discount such
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 11 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
extraordinary growth, such that the forecast tends to reflect the earlier trends. DEP also
suggested that future winter peak forecasts may be higher.
16000
15500
15000
14500
14000
13500
13000
12500
12000
11500
11000
10500
10000
Figure JFW-8. DEP Winter Peaks, Historical and Forecast
--*—winter Peak: Actual
—winter Peak: Weather Adjusted
--*—Forecast Winter no EE
- - - Linear (Weather Adjusted 13-14)
Sources: Data Requests NC Public Staff 1-3, 1-4, 1-6 and SAGE 2-9
2013 2014 2015 2016 2017 2,018 2019 2020 2021 2022 2023 2024
23. Again, the possibility of extreme loads under very rare extreme
temperatures might not result in much increase in a peak load forecast that is intended
to be a median or mean forecast.
24. Figure JFW-9 presents the DEP summer and winter peak load forecasts
and history together. As with DEC, the relationship between the summer and winter
peaks reflected in the historical data (at least as reflected for 2013-2014) is changed in
the forecast, with the winter peak having risen faster than the summer peak. Due to the
lack of sufficient historical data to establish trends, and the lack of support for a
substantial increase in the winter peak load forecast, this report draws no conclusion
with regard to the reasonableness of the DEP peak load forecasts.
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 12 of 15

16000
15500
15000
14500
14000
13500
13000
12500
12000
11500
11000
10500
10000
PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
Figure JFW-9: DEP Summer and Winter Peaks, Historical and Forecast
—Summer Peak: Weather Adjusted
—Winter Peak: Weather Adjusted
—$—Forecast Summer Peak no EE
Forecast Winter Peak no EE
Sources: Data Requests NC Public Staff 1-3, 1-4,1-6 and SALE 2-9
2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
VI. SUMMARY AND RECOMMENDATIONS
25. To summarize the observations with regard to the peak load forecasts:
a. DEC's summer peak load forecast falls within a reasonable range,
especially if it includes the 540 MW NCEMC backstand agreement that is
not reflected in the historical data.
b. DEC's winter peak load forecast seems somewhat high, even considering
the NCEMC backstand agreement.
c. There is insufficient information to come to a conclusion about the DEP
peak load forecasts.
26. The very high loads that have occurred on recent, extremely cold winter
days occur for very few days and hours; loads in other hours and on other days are
much lower. Peak load forecasts intended to represent median or mean values should
be relatively unaffected by such rare events.
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 13 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
27. If the DEC peak load forecasts include the 540 MW NCEMC backstand
agreement that is rarely invoked, they are no longer median or 50-50 peak load
forecasts; nor do they represent an average or mean of the possible peak values. This
has implications for how the forecasts are used to evaluate resource adequacy and
determine capacity needs as discussed in the Reserve Margin Report.
28. Finally, this evaluation leads to the following suggestions for future IRP
proceedings:
a. The companies should research the drivers of load spikes on extremely
cold winter mornings.
b. Future IRP filings should more clearly document wholesale contract
arrangements and how they are reflected in the forecasts and in the
historical load data. The forecasts for the various wholesale
arrangements should be provided separately from the utility load
forecasts.
c. The IRP filings should clearly specify whether the peak load forecasts are
intended to be median or mean forecasts, or what portion of the forecast
(such as, net of wholesale contracts) is intended to be a median or mean
value.
d. The IRP filings would also benefit from data and explanation of how the
forecasts are or are not consistent with recent trends in actual and
weather -adjusted peak loads.
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 14 of 15

PUBLIC VERSION—CONFIDENTIAL INFORMATION REDACTED
APPENDIX: QUALIFICATIONS OF JAMES F. WILSON
James F. Wilson is an economist and independent consultant doing business as
Wilson Energy Economics, with a business address of 4800 Hampden Lane Suite 200,
Bethesda, Maryland 20814. Mr. Wilson has over 30 years of consulting experience,
primarily in the electric power and natural gas industries. Many of his consulting
assignments have pertained to the economic and policy issues arising from the interplay
of competition and regulation in these industries, including restructuring policies,
market design, market analysis and market power. Other recent engagements have
involved resource adequacy and capacity markets, contract litigation and damages,
forecasting and market evaluation, pipeline rate cases and evaluating allegations of
market manipulation. His experience and qualifications are further detailed in his CV,
available at www.wilsonenec.com.
Wilson Evaluation of Duke IRP Peak Load Forecasts Page 15 of 15

