The Impact of Electricity Sector Restructuring on Coal-fired Power Plants in India.
This paper examines the impact of electricity sector restructuring on the operating efficiency of thermal power plants in India. Between 1996 and 2009, 85 percent of the coal-based generation capacity owned by state governments was unbundled from vertically-integrated State Electricity Boards (SEBs) into newly created state generation companies. The restructuring sought to expand generation capacity and reduce costs by encouraging the entry of independent power producers and by "corporatizing" unbundled generation companies. Although government owned, these companies were granted formal autonomy in technical, financial and managerial decisions. We examine whether greater managerial discretion and specialization in generation increased operating reliability and thermal efficiency at unbundled power plants.
To examine the impact of restructuring on the operating efficiency of state owned power plants, we use electricity generating unit (EGU)-level data on measures of operating reliability and plant-level data on thermal efficiency as outcome variables. Operating reliability is measured by the percentage of time in a year an EGU is available to generate electricity (unit availability), and the percentage of time a unit is forced to shut down due to equipment failures (forced outages). (1) Thermal efficiency is measured by coal consumption per kWh and by operating heat rate--the energy used to generate a kilowatt-hour (kWh) of saleable electricity. We also estimate the impact of reform on the capacity utilization of the EGU, i.e., the percent of time the EGU generates electricity.
To investigate the impact of reforms in the Indian electricity sector we construct a panel data set of coal-based EGUs for the years 1988-2009. The variation in the timing of reforms across states allows us to estimate the impact of unbundling on EGU reliability and plant thermal efficiency. Our difference-in-difference specification assumes that conditional on control variables--EGU/plant characteristics, EGU and year fixed effects, and state-specific linear time trends--the assignment of the timing of reforms (including not to reform) is exogenous. Under this assumption, these models identify the effect of reforms from a comparison of the performance of plants in states that unbundled with plants in states that had not yet unbundled.
To eliminate the possibility of state-year shocks affecting our estimates of average treatment effects, we also present results from a triple-difference specification that uses EGUs operated by central-government-owned generation companies as an additional control group. These companies operate outside the purview of state governments and thus were not directly affected by the reorganization of the SEBs.
Our results suggest that the gains from unbundling of generation from transmission and distribution were limited to the states that reformed before the Electricity Act of 2003. In these states, on average, EGUs at state-owned plants experienced a 5 percentage point reduction in forced outages as result of unbundling--roughly a 25 percent reduction compared to the 1995 average. The decrease in forced outages was accompanied by a 6 percentage point increase in availability. These results are driven largely by the improvements in operating reliability at EGUs with lower nameplate capacity. Our results are not driven by the decommissioning of old and inefficient EGUs or a commissioning of new more efficient ones, and thus represent an improvement in existing capacity. This is an important distinction as increasing reliability at existing units can likely be achieved more cheaply than by installing new capital equipment. (2)
On average, there is no evidence of an improvement in capacity utilization due to restructuring, although the results suggest a statistically significant increase at some EGUs. For states that unbundled prior to 2003, we find that unbundling led to a significant improvement in electricity generation at smaller generating units--a 9.4 percentage point increase in capacity utilization at 110/120 MW units. Importantly, our results show no evidence that unbundling of SEBs led to the improvement in thermal efficiency at state-owned power plants.
In summary, our analysis points to modest gains from reform. Operating reliability increased at EGUs in states that unbundled prior to 2003; but there is no evidence of an improvement in thermal efficiency. Our failure to find a larger impact from restructuring than reported in the US (Bushnell and Wolfram 2007; Chan et al. 2013) may also reflect the path that reform has taken in India thus far. In the United States unbundling resulted in independent power producers (IPPs) entering the market for generation. This has not yet occurred on a large scale in India. It may also reflect the way in which power plants are compensated for the electricity they generate. Under the 2003 Electricity Act compensation for energy used in generation is to be based on scheduled generation and to depend on operating heat rate. There is evidence that state electricity regulatory commissions have set compensation for fuel use based on very high estimates of operating heat rate, suggesting that this may not provide much of an incentive for plants to improve thermal efficiency.
The rest of the paper is organized as follows. Section 2 provides background on the Indian power sector and the nature of reforms and places our paper in the context of the literature. Section 3 describes the empirical approach taken. In section 4, we discuss econometric issues. Section 5 describes the data used in the study and section 6 our results. Section 7 concludes.
2.1 Overview of the Indian Electricity Sector
Most generating capacity in India is government owned. The 1948 Electricity Supply Act created State Electricity Boards (SEBs) and gave them responsibility for the generation, transmission, and distribution of power, as well as the authority to set tariffs. SEBs operated on soft budgets, with revenue shortfalls made up by state governments (Thakur et al. 2005). Electricity tariffs set by SEBs failed to cover costs, generating capacity expanded slowly in the 1960s and 1970s, and blackouts were common. To increase generating capacity, the Government of India in 1975 established the National Hydroelectric Power Corporation and the National Thermal Power Corporation (NTPC), which built generating capacity and transmission lines that fed into the SEB systems. In 1990, prior to reforms, 63 percent of installed capacity in the electricity sector in India was owned by SEBs, 33 percent by the central government, and 4 percent by private companies (Tongia 2003).
Our analysis focuses on coal-fired power plants, which have, for the past two decades, provided approximately 70% of the electricity generated in India. (3) Coal-fired power plants in India are, in general, less efficient than their counterparts in the US. Over the period 1988-1991 the average operating heat rate--the heat input (in kcal) required to produce a kWh of electricity--of state-owned Indian plants was, on average, 13.7% higher than publicly-owned US plants, controlling for differences in age and nameplate capacity (Chan, Cropper and Malik 2014). (4)
The higher average operating heat rates of Indian plants are due in part to the poor quality of Indian coal, but also to inefficiencies in management. The design heat rate of generating units that use coal with high moisture and/or high ash content is higher than for units with low moisture and ash content (MIT 2007). The ash content of Indian coal is between 30 and 50% (Khanna and Zilberman 1999; CEA 2011). This implies that Indian plants will require more energy to produce a kWh of electricity than comparable plants in the US. The operating heat rate of the plant may be higher than the design heat rate if the plant is poorly maintained or experiences frequent outages. (5) Pre-reform, operating heat rates at state-owned plants were, on average, 31% higher than design heat rates (Cropper et al. 2011).
State plants have, historically, been operated less efficiently than plants owned by the central government: they have had higher forced outages and lower capacity utilization (Thakur et al. 2006). Figures 1A-1C illustrate trends in the average percent of time state and central plants were available to generate electricity (plant availability), the average percent of time plants were shut down due to forced outages, and the average percent of time the plant was used to generate electricity (capacity utilization). State power plants have, on average, had lower availability and capacity utilization than central-government-owned plants and higher forced outages throughout the 1988-2009 period.
2.2 History of Power Sector Reforms
Electricity sector reforms in India were prompted by the poor performance of state-owned power plants, by large transmission and distribution losses, and by problems with the SEBs' tariff structure (Thakur et al. 2005). The tariff structure, which sold electricity cheaply to households and farmers and compensated by charging higher prices to industry, prompted firms to generate their own power rather than purchasing the expensive and unreliable electricity from the grid, an outcome that further reduced the revenues of SEBs. The result was that most SEBs failed to cover the costs of electricity production. Reform of the distribution network was necessary because of the extremely large power losses associated with the transmission and distribution of electric power--both technical losses and losses due to theft (Tongia 2003).
Beginning in 1991, the Government of India instituted reforms to increase investment in power generation, reform the electricity tariff structure, and improve the distribution network. Under the Electricity Laws Act of 1991, IPPs were allowed to invest in generating capacity. They were guaranteed a fair rate of return on their investments, with tariffs regulated by Central Electricity Authority. (6) The Electricity Regulatory Commissions Act of 1998 made it possible for the states to create State Electricity Regulatory Commissions (SERCs) to set electricity tariffs. States were to sign memoranda of understanding with the federal government, agreeing to set up SERCs and receiving, in return, technical assistance to reduce transmission and distribution losses. The Electricity Act of 2003 made the establishment of SERCs mandatory and required the unbundling of generation, transmission, and distribution (Singh 2006).
Another objective of the 2003 Electricity Act was to reform the electricity tariff structure--both for end users and for generators. SERCs are to follow the Central Electricity Regulatory Commission (CERC) guidelines in compensating generators. The CERC compensates the power plants under its jurisdiction based on performance. Compensation for energy used in generation is paid based on scheduled generation and depends on operating heat rate. Compensation for fixed costs (depreciation, interest on loans and finance charges, return on equity, operation and maintenance expenses, interest on working capital, and taxes) is based on plant availability. In addition, an availability-based tariff (ABT) was instituted in 2002 to regulate the supply of power to the grid. If a generator deviates from scheduled generation, the ABT imposes a tariff that depends on system frequency (Chikkatur et al. 2007).
There were two distinct waves of unbundling reforms in India. Table 1 shows the year in which the SERC became operational in each state and the year in which generation, transmission, and distribution were unbundled. (7) The first wave, between 1996 and 2002, took place prior to the Electricity Act of 2003. The second wave began in 2004 and continued through the end of our sample period (2009). (8) We refer to these as Phase 1 (unbundling prior to 2003) and Phase 2 (unbundling between 2004 and 2009) states. The remaining states in our sample (Phase 3 states) unbundled outside of our sample period (Table 1).
