Energy Policy 109 (2017) 659–665

Contents lists available at ScienceDirect

Energy Policy
journal homepage: www.elsevier.com/locate/enpol

Are small firms willing to pay for improved power supply? Evidence from a MARK
contingent valuation study in India
Ranjan Ghosha,⁎, Yugank Goyalb, Jens Rommelc, Julian Sagebield
a
Indian Institute of Management Ahmedabad, Vastrapur, Ahmedabad, Gujarat 380015, India
b
OP Jindal Global University, Sonipat Narela Road, Sonipat 131001, Haryana, India
c
Leibniz Centre for Agricultural Landscape Research, Eberswalder Straße 84, 15374 Müncheberg, Germany
d
Institute for Ecological Economy Research, Potsdamer Str. 105, 10785 Berlin, Germany

A R T I C L E I N F O A B S T R A C T

Keywords: This paper provides new estimates on Indian small-scale manufacturing firms’ willingness-to-pay (WTP) for
Small-scale industry reliable power supply. Almost half of Indian manufacturing lies in the small-scale sector, and its productivity is
Power outages severely affected by power outages. However, there is a surprising paucity of research on small firms’ WTP for
Reliability avoiding outages. We conduct a double-bounded dichotomous choice contingent valuation experiment with a
WTP
random sample of 260 small-scale firms in the region around Hyderabad. We find that on average, firms are
India
willing to pay approximately 20% more for uninterrupted power supply. The WTP estimates and the explanatory
factors for the firms’ decisions were tested for robustness using both probit and bivariate probit models. In
addition, a two-step Heckman correction was used to control for selection bias induced by protest responses. Our
results are vital to understand behavior of small firms, which are crucial to India's economic growth. Further, the
government's continued emphasis on power sector reforms makes the paper even more important as it provides
realistic estimates for designing tariffs while keeping in mind the preferences of the small-scale industry.

1. Introduction
Despite high electricity rates for the industrial sector, unscheduled
and scheduled power outages frequently occur in India. The lack of
power supply for manufacturers causes a significant decline in output
(Hansen, 2008; Hanisch et al., 2010; Allcott et al., 2016; Fisher-Vanden
et al., 2015). Indeed, economic costs of power outages in the context of
a developing country are substantial, with variable impact depending
on industry type and country. For instance, in Sri Lanka, the costs of
outages in the industrial sector can mount to 0.9% of the GDP
(Wijayatunga and Jayalath, 2004). In Pakistan, overall outages reduced
the GDP by 1.8% (Pasha et al., 1989). In India, Bose et al. (2006) stu-
died the state of Karnataka and pegged the loss value in high tension
(HT) industries to range from 0.09% to 0.17% of state GDP. Indian
industrial productivity is particularly undermined as a result. A recent
estimate suggests that the reported level of electricity shortages in India
lead to a reduction in plant revenues and producer surplus by 5–10%
(Allcott et al., 2016).
India's small-scale industrial sector, also known as the micro, small,
and medium scale enterprises (MSMEs), bears a heavy burden of these
power outages given their liquidity constraints in setting up a captive
power plant (Ghosh and Kathuria, 2014). As the backbone of Indian
economic growth, MSMEs employ 40% of the Indian workforce, con-
tributing to 45% of Indian manufacturing output and 40% of India's
exports (Goyal, 2013). Yet, they contribute to a mere 17% of the GDP
due to poor productivity (ibid). Part of the productivity losses can be
attributed to interruptions in electricity supply.1 Most of the MSMEs
operate in sectors in which production is highly sensitive to electricity
supply, such as food and beverages, fabricated metal products, apparel
and textiles, or pharmaceuticals. Since MSMEs suffer a dis-
proportionately higher cost of power interruptions compared to larger
firms, their productivity is highly elastic to power supply. An estimation
of these costs, specifically to MSMEs, therefore merits serious research.
Recent studies that engage with the issue (Allcott et al., 2016; Ghosh
and Kathuria, 2014; Kim and Cho, 2017) do not single out effects of
power outages on MSMEs. This dilutes the severity of its impact. Fur-
thermore, there is little research that attempts to estimate cost of power
outages by observing willingness-to-pay (WTP). In this paper, we

⁎
Corresponding author.
E-mail addresses: ranjang@iima.ac.in (R. Ghosh), ygoyal@jgu.edu.in (Y. Goyal), jens.rommel@zalf.de (J. Rommel), julian.sagebiel@ioew.de (J. Sagebiel).
1
Gujarat is an interesting example. Between 2007 and 2013, the number of SME clusters in this western state of India increased from 115 to 369 (highest in India) (Goyal, 2013). This
has been, inter alia, a response to increasing access and supply of electricity in the state during this period. Today, Gujarat has one of the highest installed capacities of electricity and also
the highest number of SMEs in any Indian state.

