_{1}

^{*}

This investigation researches how industry institutional regimes can affect the pattern and volatility of stock prices and returns. This paper searches for information signals of regulatory policy in US electric public utility company stock returns and also tests for volatility changes from the buffering effect from deregulation. Utility stock returns asymmetry in up and down markets is modeled for evidence of investor information signals of regulatory behavior. Lax regulation should lead to utility stock returns that react strongly to up markets due to weakly-constrained expected upside profits. Utility stock returns should have a small response to down markets. Stringent regulation should generate the opposite result. Since stock returns distributions typically have skewness and kurtosis, this study applies flexible probability density function (pdf) regressions methods that accommodate skewness and kurtosis. This paper concludes that since utility stock returns have a strong response to down markets relative to up markets, there is down market asymmetry in price and returns volatility. This evidence suggests that investors perceive that utility profit regulation is stringent. It also suggests, surprisingly, that the buffering effect has been increased with deregulation. Lastly, robust estimation of financial models performed herein shows that regression estimation should not assume a normally distributed error term.

The electric utility industry, once an industry that supplied a steady stream of dividends and slow earnings growth, was the subject of investor and government attention due to continuing crises that commenced soon after deregulation in the late 1990’s. During the years after deregulation, California’s electricity industry collapsed, and ENRON, the largest bulk electric power marketer, collapsed, and two of the three investor-owned public utilities in California declared bankruptcy. Additionally, $238 billion market capitalization was destroyed (Seeholzer [^{1} of regulation leads to reforms toward deregulation that are associated with these events. The industry is partially deregulated as the generation component of the business is unregulated. Additionally, utilities’ bulk power lines or transmission lines had been opened to alternative purveyors of electric power. The electric utility industry remains substantially regulated as the rates of return on all but generation assets are regulated throughout the US.

Utility regulatory policy from a shareholder’s perspective can be viewed as stringent (unfavorable) or lax (favorable). There is no consensus on the characterization of utility regulatory policy in the literature and among practitioners.

This paper searches for information signals of investors on their perspectives toward regulatory policy in electric utility company stock price volatility soon after the deregulation of the US electric public utility industry. Additionally, utility stock returns responses to up and down markets are modeled for evidence of investor information signals of regulatory behavior. Lax regulation should lead to utility stock returns that react strongly to up markets due to weakly-constrained expected upside profits. Utility stock returns should have a small response to down markets with lax regulation. Lax regulation results in earnings levels that have a strong floor that is independent of systematic risk associated with business and earnings cycles. Stringent regulation should generate a large response of utility stock returns to down markets and a small or no response to up markets for the opposite reasons.

Stock returns data and their error distributions usually have skewness and leptokurtosis, or, thick tails, which is well established in the literature. These characteristics cause intercept bias and inefficiency in slope estimates which is discussed in this paper. Therefore, this study applies skewed generalized t (SGT) probability density function (pdf) regression methods that accommodate skewness and kurtosis. Although there are many flexible pdfs that accommodate skewness and thick tails, the SGT has been found to be the pdf that has the best fitting regressions relative to the normal, other pdfs as well as the flexible pdfs for modeling stock returns.

Some of the literature suggests, albeit not strongly, that regulatory policy had been lax, as suggested by Binder and Norton [

Some industry observers and utility management have concluded that regulation has been stringent. They point to unattractive financial performance and the resulting lack of investment due to uncertainty regarding regulatory treatment of such assets.

Nwaeze [

When general corporate earnings levels are expected to rise, due to an upswing in the business cycle, stock returns rise in such anticipation and an up market occurs. The opposite occurs for down markets. Utility earnings and returns behave in a generally similar manner except that profits may have a strong upside constraint and an associated weak downside constraint, if regulation is stringent. The converse is true with lax regulation. Therefore, one should expect utility stock returns to be relatively more responsive to either down or up markets, depending upon regulatory policy.

Fifty states and the District of Columbia have regulatory jurisdiction of utilities’ retail electric business within their boundaries. Therefore, regulatory policy varies across jurisdictions. Additionally, the US Federal Energy Regulatory Commission regulates the returns on utilities’ wholesale (transmission) assets. In a study sponsored by the US industry trade association, the Edison Electric Institute, Hirst and Kirby [

During down markets, corporate profits are expected to fall. Falling expectations for utility profits may obtain but would be dampened under a lax regulatory or moderate (neither lax nor stringent) regulatory regime as utility profitability has a legally defined guideline found in early 20th century case laws of the Hope and Bluefield decisions.^{2} These cases are the standards referred in utility rate cases used for setting allowed rates of returns on common equity. This contrasts with expectations for unconstrained downturns in profits and consequent stock returns for un-regulated firms and markets in general.

This study tests for asymmetry in up and down market utility stock returns as evidence of regulatory policy as perceived by investors soon after deregulation occurred. The change in capital asset market beta (CAPM) associated with the advent of wholesale electric deregulation is estimated and controlled for in the regressions to ascertain how deregulation affected the buffering effect discussed below.

Bawa, Brown, and Klein [

Turning attention to regulation and systematic risk, Peltzman [

The buffering hypothesis has been subject to a number of empirical tests. Davidson, Rangan, and Rosenstein [

The buffering hypothesis was also tested by Binder and Norton [

Binder and Norton [

Nwaeze [

Besanko, Dsouza, and Thiagarajan [

Turning to robust estimation of models involving stock returns, this study empirically addresses non-normality of stock returns and associated capital asset pricing model (CAPM) regression error distributions. The distributions of stock returns and their regression errors have both leptokurtosis (thick tails relative to the normal pdf) and skewness. The importance of regression methods that accommodate kurtosis and skewness lies in the bias of the intercept and the inefficiency of the estimates and their resulting statistical tests. Chan and Lakonishok [

The regression methods used to estimate the models herein use flexible probability density functions that accommodate skewness and leptokurtosis to search for the best fit of the data. Robust estimation of CAPM beta is discussed in Butler, McDonald, Nelson, and White [

Although this investigation has performed the estimations with all of the above mentioned pdf’s along with the symmetric distributions such as the normal, Laplace and the centralized t, the SGT is the pdf chosen for the final estimates, which is highly consistent with much of the aforementioned literature. The SGT estimates are the only set of presented estimates for the sake of brevity. The results of the other pdf’s are available upon request.