2.3 Studies of Electricity Sector Reforms
Over the past two decades, many member countries of the OECD and more than 70 developing countries have taken steps to reform their electricity sectors (Bacon and Besant-Jones 2001; Khanna and Rao 2009). A large literature uses cross-country data to examine factors conducive to reform and the nature of reforms undertaken (Bacon and Besant-Jones 2001; Erdogdu 2013). Studies have also examined the impacts of reforms on the efficiency of generation and distribution and on electricity pricing (Jamasb et al. 2005). Much of this literature, which is summarized by Jamasb et al. (2005) and by Khanna and Rao (2009), focuses on the impact of privatization on performance and uses cross-country panel data. A related literature uses plant-level data to control for within-country variations in regional and/or plant-level characteristics to estimate the impact of reforms. Below, we discuss studies that examine the impact of reforms on generation efficiency using plant-level data.
Most of the studies that have examined the impact of reforms on generation efficiency using plant-level data employ either stochastic frontier or data envelopment analysis methods. Jamasb et al. (2005) summarize and critique four such studies in developing countries. (9) In the United States, Knittel (2002) and Hiebert (2002) use stochastic frontier analysis to study the impact of reforms on generation efficiency. Knittel (2002) estimates a stochastic production frontier that allows the mean of the efficiency component of the error term to depend on the compensation program that the generator faces. (10) He finds greater production efficiency for plants that operate under programs that provide direct incentives for increased efficiency by compensating generators based on heat rate and plant availability (compared with plants compensated on a cost-plus basis).
Hiebert (2002) estimates a stochastic frontier cost function to examine the efficiency impacts of unbundling and open access to transmission and generation using U.S. data for the period 1988-1997. As in Knittel (2002), he jointly estimates the parameters of the stochastic frontier and the factors determining the efficiency component of the error term. His analysis shows that investor-owned utilities and cooperatively-owned plants are more efficient than publicly-owned municipal plants. Hiebert adds dummy variables for states that unbundled generation from transmission and distribution in 1996 and 1997. The results indicate efficiency gains in 1996 (but not 1997) for coal-fired plants that were operating in states that implemented reforms.
Fabrizio, Rose, and Wolfram (2007) study the impact of electricity restructuring on generation efficiency in the United States using a difference-in-difference approach to measuring efficient input use. Using a plant-level panel (1981-1999) of gas- and coal-fired thermal power plants, the authors estimate cost-minimizing input demands as a function of plant characteristics while controlling for the regulatory regime. They show that privately owned utilities in restructuring states experienced greater gains in efficiency of nonfuel input use compared to similar utilities in non-restructuring states and cooperatively or publicly owned generators that were insulated from the reforms. Because of the nature of the restructuring process in the United States, their restructuring measure combines the effect of unbundling of generation from transmission and distribution with opening the generation sector to retail competition. The authors, however, attribute most of their impact to the unbundling of generation, as retail competition was limited to only seven states during the period of analysis.
Although the literature examining the impact of reforms in the Indian electricity sector is growing (e.g., Thakur et al. 2005; Singh 2006; Chikkatur et al. 2007), the only econometric study that attempts to estimate ex-post generation efficiency gains is Sen and Jamasb (2012). (11) The authors use panel data at the state level for the period 1990-2007 to test the impact of reforms on plant load factor (PLF), gross generation and transmission, and distribution losses. (12) Specifically, they explain the three performance measures as functions of six regulatory dummy variables and state and year fixed effects. (13) They find that the unbundling and tariff order dummy variables show a strong positive effect on PLF (i.e., capacity utilization), as does the ratio of industrial to agricultural electricity prices. They also find that the SERC, unbundling, and privatization dummies have increased transmission and distribution losses, possibly due to the reduced ability to hide existing losses after reform.
In contrast to the state-level approach of Sen and Jamasb (2012), we use data at the EGU level to examine the effect of unbundling on the performance of state-owned power plants. The use of disaggregated data allows us to control for the heterogeneity in EGU-specific characteristics in our estimations, thus limiting the scope of omitted variable bias as compared to studies using more aggregated data. We argue that, conditional on our control variables, the unbundling of generation from transmission and distribution can reasonably be regarded as exogenous. We also run falsification tests to see whether reforms designed to improve the efficiency of state plants also affected centrally owned coal-fired power plants.
3. EMPIRICAL STRATEGY
To examine the impact of restructuring on the operating efficiency of state owned power plants, we use EGU-level data on measures of operating reliability and plant-level data on thermal efficiency as outcome variables. Operating reliability is measured by the percentage of time in a year an EGU is available to generate electricity (unit availability), and the percentage of time a unit is forced to shut down due to equipment failures (forced outage). (14) Thermal efficiency is measured by coal consumption per kWh and by operating heat rate. We also estimate the impact of reform on the capacity utilization of the EGU (percent of time the EGU generates electricity).
The time variation in restructuring across states allows us to use a difference-in-difference (DD) estimator. Figure 2 shows the proportion of EGUs in states that have restructured, by year. With data at the EGU-level, we estimate the impact of unbundling on generation efficiency controlling for time-invariant characteristics of EGUs, year fixed effects and linear time trends specific to each state. The baseline model is estimated using the following specification,
[mathematical expression not reproducible] (1)
where [Y.sub.ist] is the measure of generation efficiency for EGU i in state s in year t. In the thermal efficiency models, i refers to
the plant, as data for operating heat rate and specific coal consumption are available only at the plant level. The variable of interest is 1[[Unbundled].sub.st], a policy indicator that takes a value of 1 starting in the year after state s unbundles its SEB; [empty set] thus estimates the average effect of the policy (averaged over time and across states). A positive and statistically significant estimate of [empty set] for unit availability and capacity utilization and a significant negative estimate for forced outage, specific coal consumption and heat rate is evidence of an average improvement in the efficiency of generation as a result of reform.
All baseline specifications estimate the impact of reforms controlling for EGU/plant fixed effects, [[theta].sub.i], and year fixed effects, [[tau].sub.t]. The inclusion of fixed effects controls for all time-invariant characteristics that affect the generation performance of an EGU or plant. The inclusion of year dummies captures macroeconomic conditions and changes in electricity sector policy that affect generation in the country as a whole. (15) The upward trend in operating reliability at both state and central plants throughout the sample period (see Figure 1) implies that without year fixed effects estimates of the impact of unbundling would be overestimated. Estimates of the effects of unbundling may also be biased due to differing pre-reform trends between states that restructured their SEBs and those that did not. To control for this, the baseline specifications include state-specific time trends, [TREND.sub.st].
The estimated models also control for EGU and plant level characteristics that directly affect generation performance. The EGU models include a quadratic age term. (16) The thermal efficiency regressions include average unit capacity in the plant, the heating content of coal (gross calorific value per kg), the average design heat rate and a quadratic term in average plant age.
To examine whether the impact of unbundling varies with the phase of unbundling, we estimate a variant of (1) that interacts the unbundled variable with indicators for Phase 1 and Phase 2 states,
[mathematical expression not reproducible] (2)
1[[PhaseUnb].sub.kst] takes the value of 1 after unbundling of the SEB in state s belonging to group k (k = Phase 1, Phase 2) and [[empty set].sub.k] is the estimate of the impact of unbundling for state-group k relative to the counterfactual of not having unbundled by 2009--the last year of the data.
In addition to examining heterogeneous treatment effects, we test for persistence in reform impacts over time. To do this, we interact the unbundled variable with a set of biennial dummy variables post reform; these measure the impact of reform 1-2 years after reform, 3-4 years after reform, and so on. Estimation of dynamic duration effects is of interest for two reasons. First, it is important to check whether reforms result in a persistent change in operating efficiency at unbundled power plants. A temporary increase in efficiency followed by a reversion to the mean may still yield a positive, significant average treatment effect in the short-term.
Second, Wolfers (2006) points out the potential for bias in estimating average treatment effects when panel-specific trends are included in a difference-in-difference analysis. Since the average treatment effect captures the average deviation from trends in the post-treatment period, incorrectly estimated pre-treatment trends cause the estimate to be biased. The likelihood of bias is increased when the estimation sample contains a relatively short pre-treatment period. In this case, a reversal of the trend in the post-treatment period would have a disproportionate effect on estimates of the trend coefficients. Thus, allowing full flexibility in post-treatment impacts (dynamic effects) enables the trend slope coefficients to be determined by the pre-treatment period trends and allows us to examine the evolution of efficiency increases after unbundling reform.
The estimate of dynamic effects of reform relies on the following specification,
[mathematical expression not reproducible] (3)
In equation (3) the unbundling variable is multiplied by a set of indicator variables that represent the number of years since the reform. [mathematical expression not reproducible] if between t and (t + 1) years have elapsed since the reform and [mathematical expression not reproducible] estimates the average impact for the same time period.
4. ECONOMETRIC ISSUES AND IDENTIFICATION
An obvious concern in estimating the impacts of reform is that the adoption of reform across states may be endogenous, thus biasing estimated impacts. Endogeneity may result from state officials explicitly considering potential efficiency improvements in deciding when to implement reform, or from unobserved heterogeneity across states that drives both the decision to reform and improvements in power plant performance. If states where power plants were likely to gain most from reform were more likely to reform first, the estimated coefficient on the reform dummy would be biased upward. Alternatively, states with greater institutional capacity may be quicker to reform and more likely to benefit from it--also resulting in a positive bias. Although it is impossible to rule out all sources of bias, our estimation strategy and the institutional context of power sector reforms in India should reduce endogeneity concerns.
