9 June 2015 An Econometric Assessment of Electricity Demand in the United States Using Panel Data and the Impact of Retail Competition on Prices By Dr. Agustin J. Ros This paper was originally presented at the Rutgers University Center for Research in Regulated Industries, 34th Annual Eastern Conference. Introduction Since the early 1970s electricity demand in the United States has been growing at an average annual rate of approximately 2%. In that period there have been major developments in the electricity sector including significant technological changes in generation services and the development of wholesale and retail competition. In this paper I use panel data covering 72 electricity distribution companies in the United States during the period from 1972–2009 to econometrically estimate structural demand equations separately for residential, commercial, and industrial customers and to examine the impact that retail competition has had on electricity prices. I find the own-price elasticity of demand for residential, commercial, and industrial consumers that are generally consistent with the published economics literature, ranging between -0.382 and -0.613 for residential demand, -0.747 for commercial demand, and ranging between -0.522 and -0.868 for industrial demand. Regarding retail electricity competition, I econometrically examine the impact of the restructuring of the retail electricity sector in the US from the mid-1990s. Since this period and up to 2009, a total of 21 states permitted retail customers (some states permitting only large industrial customers and some states also permitting smaller commercial and residential customers) to select their electricity generation supplier (retail competition) from a firm other than the incumbent electricity distribution company. As of 2009, 17 and 15 states still permitted retail competition for large and smaller customers, respectively. I estimate reduced-form static and dynamic price equations controlling for demand and supply factors, and include a binary variable for those states and time periods where retail competition was permitted. I test the null hypothesis that retail competition had no statistically significant impact on real electricity prices. I find that retail electricity competition is associated with lower electricity prices for each customer class with the magnitude of the impact being greater for the larger customer classes. www.nera.com 2 Literature Review There is a large economic literature on electricity demand in the US and other countries. Using data on US residential electricity demand for 1949–1993, Silk and Joutz (1997) find that a 1% increase in electricity prices reduces electricity consumption by -0.62%. With respect to disposable income, they find a 1% increase in income leads to a 0.82% increase in electricity consumption. Using data on residential demand for electricity in the US, Dergiades and Tsoulfidis (2008) find short- and long-run price elasticities of demand to be -0.386% and -1.065%, respectively. With regard to income, they find short- and long-run income elasticities of demand to be 0.101% and 0.273%, respectively. Paul, Myers, and Palmer (2009) find US short-run price elasticities of demand ranging between -0.04 and -0.32 (depending on customer class and region of the country), and long-run price elasticities of demand ranging between -0.02 and -1.15 (depending on customer class and region of the country). In the same paper, they summarize the results from previous studies, which I summarize below in Table 1. Table 1. Summary of Own-Price Elasticities of Demand from the Literature Customer Class Reference Short Run Long Run Residential Bohi and Zimmerman (1984) (consensus) -0.2 -0.7 Dahl and Roman (2004) -0.23 -0.43 Supawat (2000) -0.21 -0.98 Espey and Espey (2004) -0.35 -0.85 Bernstein and Griffin (2005) -0.24 -0.32 Commercial Bohi and Zimmerman (1984) 0 -0.26 Bernstein and Griffin (2005) -0.21 -0.97 Industrial Bohi and Zimmerman (1984) -0.11 -3.26 Dahl and Roman (2004) -0.14 -0.56 Taylor (1977) -0.22 -1.63 All Dahl and Roman (2004) -0.14 -0.32 Source: Paul, Myers and Palmer (2009), Table 5 Finally, using data from the Korean service sector, Lim and Lim (2014) find the shortand long-run price elasticities of electricity demand to be -0.421 and -1.002, respectively, and find short- and long-run income elasticities of electricity demand to be 0.855 and 1.090, respectively. With regard to the impact of electricity competition, a number of studies are focused on the wholesale sector but few studies focus on the retail sector. Regarding the former, Kleit