The estimation method herein applies robust estimation methods that nests the family of symmetric distributions yet are more general and flexible as they accommodate skewness and kurtosis in the pdf returns data and regression residuals.

The remainder of this paper is organized as follows. Section 3 discusses the returns response models. Section 4 reviews the robust probability density functions for estimation. Section 5 presents the data and estimation results. Section 6 is the conclusion.

The model for estimating asymmetric response of utility stock returns to the market is the asymmetric response model initially developed by Bawa, Brown, and Klein [

R i , t = α i + β 1 R m , t − + β 2 R m , t + + π i δ t + μ i , t (1)

where R i , t = r i , t − r f , t is excess return for the stock of utility i at period t, r_{i,t} is the total return on the stock of utility i as defined in more detail below, r_{f,t} is the risk free rate for period t, R m , t = r m , t − r f , t is the excess market return, r_{m,t} is the total market return for period t; R m , t − = r m , t − r f , t When r m , t − r f , t < 0 and zero otherwise, and R m , t + = r m , t − r f , t when r m , t − r f , t > 0 . And zero otherwise, and δ_{t} = 1 when r m , t − r f , t > 0 and zero otherwise,

π i = φ ( β 1 − β 2 ) , φ = E { r m , t − r f , t | r m , t − r f , t > 0 }

and μ i , t is an error term.

The asymmetric responses of stock returns to down and up markets are estimated by tests for difference between β_{1} and β_{2}.

The model for estimating the returns response to the market and deregulation is:

R i . t = α i + α n , i D n , t + α r , i D r , t + ( β i + β i , n D n , t + β i , r D r , t ) R m , t + μ i , t (2)

R m , t − = R m , t for all negative R_{m}_{,t}’s and zero otherwise; R m , t + = R m , t for all non-negative R_{m,t}’s and zero otherwise; D_{n,t} = 1 for all negative R_{m,t}’s and zero otherwise, and D_{r,t} = 1 for the post-deregulation period (after April 1996) and zero otherwise. ì_{i,t} is the error term. This model is the same as Equation (1) discussed above except for the augmentation of the model to include the slope and intercept post-deregulation dummy variables to estimate the buffering effect (see Appendix 2).

Although reform toward introducing competition to the electric utility industry is a process that started in 1978, this paper identifies the issuance of Order 888 on April 24, 1996 by the Federal Energy Regulatory Commission to be the key event in moving the industry toward competition. The main trend toward deregulation occurred between 1990 and 2002 and therefore is the time frame used for data observation. Next the development of the stock total returns for the utilities is discussed.

The utility total stock return, r_{i}, for period t − 1 to t is

r t = Δ P t + D t P t − 1 = ( P t + D t − P t − 1 P t − 1 ) = ( P t + D t P t − 1 ) − 1

where Δ P t = P t − P t − 1 is the price change of the share from time t − 1 to t, and D_{t} is the dividend for the period paid at time t.^{3} The return can be decomposed into its two components, the capital gain and the dividend yields as follows:

1 + r t = P t + D t P t − 1 = ( P t P t − 1 ) ( P t + D t P t ) = ( 1 + Δ P t P t ) ( 1 + D t P t ) = ( 1 + k t ) ( 1 + d t )

where k t = Δ P t / P t − 1 is the capital gain yield or the percent of growth of the stock price from time t − 1 to t and d t = D t / P t is the percent of dividend with respect to the price at the time of payment.

The log-return or continuously compounded return for the period is defined as

l r t = ln ( P t + D t P t − 1 ) = ln ( P t P t − 1 ) + ln ( P t + D t P t ) = ln ( 1 + r t ) = ln ( 1 + k t ) + ln ( 1 + d t ) = l k t + l d t

where l k t = ln ( P t / P t − 1 ) = ln ( 1 + k t ) is the log-capital gain yield and l d t = ln ( 1 + D t / P t ) = ln ( 1 + d t ) is log-dividend yield for the stock. Note that the following relationships exist between the return and log-return measures

r t = e l r t − 1 , k t = e l k t − 1 and d t = e l d t − 1.

The return and log-return for the multi period 0 to s, denoted by r_{0,s} and lr_{0,s}, can be computed from the single period returns using the formulas

1 + r 0 , s = ( 1 + r 1 ) ( 1 + r 2 ) ⋯ ( 1 + r s ) = ∏ t = 1 s ( 1 + r t )

and

l r 0 , t = ln ( 1 + r 0 , s ) = ln ∏ t = 1 s ( 1 + r t ) = ∑ t = 1 s ln ( 1 + r t ) = ∑ t = 1 s ln ( 1 + k t ) + ∑ t = 1 s ln ( 1 + d t ) = ∑ t = 1 s l k t + ∑ t = 1 s l d t = l k 0 , s + l d 0 , s

The above formulas assume that the dividend is reinvested each time in the stock of the company immediately after its payment. In such a case, the value of an investment in the stock s periods from now will be

V s = P 0 ( 1 + r 0 , s ) = P 0 exp ( ln ( 1 + r 0 , s ) )

P 0 exp ( ∑ t = 1 s l k t + ∑ t = 1 s l d t ) = P 0 exp ( ∑ t = 1 s l r t )