First, the inclusion of EGU fixed effects controls for any time-invariant differences across EGUs, including factors such as state location (vis-a`-vis coal mines and the transmission grid) and institutional capacity (which may be regarded as fixed over the sample period). The inclusion of state-specific time trends controls for any linear time-varying unobserved differences across states and addresses the concern that adoption of reform may be associated with pre-existing trends in power plant performance.
Second, the adoption of reform was a decision taken at the state level by bureaucrats and politicians. It is more likely that political factors determined the decision to restructure state electric utilities than beliefs about generation efficiency (Erdogdu 2013). Tongia (2003) cites opposition from the agricultural sector as a factor that delayed the adoption of reforms by some states, given that one objective of reforms was to reduce subsidies to agricultural consumers. The political importance of agricultural constituencies may have delayed the adoption of even the initial stages of reform (i.e., unbundling); (17) however, this is unlikely to bias estimates of generation efficiency.
A third econometric concern is that the coefficient on unbundling may be capturing nonlinear time-varying factors that are specific to the state but not related to unbundling. To account for this possibility we take advantage of the presence of power plants owned by the central government that operate in many states across the country. These power plants are owned and operated by the National Thermal Power Corporation (NTPC) and the Damodar Valley Corporation (DVC). They operate outside the structure of the SEBs and are thus not directly affected by restructuring. (18)
To account for state-specific non-linear year shocks, we employ a triple-difference (DDD) specification that includes central power plants and uses state-year dummy variables,
[mathematical expression not reproducible] (4)
In equation (4), [Y.sub.isot] is the outcome at EGU i in state s under ownership o in year t. [h.sub.ot] represents the full set of ownership (state/central) year effects and [[psi].sub.st] represents the full set of state-year effects. The specification thus controls for time effects in each state and time effects for each ownership type. The estimate of the impact of unbundling, [empty set], is identified by the variation in ownership-state-year (as compared to state-year variation that identifies the estimate in the DD specification).
The DDD estimate takes the following form,
[mathematical expression not reproducible] (5)
where [[DELTA].sup.t] [Y.sub.U] is the change in the outcome post reform for states that unbundle and [[DELTA].sup.t] [Y.sub.B] is the corresponding change for non-reforming states. The difference of these values for center-owned EGUs is subtracted from the difference for state-owned EGUs to obtain the estimate of the impact of unbundling reform.
We use data from the Central Electricity Authority (CEA) of India's Performance Review of Thermal Power Stations (CEA various years) to construct an unbalanced panel of 385 EGUs for the years 1988-2009. (19) Of the 385 EGUs, 270 operate in 60 state-owned generation plants and 115 are in 23 central-government-owned plants. The units in the dataset constitute 83 percent of the total installed coal-fired generation capacity in the country in the year 2009-2010. (20) Additional information on the date that the SERCs were established, the date of the unbundling reforms for
each state and ownership information for each power plant was obtained from the websites of the individual SERCs and the CEA.
Tables 2A and 2B present summary statistics that compare state EGUs (Table 2A) and plants (Table 2B) by phase of reform in the period prior to restructuring (1988-1995) and at the end of the sample period (2006-2009). Tables 3A and 3B present similar comparisons between state and central EGUs (Table 3A) and plants (Table 3B).
Prior to the first unbundling reforms in 1996, Phase 1 states were performing slightly worse than other states. The EGUs in these states were older, smaller, had higher forced outages, slightly lower availability and lower thermal efficiency compared to Phase 2 states. This pattern was reversed by 2006-09: Phase 1 states were now statistically indistinguishable in terms of performance measures--forced outages, availability, capacity utilization--from Phase 2 states. (21) Operating heat rate at plants in Phase 1 states was also slightly below operating heat rate in Phase 2 states by 2006-09, although the difference is not statistically distinguishable. This suggests that between 1996 and 2006 the states that unbundled early (Phase 1 states) outperformed the states that were just beginning to unbundle their SEBs in 2004 (Phase 2 states). The tables also show a drop in the average design heat rate of plants in Phase 1 states, which implies that at least a part of the gains in average performance measures are due to the addition of newer and more efficient units.
The comparison between state and central plants in Tables 3A and 3B confirms that central plants were significantly more efficient than state plants throughout the sample period. Over the years 1988-1995, the average capacity utilization of state EGUs was about 10 percentage points lower than EGUs at centrally owned plants. Coal consumption per kWh was about 7 percent higher at state plants. A comparison of operating heat rates at state and central plants is more difficult, as data are often missing for plants operated by the National Thermal Power Corporation (NTPC).
During the sample period, both state and central plants improved in reliability, but showed little improvement in thermal efficiency. Table 3 indicates that EGUs in both sets of plants have experienced large gains in capacity utilization (an average increase of 19 percentage points for state and 25 percentage points for central plants) and smaller gains in plant availability (an average increase of 13 percentage points for both central and state plants). Forced outages also decreased substantially at both sets of plants. There was, in contrast, little change in coal consumption per kWh.
6.1 Difference-in-Difference Results for Thermal Efficiency
We measure the impacts of unbundling on thermal efficiency using both specific coal consumption (kg/kWh) and operating heat rate (kcal/kWh). The models are estimated using plant-level data. Plants owned by the central government cannot be used as controls since data on thermal efficiency are often missing for these plants.
Coal burned per kWh depends on the design heat rate of the boiler (e.g., boilers designed to burn high-ash coal have higher design heat rates and thus require more coal), the heating value of the coal burned, and the age and capacity of the boiler (Joskow and Schmalensee 1987). Coal consumption per kWh should decrease with the heating value of the coal and capacity of the boiler and should increase with boiler age. (22) In estimating models of coal consumption we treat the heating value of the coal as exogenous to the plant. Given the structure of the Indian coal market, plant managers cannot choose coal quality. Power plants are linked to coal mines by a central government committee and thus have little leeway in determining the quality of the coal received. (23)
Operating heat rate (OPHR) is the sum of coal burned per kWh, multiplied by the heating value of the coal, plus oil burned per kWh, multiplied by the heating value of the oil. Although OPHR captures oil as well as coal usage, we expect the impact of unbundling on operating heat rate to be similar to its impact on coal consumption per kWh. (24) One way in which restructuring could reduce coal consumption and operating heat rate are through the purchase of newer generating equipment. This should improve thermal efficiency because boilers generally deteriorate as they age and, new boilers embody technical improvements. It is also possible to improve thermal efficiency by pulverizing coal before it is burned and by performing regular maintenance of boilers. By holding equipment age constant in our thermal efficiency models we focus on the change in efficiency due to managerial factors.
Table 4 indicates that after controlling for plant characteristics, year dummy variables and state-level trends, there is no evidence to support the hypothesis that unbundling improved the thermal efficiency of state-owned power plants. Plant characteristics have the expected signs; however, average treatment effects in columns  and  show no significant impact of unbundling on operating heat rate and a significant positive impact on specific coal consumption. Examining the heterogeneous impacts in column  and  reveals that plants in Phase 2 states experience a statistically significant worsening in thermal efficiency post unbundling reforms--this is also what drives the average impact of specific coal consumption in column . This result is consistent with large increases in specific coal consumption observed in Gujarat and Maharashtra beginning in 2005. These increase could be due to idiosyncratic shocks to the quality of coal (e.g., to its ash and moisture content) for which we do not have data.
Our results, which show no significant improvement in thermal efficiency as a result of restructuring, are consistent with the results of Hiebert (2002) and Fabrizio et al. (2007). Hiebert find mixed effects of restructuring on the technical efficiency of coal-fired power plants in US states that restructured their electricity sectors (improvements in 1996 but not in 1997). Fabrizio et al. (2007) find no improvement in fuel input usage at plants in states that restructured their electricity sectors. It should, however, be noted that both studies look at the impacts of restructuring shortly after states separated generation from distribution. Our panel follows plants in Phase 1 states for an average of 10 years after unbundling.
6.2 Difference-in-Difference Results for Operating Reliability
Columns  and  of Table 5 show the average effect of unbundling of SEBs on EGU availability and forced outage. Availability is the percentage of hours in a year that the EGU is available to produce electricity; forced outage is the percentage of time that the EGU is forced to shut down due to breakdowns and mechanical failures. The results in Column  and  indicate that the average impact of unbundling on state EGUs is statistically insignificant from zero.
Columns  and  of Table 5, however, show that states that unbundled prior to the Electricity Act of 2003 experienced a statistically significant improvement in operating reliability: average EGU availability increased by 6.8 percentage points. This increase represents a 10 percent increase over 1995 levels. The improvements in availability were largely driven by a reduction in forced outages. The unbundling of generation resulted in a 5.1 percentage point reduction in the time lost from breakdowns, a 25 percent reduction from average forced outage for these states in 1995.
Column  shows a decline in EGU availability in Phase 2 states due to unbundling that is significant at the 10 percent level, but no statistically significant impact on forced outages. Because plant availability, forced outages and planned maintenance must sum to 100 percent, this implies that the reduction in availability is due to increased plant maintenance. This is a very different outcome than an increase in forced outages and need not represent a decline in efficiency.
Table 6 presents robustness checks for the operating reliability models. These indicate that the reduction in forced outages in Phase 1 states is robust to sample specification and representation of time trends. For Phase 1 states the increase in EGU availability and reduction in forced outages is affected only slightly by dropping Phase 2 states from the models (i.e., to using only states that did not restructure during the sample period as a control group). This is also the case when state time trends are replaced by time trends for the three phases of unbundling.