Attachment B
Review and Evaluation of the Reserve Margin Determinations for the
Duke Energy Carolinas and Duke Energy Progress 2016 Integrated Resource Plans
James F. Wilson, Wilson Energy Economics
Prepared on behalf of Natural Resources Defense Council, Southern Alliance for Clean
Energy, and the Sierra Club
February 17, 2017
I. INTRODUCTION
1. Duke Energy Progress, LLC ("DEP") and Duke Energy Carolinas, LLC
("DEC") filed their 2016 Integrated Resource Plans ("2016 IRP") on September 1, 2016 in
Docket No. E-100 Sub 147. The peak load forecasts and reserve margins serve as the
basis for each utility's determination of the total generating capacity required over the
IRP planning horizon. The reserve margins used in the 2016 IRPs were based upon
recommendations in the DEC and DEP 2016 Resource Adequacy Studies, November 17,
2016 ("DEC RA Study", "DEP RA Study") prepared by Astrape Consulting and provided in
response to data request SACE 1-8. This report evaluates the adopted reserve margins
and the RA Studies that are the basis for the adopted reserve margins. The load
forecasts used in the 2016 IRPs are the subject of a separate Wilson Energy Economics
report ("Load Forecast Report").
II. THE 2016 RESOURCE ADEQUACY STUDIES: OVERVIEW AND RECOMMENDATION
2. The DEC and DEP RA Studies both recommend a 15% installed reserve
margin relative to summer peak demand, which provides a 17% winter installed reserve
margin (DEC RA Study pp. 6-7). This is an increase from the 14.5% installed reserve
margin relative to summer peak demand recommended in similar resource adequacy
studies in 2012 and reflected in prior IRPs. According to both RA Studies (p. 2), the
increase in the recommend reserve margins reflects "A re-evaluation of seasonal risk
after the Polar Vortex in 2014 and cold weather during 2015 resulted in a significant
shift to winter reliability issues."
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 1 of 22

3. The RA Studies document a probabilistic simulation of load and resources
to find the planning reserve margin required to satisfy a "one day in ten years" resource
adequacy criterion, equivalent to an annual Loss of Load Expectation ("LOLE") of 0.1. In
addition to this analysis of "physical reliability", the RA Studies also include evaluations
of "economic reliability", and a preliminary review of the assumptions underlying this
economic analysis raises many questions and doubts. However, the recommended
reserve margins are based on the physical reliability results, so this review was limited
to the physical reliability results.
4. The evaluation performed for this report was limited by insufficient
information provided with regard to the details of the studies, discussed in Appendix A
to this report. Accordingly, the evaluation focused on three issues having to do with
how loads were represented in the RA Studies, and that were found to be inaccurate
and unsupported:
a. First, the RA Studies extrapolated the relationship between cold
temperatures and winter loads that occurred in some hours in recent
years over much lower temperatures that have not occurred for decades
in a manner that greatly exaggerates the magnitude of the loads likely to
occur under extreme cold conditions.
b. Second, the "economic load forecast uncertainty" that was layered on
top of the weather-related load distributions was also exaggerated, and is
not supported by the underlying data it was based upon.
c. Third, the RA Studies relied upon the DEC and DEP peak load forecasts,
and treated them as forecasts of mean or average peak loads; however,
at least in the case of DEC, the forecast value apparently was not a mean
value, and was likely several hundred MW in excess of the mean forecast,
which would bias the reserve margin higher.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 2 of 22

S. The review of these three issues leads to the conclusion that the risk of
very high loads, especially in winter, was substantially exaggerated in the RA Studies,
and, therefore, the recommended increases in the DEC and DEP reserve margins are
unsupported and should be rejected.
6. These three issues are discussed in the next sections of this report.
I1:741:20*4I►hIl►[011:ISI►&17_Ngo] 2*:41:T4LTAI: Le]4OXOVL►JI 1►1114:4119L1*1
7. In recent years, very extreme cold conditions have in a few instances
resulted in very high loads on the DEC and DEP systems, as further discussed in the Load
Forecast Report. To accurately evaluate winter period resource adequacy, it was
necessary for the RA Studies to model extreme cold and its impact on load levels.
8. 2014 and 2015 were years characterized by days colder than any that had
occurred since 1996. Based on the temperature data used in the RA Study (response to
SACE 2-16), 2014 and 2015 each had two days in which temperatures dropped below 10
degrees Fahrenheit; in the years before 2014, temperatures had not dropped to even 11
degrees since 1996. However, the RA Studies used 36 years of historical weather data
back to 1980, and even lower temperatures were seen in some years in the 1980s and
1990s (3, 4, and 5 degrees in 1982, 1984, and 1986, respectively, and minus 5 in 1985).
Therefore, to use the 36 years of weather data it was necessary to model loads under
extremely cold conditions that have not been seen in over 20 years.
9. The most extreme recent winter loads have occurred on extremely cold
mornings at about 8 AM, as shown in the Load Forecast Report. Arguably, once
temperatures drop to the teens, customers may have turned on all of the equipment
that will help them stay warm, and further declines in temperature may not further
increase loads very much. However, the companies have not performed any analysis to
determine what end uses are specifically contributing to these load spikes experienced
on extremely cold winter mornings (response to SACE 2-11).
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 3 of 22