Given a sequence, V 0 = P 0 , V 1 , V 2 , ⋯ , V s , ⋯ , the return and log-return for each period can be computed easily from them using the equations

r t = V t V t − 1 − 1 and l r t = ln ( V t V t − 1 )

The flexible pdf’s used for the estimations accommodate skewness and kurtosis. Although not presented in this paper, the normal, centralized t, Laplace, and the flexible pdf’s, exponential beta of the second kind, (EGB2), skewed generalized error distribution (SGED), skewed generalized t (SGT) and the inverse hyperbolic sine (IHS) were used for the regression estimations for comparison and robustness checks. The normal pdf is not appropriate when the error term exhibits a distribution with skewness or thick tails and no symmetric pdf is appropriate when skewness is present in the data or error term. The shape of the normal pdf is completely described by the first and second moments, namely mean and variance. Pdf’s that accommodate flexible values for the third and fourth moments, skewness and kurtosis offer efficient estimates for financial models relative to OLS where skewness or kurtosis is present. Harvey and Siddique [

Hein and Westfall [

A description of the SGT pdf as well as the other flexible pdf’s considered for the final estimations are found in Hansen, McDonald, and Theodossiou [^{4}

The SGT pdf is specified as:

S G T ( y ; μ , σ , n , k , λ ) = C σ ( 1 + 1 ( ( n − 2 ) / k ) ( 1 + s i g n ( ε ) λ ) k θ k σ k | ε | k ) − ( n + 1 ) / k (3)

where

C = k / ( ( 2 ( n − 2 ) / k ) 1 / k θ B ( 1 / k , n / k ) ) ,

θ = ( k / ( n − 2 ) ) 1 / k B ( 1 / k , n / k ) 0.5 B ( 3 / k , ( n − 2 ) / k ) 0.5 S ( λ ) − 1 ,

S ( λ ) = 1 + 3 ( λ ) 2 − 4 A 2 λ 2 ,

A = B ( ( 2 / k , ( n − 1 ) / k ) ) B ( 1 / k , n / k ) − 0.5 B ( 3 / k , ( n − 2 ) / k ) 0.5 ,

δ = 2 λ A S ( λ ) − 1 ,

B( ) is the beta function, μ and σ are the mean and standard deviation of y, n and k are kurtosis parameters, λ is a skewness parameter obeying the constraint | λ | < 1 , sign is the sign function, and μ = y − μ + δσ is the deviation of y from its mode μ − δσ. Positive values of λ result a positively skewed pdf and vice versa. Note that the parameter δ is the skewness measure SK = (μ − mode(y))/σ. For λ = 0 the SGT yields the GT of McDonald and Newey [

The asymmetric response models were estimated using maximum likelihood estimation.

The sample includes all electric and electric and gas combination companies listed in Compust at database as of September 26, 1996 that were publicly traded between January 1, 1990 and July 31, 2002. These include all publicly-traded companies with SIC codes 4911 and 4931.

Opening, high, low, and close daily stock prices were obtained for the period beginning the first trading day in January 1990 and ending on December July 31, 2002 from www.msn.com. The time frame of data observation from 1990 to 2002 was purposely used as that was the transition period from regulation to deregulation in the generation portion of the US electric utility industry. Utility company paid dividends were obtained from www.msn.com. There are 3026 daily total return observations for each utility and the S & P 500 total returns index. Value Line Investment Survey [

The sample frame period had been an era of mergers and acquisitions as electric utilities consolidated as a response to growing competition and deregulation in the generation portion of the electric utility industry. During the mid 1980’s there were roughly 110 publicly traded electric and electric and gas combination utilities and as of the 2^{nd} quarter of 2002 there were 63, approaching 50 percent industry consolidation in the number of traded utilities.

Market and utility returns are daily total stock returns. The market is defined by the S & P 500. The total returns index for the S & P 500 was obtained from http://www.globalfinancialdata.com/. Utility stock total returns were calculated as described above. Total returns (rather than price changes only) were used for all analysis since utilities’ dividend yields are relatively high and dividend payments consistent, therefore dividends are a substantial component of investor returns. The inclusion of dividends is also important since dividend yields dampen reductions in total returns when stock prices decline. As the stock price falls, the dividend yield rises. The risk-free rate is the daily 30-day yield on 90 day US Treasury Bills obtained from Global Financial Data, Inc.’s web site, www.globalfinancialdata.com.

Consistently positive and large skewness for utilities compared to the S & P 500 indicates that utilities’ total returns have a higher probability of being large and positive compared to the S & P 500. Kurtosis is similar for utilities with a utility mean of 11.21 (median is 7.74) compared to the S & P 500 kurtosis of 7.93. Kurtosis for the normal pdf is 3. The kurtosis of the utilities and the S & P 500 provide evidence that the pdf’s of their returns are leptokurtic.

As an initial test of the normality of the regression residuals, the standard econometric version of the CAPM was estimated (with excess returns) using the flexible pdf’s, the centralized T, and Laplace with maximum likelihood, and the normal pdf using OLS. Although it is beyond the scope of this paper to present and discuss many of the inter-pdf characteristics of the results, the SGT pdf was chosen as the main choice of estimation among the flexible pdf’s as discussed in footnote 5. The SGT nests the normal pdf as a special case therefore comparison and testing of the SGT against the normal can be done with a chi-squared distributed likelihood ratio test. The model estimations are discussed in the next subsection.