Table 6 also investigates the impact of the decommissioning and commissioning of EGUs on our results. Columns  and  re-estimate the models dropping observations for the EGUs that were shut down during the sample period. This eliminates the possibility that units that were shut down are driving the results in Table 5. This slightly reduces the impact of unbundling on forced outages and plant availability, to -3.7 and 4.9 percentage points, respectively. To test whether it is new EGUs that are driving the results we estimate the models using EGUs that were installed pre-reform and remain in the dataset through 2009 (columns  and ). Columns  and  suggest that unbundling significantly improved the performance of equipment that was installed before unbundling in Phase 1 states, reducing forced outages by about 5 percentage points and increasing availability by about 6 percentage points.
As is the case for Phase 1 states, results for Phase 2 states are also robust to choice of sample. The reduction in availability at Phase 2 plants remains statistically significant and is associated with increased restoration and maintenance of EGUs, rather than an increase in forced outages.
6.3 Triple-difference Estimates of Operating Reliability
The triple-difference (DDD) specifications include EGUs at central power plants as an additional control group. The validity of central power plants as a control group rests partly on SEB reforms having no impact on the operating reliability of these plants. To test this, we estimate a model of the impacts of SEB restructuring on EGUs at central power plants. The results, presented in the Appendix, show that there is no evidence of unbundling reforms on operating availability or forced outages at central EGUs--the magnitude of the coefficients is small and the standard errors are large.
Table 7 presents the results from the DDD estimation of the impact of unbundling, by phase. The results in Table 7 are qualitatively similar to those in Table 5 for the DD specification. The coefficient estimates in columns  and  show a statistically significant increase in availability of 6 percentage points--equivalent to an additional 700 MW becoming available for electricity production--and a decrease in forced outage for EGUs in Phase 1 states of 5 percentage points. These results are robust to dropping from the sample units that were shut down (columns  and ). Results for Phase 2 states, although qualitatively similar to Table 5, are no longer statistically significant. When the DDD model is estimated using EGUs that were installed pre-reform and remain in the dataset through 2009, the impact of unbundling on forced outages is unaffected, suggesting that reforms improved existing capacity; however, the impact on availability is estimated less precisely.
6.4 Dynamic Effects of the Impact of Unbundling
The estimated average treatment effects for units in Phase 1 states could reflect an initial impact of reform that declined over time. Our analysis of the dynamic impacts of restructuring suggests that this is not the case. Using equation (3), we estimate the impact of unbundling by interacting a series of biennial dummy variables with the unbundling variables. Figures 4A to 4D plot the estimated coefficients of time dummy variables that represent two-year intervals after reform for Phase 1 states. (25)
Figures 3A and 3B show a similar pattern of the impact on forced outage for both DD (Figure 3A) and DDD (Figure 3B) specifications. The DD coefficients are, however, less precisely estimated. The DDD estimates in Figure 3B suggest a lag in the reduction of forced outages after unbundling for Phase 1 states. The impact is significant starting 3 years after unbundling, and is largest 3, 5 and 9 (or more) years after reform.
Figures 3C and 3D plot the results from a more flexible specification of the DDD model. Here, we allow both the pre- and post-reform time effects for state-owned EGUs to vary non-parametrically. (26) Figure 3C shows that the flexible estimation of the pre-reform trend in forced outage at state-owned EGUs yields a flat trend, conditional on covariates. The evolution of the impact after unbundling is the same as in figure 3B above. Figure 3D indicates that the significant reform impacts on availability for Phase 1 states persist for the duration of the sample.
6.5 Impacts on Capacity Utilization
Since benefits from improved electricity sector performance are primarily delivered through an increase in electricity generated from existing resources, it is important to ask whether the estimated improvement in operating performance at EGUs in Phase 1 states result in greater electricity generation. We check this by estimating the impact of unbundling on (the rate of) capacity utilization of EGUs. Table 8 suggests that, on average, increases in availability were not reflected in increased capacity utilization of state-owned EGUs. Column  and column  report the impacts, by phase, from the DD and DDD specifications. We find no evidence to suggest that, on average, unbundling generation from transmission and distribution led to an increase in capacity utilization at state EGUs.
This result is at variance with the results of Sen and Jamasb (2012) who, using state-level data, find that unbundling resulted in a 26 percentage point increase in capacity utilization at state-owned power plants. Interestingly, average capacity utilization at state-owned EGUs increased by roughly 25 percentage points from 1991 to 2009 (see Figure 1C). However, once we control for plant and year fixed effects and state time trends, this result is unrelated to unbundling.
One reason why increases in availability did not result in greater electricity generation may be that they occurred at higher cost plants. If these plants were not able to underbid lower cost plants in the merit dispatch order, increased availability would not necessarily result in increased capacity utilization. Alternatively, it could also be that there was heterogeneity in the impacts of unbundling on capacity utilization which caused the average effect to be estimated noisily. We note that the sign of the impact of unbundling on average capacity utilization in Table 8 is positive but insignificant for Phase 1 states, suggesting this possibility. (27) We examine the nature of heterogeneity in the impact of reforms by estimating models that allow for differential impacts by EGU size.
Table 9 presents difference-in-difference models which interact the Phase-specific unbundling variable with categorical variables for 4 EGU size categories--EGUs less than 100 MW, 110/120 MW, 210/220/250 and 500 MW. (28) The results show that 110/120 MW units experienced a significant positive increase in operating reliability in Phase 1 states: operating availability increases by about 12 percentage points, largely driven by a 9 percentage point reduction in time lost due to forced outages. The increase in operating availability translated into a roughly 9 percentage point increase in capacity utilization at these EGUs. Indeed, the results in Tables 5-8 appear to be driven by reductions in forced outages at small (100 MW and 110/120 MW) plants.
The estimates for Phase 2 states suggest that the impact of unbundling was to decrease EGU reliability. There is a statistically significant decline in availability which leads to a decline in capacity utilization. The estimates also show that the deterioration associated with reforms at EGUs in Phase 2 states is not due to an increase in forced outages. Thus an increase in maintenance is driving the observed decreases in availability and capacity utilization. As argued above, it is questionable whether this captures a reduction in efficiency due to reform.
This paper has examined the impact of unbundling reforms in the Indian electricity sector on the generation performance of state-owned power plants. Specifically, we have focused on the impact of unbundling of generation on the operating reliability and thermal efficiency of coal-fired power plants. Unbundling may result in an increase in generator efficiency if plant managers are given greater discretionary powers to minimize costs and are faced with hard budget constraints. Unbundling may also improve the operating performance of power plants by allowing managers to focus on decisions related solely to generation. This could result in more timely maintenance decisions and lead to reduced breakdowns and forced outages.
We find that the impacts of unbundling differ greatly between states that restructured their SEBs prior to the Electricity Act of 2003 (Phase 1 unbundlers), which made unbundling mandatory, and those that restructured in 2005 or later (Phase 2 unbundlers). Our results show that unbundling resulted in a statistically significant increase in the average availability of EGUs in states that unbundled between 1996 and 2002. We find that the increase in availability at these EGUs is mainly driven by a corresponding reduction in forced outages. There is no evidence of an impact of restructuring on average capacity utilization or improvements in thermal efficiency.
Results from a model in which plants operated by the central government serve as controls as well as plants in stated that did not unbundle generation suggest a 5.9 percentage point increase in average unit availability and a 4.9 percentage point reduction in forced outages in Phase 1 states. The reduction in forced outages represents a 25 percent reduction from the mean for these states in 1995. Examination of the duration of reform impacts, using a full set of pre- and post-reform dummies, shows that the improvements in generation reliability are not reversed in the short to medium term. Robustness checks confirm that our baseline results are not sensitive to changes in model and sample specifications.
Most of the improvements in operating reliability in Phase I states reflects improvements at small generating units that were installed prior to reforms. Smaller EGUs experienced a significant increase in operating reliability due to reform in Phase 1 states. In Phase 1 states, 110/120 MW EGUs experienced a 9.4 percentage point increase in capacity utilization, driven largely by a reduction in the time lost due to forced outages. The increase in capacity utilization represents a 24 percent increase above the 1995 average at 110/120 MW EGUs and implies an additional 2083 GWh of electricity production per year from these units. (29)
For Phase 2 states, our results suggest that the initial years following reforms were associated with a reduction in availability and capacity utilization, especially at 110/120 MW EGUs, and a decrease in thermal efficiency. The estimated coefficients are unstable and often insignificant, but suggest a worsening in generation performance across various specifications. The estimated deterioration in performance may be due to initial adjustment costs to restructuring in the states that were forced to unbundle. It should also be noted that the reductions in availability at EGUs are due to increases in planned maintenance rather than increases in forced outages.
The offsetting deterioration in Phase 2 states implies that, on average, the impact of reforms has been modest in magnitude. It is safe to say that within the period of study, the gains from unbundling reforms have been limited to an improvement in operating reliability and capacity utilization for the most inefficient plants in the states that unbundled prior to 2003.
Our results disagree with those of Sen and Jamasb (2012) who, using state-level data for India, find that unbundling increased average capacity utilization by 26 percentage points--an extremely large effect. One possible explanation for the difference is that the Sen and Jamasb (2012) may not adequately control for the strong upward trend in the capacity utilization at Indian power plants during the period of their study (see Figure 1C).