10. The RA Studies based the relationship between extreme cold
temperatures and load levels on a very simple regression analysis. A more complex
neural network approach was used to determine the relationship between temperature
and load for most hours (DEC RA Study p. 12). However, the response to a data request
(NC Public Staff 8-9) stated that "since neural networks do not do a good job of
extrapolating relationships beyond conditions seen in the training data, or identifying
relationships for rare conditions, we do not rely on the neural network relationships for
extreme conditions." Instead, the RA Studies used regression analysis. The regression
was provided in response to the same data request (Public Staff 8-9), and the equation
used to represent the relationship between temperature and DEC load, under extreme
cold conditions, was as follows:
DEC Load = -231 * (Temperature) + 20,372.
11. This equation means that under extreme cold conditions, for each degree
the temperature falls, DEC's load is assumed to increase by 231 MW (roughly 1.3%).
Four additional degrees results in 924 MW of additional load (over 5% increase).
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 4 of 22

12. This equation, and the critical 231 MW per degree assumption, were
simply based on a regression (Microsoft Excel linear trend line) to fit all recent
observations of DEC load and temperature 21.9 degrees and below (25 observations), as
shown in the response to the data request. Figure JFW-1 shows the analysis as provided
in the data request (the regression is shown in the blue dashed line and associated
equation), with some reformatting and additional analysis.
Figure JFW-1: DEC Winter Peak Temperature -Load Regression
20,000
1q,00o
•t
18,000 '
y = -61x + 18,499;' "
17,000 _
Y = -108x+ 18,914
0 16,000
♦ Original series(Astrape) y= -231x+20,372
■ Temp under 17
Is,aoa ♦ ♦
• Temp under 16
-----Linear (Original series (Astrape)j
14,000
-----Linear (Temp under 17)
---- Linear (Temp under 16)
13,000
Source: Response to Public Staff 8-9.
12,000
5 10 15 20 25
Temperature F
13. There is no documentation of why this subset of the data was chosen,
and the result is highly sensitive to the chosen range. Using observations under 20.4
degrees rather than 21.9 reduces the 231 MW value to 188 MW.
14. More important, however, the relationship between temperature and
load in the more moderate temperature range is of questionable relevance to
understanding the relationship for temperatures under extreme cold conditions. As
suggested above, while declining temperatures lead to increasing loads, once
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 5 of 22

temperatures drop to very low levels a saturation effect may take place, if nearly all
appliances are already deployed.
15. To better understand the relationship between extremely low
temperatures and load, a regression was performed focused instead on temperatures
under 17 degrees; this changed the resulting relationship between temperature and
load from 231 MW per degree to 108 MW per degree, reducing the impact of further
cold by over half. The result of this regression is shown by the red dashed line and
equation in Figure JFW-1. Further limiting the regression to temperatures under 16
degrees resulted in just 61 MW per degree (the green dashed line and equation in
Figure JFW-1), a relationship between extreme cold and load almost four times weaker
than the RA Study assumed.
16. This additional analysis demonstrates two things. First, it demonstrates
that the critical 231 MW per degree assumption used in the DEC RA Study was arbitrary,
as it reflects the particular subset of data used in the regression. Different subsets
based on different temperature ranges give very different results; the results are highly
sensitive to the choice of temperature range. Second, and more important, this analysis
suggests that the relationship between extreme cold and load is much weaker than 231
MW per degree; the 61 MW or 108 MW per degree estimates, more appropriately
focused on the coldest observations, are likely more accurate estimates of the
relationship between temperature and load under extreme cold conditions.
17. A similar analysis was conducted for DEP East, which leads to very similar
results and the same conclusions. This analysis is shown in Figure JFW-2. The RA Study
used 228 MW per degree (shown in blue), based on temperatures up to 26 degrees.
Again, it is unclear why 26 degrees was chosen, and the relevance of observations at
temperatures up to 26 degrees is doubtful. Focusing the regression on temperatures
under 19 degrees results in 153 MW per degree (shown in red); and focusing the
regression on temperatures under 18 degrees (shown in green; for which there were
only three observations) results in only 12 MW per degree. Similarly, for the much
smaller (and generally colder) DEP West zone, the RA Study based the regression on
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 6 of 22