^{2} distributed likelihood ratio test (LR-Normal) for comparing the fit of the normal and the SGT pdf’s is statistically significant for every utility estimate, indicating that the SGT regressions provide a better fit of the data. The skewness (SK) and kurtosis (KU) parameters for the CAPM regression error terms indicates that the distribution of the returns residuals is non-normal. Skewness is not defined for the SGT pdf for all pdf parameter ranges (at n < 3) therefore a few utilities do not have an estimate for skewness. Although most of the estimates of the skewness parameter are positive, they are generally quite small. The skewness parameter estimates of the SGT, λ, are positive, only three of them are statistically significant.

μ | σ | Skewness | Kurtosis | |
---|---|---|---|---|

S & P 500 | 0.04214 | 1.02090 | −0.03802 | 7.93318 |

AEP | 0.03092 | 1.28463 | 0.08322 | 7.43292 |

CIN | 0.04791 | 1.28907 | 0.11355 | 6.55949 |

CMS | 0.00373 | 1.72070 | 0.06856 | |

D | 0.05308 | 1.17777 | 0.06159 | 9.73302 |

DPL | 0.06195 | 1.37598 | 0.14146 | 6.66801 |

DTE | 0.05000 | 1.26403 | 0.10753 | 5.91875 |

DUK | 0.05085 | 1.44103 | 0.12134 | 7.29301 |

ED | 0.04741 | 1.30253 | −0.04474 | 5.97970 |

EDE | 0.04039 | 1.24469 | 0.12190 | 7.97978 |

EIX | 0.01303 | 1.75458 | 0.04157 | 14.51847 |

ETR | 0.06013 | 1.45669 | 0.23370 | 17.75511 |

FPL | 0.05005 | 1.26054 | 0.01300 | |

GMP | 0.00989 | 1.38778 | 0.03994 | 14.18595 |

HE | 0.03362 | 1.06177 | 0.10635 | 7.19879 |

IDA | 0.02748 | 1.27363 | 0.08635 | 8.98244 |

NI | 0.05315 | 1.31132 | 0.14622 | 7.72411 |

NU | 0.03773 | 1.48852 | 0.09106 | 7.61698 |

OGE | 0.06398 | 1.32225 | 0.18221 | 8.15353 |

PCG | 0.00727 | 1.69361 | 0.02540 | 15.45331 |

PEG | 0.04365 | 1.33727 | −0.17516 | 34.07430 |

PGN | 0.03804 | 1.25055 | 0.15202 | 48.75561 |

PNM | 0.02811 | 1.76315 | 0.05144 | 6.73357 |

PNW | 0.05630 | 1.56488 | 0.14052 | 8.96838 |

POM | 0.02646 | 1.33289 | 0.07297 | 7.76233 |

PPL | 0.04909 | 1.44138 | 0.13619 | 8.84195 |

PSD | 0.03719 | 1.22748 | 0.10130 | 7.09781 |

SCG | 0.04488 | 1.12971 | 0.14018 | 7.51896 |

SO | 0.06908 | 1.37276 | 0.18143 | 18.41280 |

SRP | 0.06473 | 1.52926 | 0.25138 | 14.18762 |

TE | 0.04298 | 1.22600 | 0.11201 | 6.87562 |

TXU | 0.05537 | 1.30845 | 0.20068 | 18.78902 |

UIL | 0.04499 | 1.10982 | 0.11882 | 6.25387 |

UTL | 0.03428 | 1.46703 | 0.08213 | 7.35301 |

WEC | 0.03727 | 1.18433 | 0.08613 | 5.89381 |

WPS | 0.04212 | 1.05327 | 0.11306 | 6.02368 |

WR | 0.02219 | 1.34276 | 0.06690 | 8.66620 |

Util. Mean | 0.041092 | 1.354225 | 0.099228 | 11.21663 |

Util. Median | 0.043315 | 1.316785 | 0.106940 | 7.74322 |

Data | AEP | CIN | CMS | D | DPL | DTE | DUK | ED | EDE | EIX | ETR | FPL |
---|---|---|---|---|---|---|---|---|---|---|---|---|

α | 0.01264 (0.58) | 0.03244 (1.46) | −0.00834 (−0.29) | 0.04115 (2.03)* | 0.05275 (2.22)* | 0.03828 (1.77) | 0.03128 (1.26) | 0.03012 (1.35) | 0.03244 (1.49) | 0.00473 (0.15) | 0.04626 (1.82) | 0.03469 (1.62) |

β | 0.37872 (22.89)** | 0.36077 (21.43)** | 0.28526 (13.94)** | 0.35567 (23.7)** | 0.33977 (18.13)** | 0.41714 (24.75)** | 0.42805 (23.91)** | 0.41623 (22.8)** | 0.15795 (9.83)** | 0.47156 (22.42)** | 0.31852 (15.98)** | 0.33321 (20.78)** |

σ ^ | 1.23089 (36.98)** | 1.22791 (44.86)** | 1.7489 (16.4)** | 1.15796 (25.33)** | 1.32319 (39.91)** | 1.20056 (39.89)** | 1.37718 (40.83)** | 1.24393 (43.33)** | 1.2053 (43.98)** | 2.52876 (5.28)** | 1.43755 (35.53)** | 1.22625 (26.4)** |

K | 1.57664 (11.89)** | 1.3988 (10.66)** | 2.13438 (11.04)** | 1.73995 (10.61)** | 1.86872 (12.29)** | 1.80353 (11.3)** | 1.35557 (12.36)** | 1.94618 (9.78)** | 1.38038 (16.15)** | 2.09635 (11)** | 1.92125 (11.75)** | 1.9214 (10.78)** |