Our finding that unbundling per se did not improve thermal efficiency at power plants agrees with studies in the US. Fabrizio et al. (2007) do not find evidence that unbundling affected the thermal efficiency of power plants, although they do find significant reductions in non-fuel expenditures. Studies by Bushnell and Wolfram (2005) and Chan et al. (2013) do, however, find that the sale of generation to IPPs slightly improved thermal efficiency at US power plants. Bushnell and Wolfram (2005) estimate that the divestiture of utilities in the US improved thermal efficiency by about 2%; Chan et al. (2013) find that restructuring led to 1.4% increase in fuel efficiency at investor-owned plants in states that restructured their utility sectors.
The failure to find more widespread impacts from restructuring may reflect the nature and progress of electricity reform in India. Ruet (2005) argues that unbundling and subsequent corporatization has failed to increase the technical and financial autonomy of power plant managers to the extent envisaged at the start of reforms. Executive orders from state governments continue to drive some of the important decisions of generation companies, which may be contrary to cost-minimization objectives.
The incentives for improving fuel efficiency and maintaining equipment to prevent breakdowns also depends on how plants are compensated. Under the 2003 Electricity Act SERCs are to follow the CERC's guidelines in compensating generators. The CERC compensates the power plants under its jurisdiction based on performance. Compensation for energy used in generation is paid based on scheduled generation and depends on operating heat rate. Compensation for fixed costs (depreciation, interest on loans and finance charges, return on equity, operation and maintenance expenses, interest on working capital, and taxes) is based on plant availability. How have SERCs actually compensated power plants? There is evidence that SERCs have set compensation for fuel use based on very high estimates of operating heat rate, suggesting that this may not provide much of an incentive for plants to improve thermal efficiency (Crisil Ltd. 2010).
Bacon and Besant-Jones (2001) emphasize that separating generation from transmission and distribution is likely to be most successful when it is accompanied by tariff reform and when it induces competition in generation. Tariff reform that promotes cost recovery in the electricity sector is needed to make generation profitable. Although tariff reform has begun, in 2006 only 3 of the 10 states that had unbundled were making positive profits (The Energy and Resources Institute 2009, Table 1.80). Another way in which unbundling may increase generation efficiency is through increased competitive pressure from the entry of IPPs into the electricity market. Such an effect followed the restructuring of the US electricity sector, but IPP entry has been slow to develop in India after the initial setbacks following the 1991 reforms. More recently, Indian companies such as TATA, Reliance and Adani have set up large thermal power plants. As more data become available, the estimation of the impact of their entry into the Indian electricity sector will be an interesting area of further research.
This research was supported by grants from the World Bank and Resources for the Future. The paper has been greatly improved by comments from an anonymous referee. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not represent the views of the World Bank, its Executive Directors, or the countries they represent.
Arocena, Pablo, and Catherine Waddams Price (2002). "Generating Efficiency: Economic and Environmental Regulation of Public and Private Electricity Generators in Spain." International Journal of Industrial Organization 20(1): 41-69. http://dx.doi.org/10.1016/S0167-7187(00)00073-4.
Bacon, R.W., and J. Besant-Jones (2001). "Global Electric Power Reform, Privatization, and Liberalization of the Electric Power Industry in Developing Countries." Annual Review of Energy and the Environment 26: 331-59. http://dx.doi.org/10.1146/annurev.energy.26.1.331.
Bushnell, James B., and Catherine Wolfram (2005). " Ownership Change, Incentives and Plant Efficiency: The Divestiture of US Electric Generation" http://economics.ucdavis.edu/people/bushnell/site/files-and-images/csemwp140.pdf
CEA. (2006). All India Electricity Statistics, General Review 2006. Central Electricity Authority, New Delhi, India.
CEA. Various years. Review of Performance of Thermal Power Stations. Central Electricity Authority, Ministry of Power, New Delhi, India.
Chan, Hei Sing, Maureen L. Cropper, and Kabir Malik (2014). "Why Are Power Plants in India Less Efficient than Power Plants in the United States?" American Economic Review Papers and Proceedings 104(5): 586-590. http://dx.doi.org/10.1257/aer.104.5.586.
Chan, Hei Sing, H. Fell, I. Lange, and S. Li (2013). "Efficiency and Environmental Impacts of Electricity Restructuring on Coal-fired Power Plants." Working paper.
Chikkatur, Ananth P., Ambuj D. Sagar, Nikit Abhyankar, and N. Sreekumar (2007). "Tariff-Based Incentives for Improving Coal-Power-Plant Efficiencies in India." Energy Policy 35(7): 3744-3758. http://dx.doi.org/10.1016/j.enpol.2007.01.006.
Chitkara, P. (1999). "A Data Envelopment Analysis Approach to Evaluation of Operational Inefficiencies in Power Generating Units: A Case Study of Indian Power Plants." IEEE Transactions on Power Systems 14(2): 419-425. http://dx.doi.org/10.1109/59.761859.
Crisil Risk and Infrastructure Solutions, Ltd. (2010). Study on Analysis of Tariff Orders and Other Orders of State Electricity Regulatory Commissions.
Cropper, Maureen L., Kabir Malik, Alex Limonov and Anoop Singh (2011). "Estimating the Impact of Restructuring on Electricity Generation Efficiency: The Case of the Indian Thermal Power Sector." NBER Working Paper 17383, September 2011.
Delmas, M., and Y. Tokat (2005). "Deregulation, Governance Structures, and Efficiency: The U.S. Electric Utility Sector." Strategic Management Journal 26(5): 441-460. http://dx.doi.org/10.1002/smj.456.
Erdogdu, Erkan (2013). "A Cross-country Analysis of Electricity Market Reforms: Potential Contribution of New Institutional Economics." Energy Economics 39: 239-251. http://dx.doi.org/10.1016/j.eneco.2013.05.012.
Fabrizio, Kira R., Nancy L. Rose, and Catherine D. Wolfram (2007). "Do Markets Reduce Costs? Assessing the Impact of Regulatory Restructuring on U.S. Electric Generation Efficiency." American Economic Review 97(4): 1250-77. http://dx.doi.org/10.1257/aer.97.4.1250.
Hattori, T. (1999) "Parametric Tests of Cost Efficiency for Japanese Electric Utilities: Before and After the Regulatory Reform in 1995." Unpublished paper.
Hiebert, L. Dean (2002). "The Determinants of the Cost Efficiency of Electric Generating Plants: A Stochastic Frontier Approach." Southern Economic Journal 68(4): 935-46. http://dx.doi.org/10.2307/1061501.
Jamasb, Tooraj, Raffaella Mota, David Newbery, and Michael Pollitt (2005). "Electricity Sector Reform in Developing Countries: A Survey of Empirical Evidence on Determinants and Performance." World Bank Policy Research Working Paper 3549. Washington, DC: The World Bank.
Joskow, Paul L., and Richard Schmalensee (1987). "The Performance of Coal-Burning Electric Generating Units in the United States: 1960-1980." Journal of Applied Econometrics 2(2): 85-109. http://dx.doi.org/10.1002/jae.3950020203.
Khanna, M., and N.D. Rao (2009). "Supply and Demand of Electricity in the Developing World." Annual Review of Resource Economics 1: 567-595. http://dx.doi.org/10.1146/annurev.resource.050708.144230.
Khanna, M., and D. Zilberman (1999). "Barriers to Energy-Efficiency in Electricity Generation in India." The Energy Journal 20(1): 25-41. http://dx.doi.org/10.5547/ISSN0195-6574-EJ-Vol20-No1-2.
Knittel, Christopher R. (2002). "Alternative Regulatory Methods and Firm Efficiency: Stochastic Frontier Evidence from the U.S. Electricity Industry." The Review of Economics and Statistics 84(3): 530-40. http://dx.doi.org/10.1162/003465302320259529.
MIT (2007). "The Future of Coal: An Interdisciplinary MIT Study," Cambridge, MA: Massachusetts Institute of Technology,
Plane, P. (1999). "Privatization, Technical Efficiency and Welfare Consequences: The Case of the Cote D'Ivoire Electricity Company (Cie)." World Development 27(2): 343-60. http://dx.doi.org/10.1016/S0305-750X(98)00139-9.
Pombo, Carlos, and Manuel Ramirez (2003). "Privatization in Colombia: A Plant Performance Analysis." Research Network Working Paper 166. Washington, DC: Inter-American Development Bank.
Rajan, A. Thillai (2000). "Power Sector Reform in Orissa: A Ex-Post Analysis of the Causal Factors." Energy Policy 28: 657-669. http://dx.doi.org/10.1016/S0301-4215(00)00063-X.
Ruet, Joel (2005). Privatising Power Cuts? New Delhi, India: Centre de Sciences Humaines.
Sen, A., and T. Jamasb (2012). "Diversity in Unity: An Empirical Analysis of Electricity Deregulation in Indian States." The Energy Journal 33(1): 83-130. http://dx.doi.org/10.5547/ISSN0195-6574-EJ-Vol33-No1-4.
Shanmugam, K.R., and Praveen Kulshreshtha (2005). "Efficiency Analysis of Coal-Based Thermal Power Generation in India During Post-Reform Era." International Journal of Global Energy Issues 23(1): 15-28. http://dx.doi.org/10.1504/IJGEI.2005.006408.
Singh, Anoop (2006). "Power Sector Reform in India: Current Issues and Prospects." Energy Policy 34(16): 2480-90.
Singh, Joga (1991), "Plant Size and Technical Efficiency in the Indian Thermal Power Industry." Indian Economic Review 26(2): 239-252. http://dx.doi.org/10.1016/j.enpol.2004.08.013.