http://dx.doi.org/10.1016/j.enpol.2017.07.046
Received 2 February 2017; Received in revised form 14 July 2017; Accepted 19 July 2017
0301-4215/ © 2017 Elsevier Ltd. All rights reserved.
R. Ghosh et al.
centralize the object of our analysis as the MSME, singling out the
impact on small firms in India. In doing so, we produce most recent and
robust results for estimating the cost of outages.
In the literature, different methods have been proposed to estimate
the cost of power outages (Baarsma and Hop, 2009). Most studies have
based their estimates on observed losses in output, the cost of coping
strategies, or stated preferences methods. The stated preferences
method is particularly useful since it includes the full range of costs. As
revealed costs for outages are not easy to discover, several studies have
previously used this approach at the household level (Carlsson and
Martinsson, 2007, 2008; Carlsson et al., 2011; Blass et al., 2010). While
a number of studies have investigated the costs of outages in the in-
dustrialized economies (Morrison and Nalder, 2009; Baarsma and Hop,
2009; Goett et al., 2000), evidence on developing economies is sparse.
Previous studies on India have focused on coping strategies and output
analysis (Gulyani, 1999; Sari, 2003; Allcott et al., 2016). These studies
typically use data that do not include the full range of costs (Allcott
et al., 2016; Fisher-Vanden et al., 2015). We address these shortcomings
and use a stated preferences approach, namely the contingent valuation
method, to estimate the full range of outage costs.
Contributing to a rather thin literature in this area, we make three
new interventions in this study: first, by investigating heterogeneity in
WTP for reduced power outages, we are able to distinguish WTP values
for different types of MSME firms. This allows policy-makers and reg-
ulators to assess which firms should be prioritized and to what extent
tariffs should discriminate between firms. Second, instead of asking
‘how much’ of a tariff would a firm be willing to pay for uninterrupted
power supply, we ask ‘how much extra’ it would be willing to pay. This
encourages the respondents to focus on the marginal costs (and bene-
fits), making it a more reliable and accurate indicator of their pre-
ferences and costs. This difference leads to more realistic estimates as
compared to the previous work (Bose et al., 2005). Third, we use probit,
bivariate probit, and also Heckman models to ensure robust results,
making the analysis richer and more rigorous.
Our findings show that firms are willing to pay 20% in addition to
the prevailing tariff for a reduction of scheduled and unscheduled
power outages to zero. The current estimates are significantly lower
than those of Bose et al. (2006) who pegged this value at 37%. The
higher conservativeness and, arguably, greater reliability of our results
owes itself to the design of the present questionnaire. This will be
further discussed in Section 3.
2. Survey design and data
The experiments were conducted with MSMEs in and around
Hyderabad, the joint capital of the Indian states of Andhra Pradesh and
Telangana. The region has over 18,000 industrial units employing more
than 220,000 people, which makes it a prominent location for small and
medium scale manufacturers in south India. It houses several key in-
dustrial clusters.2 Most notably, metallurgy, paper and printing, plastic
and rubber, engineering machinery, food processing, wood, chemical,
and repair and services dominate. With a consistent growth in the
number of MSMEs in the previous decade, the region offers a valuable
setting to examine the small-scale firms’ WTP for a reliable power
supply. Our sample consisted of 260 small-scale firms and was geo-
graphically stratified. In the first step, we pre-selected four different
industrial areas in the Greater Hyderabad Municipal Area. The selection
was based on expert interviews. In the second step, using detailed maps
of the selected areas, we did a random walk to approach industrial
units. In order to administer the survey, we trained enumerators for two
2
See, Brief Industrial Profile of Hyderabad and Brief Industrial Profile of Medak published
by MSME Development Institute, Ministry of MSME, Government of India, available at
http://dcmsme.gov.in/dips/hyd%20profile.pdf and http://dcmsme.gov.in/dips/medak.
pdf respectively (last accessed on 21 January 2017).
660
Energy Policy 109 (2017) 659–665
days together with a local consultant who possessed vast experience
with industrial surveys. The enumerators were graduates in social sci-
ences. After the training, enumerators were sent to industrial estates
where they approached firms based on the pre-determined random
walk. We asked enumerators to approach suitable management staff
members that (a) have a clearly defined management role in the firm
and are able to make firm relevant decisions and (b) have the necessary
key information on electricity usage of the firm. Wherever such suitable
management staff was unavailable, was not willing to respond, or was
not involved in electricity-related decisions, the enumerators were in-
structed not to conduct the survey at the firm. Such firms were marked
on a separate list of ‘no response’ data. For the purposes of monitoring
and verification, we asked enumerators to collect the individual re-
spondents’ business cards. Sometimes, one of the authors would ac-
company the enumerators randomly.
The survey questions were developed by a team of Indian and German
researchers from Humboldt-Universität zu Berlin and The Energy Resource
Institute (TERI) in Delhi. The survey was conducted during September-
December 2010. We ran two pretests on the survey and adjusted the
questionnaire before the final round. The 14-page questionnaire contained
four modules: (a) general questions on the company, e.g., the number of
employees, annual turnover, and productions cycles; (b) energy topics,
such as energy carriers, changes in energy use, the use of captive gen-
eration, and the frequency of power outages; (c) the contingent valuation
study; and (d) attitudes and knowledge.
There is a long debate among economists on the best practice of
contingent valuation method (CVM) studies (Welsh and Poe, 1998;
Hanemann, 1994; Carson et al., 2003; Johnston et al., 2017). The
contingent valuation part of our survey generally followed an up-to-
date methodology. A text describing a specific scenario was read aloud
to respondents. We chose this procedure to ensure creating homo-
geneous conditions across respondents and enumerators. Respondents
were offered a hypothetical improvement in electricity services with a
reduction of all scheduled and unscheduled power outages to zero. At
the end of the text, a budget reminder was included. Then, respondents
were asked if they were generally willing to pay for the described im-
provement (cf. question 1.1 in Appendix C). If they replied ‘no,’ they
were asked why they were unwilling to pay in order to distinguish
protest zeros from true zeros (Yu and Abler, 2010).3 If respondents
replied ‘yes,’ the enumerators continued with the double-bounded di-
chotomous choice questions as described above. (We provide the exact
wording and questions in Appendix C.)
The bid vector must be carefully chosen in a dichotomous choice
CVM study (Johnston et al., 2017). Different approaches exist for de-
signing bid vectors. In a recent study, Chung and Chiou (2017) test the
validity of the CVM method using a triple-bound dichotomous choice
model with multiple follow-up questions. In general, plausibility and
statistical efficiency guide the choice. Plausibility means that bids
should be realistic, credible, and accepted by respondents (Arrow et al.,
1993; Bateman et al., 2002; Johnston et al., 2017). If, for example, bid
values are unrealistically high, the probability of protest increases. Effi-
ciency refers to the statistical properties of the bid vector. Especially in
small samples, standard errors should be as small as possible. Bids should
center around the median WTP (Alberini, 1995a, b; Cooper, 1993). From a
statistical perspective, an equal distribution of yes and no responses is
desirable. Thus, we used data from a previous study (Hanisch et al., 2010)
3
The literature discusses different ways to treat those observations (Meyerhoff and
Liebe, 2006; Yu and Abler, 2010). Because the determinants of a general willingness-to-
pay (e.g., missing trust) differ from those of the amount (e.g., size of the firm or de-
pendency on reliable supply), it is not advisable to jointly estimate one coefficient per
variable for both decisions. In addition, different protest motives may have different
determinants or at least coefficients of different size. One way to deal with the problem is
to estimate a multi-step hurdle model (Yu and Abler, 2010). However, a larger number of
observations would be needed to accurately distinguish between the different protest
motives and their determinants in our case. Therefore, we excluded “true zeros” from the
analysis.
R. Ghosh et al. Energy Policy 109 (2017) 659–665