λ | 0.01993 (0.88) | 0.02316 (1.08) | 0.01796 (0.7) | 0.01808 (0.78) | 0.05783 (2.34)* | 0.0368 (1.52) | 0.00742 (0.35) | 0.00132 (0.05) | 0.0375 (1.71) | −0.00666 (−0.27) | 0.04697 (1.9) | 0.00514 (0.21) |

n | 4.7153 (6.06)** | 8.14022 (2.68)** | 2.83973 (11.66)** | 3.8967 (6.11)** | 4.46409 (7.78)** | 4.8989 (5.85)** | 6.1737 (4.48)** | 5.4031 (4.51)** | 5.91206 (7.83)** | 2.31309 (13.38)** | 4.04235 (8.17)** | 3.65427 (7.51)** |

SK | 0.12846 | 0.10119 | n/a | 0.16105 | 0.33563 | 0.189 | 0.04159 | 0.0054 | 0.21373 | n/a | 0.33541 | 0.05259 |

KU | 16.74093 | 6.87491 | n/a | n/a | 18.00594 | 11.2943 | 10.03874 | 7.53358 | 10.57564 | n/a | 157.40163 | n/a |

LogL | −4631.58 | −4731.1 | −5321.66 | −4384.13 | −4915.63 | −4650.79 | −4982.82 | −4827.62 | −4574.69 | −5524.25 | −5122.28 | −4562.86 |

LR-N. | 626.77** | 356.77** | 1376.83** | 570.44** | 555.16** | 387.37** | 603.04** | 219.9** | 1064.3** | 2342.45** | 617.93** | 625.12** |

R2 | 0.08796 | 0.08452 | 0.02661 | 0.09582 | 0.06226 | 0.11384 | 0.09067 | 0.10726 | 0.01501 | 0.04539 | 0.04814 | 0.07417 |

OBS | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 |

Data | GMP | HE | IDA | NI | NU | OGE | PCG | PEG | PGN | PNM | PNW | POM | ||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|

α | 0.01708 (0.63) | 0.02526 (1.38) | 0.01703 (0.77) | 0.04233 (1.86) | 0.02871 (1.1) | 0.05375 (2.35)* | 0.02504 (0.78) | 0.02812 (1.24) | 0.01937 (0.92) | 0.01434 (0.48) | 0.04156 (1.54) | 0.01162 (0.49) | ||||||

β | 0 (0) | 0.25115 (17.8)** | 0.28275 (17.02)** | 0.2745 (16.28)** | 0.26276 (13.39)** | 0.26909 (15.83)** | 0.45387 (23.01)** | 0.39926 (22.84)** | 0.40404 (24.81)** | 0.36322 (15.33)** | 0.28999 (14.04)** | 0.33895 (18.49)** | ||||||

σ ^ | 1.51031 (37.36)** | 1.03229 (31.91)** | 1.28364 (24.93)** | 1.27692 (36.04)** | 1.48315 (29.76)** | 1.29844 (32.69)** | 3.70063 (33.23)** | 1.27678 (33.74)** | 1.19715 (40.24)** | 1.70006 (40.4)** | 1.54067 (35.22)** | 1.32433 (34.1)** | ||||||

K | 0.89041 (12.15)** | 1.7801 (10.69)** | 1.89576 (11.65)** | 1.61926 (12.38)** | 1.7651 (11.54)** | 1.64702 (12.09)** | 2.59563 (14.12)** | 1.72383 (10.76)** | 1.86959 (16.81)** | 2.20637 (11.67)** | 1.91113 (12.84)** | 1.72963 (10.93)** | ||||||

λ | 0.00893 (0.53) | 0.01915 (0.81) | 0.00958 (0.4) | 0.03129 (1.37) | 0.03459 (1.47) | 0.04544 (1.98)* | 0.02206 (0.85) | −0.01319 (−0.57) | 0.03266 (1.31) | 0.06899 (2.62)** | 0.02404 (0.98) | 0.02423 (1.03) | ||||||

n | 8.47444 (3.16)** | 4.29366 (5.9)** | 3.40655 (9.16)** | 4.45892 (6.83)** | 3.9108 (7.18)** | 4.14196 (7.16)** | n/a | 4.5781 (5.38)** | 3.82797 (20.82)** | 4.09302 (9.09)** | 3.80983 (10.07)** | 4.43101 (5.87)** | ||||||

SK | 0.07244 | 0.1281 | 0.14737 | 0.21603 | 0.29957 | 0.36192 | n/a | −0.08038 | 0.28550 | 0.41152 | 0.20891 | 0.15657 | ||||||

KU | 15.25315 | 28.4597 | n/a | 23.03044 | n/a | 65.12891 | n/a | 16.91787 | n/a | 60.11703 | n/a | 21.4893 | ||||||

LogL | −4800.23 | −4120.4 | −4617.66 | −4727.17 | −5144.01 | −4738.13 | −5421.32 | −4782.28 | −4512.1 | −5696.39 | −5286.74 | −4876.01 | ||||||

LR-N. | 1763.69** | 468.92** | 938.52** | 710.16** | 665.23** | 766.09** | 2532.76** | 449.98** | 18181.1** | 519.5** | 1046.42** | 503.67** | ||||||

R2 | 0 | 0.05861 | 0.04686 | 0.04484 | 0.03197 | 0.04206 | 0.04261 | 0.09575 | 0.00037 | 0.04413 | 0.03124 | 0.06583 | ||||||

OBS | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | ||||||

Data | PPL | PSD | SCG | SO | SRP | TE | TXU | UIL | UTL | WEC | WPS | WR | ||||||

α | 0.03794 (1.51) | 0.0281 (1.32) | 0.03463 (1.75) | 0.05432 (2.3)* | 0.01638 (0.58) | 0.03161 (1.48) | 0.04145 (1.83) | 0.03472 (1.81) | 0.03334 (1.07) | 0.02829 (1.36) | 0.03052 (1.69) | 0.00937 (0.39) | ||||||