The Energy and Resources Institute (2009). TERI Energy Data Directory and Yearbook. New Delhi, India: The Energy and Resources Institute.
Thakur, Tripta, S.G. Deshmukh, and S.C. Kaushik (2006). "Efficiency Evaluation of the State Owned Electric Utilities in India." Energy Policy (34): 2788-2804. http://dx.doi.org/10.1016/j.enpol.2005.03.022.
Thakur, Tripta, S.G. Deshmukh, S.C. Kaushik, and Mukul Kulshrestha (2005). "Impact Assessment of the Electricity Act 2003 on the Indian Power Sector." Energy Policy (33): 1187-1198. http://dx.doi.org/10.1016/j.enpol.2003.11.016.
Tongia, R. (2003). "The Political Economy of Indian Power Sector Reforms." Program on Energy and Sustainable Development Working Paper no. 4. Stanford University.
Wolfers, Justin (2006). "Did Unilateral Divorce Laws Raise Divorce Rates? A Reconciliation and New Results". American Economic Review 96 (5): 1802-20. http://dx.doi.org/10.1257/aer.96.5.1802.
Table 1: Falsification--Impact of Unbundling on Central EGUs    Availability Forced Outages Availability [Unbundled] -1.516 -1.504 (2.276) (2.407) [Phase-I (*) Unbundled] -1.845 (3.306) [Phase-II (*) Unbundled] -0.681 (2.104) Time Trend State State State Unit FE Yes Yes Yes Year FE Yes Yes Yes  Forced Outages [Unbundled] [Phase-I (*) Unbundled] -2.175 (3.193) [Phase-II (*) Unbundled] 0.196 (2.515) Time Trend State Unit FE Yes Year FE Yes Notes: Std. errors in parentheses, clustered at state level. (***) p<0.01, (**) p<0.05, (*) p<0.1. All equations control for a quadratic for EGU age, year and plant fixed effects and state time trends. Number of observations = 1756 (119 Units).
Kabir Malik (a), Maureen Cropper (b*), Alexander Limonov (c), and Anoop Singh (d)
(a) The World Bank, 1818 H Street NW, Washington, DC 20433. E-mail: KMalik@worldbank.org.
(b) Department of Economics, University of Maryland, and Resources for the Future, 3114 Tydings Hall, College Park, MD 20742.
(c) Kellogg School of Management, Northwestern University. E-mail: email@example.com.
(d) Department of Industrial & Management Engineering, Indian Institute of Technology Kanpur. E-mail: firstname.lastname@example.org.
(*) Corresponding author, with Angela Harmon. E-mail: email@example.com; Harmon@econ.umd.edu.
(1.) The percentage of time a unit is available equals 100 percent minus the percent of time spent on planned maintenance and the percent of time lost due to forced outages.
(2.) We cannot state this with certainty as we do not have data on operating costs for power plants.
(3.) In 2009-10 (CEA 2010) 53% of installed capacity connected to the grid was coal-fired, 11% fired by natural gas, 23% hydro, 3% nuclear and the remainder renewables; however 70% of electricity was generated by coal-fired power plants.
(4.) We focus on the operating heat rate of state-owned plants, as data on operating heat rate of central-government-owned plants are often not reported in the Central Electricity Authority's Thermal Power Reports (various years).
(5.) Whenever a plant is started up after an outage, more coal is burned than during the normal operation of the plant.
(6.) As a referee has pointed out, some of the IPPs that initially entered the Indian electricity market (e.g., Enron and Cogentrix) exited quickly, to cut their losses.
(7.) Table 1 lists only those states containing thermal power plants. Our study focuses on coal- and lignite-fueled plants.
(8.) Assam unbundled in 2004, but its only coal-fired power plant was decommissioned in 2001-02. We retain Assam in the dataset; however, for Phase 2 plants, the first year of unbundling is, effectively, 2005, the year in which Maharashtra unbundled.
(9.) The studies are Plane (1999), Arocena and Waddams (2002), Hattori (1999), and Delmas and Tokat (2005). See also Pombo and Ramirez (2005).
(10.) Knittel examines six different programs: compensation based on heat rate, compensation based on an equivalent availability factor, price-cap programs, rate-of-return range programs, fuel-cost pass-through programs, and revenue-decoupling programs. His sample includes both gas- and coal-fired power plants.
(11.) There are several studies that examine the technical/thermal efficiency of Indian power plants, but do not examine the impact of electricity sector reforms on efficiency. Singh (1991), Chitkara (1999), Shanmugam and Kulshreshta (2005) and Thakur et al. (2006) examine how far individual plants are from the production frontier. Khanna and Zilberman (1999) explain variation in the thermal efficiency of plants as a function of plant ownership.
(12.) The analysis reported is for 245 observations across 18 states and 17 years. Variables are defined at the state level, so the analysis measures the impact of reforms on all power plants--state-owned, privately owned, and centrally owned--within a state.
(13.) The regulatory dummies are: presence of independent power producers, establishment of a SERC, unbundling of generation from transmission and distribution, passing of a tariff order, open access to transmission facilities, and privatization of distribution.
(14.) The percentage of time a unit is available equals the 100 percent minus the percent of time spent on planned maintenance and the percent of time lost due to forced outages.
(15.) In 2003 an Unscheduled Interchange charge was instituted throughout the country to compensate (penalize) plants supplying unscheduled electricity to the grid when there is excess demand (supply).
(16.) Other characteristics such as capacity, vintage and make of boiler/EGU also impact generation performance, but are time-invariant and thus subsumed by the EGU fixed effects.
(17.) It is not surprising that Orissa was the first state to reform, given the small electricity load in agriculture and weak farmer lobby in the state (Rajan 2000).
(18.) To confirm this, we conduct a falsification test to estimate the impact of state SEB unbundling on operating reliability of central EGUs using equations (1) and (2). The impact is statistically indistinguishable from zero.
(19.) The CEA reports are not available for the years 1992 and 1993. These years are thus omitted from our data. A year in the dataset is an Indian fiscal year. Thus, 1994 refers to the time period April 1, 1994, through March 30, 1995.
(20.) In 2009-10, 9% of coal-fired generating capacity was privately owned, 53% state owned and 38% owned by the central government.
(21.) Average forced outage was lower in Phase 1 states compared to Phase 2 in the period 2006-09; however, the difference in means is not statistically significant.
(22.) Because our models are estimated at the plant level, variables measured at the level of the EGU (such as age) have been aggregated to the plant level by weighting each unit by its nameplate capacity. The average nameplate capacity is a simple average of EGU capacity in the plant.
(23.) The use of washed (beneficiated) coal, which has a higher heating value, is also mandated through regulation and not determined by plant managers.
(24.) Because coal constitutes most of the kcal used to generate electricity, OPHR [approximately equal to] (Coal per kWh)*(Heating Value of Coal). It follows that the coefficient of ln(Heating Value of Coal) in the ln(OPHR) equation should approximately equal 1 plus the coefficient of ln(Heating Value of Coal) in the ln(Coal Consumption per kWh) equation. Our results confirm this.
(25.) The dummy year categories are 1-2 years, 3-4 years, 5-6 years, 6-7 years and 9 + years since unbundling. The last category captures up to 13 years after unbundling in the case of Orissa. We combine years greater than 9 into one dummy because the number of observations is too low to estimate finer categories.
(26.) This is similar to an event study specification.
(27.) The magnitude of the average term may be reduced due to gains in capacity utilization (or reliability) at some generators and possible deterioration at others--e.g. due to adjustment costs of restructuring.
(28.) We define each group based on a range of nameplate capacities that is largely composed of these capacities.
(29.) State-owned thermal power plants generated 240.8 TWh ([10.sup.3] GWh) of electricity in 2005 (CEA 2006). This figure includes gas-fired plants.
