Document [original]

Energy Policy 179 (2023) 113574

Available online 17 May 2023

nc/4.0/).

Contents lists available at ScienceDirect

Energy Policy

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

Nurturing national champions? Local content in solar auctions and firm

innovation✩

Florian Anselm Münch∗, Fabian Scheifele

Technische Universität Berlin, Chair of Innovation Economics, Straße des 17.Juni 135, 10623, Berlin, Germany

ARTICLE INFO

Dataset link: https://doi.org/10.7910/DVN/UW

06US

Keywords:

Local content requirements

Solar energy

Auctions

Green industrial policy

Staggered difference-in-difference

Propensity score matching

ABSTRACT

Rather than by an invisible hand, many industries are kick-started by a government policy. Despite little robust

evidence, local content requirements are increasingly used to incentivize domestic manufacturing if imports are

cheaper. To examine the effect of local content, we explore an unintended quasi-policy-experiment. Starting

in 2013, the Indian government simultaneously held solar auctions with and without local content, providing

an otherwise unobserved counterfactual. We digitize the results from the 41 auctions worth 8.65 billion $ in

solar module demand and collect annual revenue and solar patents of the 113 participating firms between

2004–2020. For causal identification, we compare winners of local content with similar open auction winners

in a staggered difference-in-difference estimation. While we observe an insignificant increase in the same and

the following year after firms win LCR auctions, overall, we find winning local content auctions does not

significantly increase firms’ solar patents or sales. We identify three reasons why the policy did not create

stronger, lasting effects. First, local content did not create sufficient production to enable learning by doing.

Second, local content did not generate enough revenue for re-investment into R&D. Third, local content reduced

competition in auctions. The analysis underlines the predicament countries face as open auction winners,

despite having won 9 times as much capacity, do not patent much (more).

1. Introduction

There is a long-standing, heated debate in economics about indus-

trial policy, in particular, infant industry or import substitution policies,

such as local content requirements (LCR) (Lane,2020;Grossman,1981;

Irwin,2021;Chang,2003;Rodrik,2008;Panagariya,2011;Krueger,

1997;Altenburg and Rodrik,2017). LCR aim to incentivize local pro-

duction if foreign products are cheaper than domestic ones (Irwin,

2021;OECD,2015).

The historical economic issue is reflected in contemporary renew-

able energy auctions. More than 100 countries worldwide have adopted

auctions in the last decade to deploy renewable energy (IRENA,2019).

In these auctions, governments offer contracts, so-called power pur-

chasing agreements, to the private bidder committing to build a re-

newable energy plant at the lowest price (Dobrotkova et al.,2018;

Bayer,2018;Del Río and Kiefer,2022). While auctions have been

credited with strong price reductions as they open electricity generation

to competition, private investment, and multinational companies (Win-

kler et al.,2018;Dobrotkova et al.,2018), auctions also incentivize

bidders in countries without existing solar module industries to import

Abbreviations: LCR, Local Content Requirements; NSM, India’s National Solar Mission; PSM, Propensity Score Matching; PS, Propensity Score

✩This research is in parts a result of a project funded by the German Institute for Metrology (PTB).

∗Corresponding author.

E-mail addresses: [email protected] (F.A. Münch), [email protected] (F. Scheifele).

solar components. As the global solar photovoltaic industry is highly

concentrated, 87% of component manufacturing and 85% of patents

are produced in just six countries (IEA,2021,2022;Luan et al.,2021),

the rest of the world imports solar components. Governments have

applied LCR to counteract the price incentive to import in 145 instances

since 2008, including 28 times in renewable energy auctions and most

recently in the US Inflation Reduction Act (OECD,2015;PIIE,2021).

