|
Observers
have likened recent increases in power prices in some regions
to a "perfect storm"—a quirky confluence of tighter reserve margins,
higher prices for generation fuel stocks such as natural gas, and (dare
I say it) exploitation of short-term efficiencies by energy marketers.
These and other events have
driven retail energy prices up around the country.1 Yet, at
the same time, an environment of historically low interest rates partially
has offset the magnitude of these commodity-based increases for retail
customers by exerting downward pressure on rates for electric transmission
and distribution (T&D) services. A strong correlation between interest
rates and prices for these "wires" services implies a direct financial
impact for every customer of a U.S. investor-owned utility.
The Interest
Rate Connection
Interest rates drive retail
T&D rates in a variety of different ways, from cost-of-debt to option
valuation for hedging purposes. One of the more direct effects, and the
one that will be examined, is the interest rate captured in the allowed
return on equity (AROE) for transmission and distribution prices. T&D
prices in most jurisdictions currently are determined by cost-of-service
(COS) ratemaking and are expected to remain so into the future. The COS
method dictates that a utility can recover all its costs plus a reasonable
rate of return on invested capital.
T&D Prices
= f(Operations and Maintenance Costs + Depreciation + Taxes +
Debt Financing Expense + Allowed Profit)
Allowed Profit is a function
of the return on equity the regulator feels is appropriate for a utility's
risk profile and the amount of net capital the utility has on its books.
Regulators have several methods at their disposal when determining the
appropriate AROE for a utility. The Capital Asset Pricing Model (CAPM),
the Equity Risk Premium (ERP) method, and the Discounted Cash Flow (DCF)
technique are three commonly used approaches that regulators use for AROE
calculation. CAPM and ERP are direct functions of the prevailing interest
rates. Interest rates have a second order effect2 on the DCF
approach.
CAPM:
AROE = Treasury Interest Rate + Equity Beta (Equity Market Return -
Treasury Interest Rate)
Equity Risk
Premium: AROE = Debt Interest Rate + (Historical Equity Return - Historical
Debt Return)
DCF Method:
Asset Pricet = Dividendt+1 /(AROE - Dividend Growth
Rate)
Considering that the techniques
regulators use to determine AROE have either a direct or second-order
effect on the calculated result, one would expect to find a strong correlation
between interest rates and utilities' AROE. To test this hypothesis, a
range of interest rate series was regressed against the U.S. average AROE
for electric utilities3. Strong relationships were discovered
between the AROE and the LIBOR (London Interbank Offer Rate) five-and
10-year swaps with a one-year lag, and the five and 10-year Treasury rates
with a one-year lag.4
Financial Implications
This strong correlation implies
a financial impact for any firm that is long (e.g. a wires company) or
short (e.g. an industrial) delivery service, and is especially relevant
to retail energy marketers who naturally will accumulate large short positions
in the course of building their portfolios. To demonstrate this effect,
consider how future changes in interest rates might affect future revenues
expected from T&D service. We can calculate that affect, based on certain
assumptions, such as the average composition of an electric utility's
T&D revenue requirements, as shown in Table 3,5 and a typical
interest rate forecast, as shown in Figure 2.6
Using the calculated regression
equation for Treasury 10 with a one year lag:
AROE = 0.081867
+ 0.514980 * Treasury10t-1
The implied change in T&D rates
due to changes in interest rates, with all other things remaining equal,
can be calculated as follows:
Calculated
AROE for 2000 = 0.081867 + (.0514980 * 0.0569) = 11.12%
Calculated
AROE for 2006 = 0.081867 + (.0514980 * 0.0536) = 10.93%
The calculation is by no means
a prognostication, since interest rates are volatile and likely will have
moved from the time of writing. The purpose is to show the relationship.
The relationship implies that
a 1 percent change in the previous year's7 10-year Treasury
rate translates into a 0.25 percent average change in AROE, or a 0.05
percent change in average U.S. delivery rates.8 Each sustained
change of 0.05 percent is equivalent to a present value effect to end-users
over ten years of approximately $330 million.9
Future Expectations
There are many factors that
can move future electric transmission and distribution rates other than
interest rates. Besides the obvious variables listed in Diagram 1, there
are potential structural paradigm shifts. For example, as states continue
to mandate delivery and generation unbundling, pure wires companies should
expect to have a lower equity beta, a higher credit rating, and a higher
debt-to-equity ratio than combined wires and generation companies. All
will have a downward effect on T&D rates. Alternatively, transmission
prices could increase significantly if the substantive improvements were
made in the transmission grid.10
This analysis chose to focus
specifically on the interest rate-to-AROE relationship since it is part
of the largest component of revenue requirements, and one of the most
volatile pieces that can be quantified (i.e. versus regulatory uncertainty).
The analysis shows how one could expect allowed return on equity to change
as interest rates change and the resulting financial effect on endusers.
