Forecasting Workshop
supporting the
DPL IRP
Economics and Forecasting Group
Pepco Holdings Inc.
September 15, 2009
1
Summary of Suggestions from Respondents
to DPL’s DE IRP
2
Stakeholder Recommendations
• Staff
• Update load and EE forecast to account for the current economic downturn and to
inform specific supply procurement needs.
• DPA
• DPL should develop an in-house long-term energy and demand requirements
forecast for its various customer classes in-house and on an annual basis.
• DPL should develop full documentation of the forecast expressly noting all
assumptions and key inputs and providing the forecasted outputs.
• DPL should conduct a more detailed analysis of key factors impacting energy and
demand. As part of its forecasting process DPL should consider economic factors
other than employment for forecasting energy consumption. In addition, DPL should
project trends in customer migration and support forecasted migration rates as part of
the assumptions to the forecast.
• DPL should test whether using different weather data for the southern portions of the
state (perhaps using Dover or Georgetown data) would better predict load for that
portion of its customers.
3
Consistency With Recently Approved IRP
Regulations
4
Required Features of the
Forecast
• Both historical data and future estimates.
• Both winter and summer peak demand for total Delmarva Delaware load and
Delmarva Delaware SOS load by customer class.
• Weather adjustments, including consideration of climate change potential.
• Five (5) year historical loads, current year-end estimate and ten (10) year
weather adjusted forecast showing individually and aggregated Delmarva
Delaware and Delmarva Delaware SOS load, and both Delmarva Delaware and
Delmarva Delaware SOS load disaggregated by customer classes, including
both capacity (MW) and energy requirements (MWh).
• Analyses of how existing and forecast Conservation, DR, DSM, Customer-Sited
Generation, various economic and demographic factors, including the price of
electricity, will affect the consumption of electric services, and how customer
choice under Retail Competition of utility service may affect future loads.
• Description of the process the Company used to develop these forecasts.
Forecasts shall include the probability of occurrence. Within the forecasting
modeling descriptions the Company shall demonstrate how well its model
predicted past load data for the prior five (5) years.
5
DPL’s Forecasting Activities
• Forecasts are updated on a specified schedule.
• Each September-December the benchmark Budget and
Planning Forecast is prepared and approved.
– Any additional forecasting work must be consistent.
– On-going monitoring for material changes.
– To support DE IRP modeling, IRP forecasts must be available by
January 1, 2010.
6
DPL Forecast Methodology
• Forecasting at DPL must also address the needs of the 1.7
million customers that PHI serves in six jurisdictions in three
states and the District of Columbia.
– The forecasts we prepare must be carefully crafted to avoid real or
apparent contradictions.
– They must also meet a very high technical standard because of
regulatory scrutiny (and the corporate need for a high quality
forecast).
– While we don’t have time to write long narratives as part of a routine,
we do insure that our forecasting work is documented, archived,
reproducible and auditable.
7
Forecast Consistency
• With such a wide range of service areas, zones and
planning needs, it is of the utmost importance that the
forecast “hang together.” Forecast components must be
“mutually confirming.”
– Growth rates will compare reasonably between history and forecast,
between jurisdictions, and against our experience.
– Implied weather normalization factors, price elasticities, income
elasticities, residential usage per customer and other metrics must
compare reasonably across our jurisdictions, with what little literature
exists on the subject, with the experience reported by other utilities
and against our own experience elsewhere.
8
Checks and Balances
• For the Delmarva zone we prepare independent forecasts of DPL Gross
Retail Output, zonal Net Sendout (NSO), and zonal MW demand.
• We expect that Billed Sales will:
– Exhibit reasonable implied price and income elasticities.
– Exhibit reasonable growth rates.
– Produce reasonable usage per customer numbers.
– The ratio of RES sales to total residential customers will be well behaved.
– The ratio of residential to commercial customers will stay within a
reasonable range.
• We expect that sales, GRO and NSO will imply reasonable losses,
consistent with our prior experience
– Define and discuss GRO, NSO, NEL and MW.
• We expect that NSO and MW demand will imply reasonable load
factors, consistent with our prior experience.
• As new information becomes available, it serves to either confirm our
outlook or cause us to re-examine our forecast.
9
Factors to Consider In
Forecasting
• DPL considers many other factors besides employment
and income.
• As a rule of thumb, employment growth determines new
customer formation (breadth of market), while real income
per household and the real price of electricity determines
customer usage (depth of market).