to determine bid levels that asked for WTP in an open-ended format. Table 1
Specifically, we used rounded quintiles from this study as bid levels for the Summary statistics.
Source: Own calculations
present study to ensure both plausibility and efficiency.
Five bid levels with a per-unit increase at INR4 0.50, 0.80, 1.20, Variable Description Mean Std. Dev. Min. Max. N
1.80, and 2.50 were used in three different versions of the ques-
tionnaire, each of which contained one out of the three different TARIFF Average tariff paid in 4.654 0.983 3 8.5 260
INR
starting-bid values. The first question started with either INR 0.80,
BACKUP 1 = firm has backup 0.538 0.499 0 1 260
1.20, or 1.80. Depending upon the answer, the bid value of the follow- power
up question was shifted to the upper or lower neighbor. For example, if TURNOVER Yearly turnover in 51.442 104.586 0.05 540 257
the service improvement was offered at INR 0.80 and the respondent million INR
rejected, then the follow-up bid would be INR 0.50. In case the re- PRODHOURS Average production 11.258 4.942 6 24 252
hours per day
spondent accepted, the follow-up bid would have been INR 1.20.
UNSCHED Hours of unscheduled 3.052 2.882 0 10.667 260
power cuts per day
3. Results and discussion SCHED Hours of scheduled 1.171 1.364 0 10.667 260
power cuts per day