β | 0.32879 (18.5)** | 0.26917 (16.52)** | 0.27145 (19.03)** | 0.38496 (20.61)** | 0.27807 (15.09)** | 0.26518 (16.21)** | 0.3327 (18.89)** | 0.25387 (16.27)** | 0 (0) | 0.29023 (17.65)** | 0.27091 (18.88)** | 0.25373 (14.76)** | ||||||

σ ^ | 1.44237 (25.6)** | 1.1919 (36.73)** | 1.10238 (33.34)** | 1.34322 (41.04)** | 1.90986 (8.62)** | 1.17319 (43.16)** | 1.28233 (27.04)** | 1.06128 (49.29)** | 1.50574 (32.71)** | 1.15014 (38.41)** | 0.99239 (52.91)** | 1.4101 (22.38)** | ||||||

K | 1.57153 (11.37)** | 1.68135 (11.47)** | 1.40588 (10.74)** | 2.0868 (13.57)** | 1.96469 (11.25)** | 1.39883 (10.76)** | 2.03412 (10.23)** | 1.67291 (13.42)** | 0.99411 (12.11)** | 1.88553 (10.62)** | 1.55622 (12.9)** | 1.65203 (11.1)** | ||||||

λ | 0.04057 (1.83) | −0.00542 (−0.23) | 0.02084 (0.98) | 0.04126 (1.6) | 0.01396 (0.58) | 0.02309 (1.07) | 0.02145 (0.85) | −0.00447 (−0.19) | 0.01642 (0.73) | 0.05015 (2.02)* | 0.04414 (1.93) | 0.00421 (0.19) | ||||||

n | 3.95257 (6.2)** | 4.67901 (5.89)** | 5.41577 (3.91)** | 3.97617 (12.14)** | 2.51462 (11.78)** | 7.98421 (2.66)** | 3.81108 (6.78)** | 6.2272 (6.02)** | 12.34229 (1.58) | 4.88652 (5.3)** | 8.61146 (3.75)** | 3.61687 (6.81)** | ||||||

SK | 0.38868 | −0.03273 | 0.12786 | 0.28395 | n/a | 0.10206 | 0.17479 | −0.01885 | 0.09581 | 0.24506 | 0.1615 | 0.05299 | ||||||

KU | n/a | 15.52784 | 12.43875 | n/a | n/a | 6.98832 | n/a | 7.13074 | 9.1353 | 10.72751 | 5.71448 | n/a | ||||||

LogL | −4984.19 | −4571.21 | −4283.62 | −4946.5 | −5125.39 | −4589.18 | −4764.81 | −4324.27 | −4143.2 | −4539.15 | −4148.69 | −4871.91 | ||||||

LR-N. | 747.67** | 510.65** | 562.61** | 1770.18** | 1909.21** | 356.45** | 435.13** | 411.35** | 560.08** | 309.54** | 261.32** | 782.88** | ||||||

R2 | 0.05309 | 0.05044 | 0.06017 | 0.05387 | 0.02391 | 0.05059 | 0.06869 | 0.05425 | 0 | 0.06336 | 0.07138 | 0.0341 | ||||||

OBS | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 3026 | 2467 | 3026 | 3025 | 3026 | ||||||

The results in

The asymmetric model estimation results for the SGT are in

Utility | β i , n = ( β i − − β i + ) | LR-Normal | Regression Error Term | |
---|---|---|---|---|

Skewness | Kurtosis | |||

AEP | 0.63591** | 644.99** | 0.06714 | 20.09796 |

CIN | 0.47909** | 345.79** | 0.14922 | 6.78236 |

CMS | 0.33106** | 1383.21** | ||

0.42911** | 566.34** | 0.13054 | ||

DPL | 0.41650** | 553.43** | 0.45028 | 42.55113 |

DTE | 0.65440** | 390.64** | 0.16753 | 17.70259 |

DUK | 0.62964** | 609.71** | 0.08367 | 10.97236 |

ED | 0.74933** | 216.75** | 0.00737 | 7.20859 |

EDE | −0.00120 | 1081.50** | 0.26244 | 11.34728 |

EIX | 0.64973** | 2343.33** | ||

ETR | 0.51522** | 631.41** | 0.28481 | |

FPL | 0.52046** | 631.53** | 0.20061 | |

GMP | 0.00113 | 1901.47** | 0.11253 | 23.29952 |

HE | 0.31769** | 459.53** | 0.12574 | 20.51400 |

IDA | 0.34791** | 933.80** | 0.17406 | |

NI | 0.36974** | 704.70** | 0.26356 | 22.2486 |

NU | 0.34370** | 670.02** | 0.47053 | |

OGE | 0.36356** | 763.45** | 0.41151 | 55.11190 |

PCG | 0.67678** | 2558.41** | ||

PEG | 0.61655** | 461.34** | −0.17928 | 40.07774 |

PGN | 0.65285** | 18,117.66** | 0.37742 | |

PNM | 0.42358** | 517.25** | 0.44438 | 105.94479 |

PNW | 0.48876** | 1037.71** | 0.31613 | |

POM | 0.38260** | 498.89** | 0.25469 | 30.87543 |

PPL | 0.32531** | 736.91** | 0.39597 | 720.96758 |

PSD | 0.33936** | 509.28** | −0.04004 | 20.7707 |

SCG | 0.28571** | 559.55** | 0.20312 | 14.26283 |

SO | 0.62606** | 1817.25** | 0.32385 | |

SRP | 0.29644** | 1893.73** | ||

TE | 0.34318** | 345.94** | 0.15439 | 8.06323 |

TXU | 0.53978** | 439.85** | 0.03524 | |

UIL | 0.22461** | 407.47** | −0.01053 | 7.24365 |

UTL | 0.00032 | 659.85** | 0.20288 | 26.12142 |

WEC | 0.33071** | 300.99** | 0.29048 | 12.58194 |

WPS | 0.31888** | 256.06** | 0.19964 | 5.60454 |

WR | 0.34982** | 787.05** | 0.20040 |

**Statistically significant at 99.99%. Missing skewness and kurtosis indicate that they are not defined for the SGT and do not exist. The results above are estimates of the model (Equation (2)) as shown below using the SGT density: R i . t = α i + α n , i D n , t + α r , i D r , t + ( β i + β i , n D n , t + β i , r D r , t ) R m , t + μ i , t .