and Terrell (2001) find that by eliminating technical inefficiencies, gas-fired generation plant could reduce costs by up to 13%. Fabrizio et al. (2007) found evidence of reduced fuel and nonfuel expenses in fossil-fueled plants in states that restructured their wholesale markets. Zhang (2007) finds that, in states that have restructured nuclear-fueled plants, utilization is higher and operating costs are lower. www.nera.com 3 Regarding the impact of retail competition, a recent paper by Su (2014) uses a difference-indifference approach to estimate the policy impact for US states that restructured their electricity retail markets. Su finds that “only residential customers have benefitted from significantly lower prices but not commercial or industrial customers. Furthermore, this benefit is transitory and disappears in the long run.” Swadley and Yücel (2011) find that retail competition makes the market more efficient by lowering the markup of retail prices over wholesale costs, and it generally appears to lower prices in states with higher customer participation rates in retail choice. Data I use several different data sources for my study: (I) FERC Form 1 data that contains information on residential, commercial, and industrial revenues and sales volume; (II) data from the bureau of labor statistics on price indices used for deflating prices and other relevant variables; and (III) inputs and results from a total factor productivity study containing information on cost indices, and total factor productivity for the 72 US electricity firms that I use in this study.1 In Table 2 below, I provide a description of the variables and data sources used for this study using average revenue per unit as a proxy for price.2 Table 2. Description of Variables Variable Name Description Data Source Lntfp Natural log of index of total factor productivity TFP Study ln_rpres Natural log of deflated residential revenue per unit sales volume using US census region cpi and urban consumer as deflator FERC Form 1, TFP Study, and author’s calculations ln_rpcom Natural log of deflated commercial revenue per unit sales volume using US census region cpi and urban consumer as deflator FERC Form 1, TFP Study, and author’s calculations ln_rpindus Natural log of deflated industrial revenue per unit sales volume using US census region cpi and urban consumer as deflator FERC Form 1, TFP Study, and author’s calculations ln_qres Natural log of residential sales volume FERC Form 1, TFP Study, and author’s calculations ln_qcom Natural log of commercial sales volume FERC Form 1, TFP Study, and author’s calculations ln_qindus Natural log of industrial sales volume FERC Form 1, TFP Study, and author’s calculations labprcindex Real labor cost index FERC Form 1 and TFP Study caprcindex Real capital cost index FERC Form 1 and TFP Study t_hdd State heating degree day index National Oceanic and Atmospheric Administration t_cdd State cooling degree day index National Oceanic and Atmospheric Administration ln_pop Natural log of state population Bureau of Economic Analysis ln_income Natural log of real state personal income using US census region cpi and urban consumer as deflator Bureau of Economic Analysis ln_price_natural gas Natural log of real state natural gas price index using US census region cpi and urban consumer as deflator US Energy Information Administration compl Indicator variable 1 if competition for large industrial customers is permitted Author’s construct comps Indicator variable 1 if competition for residential and commercial customers is permitted Author’s construct ratecap Indicator variable 1 if state that permitted competition had a rate cap for residential and commercial customers Swadley and Yücel (2011) Table 1 www.nera.com 4 In Table 3 below, I present summary statistics of the variables. Mean values over the period for deflated residential, commercial, and industry prices were ¢10.02 kWh, ¢8.91 kWh, and ¢6.16 kWh, respectively. And over the time period, there was a downward trend in deflated prices for all customer classes beginning in the late 1980s and lasting through the early 2000s. Approximately 14% and 13% of observations of the data reflect retail competition for large and residential and commercial customers, respectively. Table 3. Summary Statistics Variable Obs Mean Std. Dev. Min Max lntfp 2736 .5056051 .3717154 -.8342119 1.662097 unem 2448 6.16156 2.008683 2.3 15.6 ln_pop 2736 15.59572 .9018602 13.04594 17.42538 ln_rpres 2736 -2.300279 .2935026 -3.328425 -1.22254 ln_rpcom 2736 -2.417624 .3220173 -3.37734 -1.371787 ln_rpindus 2697 -2.786458 .3775694 -4.643488 -1.363757 ln_qres 2736 15.19656 .9528475 12.62801 17.82536 ln_qcom 2736 15.03261 1.052372 11.754 17.66853 ln_qindus 2736 15.02367 1.051302 11.01396 17.18522 labprcindex 2736 .8858641 .2957938 .2484632 3.048455 caprcindex 2736 1.148668 .5397295 .2104876 11.42236 t_hdd 2736 5162.852 2064.818 430 10810 t_cdd 2736 1100.366 862.2685 74 3827 ln_income 2736 10.1229 .2125993 9.519778 10.71989 ln_price_natural gas 2367 2.091865 .1812828 1.358819 2.655835 Compl 2736 .1425439 .349671 0 1 Comps 2736 .1312135 .3376954 0 1 Rate cap 2736 .0811404 .2731004 0 1 Econometric Models Econometric estimation of US electricity demand I estimate demand equations for US electricity distribution services for a 72-company panel sample from 1972 to 2009 to determine the price elasticity of demand, as well as the effects of other factors. I estimate demand equations separately for residential, commercial, and industrial electricity demand. In these demand equations, output (sales volume) is the left-hand side dependent variable. I am measuring how electricity output changes when other variables, such as price and income, change. The basic model is of the form: (1) yit = Yit γ + Xitβ + μi + υit www.nera.com 5 Where yit is the dependent variable (electricity consumption), Yit is a 1 x g2 vector of observations on g2 endogenous variables included as covariates (in my demand models, g2 =1 since I assume that price is the only endogenous variables), and these variables are allowed to be correlated with the υit , Xit is a 1 x k1 vector of observations on the exogenous variables included as covariates (such as income, population, price of natural gas, and heating and cooling degree days), γ is a g2 x 1 vector of coefficients, β is a k1 x 1 vector of coefficients, μi is the individual-level effect (i.e., the unobservable company-level effects), and υit is the disturbance term. I estimate demand models for each type of customer using four different estimators: fixedeffects, random-effects, first-difference, and the Arellano-Bond estimator for dynamic models. The fixed-effect estimator fits the model after sweeping out the μi by removing the panel-level means from each variable. The random-effects estimator treats the μi as random variables that are independent and identically distributed (i.i.d.) over the panels. The first-difference estimator removes the μi by fitting the model in first differences. The Arellano-Bond model is a linear dynamic panel-data model that includes p lags of the dependent variables as covariates and contains unobserved panel-level effects, fixed or random.3 I expect price to be endogenous, and for instruments I use data from the TFP Study of US distribution companies. Specifically, I use a deflated labor price index and a deflated capital price index. These two variables reflect changes in input costs for the 72 distribution companies over the period and are thus good candidates for instruments. In addition to price, I expect electricity demand to be positively related to population, real income, the state heating degree day index, the state cooling degree day index, and the deflated price of natural gas.4 To capture the possibility of changing demand preferences over time, I include three decade binary variables. Residential demand models The first three models are static demand models, while the Arellano-Bond model is a dynamic demand model. Table 4 contains the results of my static and dynamic residential demand equations. Since the left-hand side dependent variable is the log of residential demand, the coefficient on residential prices is the price elasticity of demand. The fixed-effects and random-effects estimators provide very similar results for the price elasticity of demand and for all the variables in the model. The price elasticity of demand using the fixed-effects estimator is estimated to be -0.440 and statistically significant (t statistic = -3.65). The price elasticity of demand using the random-effects estimator is estimated to be -0.430 and statistically significant (t statistic = -3.57). When I use the first-difference estimator, the price elasticity of demand increases to -0.613 (t statistic = -3.23). Finally, when I use the Arellano-Bond dynamic estimator, the price elasticity of demand is estimated to be -0.382, and is estimated very precisely.5 The econometric evidence, therefore, supports a price elasticity of demand for residential customers ranging between -0.382 and -0.613.6 These estimates are within the range found in the economic literature (see Section 2 above). With respect to the income elasticity of demand, the fixed-effects