temperatures under 22 degrees, and focusing the regression on colder observations
results in a substantially weaker relationship between temperature and peak load.
Figure JFW-2: DEP East Winter Peak Temperature -Load Regression
14,000
13,500
Y=-12x+13,305--=•ct-----------!
13,000
♦
y- -153x+ 15,138
12,500
12,000 +�
♦ Original series (Astrape)
11,500 Temp under 19
Ternp under 18
Linear (Original series (Astrape)) !A ♦ `,
11,000
-----Linear (Temp under 19)
-----Linear (Temp under 18) y= -228x+ 15,193+ •�•, +♦�
10,500
Source: Response to Public5taff8-9.
10,000
5 10 15 20 25 30
Temperature F
18. The 231 MW per degree assumption for DEC, and 228 MW per degree
assumption for DEP East, result in some very extreme peaks under the very cold
conditions represented in some of the 36 weather years. Figure JFW-3 shows a graphic
from the DEC RA Study illustrating how high winter peaks are assumed to go, as a result
of the 231 MW per degree assumption. While the extreme cold in 2014 and 2015
resulted in extreme peak loads roughly 5% to 8% above the anticipated, normal winter
peak loads in those years, the 231 MW per degree assumption results in modeling peaks
in the 1982 weather year 18% above the anticipated winter peak (for 2019, the year
that is the focus of the RA Studies, 18% equates to over 3,300 additional MW).
Modeling such extreme peaks will, of course, drive the winter reserve margin higher.
Using the more realistic 61 or 108 MW per degree estimates would bring these extreme
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 7 of 22

peaks down considerably. Figure JFW-3 also shows the similar graphic from the DEP RA
Study, which also reflects very extreme winter peaks (over 20% above the normal winter
peaks) based on the unrealistic estimates of the relationship between extreme cold and
load.
Figure JFW-3: Figure 3 from the DEC and DEP RA Studies
Figare I DEC Wcoter Peak Weatker ViLriAbibrr
C—ee At �n'O a -ill :�' �CiT �I r+ R>t d' �rin4'V1 pf-R'7 =z— * n
Fipre 3 DIP Witter Peak Weeder Varishbtr
-1S1h
-1SX
19. The critical assumptions about the impact of extreme cold on load levels
were chosen based on simple regressions over rather arbitrarily -chosen temperature
ranges, despite the high sensitivity of the results to the chosen ranges. This casual
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 8 of 22

approach stands in contrast to the rigorous process and analysis that the load
forecasters at PJM Interconnection, LLC, who prepare peak load forecasts and evaluate
reserve requirements for all or part of 13 states (including North Carolina) and the
District of Columbia, underwent to enhance their load forecasting methodology
following the polar vortex experience. The PJM load forecasters developed
enhancements to more accurately represent the relationship between loads and
extreme temperatures. The proposed changes were discussed with stakeholders over
multiple meetings of the PJM Load Analysis Subcommittee before the changes were
approved and implemented. PJM's enhanced methodology now employs additional
"weather splines" (essentially, regressions over ranges of temperatures), in order to
more accurately capture the relationships between load and temperature over different
temperature ranges, including extreme hot or cold conditions.'
1 See, for instance, PJM, Item 4 — Forecast Update, Load Analysis Subcommittee meeting September 2,
2015, slides 2-23 (describing use of four temperature ranges each for summer and for winter splines).
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 9 of 22

20. The arbitrary and inaccurate assumptions about the relationship between
extreme cold and load results in modeling some very extreme winter loads in the RA
Study simulations, driving the reserve margin results higher. In response to a data
request (Public Staff 8-13), the winter LOLE values across the 36 simulated weather
years (1980 through 2015) were provided. As noted above, 2014 and 2015 were years
characterized by days colder than any that had occurred since 1996. Therefore, it
should be expected that 2014 and 2015 would contribute significantly to the total
winter LOLE across the 36 weather years. However, instead these two years play a small
role. In the DEC RA Study, 2014 and 2015 contribute only 1.6% of the total winter LOLE
across the 36 weather years (two average years out of 36 would be expected to
contribute 5.6% of the total). The four years with the most exaggerated extreme loads
shown in Figure JFW-3 (1982, 1985, 1994 and 1996) account for 64% of the total winter
LOLE. 87% of the LOLE occurs in the 17 weather years from 1980 through 1996, while
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 10 of 22