10
11
May 09
Jul-14
Jan-14
Jul-13
Apr 09
Jan-13
Jul-12
Jan-12
Jul-11
Mar 09
Jan-11
Jul-10
Jan-10
DE Employment Outlook
Aug 09
Feb 09
Jul-09
Jan-09
Jul-08
Jan 09
Jan-08
Jul 09
Jul-07
Jan-07
Jul-06
Dec 08
Jun 09
Jan-06
Jul-05
Jan-05
520
510
500
490
480
470
460
450
440
430
420
(1,000s)
12
Jul-14
May 09
Jan-14
Jul-13
Jan-13
Apr 09
Jul-12
DE Real Disp. Income/Employee
Jan-12
Jul-11
Jan-11
Mar 09
Jul-10
Jan-10
Jul-09
Feb 09
Aug 09
Jan-09
Jul-08
Jan-08
Jan 09
Jul-07
Jul 09
Jan-07
Jul-06
Jan-06
Dec 08
Jun 09
Jul-05
Jan-05
84
82
80
78
76
74
72
70
68
(1,000s, 2008$)
Other Types Of Data
• It often seems reasonable to include many other
kinds of data.
– We look at everything – distribute Housing Survey.
– Candidate data concepts must not contain large
measurement errors.
– Candidate data concepts must be forecastable,
and someone has to actually forecast them.
– Candidate data concepts must not contribute
unduly to forecast risk.
13
Models Don’t Forecast …
• The model doesn’t do the forecast, it’s the
forecaster.
– Good forecasters are always very aware of all of
these different pieces of information, and consider
them carefully.
– Example – DE Residential Usage.
• Discuss recent patterns of customer usage, referring back to
usage slides (p. 9).
• Discuss how difficult they are to explain.
– Technology
– Real income
– Wealth (double whammy of corporate equities and home
equities)
– Price (compare effect to implied elasticities)
14
What Has Caused the Recent
Decline in Usage?
DPL Residential - (WC Metered)
12.3
12.1
usage per customer
11.9
11.7
11.5
11.3
11.1
10.9
10.7
2002 2003 2004 2005 2006 2007 2008
15
Is an Implied Price Elasticity
of Approx. -0.12 Reasonable?
Real DPL DE Residential Prices
(Average Revenue per kWh, 2008$)
0.16
0.14
0.12
0.10
0.08
0.06
0.04
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
Note: The “price” illustrated here is the total all-in price of electricity paid by the
consumer, inclusive of all taxes and surcharges and the commodity cost of power.
16
The Choice of Weather
Stations
• The choice of weather stations and the choice of the correct weather
metrics is one of the central issues in utility sales forecasting.
– There has never been a time at DPL when this subject was not under
review.
– I have personally been involved in discussions of this subject at DPL since
1982.
• In order for weather data to be useful in analysis it must have several
characteristics.
– It must go back a long way without interruptions.
– The weather observations must be consistent over time.
• Our current practice is to use Wilmington Airport.
– In the early years, before statistical forecasting, Company engineers
watched the weather at the Wallops Island, VA Naval Air Station very
carefully, as a kind of early warning system.
– As statistical forecasting came into vogue forecasters found it useful to use
weather measured at Wallops Island and the Philadelphia Airport.
– Beginning in 2005, I verified (by inspection) that the data from the
Wilmington Airport offers more than 30 years of hourly history that appears
to be of good quality.
17
Weather Scenarios
• DPL re-estimates its weather normalization adjustments annually.
– Discuss 2008 weather normalization study.
• The DPL energy and demand forecasts are weather normalized.
– Discuss weather normalization of summer and winter peak.
• Climate change scenarios are controversial.
– Because climate change occurs slowly over time, DPL has not prepared
scenarios considering climate change potential, except for the 2008 DE IRP.
– There is a problem of interpretation in climate change scenarios – i.e., the
“forecast” becomes a bounding case.
– Discuss the 2008 Extreme Weather Scenario (this is a banding scenario).
– Discuss changes to this approach for our 2010 submission.
– One possibility would be to construct a banding scenario using weather
metrics that are +/- two standard deviations from normal.
18
Using Additional Weather
Stations
• WLM is the load center.
• Our sales data is not readily available month-to-month on a
sub-jurisdictional basis (e.g., by operating region)
– A sub-jurisdictional forecast methodology will not make the
jurisdictional forecast more accurate.
• While we believe the current jurisdictional forecast is very
accurate, we’re willing to try working with alternative
weather stations. To do so, we would like for the group to:
– Agree on a set of weather stations.
– Agree on the experiment to be conducted.
– Agree on how the results will be evaluated.
– Conduct the analytical work between November 2009 and June
2010 (allowing us to schedule the tasks).