and random-effects estimators also provide very similar results. The income elasticity of demand using the fixed-effects estimator is estimated to be 0.366 and statistically significant (t statistic = 5.51). The income elasticity of demand using the random-effects estimator is estimated to be 0.372 and statistically significant (t statistic = 5.64). When I use the first-difference estimator, the www.nera.com 6 income elasticity of demand decreases to 0.105 (t statistic = 1.90). Finally, when I use the Arellano-Bond dynamic estimator, the income elasticity of demand is estimated to be 0.411 and is estimated very precisely. To control for changes in quantity demanded over time, I included three decade binary variables. All the decade variables are positive and statistically significant. Specifically, using the results from the fixed-effects estimator I find that, compared to the 1970s and holding all factors constant, residential demand in the 1980s, 1990s, and 2000s was approximately 10%, 12%, and 13% higher, respectively. The similarity in the coefficients suggests that, holding all factors in the model constant, residential demand has not changed much since 1980. Other variables in the model are population, the price of natural gas, and the heating and cooling degree day indices. With respect to population, the four models provide similar results: a 1% increase in population results in an increase in residential demand ranging between 0.786% and 0.873% and is estimated very precisely (in each case a p-value of less than 0.001). I find natural gas to be a substitute for residential electricity consumption. A 1% increase in the real price of natural gas results in an increase in residential electricity consumption of approximately 0.090%. Finally, I find that the heating and cooling degree days index positively affects the demand for residential electricity. Table 4. Estimation of Static 2SLS and Dynamic Residential Demand Equations Using Panel Data Variable Fixed Effects Random Effects First Difference Arellano-Bond ln_rpres -.43983922*** -.42958065*** -.10504817*** decade_80s .09833813*** .09856919*** .01954227*** decade_90s .11720025*** .12003327*** .02868456*** decade_2000s .12444248*** .12918732*** .02141042** ln_pop .87328382*** .85569493*** .21840973*** t_hdd .00003234*** .00003248*** .00002846*** t_cdd .00011367*** .00011673*** .00017076*** ln_income .36623779*** .37236283*** .11290548*** ln_price natural gas .08934869*** .09028773*** .00541858 ln_rpres_cpi D1. -.61326955** decade_80s D1. .0111662 decade_90s D1. .0131513 decade_2000s D1. .01895704 ln_pop D1. .78628219*** t_hdd D1. .00003095*** t_cdd D1. .00013223*** ln_income D1. .10526433 ln_price natural gas D1. .02681842 ln_qres L1. .72511645*** _cons -3.6963764*** -3.4667861*** .00727446*** -.95904616*** N 2367 2367 2295 2295 Wald χ2 Prob > χ2 0.0000 Prob > χ2 0.0000 Prob > χ2 0.0000 Prob > χ2 0.0000 *p<0.05; **p<0.01; ***p<0.001 www.nera.com 7 Commercial demand models Table 5 contains the results of my static and dynamic commercial demand equations. The fixed-effects and random-effects estimators provide very similar results for the price elasticity of demand and for all the variables in the model. The price elasticity of demand using the fixedeffects estimator is estimated to be -0.427 but is not estimated precisely (t statistic = -1.71). The price elasticity of demand using the random-effects estimator is estimated to be -0.423 and is also not statistically significant (t statistic = -1.71). When I use the first-difference estimator, the price elasticity of demand decreases to -0.084 but is estimated with very poor precision (t statistic = -0.39). Finally, when I use the Arellano-Bond dynamic estimator, the price elasticity of demand is -0.747 and estimated very precisely.7 The econometric evidence, therefore, supports a price elasticity of demand for commercial customers of -0.747, also within the findings in the economics literature (see Section 2). With respect to the income elasticity of demand, the fixed-effects and random-effects estimators also provide very similar results. The income elasticity of demand using the fixedeffects estimator is estimated to be 0.877 and statistically significant (t statistic = 3.90). The income elasticity of demand using the random-effects estimator is estimated to be 0.883 and statistically significant (t statistic = 3.98). When I use the first-difference estimator, the income elasticity of demand decreases to 0.448 (t statistic = 4.98). Finally, when I use the Arellano-Bond dynamic estimator, the income elasticity of demand is estimated to be 0.584 