Table 1: Timeline of Reforms by States under the 1998 and 2003 Electricity Acts Unbundling Phase State SERC operational SEB unbundled Orissa 1995 1996 Andhra Pradesh 1999 1998 Haryana 1998 1998 Phase 1 Karnataka 1999 1999 Uttar Pradesh 1999 1999 Rajasthan 2000 2000 Delhi 1999 2002 Madhya Pradesh Assam 1998 2001 2002 2004 Maharashtra 1999 2005 Phase 2 Gujarat 1998 2006 West Bengal 1999 2007 Chhattisgarh 2000 2008 Punjab 1999 2010 Phase 3 Tamil Nadu 1999 2010 Bihar 2005 2012 Jharkhand 2003 approved Dec 2013 Table 2A: Variable Means, State-owned EGUs, by Unbundling Phase (EGU Data) Phase-I Phase-II 1988-1995 1988-1995 Mean Std. dev. Mean Std. dev.     Nameplate capacity (MW) 117 73 146 74 Generation (GWh) 534 489 686 498 Age (yrs.) 14.8 8.0 13.5 8.2 Forced outages (%) 21.5 20.4 16.8 20.4 Planned maintenance (%) 12.2 18.7 14.2 18.7 Availability (%) 66.3 23.4 69.0 23.8 Capacity utilization (%) 50.0 21.2 49.8 20.7 2006-2009 2006-2009 Nameplate capacity (MW) 164 91 172 86 Generation (GWh) 1062 750 1052 656 Age (yrs.) 23.0 12.2 24.6 11.7 Forced outages (%) 10.8 14.7 12.6 16.3 Planned maintenance (%) 8.2 15.6 6.1 9.8 Availability (%) 81.0 19.8 81.4 18.0 Capacity utilization (%) 69.1 23.7 68.1 20.1 Phase-III 1988-1995 Mean Std. dev. Diff. in means   - - Nameplate capacity (MW) 131 60 -29 (***) -14 (***) Generation (GWh) 561 465 -152 (***) -27 Age (yrs.) 12.9 7.5 1.3 (**) 1.8 (***) Forced outages (%) 17.6 17.2 4.6 (***) 3.9 (**) Planned maintenance (%) 18.3 27.4 -2 -6.1 (***) Availability (%) 64.1 26.4 -2.6 (*) 2.2 Capacity utilization (%) 46.0 24.0 0.2 3.9 (**) 2006-2009 Nameplate capacity (MW) 159 61 -8 4 Generation (GWh) 1038 664 10 24 Age (yrs.) 24.7 9.3 -1.6 (*) -1.7 (*) Forced outages (%) 13.3 18.4 -1.8 -2.5 Planned maintenance (%) 12.6 23.1 2.1 (**) -4.4 (**) Availability (%) 74.2 27.6 -0.3 6.9 (***) Capacity utilization (%) 66.1 30.0 1 3 Notes: Phase 1 (pre-2003): Andhra Pradesh, Haryana, Karnataka, Orissa, Rajasthan, Uttar Pradesh, Delhi, and Madhya Pradesh. Phase 2 (post-2003): Gujarat, Maharashtra, West Bengal, Chhattisgarh and Assam. Phase 3 (out-of-sample): Bihar, Punjab, Tamil Nadu and Jharkhand. GWh, gigawatt-hours; MW, megawatts. 1988-1995 does not contain data for 1992 and 1993. Difference in means according to a two-sample t-test with unequal variances (***) p<0.01, (**) p<0.05, (*) p<0.1. Number of observations (1988-1995): Phase 1-466, Phase 2- 461, Phase 3-217. Number of observations (2006-2009): Phase 1- 399, Phase 2-370, Phase 3-155. Table 2B: Variable Means, State-owned Plants, by Unbundling Phase (Plant Data) Phase-I 1988-1995 Obs. Mean Std. Dev. Obs.     No. of operating units 117 3.98 3.03 118 Nameplate capacity (MW) 115 473 421 117 Heating value of coal (kcal/kg) 58 4203 617 67 Design heat rate (kcal/kWh) 36 2633 194 41 Operating heat rate (kcal/kWh) 59 3478 950 69 Specific coal cons. (kg/kWh) 98 0.83 0.15 103 2006-2009 2006-2009 No. of operating units 86 4.64 2.76 93 Nameplate capacity (MW) 86 760 551 93 Heating value of coal (kcal/kg) 48 3547 386 45 Design heat rate (kcal/kWh) 53 2405 177 66 Operating heat rate (kcal/kWh) 53 2901 642 65 Specific coal cons. (kg/kWh) 76 0.82 0.13 63 Phase-II 1988-1995 Mean Std. Dev. Obs.    No. of operating units 3.91 1.67 50 Nameplate capacity (MW) 574 383 49 Heating value of coal (kcal/kg) 4307 604 32 Design heat rate (kcal/kWh) 2438 148 12 Operating heat rate (kcal/kWh) 3135 537 32 Specific coal cons. (kg/kWh) 0.72 0.12 49 2006-2009 No. of operating units 3.98 1.90 44 Nameplate capacity (MW) 685 509 44 Heating value of coal (kcal/kg) 3673 493 29 Design heat rate (kcal/kWh) 2423 201 29 Operating heat rate (kcal/kWh) 2932 323 29 Specific coal cons. (kg/kWh) 0.78 0.09 41 Phase-III 1988-1995 Mean Std. Dev. Diff.   - No. of operating units 4.34 2.41 0.08 Nameplate capacity (MW) 580 285 -100 (*) Heating value of coal (kcal/kg) 3809 380 -104 Design heat rate (kcal/kWh) 2486 70 [95 (***) Operating heat rate (kcal/kWh) 3210 664 342 (**) Specific coal cons. (kg/kWh) 0.82 0.13 0.11 (***) No. of operating units 3.52 1.73 0.66 (*) Nameplate capacity (MW) 561 347 74.4 Heating value of coal (kcal/kg) 3773 334 -125 Design heat rate (kcal/kWh) 2383 110 -18.2 Operating heat rate (kcal/kWh) 2777 456 -31.9 Specific coal cons. (kg/kWh) 0.78 0.15 0.04 (**) in means - No. of operating units -0.36 Nameplate capacity (MW) -107 (*) Heating value of coal (kcal/kg) 394 (***) Design heat rate (kcal/kWh) [47 (***) Operating heat rate (kcal/kWh) 268 Specific coal cons. (kg/kWh) 0.01 No. of operating units 1.12 (***) Nameplate capacity (MW) [99 (**) Heating value of coal (kcal/kg) -226 (***) Design heat rate (kcal/kWh) 21.9 Operating heat rate (kcal/kWh) 123 Specific coal cons. (kg/kWh) 0.04 Notes: Phase 1 (pre-2003): Andhra Pradesh, Haryana, Karnataka, Orissa, Rajasthan, Uttar Pradesh, Delhi, and Madhya Pradesh. Phase 2 (post-2003): Gujarat, Maharashtra, West Bengal, Chhattisgarh and Assam. Phase 3 (out-of-sample): Bihar, Punjab, Tamil Nadu and Jharkhand. GWh, gigawatt-hours; MW, megawatts; kcal/kWh, kilo-calories/kilowatt-hours. 1988-1995 does not contain data for 1992 and 1993. Difference in means according to a two-sample x-test with unequal variances (***) p<0.01, (**) p<0.05, (*) p<0.1. Table 3A: Variable Means, by Sector (EGU Data) CENTER STATE 1988-1995 1988-1995 Mean St Dev Mean St Dev Diff. in means     - Nameplate capacity (MW) 194 132 131 72 62.80 (***) Generation (GWh) 1046 917 602 493 443.6 (***) Age (yrs.) 13.5 10.7 13.9 8.0 -0.36 Forced outages (%) 14.9 16.8 18.7 19.7 -3.82 (***) Planned maintenance (%) 9.4 13.9 14.2 20.7 -4.79 (***) Availability (%) 75.7 19.9 67.1 24.1 8.623 (***) Capacity utilization (%) 59.5 21.1 49.2 21.5 10.23 (***) 2006-2009 2006-2009 Nameplate capacity (MW) 259 155 166 85 93.01 (***) Generation (GWh) 1928 1281 1054 699 873.4 (***) Age (yrs.) 20.2 12.2 23.9 11.6 -3.72 (***) Forced outages (%) 5.6 9.6 11.9 16.0 -6.36 (***) Planned maintenance (%) 5.8 5.5 8.1 15.4 -2.28 (***) Availability (%) 88.7 10.5 80.0 20.8 8.642 (***) Capacity utilization (%) 84.7 14.2 68.2 23.6 16.49 (***) Notes: GWh, gigawatt-hours; MW, megawatts. 1988-1995 does not contain data for 1992 and 1993. Difference in means between State and Central plants according to a two-sample t-test with unequal variances (***) p<0.01, (**) p<0.05, (*) p<0.1. Number of observations (1988-1995): Center- 404, State- 1141. Number of observations (2006-2009): Center-435, State-924. Table 3B: Variable Means, by Sector (Plant Data) CENTER 1988-1995 Obs. Mean Std. Dev. Obs.     No. of operating units 92 4.39 2.26 285 Nameplate capacity (MW) 90 872 601 281 Heating value of coal (kcal/kg) 42 4092 543 157 Design heat rate (kcal/kWh) 12 2530 164 89 Operating heat rate (kcal/kWh) 43 2984 387 160 Specific coal cons. (kg/kWh) 67 0.73 0.12 250 2006-2009 No. of operating units 87 5.00 2.17 223 Nameplate capacity (MW) 87 1297 854 223 Heating value of coal (kcal/kg) 11 4323 267 122 Design heat rate (kcal/kWh) 23 2505 137 148 Operating heat rate (kcal/kWh) 23 3138 398 147 Specific coal cons. (kg/kWh) 74 0.71 0.07 180 STATE 1988-1995 Mean Std. Dev. Diff. in means   - No. of operating units 4.01 2.44 0.38 Nameplate capacity (MW) 534 386 338 (***) Heating value of coal (kcal/kg) 4167 598 -75 Design heat rate (kcal/kWh) 2523 185 6.73 Operating heat rate (kcal/kWh) 3276 751 -293 (***) Specific coal cons. (kg/kWh) 0.78 0.14 -0.05 (***) 2006-2009 No. of operating units 4.14 2.28 0.86 (***) Nameplate capacity (MW) 689 502 608 (***) Heating value of coal (kcal/kg) 3647 424 676 (***) Design heat rate (kcal/kWh) 2409 178 96 (***) Operating heat rate (kcal/kWh) 2890 486 247 (**) Specific coal cons. (kg/kWh) 0.80 0.12 -0.08 (***) Notes: GWh, gigawatt-hours; MW, megawatts; kcal/kWh, kilo-calories/kilowatt-hours. 