19
Winter and Summer Class
Peaks
• Load data by customer class is generally not metered.
– It is available only from market settlements data.
– The process is to take the class contribution to load at the time of the
system peak and then allocate demand to classes according to
those shares.
• In the 2008 DE IRP filing, DPL prepared estimates of
Delmarva DE SOS load by customer class using a sharing
methodology. We will continue to prepare that study.
– Discuss the 2008 filing.
20
Customer Migration Rates
• DPL believes our projected trends in migration rates
are reasonable and defensible.
– As will be shown, migration rates are a special, difficult, case
in forecasting.
– The current forecast assumes the migration rate will equal the
observed rate in a month just prior to the preparation of the
current budget.
– The current assumption in the DE IRP is that migration rates
equal their June 2008 level.
– The subject is important and under review. A scheduled
review of the customer migration forecast will begin during the
second half of September.
21
0.005
0.010
0.015
0.020
0.025
0.030
0.035
0.040
0.045
Jan-07
Feb-07
Mar-07
Apr-07
May-07
Jun-07
Jul-07
Aug-07
RES (L)
Sep-07
Oct-07
Nov-07
Dec-07
RSH (L)
Jan-08
Feb-08
Mar-08
SL (L)
Apr-08
May-08
Jun-08
COM (R)
Jul-08
Aug-08
Sep-08
Oct-08
Customer Migration Spaghetti
IND (R)
Nov-08
Dec-08
Jan-09
Feb-09
Total (R)
Mar-09
Apr-09
May-09
Jun-09
Jul-09
22
0.300
0.400
0.500
0.600
0.700
0.800
0.900
1.000
It All Seems Horizontal
• Customer migration rates by rate class are extremely
volatile (month-to-month, even).
• There are very few cases where there are trends in the
data.
– The data are volatile (up and down).
– The lines are more or less horizontal – it’s hard to argue that these
series are trending up or down.
– With the variability in the data, it would even be hard to complete a
less naïve judgmental forecast.
23
What a Migration Model Would
Look Like
• It is a daunting task to construct a forecast model for
migration rates.
– Since the migration rates are proportions, we would prefer to specify
it as a logit model.
– Since electricity is a homogeneous commodity, we think the principle
explanatory variable would be the relative real price of electricity
between DPL and a competitive supplier.
– The only difference between the DPL price and the competitor price
is the cost of the commodity (the contribution to total price by the
wires business is the same).
– We assume that our power procurement process provides results
similar to performance of the power procurement specialists
employed by competitive suppliers.
24
What a Migration Model Would
Look Like
– We’re not able to estimate the model in this thought exercise
because we don’t have data on the cost of purchased power to the
competitive supplier. We can not calculate the explanatory variable.
– We assume that over the long run, the ratio of the two delivered
prices will vary randomly around 1.0 (equally skillful procurement).
– As long as the ratio of the two prices approximates 1.0 over the long
run, consumers should be indifferent between the providers, and the
migration rate should remain unchanged.
– If the ratio is not centered on 1.0, the migration rate will quickly go to
either 0.0% or 100%, so the observed migration rates validate the
assumption.
25
Forecasts of Energy Efficiency
• DPL considers the effects of conservation both directly and
indirectly.
– Changes in real prices lead directly to conservation.
– Changes in real incomes can lead directly to “conservation.”
– Conservation can occur for ethical reasons.
– Conservation can occur because of programs – this kind of
conservation is considered “below the line.”
• As discussed above, economic and demographic factors,
including the price of electricity, are carefully considered in
the preparation of the forecast.
– Employment tends to drive customer formation.
– Real prices affect usage.
– Real incomes per household affect usage.
26
Programmatic Goals As
Forecasts
• Programmatic goals are somewhat speculative.
– PJM does not include unless traded as credits or operator
controlled.
– Future work of SEU will be clarifying.
• How does one incorporate those programmatic goals
into the forecast?
– Some utilities are considering end-use modeling.
– Remove from forecast below the line.
27
Forecast Accuracy
• Sales variances are reviewed monthly.
• Longer term forecast accuracy requires that you wait.
• Discuss forecast accuracy measurement.
• What is a reasonable standard for accuracy?
28
Documentation to Meet Needs
• DPL agrees to prepare documentation as
required.
• DPL would like to take this opportunity to
discuss our understanding of how this
documentation should be structured.
• Consistency with Recently Approved IRP
Regulations.
• Documentation can be the most expensive
aspect of forecasting; we work to prepare the
“right” amount of documentation.
29
How the Parties Can Help
30
Suggestions and Next Steps
31
Thank You!
Any Questions?
32