and is estimated very precisely. The econometric evidence thus supports an income elasticity of residential electricity demand ranging between 0.448 and 0.883. These estimates are within the range found in the economic literature (see Section 2). To control for changes in quantity demanded over time, I included three decade binary variables. All the decade variables are positive and statistically significant. Specifically, using the results from the fixed-effects estimator I find that, compared to the 1970s and holding all factors constant, commercial demand in the 1980s, 1990s, and 2000s was approximately 14%, 23%, and 20% higher, respectively. Other variables in the model are population, the price of natural gas, and the heating and cooling degree day indices. With respect to population, a 1% increase in population results in an increase in commercial demand ranging between 0.472% and 0.767%, depending on the model, and is estimated precisely, in three cases with a p-value of less than 0.001. I find evidence that natural gas is a substitute for commercial electricity demand in two of the four models, with a 1% increase in the real price of natural gas resulting in approximately a 0.10% increase in commercial demand. And I find some evidence that the heating and cooling degree days index positively impacts the demand for commercial electricity. www.nera.com 8 Table 5. Estimation of Static 2SLS and Dynamic Commercial Demand Equations Using Panel Data Variable Fixed Effects Random Effects First Difference Arellano-Bond ln_rpcom_cpi -.42668522 -.42263656 -.11108075*** decade_80s .13077288*** .13085258*** .02271251** decade_90s .20658202*** .20768634*** .02041106* decade_2000s .1796107** .18139979** -.01235798 ln_pop .66475958*** .65881121*** .11411259*** t_hdd 4.084e-06 3.800e-06 9.531e-06 t_cdd 1.707e-06 2.974e-06 .00008662*** ln_income .87698091*** .8803502*** .08685953** ln_price natural gas .09677514* .09694503* -.00966809 ln_rpcom_cpi D1. -.08351428 decade_80s D1. .00950299 decade_90s D1. .02985924 decade_2000s D1. .04058238* ln_pop D1. .47245619* t_hdd D1. .00001042* t_cdd D1. .00006827*** ln_income D1. .44778965*** ln_price natural gas D1. -.01570014 ln_qcom L1. .85125672*** _cons -5.5837047*** -5.5164864*** .0167317*** -.7947528 N 2367 2367 2295 2295 Wald χ2 Prob > χ2 0.0000 Prob > χ2 0.0000 Prob > χ2 0.0000 Prob > χ2 0.0000 *p<0.05; **p<0.01; ***p<0.001 Industrial demand models Table 6 contains the results of my static and dynamic industrial demand equations. The price elasticity of demand using the fixed- and random-effects estimator is contrary to economic theory, showing a positive price elasticity of demand. I therefore rely on the results from the first-difference and Arellano-Bond models. When I use the first-difference estimator, I find the price elasticity of demand to be -0.868 (t statistic = -2.80). When I utilize the Arellano-Bond dynamic estimator, the price elasticity of demand is estimated to be -0.522 and is estimated precisely.8 The econometric evidence, therefore, supports a price elasticity of demand for industrial customers between -0.522 and -0.868, also within the findings in the economics literature (see Section 2). With respect to the income elasticity of demand, the fixed-effects and random-effects estimators also provide very similar results. The income elasticity of demand using the fixed-effects estimator is estimated to be 1.467 and statistically significant (t statistic = 4.95). The income elasticity of demand using the random-effects estimator is estimated to be 1.456 and statistically significant (t statistic = 5.02). The income elasticity of demand using the www.nera.com 9 first-difference estimator is significantly lower at 0.720 (with a student t statistic = 4.26). When I use the Arellano-Bond dynamic estimator, the income elasticity of demand is estimated to be 1.58 and is estimated precisely and closer to the estimates found from the fixed- and random-effects estimators. To control for changes in quantity demanded over time, I included three decade binary variables. The results are mixed. Three of the four models suggest that industrial electricity demand was lower in the 1980s (than in the 1970s) but higher in the 1990s and 2000s. The remaining model, however, suggests that industrial electricity demand was lower in each decade compared to the 1970s. Other variables in the model are population, the price of natural gas, and the heating and cooling degree day indices. With respect to population, the fixed and random effects model finds that a 1% increase in population results in an increase in industrial