The positive side of this thought process is that a continuing low interest
rate environment should attenuate the effect of high commodity prices
for retail users of electricity.
David Foti works at Accenture
(formerly Andersen Consulting) and is a frequent contributor to Public
Utilities Fortnightly. Contact Mr. Foti at david.a.foti@accenture.com.
1 E.G. The Price of Primary Service increased by 11% in FRCC and 25%
in ERCOT from Feb '00 to Feb. '01. Megawatt Daily - Enron Retail
Index.
2
Higher interest rates are likely to exert downward pressure on dividend
growth rates
3
Regulatory Research Associates
4
Monthly average for Bloomberg Series US0003M; US0006M; US0012M; USSWAP2;
USSWAP5; USSWAP10; USSWAP30; GB12; GT2; GT5; GT10; GT30
5
Powerdat, 1998 FERC 1 Data for 40 IOUs
6
Bloomberg, March 29, 2001
7
Reflective of start-to-finish regulatory lag for a rate case.
8
Using a 50.5% debt-to-capitalization per aforementioned Powerdat data.
9
Using national average transmission and distribution rates from the
EIA. http://www.eia.doe.gov/oiaf/aeo/supplement/sup_elec.pdf
10
Wall Street Journal, "California Isn't the Only Place Bracing
for Electrical Shocks", 4/26/2001.
Technical
Appendix:
Regression Analysis for the Layman
|
|
Regression analysis is
a forecasting technique that identifies a relationship between a
forecasted variable and underlying explanatory variables, also known
as the dependent variable and independent variables, respectively.
The distinguishing feature of the regression method is the attempt
to identify the factors influencing a forecast, as opposed to other
techniques that focus exclusively on the computation of projected
values.
EQUATION
AND VARIABLES.
In regression models, a numeric relationship is defined through
a mathematical equation. Among other things, the equation is used
to: 1) identify the independent variables believed related to the
variable being forecasted; 2) measure the extent to which an independent
variable influences a dependent variable; and 3) summarize the mathematical
relationship between the independent variables and the dependent
variable. A linear equation is the functional form used in the interest-rate
model. The equation for a linear regression model is:
Y = a + b1X1 + b2X2
+.... bkXk + e
Where:
Y = Dependent Variable
X = Independent Variable
a = Y-Intercept
b = Coefficient
ie = Error Term
INTERCEPT
AND COEFFICIENT. Regression analysis is employed to estimate
the y-intercept and coefficient terms depicted in the preceding
equation. The coefficient measures the extent to which the dependent
variable changes in relation to a change in a given independent
variable. That is, the coefficient represents the slope of the relation
between the dependent and independent variables. Selection of independent
variables and analysis of coefficients is key to developing an effective
regression model.
DIAGNOSITIC
CHECKS. A number of diagnostic measures can be used to
evaluate coefficients in the interest-rate models. In this study,
the diagnostic measures are (1) R-squared, (2) adjusted R-squared,
(3) p value, and (4) Durbin-Watson test.
1. R-squared.
depicts the fraction of the variation in the dependent variable
(Y) explained by the regression model. In other words, R-squared
indicates how successful the model is in predicting Y relative to
the benchmark of setting the coefficients equal to zero and using
the sample mean to predict Y. R-squared is a positive value between
zero and one. A zero value implies that none of the variation is
explained by the model, while a unit value implies that all of the
variation is explained. A rule of thumb is that a "good" model should
have an R-squared value between 0.8 and 1.0. When evaluating R-squared,
it is important to note that the more data points and independent
variables used, the larger the computed R-squared value.
2. Adjusted
R-squared corrects for the data/independent variable bias inherent
with the R-squared. As such, the adjusted R-squared is useful in
comparing models based on a different number of independent variables
and/or data points. Other than the correction, the adjusted R-squared
is evaluated in the same manner as the R-squared.
3. P value
indicates the probability that the independent variable is not statistically
significant. A benchmark of 0.05 typically is used. If the P value
falls below this benchmark, then there is greater than a 95 percent
chance that the variable is statistically significant to the dependent
variable.
4. Durbin-Watson Statistic.
This term allows a diagnostic test for autocorrelation. Autocorrelation
exists in a regression equation when the error term from one time
period is dependent upon the error terms from previous periods.
Systematic variation in error terms implies that a forecast may
not properly reflect the relationship between a dependent variable
and its independent variables. Autocorrelation is not present when
the Durbin-Watson statistic is near two. A Durbin-Watson statistic
near zero implies a strong positive correlation between error terms,
while a statistic near four implies a strong negative relationship.
—D.A.F.
|
Articles found on this page are available to Internet subscribers only. For more information about obtaining a username and password, please call our Customer Service Department at 1-800-368-5001.
|
Public
Utilities Fortnightly & Spark