only 13% of the LOLE occurs in the 19 years from 1997 to 2015. This data is illustrated in
Figure JFW-4.2
Figure JFW-4; DEC Winter LOLE by Weather Year
(percent of total winter LOLE, 16% winter reserve)
25%
ear
15%
10%
Source: Response to NC Public Staff 8-13.
5%
A814RR 8888 8888888000�
.-r- r -i - N .�-i - - - - .�-c - N N N N N N N N N N N N N N N N
21. In the DEP RA Study the results are similar. 2014 and 2015 contribute
only 7.2% of the winter LOLE. The four years 1982, 1985, 1994 and 1996 account for
66% of the LOLE, and the first 17 years contribute 78%, the final 19 years only 22% of
the LOLE (Figure JFW-5).
Z The data request (Public Staff 8-13) provided results for a 16% and 18% winter reserve margin, but not
the recommended 17% winter reserve margin; the above results are for the 16% winter reserve margin.
Under the 18% reserve margin assumption, the contribution of 2014 and 2015 to the total LOLE is even
smaller, and the role of the assumed extreme loads in the earlier years is even greater.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 11 of 22

Figure JFW-5: DEP Winter LOLE by Weather Year
(percent of total winter LOLE, 16% winter reserve)
zs%
209%
15%
Source: Response to NC Public Staff 8-13.
10%
S%
0%
ff +--i ti .y ti ti ti ti N ti ti M .ti N N N N N N N N N N N N N N N N
22. Thus, the vast majority of the winter LOLE in the RA Studies occurs in
weather years long past, based on temperatures that have not been seen in decades
and highly speculative assumptions about how loads would increase due to such
temperatures, should they occur again. These assumptions, new in the 2016 RA Studies,
drive the reserve margins higher.
IV. REPRESENTING ECONOMIC LOAD FORECAST ERROR
23. In addition to the variability of load due to weather, the RA Studies
additionally include "economic load forecast error", intended to represent the possible
error in four year ahead load forecasts (DEC RA Study, p. 16). The economic load
forecast uncertainty is represented as a symmetric probability distribution (DEC RA
Study Table 4 p. 17). A 7.9% probability is assigned to both +4% and -4% shifts in load,
24% probability is assigned to both +2% and -2% shifts, and 36.3% chance is assigned to
no change due to economic load forecast error. Thus, all loads, including the extreme
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 12 of 22

weather-related load levels discussed in the prior section, are increased by an additional
4% under the highest economic load forecast error scenario, and 2% under an additional
scenario assigned a 24% probability.
24. This section of the report first explains why it is not appropriate to
include multi-year economic load forecast uncertainty in the RA Studies. It then
explains that the probability distribution of economic load forecast error used in the RA
Studies is not supported by the underlying data it was based upon, and greatly
overstates the risk of sharp increases in load due to forecast error.
25. The RA Studies rationalize adding the multi-year economic load forecast
uncertainty as follows: "Four years is an approximation for the amount of time it takes
to build a new resource or otherwise significantly change resource plans." (DEC RA
Study, p. 16) However, this is not correct; there are many short lead time actions that
can and would be taken. If load grows faster than expected, the utilities (and customers
and other market participants too) would have time to adjust their plans, if the rate of
load growth raised concern about resource adequacy. To name a few potential actions,
the development of some new resources might be accelerated; demand response or
energy efficiency programs could be increased; a planned retirement could be delayed;
firm purchases from adjacent regions could be adjusted; or wholesale sales contracts
could be allowed to expire. The RA Studies essentially assume the reserve margin and
resource plan must be chosen three years in advance, and then remain frozen, even if
load growth is much stronger than expected year after year (responses to SACE 2-22 and
2-23). This is not realistic, and assuming load can rise sharply due to multi-year forecast
error, but no adjustments to the resource mix can be made over three years, biases the
planning reserve margin upward.
26. It is notable that PJM, in its resource adequacy analyses, acknowledges
that resource plans can and would be adjusted as needed if load grows faster than
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 13 of 22