1988-1995 does not contain data for 1992 and 1993. Difference in means between State and Central plants according to a two-sample t-test with unequal variances (***) p<0.01, (**) p<0.05, (*) p<0.1. Table 4: Thermal Efficiency--Impact of Unbundling on State Plants   Log Log Heat rate Specific Coal Cn. [Unbundled] 0.0320 0.0356 (*) (0.0201) (0.0189) [Phase-I (*) Unbundled] [Phase-II (*) Unbundled] ln(Design Heat Rate) 0.491 (***) 0.483 (***) (0.157) (0.138) ln(Heating Value of Coal) 0.514 (***) -0.451 (***) (0.0890) (0.0869) Average Age 0.00578 (**) 0.00786 (**) (0.00261) (0.00347) Average Age 2 0.000139 (***) 8.20e-05 (4.35e-05) (5.04e-05) Average Nameplate Capacity -0.000953 -0.000572 (0.000698) (0.000677) Time Trend State State Plant FE Yes Yes Year FE Yes Yes Observations 478 478   Log Log Heat rate Specific Coal Cn. [Unbundled] [Phase-I (*) Unbundled] -0.0183 -0.0107 (0.0229) (0.0179) [Phase-II (*) Unbundled] 0.0820 (***) 0.0818 (***) (0.0223) (0.0207) ln(Design Heat Rate) 0.448 (***) 0.444 (***) (0.133) (0.117) ln(Heating Value of Coal) 0.508 (***) -0.457 (***) (0.0834) (0.0824) Average Age 0.00711 (**) 0.00908 (**) (0.00259) (0.00339) Average Age 2 0.000120 (**) 6.46e-05 (4.45e-05) (4.91e-05) Average Nameplate Capacity -0.000872 -0.000498 (0.000659) (0.000644) Time Trend State State Plant FE Yes Yes Year FE Yes Yes Observations 478 478 Notes: Std. errors in parentheses, clustered at state level. (***) p<0.01, (**) p<0.05, (*) p<0.1. All equations control for a quadratic for plant age, average capacity, design heat rate, heat content of coal, year and plant fixed effects and state time trends. Number of observations = 478 (46 Plants). Table 5: Operating Reliability--Impact of Unbundling on State EGUs    Avera ge Impacts Heteroge Availability Forced Outages Availability [Unbundled] 0.743 -1.824 (1.885) (1.352) [Phase-I (*) Unbundled] 6.793 (**) (2.819) [Phase-II (*) Unbundled] -5.559 (*) (2.993) Time Trend State State State Unit FE Yes Yes Yes Year FE Yes Yes Yes  neous Impacts Forced Outages [Unbundled] [Phase-I (*) Unbundled] -5.110 (***) (1.726) [Phase-II (*) Unbundled] 1.599 (2.467) Time Trend State Unit FE Yes Year FE Yes Notes: Std. errors in parentheses, clustered at state level. (***) p<0.01, (**) p<0.05, (*) p<0.1. All equations control for a quadratic for EGU age, year and plant fixed effects and state time trends. Number of observations = 4298 (270 Units). Table 6: Robustness Checks--Impact of Unbundling on State EGUs   Drop Phase 2 Forced Availability Outages 1 [[Phase-I (*)Unbundled].sub.it] 5.983 (**) -3.885 (**) (2.512) (1.447) 1 [[Phase-II (*) Unbundled].sub.it] Time Trend State State Observations 2,605 2,605 Number of units 166 166   Phase Trends Forced Availability Outages 1 [[Phase-I (*)Unbundled].sub.it] 6.711 (**) -5.258 (***) (2.870) (1.740) 1 [[Phase-II (*) Unbundled].sub.it] -6.656 (**) 1.754 (3.097) (2.350) Time Trend Phase Phase Observations 4,298 4,298 Number of units 270 270   Drop Shutdown Forced Availability Outages 1 [[Phase-I (*)Unbundled].sub.it] 4.943 (*) -3.698 (**) (2.359) (1.421) 1 [[Phase-II (*) Unbundled].sub.it] -5.415 (*) 0.987 (2.949) (2.583) Time Trend State State Observations 3,859 3,859 Number of units 236 236   Drop Enter/Exit Forced Availability Outages 1 [[Phase-I (*)Unbundled].sub.it] 6.141 (*) -5.134 (**) (3.163) (2.047) 1 [[Phase-II (*) Unbundled].sub.it] -8.501 (**) 1.434 (3.013) (2.378) Time Trend State State Observations 2,895 2,895 Number of units 147 147 Notes: Standard errors in parentheses clustered at state level. (***) p<0.01, (**) p<0.05, (*) p<0.1. All equations control for a quadratic for EGU age, and EGU and Year fixed effects. Columns - drop Phase 2 states from the estimation sample. Columns -, substitute phase-wise trends instead of state-specific trends. Columns - drop units that were decommissioned during the sample period. Columns - drop units that were either commissioned or decommissioned during the sample period. Table 7: Triple Difference Estimates (DDD)--Impact of Unbundling on State EGUs    Drop Shutdown Forced Availability Outages Availability [Phase-I (*) Unbundled] 5.959 (*) -4.938 (**) 6.284 (*) (3.12) (1.818) (3.175) [Phase-II (*) Unbundled] -3.684 3.104 -3.620 (2.233) (2.447) (2.285) Oervations 6054 6054 5,541 Number of Units 385 385 344    Drop Enter/Exit Forced Forced Outages Availability Outages [Phase-I (*) Unbundled] -4.435 (**) 7.398 -5.088 (**) (1.709) (4.500) (2.203) [Phase-II (*) Unbundled] 2.711 -4.239 1.679 (2.419) (5.589) (6.400) Oervations 5,541 4,024 4,024 Number of Units 344 203 203 Notes: Standard errors in parentheses clustered at state level. (***) p<0.01, (**) p<0.05, (*) p<0.1. All equations control for a quadratic for EGU age, and a full set of state X year, ownership X year and EGU fixed effects. Table 8: Capacity Utilization Factor--Impact of Unbundling on EGUs   Capacity Utilization DD DDD [Phase-I (*) Unbundled] 3.955 1.101 (3.475) (2.789) [Phase-II (*) Unbundled] -4.039 0.571 (3.281) (2.133) Observations 4,298 6,054 Number of units 270 385 Notes: Std. errors in parentheses, clustered at state level. (***) p<0.01, (**) p<0.05, (*) p<0.1. ) Estimations in both column  and , respectively, control for all the same controls as the earlier DD and DDD estimates. Table 9: Operating Reliability by Size of EGU  Interaction Variable Availability Phase-I [Phase-I (*) Unbundled] (*) Less than 100 MW 4.239 (3.265) [Phase-I (*) Unbundled] (*) 110/120 MW 12.26 (***) (3.041) [Phase-I (*) Unbundled] (*) 200/210 MW 6.466 (4.279) [Phase-I (*) Unbundled] (*) 500 MW 1.169 (2.178) Phase-II [Phase-II (*) Unbundled] (*) Less than 100 MW -6.098 (4.998) [Phase-II (*) Unbundled] (*) 110/120 MW -7.492 (**) (3.275) [Phase-II (*) Unbundled] (*) 200/210 MW -4.396 (2.565) [Phase-II (*) Unbundled] (*) 500 MW -10.13 (***) (2.358)  Dependent Variable Interaction Variable Forced Outages Phase-I [Phase-I (*) Unbundled] (*) Less than 100 MW -5.564 (**) (2.168) [Phase-I (*) Unbundled] (*) 110/120 MW -9.313 (**) (3.518) [Phase-I (*) Unbundled] (*) 200/210 MW -2.913 (1.748) [Phase-I (*) Unbundled] (*) 500 MW -0.192 (2.224) Phase-II [Phase-II (*) Unbundled] (*) Less than 100 MW 3.706 (4.003) [Phase-II (*) Unbundled] (*) 110/120 MW 2.764 (4.793) [Phase-II (*) Unbundled] (*) 200/210 MW 0.325 (1.487) [Phase-II (*) Unbundled] (*) 500 MW 4.658 (***) (1.413)  Interaction Variable Capacity Utilization Phase-I [Phase-I (*) Unbundled] (*) Less than 100 MW 2.141 (5.033) [Phase-I (*) Unbundled] (*) 110/120 MW 9.415 (**) (4.214) [Phase-I (*) Unbundled] (*) 200/210 MW 2.812 (4.049) [Phase-I (*) Unbundled] (*) 500 MW 1.716 (3.057) Phase-II [Phase-II (*) Unbundled] (*) Less than 100 MW 1.013 (4.129) [Phase-II (*) Unbundled] (*) 110/120 MW -7.851 (**) (3.482) [Phase-II (*) Unbundled] (*) 200/210 MW -3.514 (3.366) [Phase-II (*) Unbundled] (*) 500 MW -14.57 (***) (2.825) Notes: Number of observations for all specifications = 4298 (270 EGUs). Each column in Panel A and Panel B represents coefficients from a single DD model. Less than 100MW: all EGUs <100MW; 110/120MW: between 100MW and <150MW; 200/210/250MW: between 150MW and 300MW; and 500MW: 490 MW and above. All equations control for a quadratic for EGU age, year and EGU fixed effects and state time trends. Standard errors in parentheses clustered at the state level. (***) p<0.01, (**) p<0.05, (*) p<0.1.
|Printer friendly Cite/link Email Feedback|
|Author:||Malik, Kabir; Cropper, Maureen; Limonov, Alexander; Singh, Anoop|
|Publication:||The Energy Journal|
|Date:||Apr 1, 2016|
|Previous Article:||The Causes and Effects of Deregulation, Volume I and II.|
|Next Article:||Short-term Hedging for an Electricity Retailer.|