3.1. Descriptive evidence

3.2. Contingent valuation experiment
Our survey responses indicated that power outages occur on a daily
basis for firms in and around Hyderabad. They are highly seasonal with
The results of the double-bounded dichotomous choice contingent
summer time peaks in both scheduled (pre-announced) and un-
valuation questions are presented here.6 As the inclusion of the second bid
scheduled (unannounced) outages. During summer, there is low power
question can lead to bias in the estimated WTP (Bateman et al., 2001), the
generation from hydro plants while there is high demand for water
results of a probit model using only the first bid are also shown.
pumping, cooling, and air conditioning (Appendix Table B.1). Firms in
We can estimate WTP only for the sub-sample of respondents for
the region face a daily average of six hours of power outages. More than
whom WTP question data are available. If these observations are not
90% of our respondents experienced unscheduled power outages during
missing randomly, a selection bias would be introduced. In order to
the summer. Scheduled outages, which last less than one hour per day,
account for this bias, we estimated a two-step Heckman probit model.
play a rather minor role in this context. Even during winter and mon-
The equation of interest is the first probit equation of the dichotomous
soon months, unscheduled power outages exceed a daily average of two
choice contingent valuation question. The selection equation is the
hours, significantly eroding firms’ productivity.
generally willing question we have put in front of the WTP questions
We asked the respondents to self-assess the severity of a number of
(cf. Appendix C). The null hypothesis of the likelihood ratio test of in-
electricity-related problems on a scale from 1 to 5, where 1 denoted no
dependent equations had to be rejected (X2 = 5:35; p value = 0.0207),
problems at all and 5 denoted severe problems (Appendix Table B.2). We
indicating that a selection bias has been introduced by the generally
discovered that power outages are the most severe problem for firms with
willing question. However, while comparing the results of the Heckman
the unscheduled outages being ranked higher than the scheduled. Less se-
probit model and the simple probit model, only minor differences in
vere problems included low responsiveness of utilities to complaints, voltage
estimated parameters are found. Therefore, we will continue our ana-
fluctuations, bribes to service staff for fixing broken equipment, and diffi-
lysis with the simple probit and bivariate probit models. The results of
culties to access additional installed loads or a completely new connection.
the Heckman Probit model are presented in Appendix Table B.4.
The losses due to an outage of one hour go up to INR 1,000,000 with
Besides the bid variables, the variables summarized in Table 1 were
the average loss being a little more than INR 10,000. Only 10% of the
included in the models. The contingent valuation results from the probit
respondents suffered no losses from power outages. More than half of
(only first bid question) and the bivariate probit (both bid questions)
the respondents had facilities for back-up supply – entirely based on
are presented in Table 2. Both the probit and the bivariate probit
diesel engine electric generators. The median cost incurred (which is
models are highly significant (X2 = 35.16 in the probit and ×2 =
also the most frequent answer representing 45.9% with back-up) with
43.55 in the bivariate probit). Yet, in the probit model, none of the
generators is about INR 10 per kWh, which is more than twice what
estimated parameters except BID1 and BACKUP are significant. In the
firms usually pay for power. The installed back-up capacity ranges from
bivariate probit model all estimated parameters except SCHED are
small generators with 5 kW to larger engines with 1000 kW, and the
significant at least at the 10% level. The reason for these differences
costs per kWh ranges from INR 5–50. However, more than 25% of the
may be related to the efficiency gains obtained by including the follow-
surveyed firms were paying INR 20 or more per kWh generated by their
up question (Hanemann et al., 1991). In both models, the bid variables
back-up facilities (Appendix Table B.3). This implies that firms already
are significant and negative, which means the higher the bid amount,
pay substantial amounts to manage power outages.
the lower the probability to answer with a ‘yes.’ WTP increases when
Table 1 summarizes other important variables which we expect will
the firm (a) faces higher average tariffs, (b) uses back-up power during
impact the WTP for reduction in power outages.5 The average tariff a
power cuts, (c) has higher turnover, (d) logs fewer production hours,
firm pays (TARIFF) is INR 4.6 with a standard deviation of almost 1.
and (e) sustains more unscheduled power cuts. Finally, in the bivariate
The highest tariff reported is INR 8.5. More than half of the firms have a
Probit model, ρ is significant which means that the error terms of the
power backup system (BACKUP). The yearly turnover (TURNOVER), an
two equations are correlated, justifying the use of a bivariate Probit
indicator for the size of the firm, varies greatly. The mean turnover is
model instead of two separate probit models.
about INR 51,000 million with a standard deviation of INR 104,000
Table 3 gives the mean WTP values in INR per kWh, 95% confidence
million. Similarly, production hours per day (PRODHOURS) vary. The
intervals, and p-values for the probit and bivariate probit models. The
mean hours of production are 11.3 with a standard deviation of 4.9.
WTP depends on the values of the firm-specific exogenous variables so
Some firms produce 24 h per day.
that the mean WTP represents the sample average (Table 1). The mean
WTP in the probit model is slightly smaller than in the bivariate probit
4
INR is Indian National Rupee. During the time of our survey, 1 USD was equivalent to model (INR 1.09 and INR 1.22), but the confidence intervals overlap,
INR 44.34. By January 2017, it has increased to INR 68.11.
5
Note that for the questions on PRODHOURS and TURNOVER we did not get responses
from all firms. These firms are excluded from the regression models that use these vari-
6
ables. For an explanation of the econometric model, please see Appendix A.