expected. Accordingly, PJM represents only one year of economic load forecast error in
its resource adequacy analyses.3
27. It could be appropriate to represent multiple years of forecast
uncertainty in a more sophisticated model that is able to internally determine
contingent actions to realistically adjust the resource mix over time as the load forecast
and other resources change over time. For instance, the Electric Power Research
Institute's Over/Under capacity planning model, developed by Decision Focus
Incorporated in the 1970s, had this capability.4 However, the SERVM model that was
used in the RA Studies does not represent such contingent decisions (responses to SACE
2-22 and 2-23). To represent multi-year load forecast uncertainty, but not the actions
that would be taken to adapt resource planning over time as such uncertainty resolves,
is a flawed methodology that biases the result toward higher planning reserve margins.
28. Turning to the values used for the economic load forecast error, the DEC
RA Study states (pp. 16-17) that the probability distribution was based on the historical
forecasting errors reflected in the U.S. Congressional Budget Office ("CBO") U. S. Gross
Domestic Product ("GDP") forecasts, and applying a 0.4 elasticity of peak demand to
economic changes. The DEC and DEP load forecasts rely upon forecasts of the North
Carolina economy from Moody's Analytics, so it can be questioned whether CBO U.S.
GDP forecasting errors are a reasonable proxy for the applicable economic forecasting
errors. Moody's forecasts over the past decade have frequently been far too high;
Moody's failed to anticipate the deep recession that occurred in around 2008, and for
several years after the recession was forecasting a strong recovery that never occurred.
3 See, for instance, PJM, 2012 PJM Reserve Requirements Study, p. 20 (explaining the rationale for using a
forecast error factor representing one year of forecast error).
4 Decision Focus Incorporated, Costs and Benefits of Over/Under Capacity in Electric Power System
Planning, EPRI EA -927, Project 1107, October 1978.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 14 of 22

29. The CBO data is readily available, including the 3 -year GDP forecast errors
that were the basis for the economic load forecast error distributions used in the RA
Studies .5 Figure JFW-6 presents the full distribution of the 3 -year forward GDP forecast
errors. The right axis in Figure JFW-6 shows the distribution in economic load forecast
error terms (GDP error x 0.4 elasticity, as noted above).
30. The distribution used in the RA Studies misrepresents the distribution of
CBO forecast errors:
a. First, the CBO forecast errors, overall, are not symmetric; there is more
over -forecasting than under -forecasting. The mean error was +0.7%
over -forecast. A bias toward over -forecasting is reflected in the CBO's
5 Congressional Budget Office, CBO's Revenue Forecasting Record, November 10, 2015, and Supplemental
Data available at https://www.cbo.gov/sites/default/files/114th-congress-2015-2016/reports/50831-
RevenueForecasting-SuppData.xlsx.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 15 of 22
Figure JFW-6: Distribution of CBO 3 -Year GDP Forecast Errors
(3 -year forward GDP forecasts made in 1982 through 2012)
4.8
12
11
4.4
0
10
4.0
a
H
9
GDP overforecastby3%or more
3.6
Ga
8
occurred in 29% of the forecasts.
3.2
0
O
7
Underforecastby3%or more
28
occurred in 16% of the forecasts.
6
2.4
5
2.0
u
=
au
u
4
1.6
o
i
a
3
The largest under -forecast,
1.2
d
D
O
2
in peak load terms (GOP
0.8
c
w1
error 0. 4), was 1.84%.
0.4
N
M
0
O.D
a
m
0
-1
-0.4
`p
L
a-2
Mean error: 0.7%over-forecast
LU
1
08
C
47
-3
-1.2
m
-4
-1.6
L°
-5
2.0
M
Source: Corgressional Budget Office, C80's Revenue Forecast Record, November 2015 [supporting data).
.�+
-6
-2.4
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%
100°fo
Cumulative Frequency
30. The distribution used in the RA Studies misrepresents the distribution of
CBO forecast errors:
a. First, the CBO forecast errors, overall, are not symmetric; there is more
over -forecasting than under -forecasting. The mean error was +0.7%
over -forecast. A bias toward over -forecasting is reflected in the CBO's
5 Congressional Budget Office, CBO's Revenue Forecasting Record, November 10, 2015, and Supplemental
Data available at https://www.cbo.gov/sites/default/files/114th-congress-2015-2016/reports/50831-
RevenueForecasting-SuppData.xlsx.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 15 of 22

forecasts and also in those of other forecasters generally (for example,
the Blue Chip Consensus of multiple forecasters, as noted in the CBO
analysis of forecast errors cited above, also exhibits a bias toward over -
forecasting).
b. Second, the large magnitude errors tend to be over -forecasting errors;
under -forecasting errors tend to be smaller. Put another way, when
economic growth is stronger than expected, the error tends to be small,
but when economic growth is weaker than expected the difference can
be more substantial. This is not surprising: economic downturns can be
sudden, largely unexpected, and sharp, as recently seen in 2008.
Surprisingly strong economic growth, by contrast, would tend to develop
and accumulate more slowly over time.
31. In contrast, the economic load forecast error distribution used in the RA
Studies has a mean of zero, and assigns 7.9% and 24% probability to under -forecasting
peak load by 4 percent and 2 percent, respectively, as described above. However, over
the thirty years of CBO data, the largest 3 -year GDP under -forecast error was 4.61
percent, which translates (times 0.4) into a load forecast under -forecast of 1.84%. Thus,
the RA Studies assign almost 32% probability to under -forecast errors whose magnitude
(+4% or +2%, in load forecast terms) never happened even once in 30 years according to
the data the distribution was purportedly based upon.
32. Consequently, even accepting the inclusion of multi-year economic
forecast errors, and accepting use of the CBO data to develop the distribution, the RA
Studies have misrepresented the distribution of errors, exaggerating the risk of
substantial under -forecasting. This exaggeration of the potential for under -forecasting
of economic load growth, in addition to the exaggeration of winter peak loads, will
further bias the planning reserve margin upward.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 16 of 22