^{2} distributed with three degrees of freedom. The χ^{2} tests show that the SGT provides a better fit for the regressions than OLS. All χ^{2} statistics are significant at the 99.99 percent level. Note once again that the regression residuals’ skewness is generally positive and the kurtosis estimate indicates leptokurtic pdf’s for the residuals.

If there were no asymmetry in utility stock returns responses in down and up markets, the coefficient would be insignificant.

Turning to the impact of deregulation on systematic risk, _{i,r}) period shows a significant and consistent reduction in beta for 29 of 36 utilities and all of these slopes are significant at a minimum 95% level of significance.

The buffering impact is not supported by these results. An inspection of the variance of market returns in pre- and post-periods suggest that the volatility of the market rose more relative to utility returns to result in declining betas. Since 1996, the market had been ravaged at that time with problems such as the technology sector meltdown, a recession in the early 2000’s, and investor confidence problems associated with corporate governance. This would result in a lower level of systematic risk for the utilities if the correlation coefficient remains the same but the ratio of the standard deviations of the utility stock to market returns falls. This can be seen by inspection of the beta:

β i = ρ i , m σ i σ m ρ m 2 = ρ i , m σ i σ m

where ρ_{i,m}= correlation coefficient of the utility stock i and market returns, σ_{i} is the standard deviation of utility i’s returns and σ_{m} is the market returns standard deviation. The σ_{m} for the S & P 500 daily total return was 0.723 for the pre-deregulation period and was 1.268 for the post deregulation period.

Consistent with Blinder and Norton [

Symbol | 1990-02 | Pre-96 | Post-96 | Δ β | Dereg. Slope ( β r , i ) |
---|---|---|---|---|---|

AEP | 0.3767** | 0.5958** | 0.2676** | −0.3327 | −0.3319** |

CIN | 0.3620** | 0.4803** | 0.2859** | −0.1936 | −0.1957** |

CMS | 0.2860** | 0.4446** | 0.2299** | −0.2167 | −0.2090** |

D | 0.3550** | 0.4671** | 0.2870** | −0.1806 | −0.1700** |

DPL | 0.3403** | 0.4836** | 0.2786** | −0.2059 | −0.1892** |

DTE | 0.4145** | 0.6817** | 0.3024** | −0.3795 | −0.3594** |

DUK | 0.4259** | 0.6035** | 0.3203** | −0.2851 | −0.2865** |

ED | 0.4161** | 0.7617** | 0.2752** | −0.4833 | −0.4735** |

EDE | 0.1581** | 0.0006 | 0.2178** | 0.2167 | 0.1405** |

EIX | 0.4728** | 0.6132** | 0.3759** | −0.2317 | −0.2334** |

ETR | 0.3177** | 0.5667** | 0.1984** | −0.3683 | −0.3647** |

FPL | 0.3333** | 0.5296** | 0.2283** | −0.3013 | −0.2905** |

GMP | 0.0002 | 0.0021 | 0.0002 | 0.0000 | −0.0012 |

HE | 0.2489** | 0.2895** | 0.2363** | −0.0590 | −0.0668 |

IDA | 0.2845** | 0.3205** | 0.2787** | −0.0388 | −0.0319 |

NI | 0.2755** | 0.3505** | 0.2486** | −0.1030 | −0.1039* |

NU | 0.2669** | 0.4239** | 0.1935** | −0.2200 | −0.1991** |

OGE | 0.2699** | 0.3577** | 0.2376** | −0.1206 | −0.1254** |

PCG | 0.4541** | 0.6932** | 0.3100** | −0.3813 | −0.3788** |

PEG | 0.3984** | 0.6191** | 0.2929** | −0.3348 | −0.3249** |

PGN | 0.4020** | 0.6364** | 0.2990** | −0.3414 | −0.3346** |

PNM | 0.3639** | 0.5093** | 0.3138** | −0.1959 | −0.1845** |

PNW | 0.2893** | 0.5530** | 0.1874** | −0.3644 | −0.3793** |

POM | 0.3374** | 0.4240** | 0.2886** | −0.1418 | −0.1326** |

PPL | 0.3285** | 0.3310** | 0.3267** | −0.0058 | 0.0085 |

PSD | 0.2701** | 0.3450** | 0.2416** | −0.1026 | −0.0996* |

SCG | 0.2712** | 0.3196** | 0.2518** | −0.0710 | −0.0725* |

SO | 0.3853** | 0.6699** | 0.2439** | −0.4213 | −0.3995** |

SRP | 0.2788** | 0.2685** | 0.2875** | 0.0149 | 0.0032 |

TE | 0.2643** | 0.3599** | 0.2292** | −0.1366 | −0.1295** |

TXU | 0.3321** | 0.4801** | 0.2580** | −0.2233 | −0.2308** |

UIL | 0.2545** | 0.2244** | 0.2784** | 0.0534 | 0.0414 |

UTL | 0.0000 | 0.0003 | 0.0368 | 0.0029 | 0.0003 |

WEC | 0.2900** | 0.3969** | 0.2519** | −0.1460 | −0.1355** |

WPS | 0.2682** | 0.3307** | 0.2459** | −0.0946 | −0.0921** |

WR | 0.2519** | 0.3273** | 0.2249** | −0.1027 | −0.1209** |

Median | 0.3044 | 0.4373 | 0.2542 | −0.1871 | |

Mean | 0.3070 | 0.4293 | 0.2489 | −0.1804 |

*Statistically significant at 95%; **Statistically significant at 99%.