661
R. Ghosh et al. Energy Policy 109 (2017) 659–665

Table 2 electricity they consume. Even if we assume that their LT-cohort can be
Estimation results: probit and bivariate probit. compared to our MSME sector, our estimate for WTP is 20% higher than
the prevailing tariff, almost half of their estimate for the LT-cohort. Our
(1) (2)
Probit Bivariate Probit conservative estimate indicates more accuracy. In fact, in their sample,
almost half the firms were not willing to pay more for improved elec-
Response_BID1 tricity supply, making a 37% increase in WTP rather skewed. Further-
Constant 0.049 (0.561) 0.163 (0.364)
more, we did not ask the respondents about their WTP for improved
BID1 −0.010*** (0.002) −0.008*** (0.002)
TARIFF 0.110 (0.101) 0.145** (0.065) power supply but about their ‘desired increase’ in WTP, nudging them
BACKUP 0.601*** (0.229) 0.438*** (0.140) to focus on the problem directly.
TURNOVER −0.000 (0.001) 0.001* (0.001) Our results can be very helpful in informing policy-makers about the
PRODHOURS −0.002 (0.020) −0.025* (0.013) power consumption preferences of small-scale firms. In December 2016,
UNSCHED 0.049 (0.037) 0.050** (0.023)
the Government of India (GoI) unveiled a Draft Electricity Plan in an effort
SCHED 0.106 (0.095) 0.014 (0.056)
ρ 0.483*** (0.163) to outline the government's efforts and plans for the power sector.7 It
N 199 199 acknowledges the high share of MSMEs in total industrial energy con-
Log lik.(Null) 136.362 sumption, which is roughly 25%, and proposes an energy mapping of the
Log lik. −118.779 -228.518
sector. Even the Ministry of Micro, Small, & Medium Enterprises continues
Chi-squared 35.165 43.559
to implement an Infrastructure Development Program scheme, which lays
Standard errors in parentheses; own calculations. special importance to power distribution to MSME clusters. Our paper
*
p < 0.10. provides solid estimates which would otherwise be very difficult to obtain
**
p < 0.05. through secondary mapping exercises. The insights on behavior as per
***
p < 0.01.
industrial classification also show that impacts and preferences are dif-
ferential across firm types, and hence, policy should be more targeted in
nature. Industrial tariffs in India have increased from an average of INR
Table 3
4.16/kWh in 2007 to INR 7.64/kWh in 2015. This increase is in part
Willingness-to-pay.
because industrial electricity tariff cross-subsidizes electricity in the agri-
Mean WTP Lower Level Higher Level p-value cultural sector (Bhattacharyya and Ganguly, 2017). As a response, the
National Tariff Policy 2006 and the Electricity Amendment Bill of 2014
Probit 1.09 0.80 1.28 0.0002
aim to gradually phase out the subsidy. Findings from our study suggest
Bivariate Probit 1.22 1.07 1.39 0.0000
that more than the price, it is the supply that matters. A policy of removing
cross-subsidization must be driven by the impetus to ensure better power
which indicates that the difference in these values is statistically not supply through increased financial surplus rather than to focus only on
significant. The mean WTP is thus within a range of INR 0.80 INR and tariff reduction for industrial consumers. Perhaps addressing the issues of
INR 1.39. These WTP values make up approximately 20% of the power theft, which is responsible for a loss of 20% of electricity generated
average tariff that firms paid during the survey period. However, the in India (Gaur and Gupta, 2016), could be more useful in this context.
results also indicate that the WTP is lower than what most firms pay As our results on the WTP indicate (and given the difficulty of bu-
when using back-up power such as diesel generators. reaucracy in addressing problems of outages), an alternative solution
We also assess if WTP varies with the industry type and nature of could be the interruptible electricity supply contracts (Baldick et al.,
production. In Table B.5, we present WTP results of representative firms 2006). These contracts offer rebates to firms for accepting outages
from the prominent industry classifications in our sample of firms. We during scarcity. This triggers a trade in outages and develops a me-
control for firm size and other power supply covariates and observe that chanism to allocate them to the least affected. Allcott et al. (2016)
the WTP is highest for the pharmaceutical representative firm (INR conducted simulations on offerings of interruptible contracts in India
1.73) followed by the food and chemical firms (1.58 and 1.35 INR, and found that their nationwide implementation could substantially
respectively). They are comparatively lower for paper, electronics, and reduce impact of shortages. The interruptible contracts may be de-
machinery and tool firms. This is consistent with the results from Ghosh signed between industrial, agricultural, and household users to reap
and Kathuria (2014) who report that for firms in the same region, the further gains (Müller et al., 2016; Kimmich and Sagebiel, 2016). Since
likelihood of opting for a captive power plant (CPP) is highest for food, agriculture could be seasonal in nature, the rebates for them could be
chemical, and pharmaceutical firms. higher, and their power could then be channelized to MSMEs who
would bear a higher cost of power outages. Indeed, this would be
especially useful for small-scale firms that generally face allocative ef-
4. Conclusions and policy implications ficiency challenges owing to diseconomies of scale. Another option
could be to explore the possibility of creating group captive power
Our study augments a rather thin literature on estimating how much plants – G-CPPs, shared in an MSME cluster (Goyal and Ghosh, 2015).
Indian small-scale firms (MSMEs) are willing to pay for reliable power Here, the MSME cluster can create a Special Purpose Vehicle, invest to
supply. This is done using a randomized survey of 260 MSMEs in and establish a small power plant to satisfy the cluster's captive needs, and
around the Hyderabad region. Using the contingent valuation method, sell the surplus power to the government. From a policy perspective,
we show that MSMEs in the region are willing to pay 20% more than our study centralizes the problem of energy shortages for MSMEs in
the prevailing tariffs. The results are instructive and depart from the India. However, given the diverse experience of stakeholders in power
previous work (Bose et al., 2006) in a number of ways, not only in supply dynamics, care must be taken to generalize the results. That said,
studying the impact on MSMEs alone but also in producing a more the 20% higher price that MSMEs are willing to pay for better power
reliable estimate. Bose et al. (2006) had surveyed 501 firms across the services suggests that the government can potentially improve power
state of Karnataka (more than 300,000 units) while our examination supply even if part of the cost is shared by the industry.
was performed for 260 firms in the Hyderabad region (18,000 units),
making it more focused (and indeed more recent). In terms of the
method, we enrich the analysis further with a bivariate probit model
and a Heckman model as a robustness check. Bose et al. categorized 7
See, Draft Electricity Plan, at http://www.cea.nic.in/reports/committee/nep/nep_
firms into HT (high tension) and LT (low tension) based on how much dec.pdf (last accessed on 13 April 2017).

662
R. Ghosh et al. Energy Policy 109 (2017) 659–665

Acknowledgments
We thank Philip N Kumar, Vamsi Krishna, and their team for the sup-
port in conducting the field work. We are grateful to Veena Aggarwal,
Deepika Garg, Markus Hanisch, Kaushik Deb, Krithika Ramakrishnan, and
Kai Rommel. We also thank two anonymous referees whose comments
helped improve the paper. This work was conducted as part of the research
project ‘Sustainable Hyderabad’ and financed by the German Federal
Ministry of Education and Research (Grant Number: 01LG0506A).

Appendix A.