V. RELIANCE UPON DEC AND DEP FORECASTS AS MEAN PEAK LOADS
33. The last topic of this report has to do with the interpretation of the DEC
and DEP peak load forecasts as used in the RA Studies.
34. Peak load forecasts are generally intended to be median values (that
have an equal chance of being too high or too low compared to the actual value, when it
becomes available) or perhaps mean values (representing an average or expected value
of the possible values). The RA Studies adjust the 36 load distributions such that the
average of the peak loads equals the DEC or DEP forecast for 2019, essentially using the
company forecasts as mean values (per the response to NC Public Staff 8-10).
35. However, in a recent supplemental response to a data request, DEC
states that its peak load forecast includes 540 MW of a backstand agreement with the
North Carolina Electric Membership Corporation ("NCEMC"), and that this load would
not be reflected in the historical values except when the Catawba Nuclear Station was
not in operation (responses and follow-up responses to data requests SACE 3-2 and 3-3).
Thus, DEC has apparently added to the forecasts loads associated with a wholesale
arrangement that is very unlikely to be called upon at any time. As a result, the
forecasts no longer represent median or mean values. In particular, if the likelihood or
frequency of the backstand agreement being called at peak times is 10%, the expected
value impact of this arrangement on the peak load forecast should be 54 MW, not 540
MW.
36. For the purpose of the RA Study, any such loads should have been
removed, and either treated probabilistically or replaced by the expected values. There
may be other DEC or DEP wholesale arrangements that are reflected in the forecasts
and also should have been adjusted to mean values for the purpose of the RA Studies.
This error, which may have been due to a miscommunication or misunderstanding
between Astrape Consulting and the companies, results in exaggerating the peak loads
(for DEC by at least 500 MW, and perhaps also for DEP), which will lead to higher
calculated reserve margins.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 17 of 22

VI. SUMMARY AND RECOMMENDATIONS
37. This evaluation leads to the conclusion that the recommended increases
in the DEC and DEP reserve margins are not supported by the RA Studies and are not
necessary at this time, due to the following flaws, all of which improperly inflate the
planning reserve margins:
a. The regressions used to estimate the impact of extreme cold on load
levels overstate the impact; more accurate regressions more focused on
colder temperatures suggest a much more moderate impact of extreme
cold on load.
b. The application of multiple years of economic load forecast uncertainty is
inappropriate in a model that does not represent the contingent actions
that could be taken if load grows more rapidly than expected.
c. Even accepting the application of multiple years of economic load
forecast uncertainty, the probability distribution used, based on CBO
data, misrepresents that data, and greatly overstates the risk of sharp,
unexpected increases in economic growth and load.
d. The RA Studies have assumed that the companies' load forecasts are
mean values; but at least in DEC's case, wholesale commitments have
been added to the forecast that are very unlikely to be called upon, so
the forecast is not a mean value, and the mean value is substantially
lower.
38. It is certainly appropriate to consider both summer and winter resource
adequacy for planning purposes. However, because the risk and magnitude of extreme
winter peak loads was greatly exaggerated in the RA Studies, the suggestions in the
2016 IRPs that planning should now be winter -focused (DEC 2016 IRP pp. 4-5; DEP 2016
IRP pp. 4-5) should be rejected. While resource adequacy was indeed challenged during
the polar vortex period due to both extreme loads and very poor resource performance,
the lessons learned and practices put in place since that time have addressed the
resource performance risk, as noted in the RA Studies (DEC RA Study, p. 38).
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 18 of 22