association between falling betas and deregulation, albeit in the opposite direction proposed by Peltzman [

The Binder and Norton [

A competing theory to the buffering hypothesis, Joskow and MacAvoy [

In a presentation of a utility beta study to the US Edison Electric Institute (electric utility research and lobbying institute) member companies by the Brattle Group [

This paper investigated the asymmetric response of utility stock price volatility for information signals of investors’ perceptions of utility regulatory policy. Secondly, it estimated the impact of wholesale deregulation on systematic risk, or, the buffering effect associated with deregulation. The results indicate that utility stock returns do respond asymmetrically to daily up and down markets and that the down market response of utility stock returns dominates. This direction of the asymmetry suggests that investors’ perceptions of utility regulatory policy are stringent.

Furthermore, the impact of deregulation on utility stock returns was modeled with pre- and post-wholesale-deregulation estimates of systematic risk. This deregulation required that electric utilities open their bulk power transmission lines to alternative purveyors of electric power. Systematic risk decreased substantially with wholesale competition. The result leads to the notion that the buffering impact was swamped any of a number of possibilities effecting the industry or market. They include the relative risk of the market increased, the cost structure of utilities improved, utilities faced positive demand shocks, prices increased, or the utility industry divested generation to a level that reduced their investment risk.

Some policy implications of these results suggest that stringent or adverse regulation, which attempts to squeeze as much of the financial resources of the regulated utility for lower rates, increases the volatility of utility stocks, investment risk, and therefore the cost of common equity capital. Therefore, lax regulation or regulation that is favorable to investors by reducing risk may have a rate reducing impact as the costs of common equity are less due to lower investment risk. A future area for research is whether lax regulation with a lower cost of common equity capital results in relatively lower rates than in adverse regulatory jurisdictions.

This paper has benefited from many helpful comments of participants at the Rutgers University Center for Research in Regulated Industries Eastern Conference and the Annual Conference of the Multinational Finance Society. The author wishes to thank Panayiotis Theodossiou, Tim Mount, Gene Pilotte, John Broussard, Michael Crew and Uzi Yaari for helpful comments.

The author declares no conflicts of interest regarding the publication of this paper.

Michelfelder, R.A. (2018) Asymmetric Response of Public Utility Stock Returns Volatility to Up and Down Markets and Deregulation. Journal of Mathematical Finance, 8, 576-598. https://doi.org/10.4236/jmf.2018.83037

Symbol Utility Company Name

AEP American Electric Power

CIN Cinergy

CMS CMS Energy

D Dominion Resources

DPL Dayton Power and Light

DTE DTE Energy

DUK DUK Duke Power

ED Consolidated Edison

EDE Empire District

EIX Edison International

ETR Entergy

FPL Florida Power and Light

GMP Florida Public Service

HE Hawaii Electric

IDA Idaho Power

NI NiSource

NU Northeast Utilities

OGE Oklahoma Gas and Electric

PCG Pacific Gas and Electric

PEG Public Service Electric and Gas

PGN Progress Energy

PNM Public Service of New Mexico

PNW Pinnacle West

POM Potomac Electric Power Company

PPL Pennsylvania Power and Light

PSD Puget Power and Light

SCG SCANA Corp

SO Southern Companies

SRP Sierra Pacific Power

TE TECO Energy

TXU Texas Utilities

UIL United Illuminating

UTL Unitil Corp

WEC Wisconsin Electric

WPS WPS Resources

WR Western Resources

The Bawa, Brown, Klein (1981) asymmetric response model can be re-written as follows:

R i . t = α i + D t + + α i − D t − + β i + R m , t + + β i − R m , t − + ε i , t

where D t + = 1 for all positive excess returns and zero otherwise, D t − = 1 for all negative returns and zero otherwise. Add and subtract both sides of the above equation by α i + D t − and β i + D n , t and note

R i . t = α i + ( D t + + D t − ) + ( α i − − α i + ) D t − + β i + ( R m , t + + R m , t − ) + ( β i − − β i + ) R m , t − + ε i , t

can be re-written as:

R i . t = α i + α n , i D n , t + β i R m , t + ( β i − − β i + ) D n , t R m , t + ε i , t

or,

R i . t = α i + α n , i D n , t + ( β i + β i , n D n , t ) R m , t + ε i , t .

In the above equation note that D t + + D t − = 1 for all values of r_{m,t} and r m , t − = D n , t r m , t . Additionally, D t − is re-written as D_{n,t}. The final model for estimating asymmetry and the impact of deregulation on returns includes the above model augmented with deregulation slope and intercept dummy variables, D_{r,t}. The final model is:

R i . t = α i + α n , i D n , t + α r , i D r , t + ( β i + β i , n D n , t + β i , r D r , t ) R m , t + μ i , t

β i , n is the asymmetry coefficient and is the difference between the down and up market betas ( β i − − β i + ) , and β i , r is the coefficient that represents the change in beta due to deregulation and β i is the standard CAPM beta. It follows from the results for the above models that the intercept and slope of the model for the pre-deregulation period are respectively α i and β i and for the post-deregulation period α i + α r , i and β i + β r , i . In this respect, the coefficients α r , i and β r , i show the change in the value of the intercept and slope dummies during the post deregulation period. Statistical significance of these coefficients will indicate that the parameters are different during the two periods. More specifically, a positive β r , i implies that the deregulation period is associated with an increase in systematic risk.