Econometric model

A firm i's profit is defined as the difference of firm-specific revenues Ri and costs Ci. We assume the firm to maximize profits by adjusting the
produced quantity Xi . The maximum attainable profit is a function of exogenously given output and input prices Pi , firm-specific characteristics Si
and electricity quality qi . The profit maximization problem can be written as:

max πǐ = Ri (Xi ) − Ci (Xi ) = πǐ *(Pi , Si, qi ) (1)

Where π̌i and πǐ * denote profit and the maximum attainable profit for given prices, firm-specific characteristics, and electricity quality.
We assume pi , Xi,Si as fixed and that improved electricity quality has a non-negative impact on profits (i.e. the first derivative of the profit
function is equal to or larger than zero).

∂πǐ *
≥0
∂qi (2)
̌ i for an improvement in electricity quality from level q to q for all q < q .
Eq. (2) implies that there exists a non-negative willingness-to-pay WTP 0 1 0 1
Willingness-to-pay for improvements in electricity quality is defined as the maximum amount of money a firm would be willing to give up to become
indifferent between the two provision levels q0 and q1:
̌ i = πǐ *(p , Si, q ) − πǐ *(p , Si, q )
WTP (3)
i 0 i 1

̌ i and πǐ * by the amount a of electricity in kwh which the firm

In our survey, we are interested in the WTP per kwh. Thus we have to divide WTP
̌ i
WTP πǐ *
uses . This can be expressed as WTPi = a
and πi* = a
. With this simple transformation the relevant formula in willingness to pay per unit of kwh is
WTPi = πi*(pi , Si, q0) − πi*(pi , Si, q1) (4)
Respondents were presented with a status quo scenario with the current level of power quality q0 and an improved scenario with electricity
quality q1 in which there would be no power cuts. We then asked the respondents if they would be willing to pay a pre-given amount T (the bid) for
the improved scenario. The respondents can either accept or reject the offer. In a follow-up question, respondents get a higher (lower) price if they
accepted (rejected) the first offer. This double-bounded dichotomous contingent valuation method has been developed by Hanemann et al. (1991)
who show that efficiency is increased compared to a single-equation technique (Hanemann, 1984)
In this format, willingness-to-pay, as defined in Eq. (3), cannot be estimated directly, but we can observe whether the respondents’ willingness to
pay is larger or smaller than T . Formally, improved scenario with q1 is chosen if
πi*(P , Si, q1) − πi*(Pi , Si, q0) + T = ∆π (Pi , Si, q0 , q1, T )> 0 (5)
For simplicity, we assume the profit difference ∆π to be linear in its arguments and leave out Pi as we have no information collected on prices.
∆π (Si, q0 , q1, T ) = α + βSi + γT + ϵ (6)
where α , β, and γ are parameters and ϵ represents the random and unobserved elements of the profit function. γ requires further explanation. Any amount T
that the firm is willing to give up for improved electricity supply will come at an opportunity cost since this money cannot be used otherwise. Thus, γ is the
marginal profit lost due to this forgone opportunity to invest. Analogous to utility theory, γ can be interpreted as the marginal profit from one unit of additional
money. In other words, if the firm has one more unit of money, profits will increase by γ . Using this interpretation, we can calculate willingness-to-pay as
α + βSi
WTPi = −
γ (7)
To estimate the parameters of ∆π , we must make assumptions regarding the stochastic part of the profit function. The probability to respond a no,
i.e., not to pay for the improvement in electricity quality is
Prob(q0) = Prob(α + βSi + γT + ϵ< 0) = Prob(ϵ > α + βSi + γT (8)
And the probability to respond with a yes is
Prob(q1) = 1 − Prob(q0) (9)
Assuming normal distribution for ϵ , the cumulative distribution function is
Prob(q0) = Φ[α + βSi + γT ] (10)
This equation represents the standard probit model and can be estimated using the maximum likelihood method. Since we have asked a follow-up
question, we have two equations to be estimated, and it is highly likely that the error terms ϵ are correlated for each respondent. More precisely, the
unobserved part regarding the first question is correlated with the unobserved part of the second question which can be incorporated with a bivariate
probit model. The corresponding log-likelihood function can be maximized using the Newton-Raphson method. A more detailed explanation of the
model can be found in standard textbooks on contingent valuation (e.g., Bateman et al., 2002; Freeman et al., 2014; Haab and MacConnell, 2002).

663
R. Ghosh et al. Energy Policy 109 (2017) 659–665

Appendix B.

See Tables B.1–B.5

Table B.1
Power outages (hours/day).

Variable N Mean SD Min Max Freq.

Scheduled Summer 260 2.12 1.97 0 12 29.62%

Scheduled Winter 260 0.61 1.32 0 12 51.92%
Scheduled Monsoon 260 0.73 1.35 0 12 46.92%
Unscheduled Summer 260 3.86 3.28 0 12 8.85%
Unscheduled Winter 260 2.55 3.5 0 16 23.85%
Unscheduled Monsoon 260 2.74 3.47 0 12 16.92%

Table B.2
Self-assessment of supply issues.

Variable N Mean SD

Voltage fluctuations 244 1.92 1.18

Scheduled power cuts 247 2.44 1.52
Unscheduled power cuts 250 3.87 1.12
No response to complaints 244 1.97 1.08
Bribes required for repairs 244 1.91 1.00
Getting a new connection 240 1.54 0.84
Getting additional load 238 1.82 0.98

Table B.3
Summary statistics.