39. To the extent planning reserve margins are largely a result of brief winter
peaks that occur only under very rare extreme cold conditions, it will likely be more
economical to develop tailored peak -shaving programs focused on such events rather
than to build additional power plants to serve such rare and brief load spikes.
40. Finally, this evaluation leads to the following suggestions, for future IRPs:
a. The companies should research the drivers of sharp winter load spikes
under extreme cold conditions, and study the relationship between
extreme cold and load, to inform future resource adequacy studies.
b. The companies should also research the potential for load forecast errors
due to economic and demographic forecast errors, and the realistic
extent to which this could ultimately lead to less capacity than planned in
a delivery year, also to inform future resource adequacy studies.
c. More detailed information about the RA Studies, and thorough
validation, should be required. To start, all model reports, and a more
comprehensive set of sensitivity analyses, should be provided.
d. The companies should consider focusing on an alternative capacity
measure rather than "installed" capacity and installed reserve margin.
As the rather confusing and counter -intuitive discussions in the RA
Studies make clear (DEC RA Study pp. 3-4), with increasing amounts of
seasonal and intermittent resources such as solar on the system, the total
"installed" capacity tells little about seasonal resource adequacy, and the
required installed reserve margin for each season becomes highly
sensitive to the resource mix. More meaningful summer and winter
reserve margin measures would compare the total seasonal capacity
value of all resources, rather than total installed capacity, to the seasonal
peak load forecasts. Resources' seasonal capacity values reflect the
expected contributions to meeting seasonal peak loads, taking into
account forced outage rates, likely availability during summer and winter
peak periods (in particular, for solar), and perhaps other considerations.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 19 of 22

APPENDIX A: LIMITATIONS OF THIS REVIEW
1. Resource adequacy studies necessarily involve numerous assumptions
about loads and resources. To evaluate such a study properly requires a careful review
of the various assumptions and how they interact through the simulation to create the
study results. Of critical importance is the probabilistic representation of loads and
resources. Because the goal is to find the reserve margin to satisfy LOLE = 0.1 (one
outage event in ten years), the loss of load will occur only under extremely low -
probability combinations of load and resource conditions. Therefore, to validate such a
simulation (to gain confidence that the various assumptions are realistic in combination
and lead to realistic results) requires careful review of, among other things, the
combinations of multiple rare events that lead to the loss of load. To fully understand
and valued how the loss of load occurs, the following questions should be explored:
• When loss of load occurs, what is the day of week, hour, weather condition,
and load level?
• What conditions have combined to cause the extremely high load, if
applicable?
• Which resources are unavailable at that time and in what quantities, and why
are they unavailable? In particular, what is the state of demand response,
pumped hydro, and purchases through the interties?
These are just a few of the many questions that should be explored in a detailed
validation of a resource adequacy study.
2. In addition, a thorough review would consider the results of additional
sensitivity analyses around various assumptions, to understand the impact of the
assumptions on the results and recommendations. Sensitivity analysis will often reveal
that the results are highly sensitive to certain assumptions, which may be unrealistic and
suggest that further consideration of the particular values chosen for the assumptions is
warranted.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 20 of 22

3. However, such validation of the RA Studies could not be performed
because the required details and sensitivity analyses were not provided in response to
data requests (responses to SACE 1-9, SACE 2-18, SACE 2-26, SACE 2-27, SACE 3-4, SACE
3-18, SACE 3-19). This lack of information limited the evaluation of the RA Studies
discussed in this report.
4. Furthermore, it appears that the Astrape Consulting staff who performed
the analyses also did not complete such a validation exercise; responses to data
requests indicate that the basic model output reports that would be used in such an
effort were not even created, nor was additional sensitivity analysis performed (beyond
the few documented in the reports) (responses to SACE 3-4, SACE 3-18, SACE 3-19). The
apparent lack of basic validation of the simulation results raises concern about the
accuracy of the RA Studies and the reliability of the resulting reserve margin
recommendations.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 21 of 22

APPENDIX B: QUALIFICATIONS OF JAMES F. WILSON
James F. Wilson is an economist and independent consultant doing business as
Wilson Energy Economics, with a business address of 4800 Hampden Lane Suite 200,
Bethesda, Maryland 20814. Mr. Wilson has over 30 years of consulting experience,
primarily in the electric power and natural gas industries. Many of his consulting
assignments have pertained to the economic and policy issues arising from the interplay
of competition and regulation in these industries, including restructuring policies,
market design, market analysis and market power. Other recent engagements have
involved resource adequacy and capacity markets, contract litigation and damages,
forecasting and market evaluation, pipeline rate cases and evaluating allegations of
market manipulation. Of particular relevance to this report, Mr. Wilson recently
performed a peer review of a resource adequacy study prepared by Astrape Consulting
using the same model used in this proceeding at the request of the Eastern
Interconnection States' Planning Council (Wilson, James F., Comments on "The Economic
Ramifications of Resource Adequacy Whitepaper", prepared by Astrape Consulting for
EISPC and NARUC, March 24, 2013). Mr. Wilson's experience and qualifications are
further detailed in his CV, available at www.wilsonenec.com.
Wilson Evaluation of Duke IRP Reserve Margin Determinations Page 22 of 22