demand of 0.361% and 0.338%, respectively while for the first-difference model a 1% increase in population results in an increase in industrial demand range of 0.882%. I find evidence in the Arellano-Bond model that natural gas is a complement for industrial electricity consumption. A 1% increase in the real price of natural gas results in a decrease in industrial electricity consumption of approximately 1.236%. Finally, I do not find strong evidence that the heating and cooling degree days index impacts the demand for industrial electricity. Table 6. Estimation of static 2SLS and dynamic industrial demand equations using panel data Variable Fixed Effects Random Effects First Difference Arellano-Bond ln_rpindus_cpi 1.1752322*** 1.1937324*** -.0342152* decade_80s -.11297932** -.11340149** -.03659619*** decade_90s .13545005** .14296852** -.06222053*** decade_2000s .07849822 .0928366 -.12470984*** ln_pop .36146288** .33779817** .06730164 t_hdd -7.977e-06 -.00001413 -8.382e-06 t_cdd .00003866 .0000471 2.600e-06 ln_income 1.4671618*** 1.4585598*** .10344492* ln_price natural gas .13354912 .12156241 -.08090781* ln_rpindus_cpi D1. -.86765614** decade_80s D1. .06215827** decade_90s D1. .06364113* decade_2000s D1. .07172708* ln_pop D1. .8819554** t_hdd D1. -2.597e-06 t_cdd D1. -.0000136 ln_income D1. .71977543*** ln_price natural gas D1. -.02696871 ln_qindus L1. .93404923*** _cons -2.5161967 -1.9718117 -.03033039*** -.92398406 N 2333 2333 2262 2262 Wald χ2 Prob > χ2 0.0000 Prob > χ2 0.0000 Prob > χ2 0.0000 Prob > χ2 0.0000 *p<0.05; **p<0.01; ***p<0.001 www.nera.com 10 Econometric Estimation of US Electricity Price Equations In this section, I estimate reduced-form price equations for residential, commercial, and electricity customer classes. The left-hand side dependent variable is log of real price and is the same price variable that I used above for the demand equations. The right-hand side independent variable comps is a binary variable with a value of one if the state permitted retail competition for residential and commercial (small) customers. I use this variable for the residential and commercial price equations. For the industrial price equations, the variable compl is a binary variable with a value of one if the state permitted retail competition for industrial (large) consumers. I estimate fixed and random effects models assuming that comps and compl are exogenous. In addition, I again estimate dynamic models using the Arellano-Bond estimator and provide two results: the first treats the competition variable as exogenous, and the second assumes it is endogenous and uses the lagged values of the competition variable as instruments. Other independent variables in my reduced-form price equation include binary variables for the 1980s, 1990s, and 2000s, log of population, log of the firm’s total factor productivity (lntfp), heating and cooling degree day indices, real personal income, and the log of real price of natural gas. I also include the variable ratecap to control for the fact that some of the states that permitted competition imposed a rate cap on electricity prices for small customers (residential and commercial) for a period of time. Thus, it is important to control for this effect, otherwise the effect would be included within the comps coefficient. The variable ratecap is a binary variable with one in those states that permitted competition and had a rate cap in the year in question. Residential price equations Table 7 contains the results of my residential price model. The fixed and random effects estimators provide very similar results (and each having a p-value of less than 0.001): holding all other factors constant, residential prices were approximately 8% lower in those states that permitted retail competition for residential and commercial consumers. The Arellano-Bond estimator—assuming that comps is exogenous (model 3)—indicates that residential prices were approximately 3% lower in those states that permitted retail competition for residential and commercial consumers, but the effect was not significant. When I use the Arellano-Bond estimator and consider comps as endogenous, I find the impact is practically zero and is not estimated precisely. Based upon these results, while there is some evidence that residential electricity competition is associated with lower residential prices, additional work should be performed on finding suitable instruments for the comps variable in order to ensure that the parameter estimates for comps are unbiased. Other significant findings include the impact of lntfp and personal income on residential