Variable N Mean SD Min Max

Captive Binary 260 0.54 0.5 0 1

Paid Tariff 260 4.65 0.98 3 8.5
Stated Loss of a one hour power 260 10724.78 63261.85 0 1000000
cut in INR
Capacity in kW 133 245.65 254.83 5 1000
Costs INR/kWh 122 17.06 13.72 5 50

Table B.4
Results from Probit (1) and Probit with Heckman correction (2).

(1) (2)
Probit Heckman probit

Response_Cuts_BID1
Cuts_Bid1 −0.0103*** (0.0024) −0.0088*** (0.0020)
AVG_tariff 0.1099 (0.1011) 0.0498 (0.0917)
Have_captive 0.6011*** (0.2286) 0.2582 (0.1891)
turnoversmall −0.0003 (0.0010) −0.0006 (0.0009)
Prod_hours_day −0.0019 (0.0205) 0.0043 (0.0162)
UNSCHED 0.0490 (0.0368) −0.0354 (0.0318)
SCHED 0.1060 (0.0952) 0.1146 (0.0808)
Constant 0.0486 (0.5612) 0.8398* (0.4803)
WTP_Cuts_General
AVG_tariff 0.1676* (0.1014)
Have_captive 0.5448*** (0.2034)
turnoversmall 0.0010 (0.0010)
Prod_hours_day −0.0372** (0.0172)
UNSCHED 0.2290*** (0.0428)
SCHED −0.1970*** (0.0760)
Constant −0.1638 (0.5074)
atanh(rho)
Constant −11.9453 (66.5740)
N 199 249
pseudo R2 0.129
Log lik. −118.7795 −219.6840
Chi-squared 35.1647 31.4814

Standard errors in parentheses; own calculations.

*
p < 0.10.
**
p < 0.05.
***
p < 0.01.

664
R. Ghosh et al. Energy Policy 109 (2017) 659–665

Table B.5
WTP heterogeneity by industry-type.

Firm 1 Firm 2 Firm 3 Firm 4 Firm 5 Firm 6

Industry classification (ASIa 2-digit code) Food (10) Paper (17) Chemicals (20) Pharma (21) Electronics (26) Machinery and Tools (28)

Tariff (INR) 4 4.8 6 4 4.75 5.75

Backup (yes/no) Yes No Yes Yes No No
Turnover (M-INR) 100 17 200 100 12 100
Production hours 8 8 24 7 10 10
Scheduled PC (hours) 1 2 0 5 1 1
Unscheduled PC (hours) 4 3 0 5 4 4
WTP (INR/KWH) 1.58 1. 02 1.35 1.73 0.99 1.30

a
Annual Survey of Industries, Government of India.

Appendix C. Supporting information

Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.enpol.2017.07.046.

References
Alberini, A., 1995a. Efficiency vs bias of willingness-to-pay estimates: bivariate and in-
terval-data models. J. Environ. Econ. Manag. 29 (2), 169–180.
Alberini, A., 1995b. Optimal designs for discrete choice contingent valuation surveys:
single-bound, double-bound, and bivariate models. J. Environ. Econ. Manag. 28 (3),
287–306.
Allcott, H., Collard-Wexler, A., O'Connell, S.D., 2016. How do electricity shortages affect
industry? Evidence from India. Am. Econ. Rev. 106 (3), 587–624.
Arrow, K., Solow, R., Portney, P. R., Leamer, E. E., Radner, R., Schuman, H. 1993. Report
of the NOAA panel on contingent valuation. Federal Register, 58,10, pp. 4601–4614.
Baarsma, B.E., Hop, J.P., 2009. Pricing power outages in the Netherlands. Energy 34 (9),
1378–1386.
Baldick, R., Kolos, S., Tompaidis, S., 2006. Interruptible electricity contracts from an
electricity retailer's point of view: valuation and optimal interruption. Oper. Res. 54
(4), 627–642.
Bateman, I.J., Carson, R.T., Day, B., Hanemann, M., Hanley, N., Hett, T., Jones-Lee, M.,
Loomes, G., Mourato, S., Pearce, D., Özdemiroglu, E., et al., 2002. Economic
Valuation with Stated Preference Techniques: a Manual. Edward Elgar, Chaltenham.
Bateman, I.J., Langford, I.H., Jones, A.P., Kerr, G.N., 2001. Bound and path effects in
double and triple bounded dichotomous choice contingent valuation. Resour. Energy
Econ. 23 (3), 191–213.
Bhattacharyya, R., Ganguly, A., 2017. Cross subsidy removal in electricity pricing in
India. Energy Policy 100, 181–190.
Blass, A.A., Lach, S., Manski, C.F., 2010. Using elicited choice probabilities to estimate
random utility models: preferences for electricity reliability. Int. Econ. Rev. 51 (2),
421–440.
Bose, R.K., Shukla, M., Srivastava, L., Yaron, G., 2006. Cost of unserved power in
Karnataka, India. Energy Policy 34 (12), 1434–1447.
Carlsson, F., Martinsson, P., 2007. Willingness to pay among Swedish households to avoid
power outages: a random parameter Tobit model approach. Energy J. 28 (1), 75–89.
Carlsson, F., Martinsson, P., 2008. Does it matter when a power outage occurs?—A choice
experiment study on the willingness to pay to avoid power outages. Energy Econ. 30
(3), 1232–1245.
Carlsson, F., Martinsson, P., Akay, A., 2011. The effect of power outages and cheap talk on
willingness to pay to reduce outages. Energy Econ. 33 (5), 790–798.
Carson, R.T., Mitchell, R.C., Hanemann, M., Kopp, R.J., Presser, S., Ruud, P.A., 2003.
Contingent valuation and lost passive use: damages from the Exxon Valdez oil spill.
Environ. Resour. Econ. 25, 257–286.
Chung, Y.-S., Chiou, Y.-C., 2017. Willingness-to-pay for a bus fare reform: a contingent
valuation approach with multiple bound dichotomous choices. Transp. Res. Part A:
Policy Pract. 95, 289–304.
Cooper, J.C., 1993. Optimal bid selection for dichotomous choice contingent valuation
surveys. J. Environ. Econ. Manag. 24, 25–40.
Fisher-Vanden, K., Mansur, E.T., Wang, Q.J., 2015. Electricity shortages and firm pro-
ductivity: evidence from China's industrial firms. J. Dev. Econ. 114, 172–188.
Freeman, A.M.I.F., Herriges, J.A., Kling, C.L., 2014. The Measurement of Environmental
and Resource Values: Theory and Methods. Routledge.
Gaur, V., Gupta, E., 2016. The determinants of electricity theft: an empirical analysis of
Indian states. Energy Policy 93, 127–136.
Ghosh, R., Kathuria, V., 2014. The transaction costs driving captive power generation:
evidence from India. Energy Policy 75, 179–188.
Goett, A.A., Hudson, K., Train, K.E., 2000. Customers' choice among retail energy sup-
pliers: the willingness-to-pay for service attributes. Energy J. 21 (4), 1–28.
Goyal, M., 2013. SMEs employ close to 40% of India's workforce, but contribute only 17%
to GDP. Economic Times. 9 June.
Goyal, Y., Ghosh, R., 2015. Group Captive power plants in small and medium-scale in-
dustrial clusters in India. In: Mohieldin, Mahmoud, Petkoski, Djordjija (Eds.),
Financing Sustainable Development: Ideas for Action. World Bank Group,
Washington DC.
Gulyani, S., 1999. Innovating with infrastructure: how India's largest carmaker copes with
poor electricity supply. World Dev. 27 (10), 1749–1768.
Haab, T.C., MacConnell, K.E., 2002. Valuing Environmental and Natural Resources: the
Econometrics of Non-market Valuation. Elgar, Cheltenham.
Hanemann, M., Loomis, J., Kanninen, B., 1991. Statistical efficiency of double-bounded
dichotomous choice contingent valuation. Am. J. Agric. Econ. 73 (4), 1255–1263.
Hanemann, W.M., 1984. Welfare Evaluations in Contingent Valuation Experiments with
Discrete Responses. Am. J. Agric. Econ. 66, 332–341.
Hanemann, W.M., 1994. Valuing the environment through contingent valuation. J. Econ.
Perspect. 8, 19–43.
Hanisch, M., Kimmich, C., Rommel, J., Sagebiel, J., 2010. Coping with power scarcity in
an emerging megacity: a consumers' perspective from Hyderabad. Int. J. Glob. Energy
3–4, 189–204.
Hansen, C. J., 2008. Bottom-up Electricity Reform Using Industrial Captive Generation: A
Case Study of Gujarat, India. Oxford Institute for Energy Studies.
Johnston, R.J., Boyle, K.J., Adamowicz, W., Bennett, J., Brouwer, R., Cameron, T.A.,
Hanemann, W.M., Hanley, N., Ryan, M., Scarpa, R., 2017. Contemporary guidance
for stated preference studies. J. Assoc. Environ. Resour. Econ. 4 (2), 319–405.
Kim, K., Cho, Y., 2017. Estimation of power outage costs in the industrial sector of South
Korea. Energy Policy 101, 236–245.
Kimmich, C., Sagebiel, J., 2016. Empowering irrigation: a game-theoretic approach to
electricity utilization in Indian agriculture. Uti. Policy 43 (B), 174–185.
Meyerhoff, J., Liebe, U., 2006. Protest beliefs in contingent valuation: explaining their
motivation. Ecol. Econ. 57 (4), 583–594.
Morrison, M., Nalder, C., 2009. Willingness to pay for improved quality of electricity
supply across business type and location. Energy J. 30 (2), 117–133.
Müller, M., Rommel, J., Kimmich, C., 2016. Farmers' Adoption of Irrigation Technologies:
experimental evidence from a coordination game with Positive Network Externalities
in India. Ger. Econ. Rev. ( In Press).
Pasha, H.A., Ghaus, A., Malik, S., 1989. The economic cost of power outages in the in-
dustrial sector of Pakistan. Energy Econ. 11 (4), 301–318.
Sari, E. N., 2003. Economic Impact of Poor Power Quality on Industry: Review of Studies
(India),〈http://pdf.usaid.gov/pdf_docs/PNACX087.pdf〉.
Welsh, M.P., Poe, G.L., 1998. Elicitation effects in contingent valuation: comparisons to a
multiple bounded discrete choice approach. J. Environ. Econ. Manag. 36, 170–185.
Wijayatunga, P.D., Jayalath, M., 2004. Assessment of economic impact of electricity
supply interruptions in the Sri Lanka industrial sector. Energy Convers. Manag. 45
(2), 235–247.
Yu, X., Abler, D., 2010. Incorporating zero and missing responses into CVM with open-
ended bidding: willingness to pay for blue skies in Beijing. Environ. Dev. Econ. 15 (5),
535–556.

665