electricity prices. In each of the models, an increase in tfp of 1% results in a decrease in residential prices, ranging from -0.253% to -0.084%. In each of the models, an increase in personal income results in a decrease in prices, ranging from -0.338% to -0.186%. www.nera.com 11 Table 7. Estimation of Static and Dynamic Residential Price Equations Using Panel Data Variable Fixed Effects (1) Random Effects (2) Arellano-Bond (3) Arellano-Bond (4) comps -.089*** -.084*** -.008 -.001 ratecap -.105*** -.101*** -.085*** -.081*** decade_80s .045*** .043*** .02*** .02*** decade_90s -.057*** -.065*** .009 .009 decade_2000s -.075*** -.098*** .06*** .056*** ln_pop .053 .067*** -.016 .007 Lntfp -.22*** -.253*** -.087*** -.084*** t_hdd -2.2e-05* -1.2e-05* -6.78e-06 -6.67e-06 t_cdd -4.3e-05*** -5.2e-05*** -2.15e-05* -2.5e-05* ln_income -.338*** -.258*** -.186*** -.191*** ln_price natural gas -.079*** -.048 .042* .045* ln_rpres_cpi L1. .773*** .780*** _cons .792 .330 1.61** 1.53** N 2367 2367 2295 2295 *p<0.05; **p<0.01; ***p<0.001 Commercial price equations Table 8 contains the results of my commercial price model. The fixed- and random-effects estimators provide very similar results (each having a p-value of less than 0.001): holding all other factors constant, commercial prices were approximately 16% lower in those states that permitted retail competition for residential and commercial consumers. The Arellano-Bond estimator, assuming that comps is exogenous (model 3), indicates that commercial prices were approximately 19% lower in those states that permitted retail competition for residential and commercial consumers (with a p-value of less than 0.001). When I use the Arellano-Bond estimator and consider comps as endogenous, the impact is approximately 19%. Based upon these results, there is evidence that commercial electricity competition is associated with lower commercial prices. Other significant findings include the impact of lntfp and personal income on residential electricity prices. In each of the models, an increase in tfp results in a decrease in prices, ranging from -0.278% to -0.056%. In each of the models, an increase in personal income results in a decrease in prices, ranging from -0.68% to -0.29%. www.nera.com 12 Table 8. Estimation of Static and Dynamic Commercial Price Equations Using Panel Data Variable Fixed Effects (1) Random Effects (2) Arellano-Bond (3) Arellano-Bond (4) comps -.176*** -.171*** -.038*** -.035*** ratecap -.014 -.010* -.019 -.011 decade_80s .019 .016 .0042 .004 decade_90s -.108*** -.119*** -.008 -.007 decade_2000s -.105*** -.134*** .051*** .05*** ln_pop .075* .079*** .027 .027 lntfp -.236*** -.278*** -.058*** -.056** t_hdd -2.5e-05** -1.0e-05 -5.7e-06 -5.3e-06 t_cdd -3.2e-05 -4.7e-05* -2.2e-05 -2.3e-05 ln_income -.683*** -.567*** -.296*** -.292*** ln_price natural gas -.087** -.048 .049 .057* ln_rpcom_cpi L1. .811*** .821*** _cons 3.89*** 2.55*** 2.09*** 2.06*** N 2367 2367 2295 2295 *p<0.05; **p<0.01; ***p<0.001 Table 9 contains the results of my industrial price model. The fixed- and random-effects estimators provide very similar results (each having a p-value of less than 0.001): holding all other factors constant, industrial prices were approximately 24% lower in those states that permitted retail competition for industrial consumers. The Arellano-Bond estimator indicates that industrial prices were approximately 30% lower in those states that permitted retail competition for industrial consumers (with a p-value of less than 0.001). When I use the Arellano-Bond estimator and consider compl as endogenous, the impact is approximately 29%. Based upon these results, there is evidence that industrial electricity competition is associated with lower industrial prices. Other significant findings include the impact of tfp and personal income on residential electricity prices. In each of the models, an increase in tfp results in a decrease in prices, ranging from -0.263% to -0.067%. In each of the models, an increase in personal income results in a decrease in prices, ranging from -0.811% to -0.376%. www.nera.com 13 Table 9. Estimation of Static and Dynamic Industrial Price Equations Using Panel Data Variable Fixed Effects (1) Random Effects (2) Arellano-Bond (5) Arellano-Bond (6) Compl -.269*** -.263*** -.0741*** -.0674*** decade_80s .0861*** .0844*** .00935 .00762 decade_90s -.0779*** -.0866*** -.00336 -.00385 decade_2000s -.0404 -.0715* .0813*** .0799*** ln_pop .175*** .0967*** -.00951 -.00845 Lntfp -.193*** -.263*** -.075** -.0674** t_hdd -3.0e-05* -8.8e-06 -3.4e-06 -3.7e-06 t_cdd -4.5e-05 -5.1e-05* -1.1e-05 -1.3e-05 ln_income -.811*** -.616*** -.379*** -.376*** ln_price natural gas -.149*** -.0795 .0317 .0377 ln_rpindus_cpi L1. .791*** .803*** _cons 3.38*** 2.44*** 3.39*** 3.37*** N 2333 2333 2262 2262 *p<0.05; **p<0.01; ***p<0.001 References T. Dergiades and L. Tsoulfidis, “Estimating residential demand for electricity in the United States, 1965–2006,” Energy Economics, 2008. K. Lim and S. Lim, “Short- and long-run elasticities of electricity demand in the Korean service sector,” Energy Policy, 2014. A. Paul, E. Myers, and K. Palmer, “A Partial Adjustment Model of US Electricity Demand by Region, Season and Sector,” Resources for the Future Discussion Paper, 2009. Available at http://www.rff.org/documents/rff-dp-08-50.pdf. J. Silk and F. Joutz,“Short and long run elasticities in US residential electricity demand: a co-integration approach,” Energy Economics, 1997. X. Su, “Have Customers Benefited from Electricity Retail Competition,” Journal of Regulatory Economics, 2015 (forthcoming). A. Swadley and M. Yücel, “Did residential electricity rates fall after retail competition? A dynamic panel analysis,” Energy Policy, 2011. Notes 1 I was the co-author (with Jeff Makholm) of an expert report entitled Total Factor Productivity in the United States Electricity Sector from 1972–2009. We were expert witnesses on behalf of the Alberta Public Utility Commission in a proceeding in 2012 whose objective was setting tariffs for electricity and gas distribution companies. Our analysis was used to establish the X-factor in a price cap plan. Throughout this paper, I refer to the study as “TFP Study.” 2 I did not have data on actual tariffs for the different customer classes over the time period. Instead, I use average revenue per unit as a proxy for price and assume that some errors exist in variable (EIV). Nevertheless, EIV should not present a significant problem because I treat the price variable as (jointly) endogenous in my structural demand equations, and I explicitly model price in my reduced-form price equations. 3 The model is constructed so that by definition the unobserved panel-level effects are correlated with the lagged dependent variables, thus making standard estimators inconsistent. Arellano and Bond derived a consistent generalized method-of-moments (GMM) estimator for the parameters of the model. 4 I consider natural gas as a substitute for electricity and expect the cross-price elasticity of demand to be positive. 5 The total effect is -.105048/(1-.725116). 6 I also estimate every model by including interaction terms between the price variable and each decade variable to test whether there is evidence that the price elasticity of residential electricity demand changed significantly during the decades. For the fixed-effects model, the price elasticity of demand for the 1980s, 1990s, and 2000s ranged between -0.383 to -0.474, similar to the models without the interaction effects but the parameters were not estimated precisely, none being significant at the 5% level of statistical significance. Results for the random-effects estimator are similar. 7 The total effect is -.11108075/(1-.85125672). 8 The total effect is -.0342152/(1-.93404923). About NERA NERA Economic Consulting (www.nera.com) is a global firm of experts dedicated to applying economic, finance, and quantitative principles to complex business and legal challenges. For over half a century, NERA’s economists have been creating strategies, studies, reports, expert testimony, and policy recommendations for government authorities and the world’s leading law firms and corporations. We bring academic rigor, objectivity, and real world industry experience to bear on issues arising from competition, regulation, public policy, strategy, finance, and litigation. NERA’s clients value our ability to apply and communicate state-of-the-art approaches clearly and convincingly, our commitment to deliver unbiased findings, and our reputation for quality and independence. Our clients rely on the integrity and skills of our unparalleled team of economists and other experts backed by the resources and reliability of one of the world’s largest economic consultancies. With its main office in New York City, NERA serves clients from more than 25 offices across North America, Europe, and Asia Pacific. Contact For further information and questions, please contact the author: Dr. Agustin Ros Vice President +1 617 927 4506 agustin.ros@nera.com The opinions expressed herein do not necessarily represent the views of NERA Economic Consulting or any other NERA consultant. Please do not cite without explicit permission from the author.