Proponents of LCR argue that they are temporarily necessary to

protect and nurture a local, globally competitive industry (Chang,

2003;Wade,2018). Opponents contend that LCR induce rent-seeking,

perpetuate inefficiencies, and more often than not become permanent

protection schemes (Hufbauer et al.,2013;Panagariya,2011;Krueger,

1990). While there is a rich theoretical and conceptual literature about

industrial policies in general and local content more specifically, Lane

(2020) points out that ‘‘past empirical evidence is not mixed. It is

often vacuous’’. Hansen et al. (2020), Kuntze and Moerenhout (2012)

and Veloso (2001) come to the same conclusion regarding local content

policies.

In this paper, we empirically examine the following research ques-

tion. Can local content in solar auctions incentivize local production

https://doi.org/10.1016/j.enpol.2023.113574

Received 30 July 2022; Received in revised form 27 March 2023; Accepted 2 April 2023

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

and create an innovative local, globally competitive solar PV industry?

Innovation is arguably the crucial outcome to guarantee LCR are a

temporary and not a permanent policy. Without innovation, protection

through LCR remains permanently necessary to secure demand for

local products that would otherwise be replaced with cheaper imports.

We study this question with the example of the LCR introduced in

India’s National Solar Mission (NSM). This national flagship policy

turned India from a country with barely any solar energy generation

in 2010 into the fifth-largest solar market in 2022. The LCR in India

targeted cells and modules and were designed to create manufacturing

jobs, develop strategic technological autonomy, and global technology

leadership (Hufbauer et al.,2013;Johnson,2016;Shrimali and Sahoo,

2014;Shrimali et al.,2016,2017;Singh and Pandey,2021;MnRE,

2009).

This paper makes two novel contributions. First, we analyze the

effect of LCR at the firm level. Previous studies of renewable energy

auctions analyzed industry- or auction-level outcomes (Hansen et al.,

2020;Lewis and Wiser,2007;Probst et al.,2021;Matthäus et al.,2021;

Hochberg and Poudineh,2018;Del Río and Linares,2014;Bayer,2018;

Winkler et al.,2018;Kruger and Eberhard,2018), rather than firm-

level outcomes, such as innovation (Del Río and Kiefer,2022;Edler

and Uyarra,2013;Mazzucato,2018;Atkin et al.,2017). We collect a

unique panel data set of annual patents, revenue, and auction outcomes

for the 113 firms that participated in 41 national solar auctions for

the period 2004–2020. Second, we follow a series of recent studies

that apply new methods for causal inference to examine industrial

policy (Lane,2020;Juhász,2018) to provide the first causal estimate of

local content on firm-level innovation and production. We exploit that

the Indian government held simultaneously some auctions with and

without LCR (Probst et al.,2020), which created an otherwise unob-

served counterfactual or control group. As firms self-selected into open

or LCR auctions, there is a concern about selection bias. We overcome

this concern by comparing annual solar patents and revenue of firms

that won at least one LCR auction (treatment group) with firms that

won only in open auctions (control group) in a staggered difference-

in-difference design (Callaway and Sant’Anna,2021). Moreover, we

improve the causal identification by restricting the comparison to firms

that were similar in the characteristics, which predict participation in

LCR auctions and solar patenting, using propensity score matching.

While we show that there has been an overall increase in patent

applications in India with the onset of the National Solar Mission, we

find that winning auctions with LCR does not lead to significantly more

innovation or revenue than winning auctions without LCR. Although

we observe an insignificant increase in the same and the following

year after firms win LCR auctions in solar patents and revenue, the

positive point estimates turn negative in the following years but remain

insignificant throughout the observed period.

We also analyze why the policy did not induce more innovation.

For this purpose, we conduct a falsification exercise to determine

which mechanisms did not materialize and caused the policy to be

ineffective. Finally, we estimate that the total additional costs of the

local content policy, relative to the price paid in the auction scheme,

were quite low (in total, approximately 20 million $). We also illustrate

that the government would have had to earmark at least around 30%

instead of the actual 5% of the total demand for local content to foster

patenting. Given firms participating exclusively in open auctions, which

auctioned approximately 9 times as much capacity, did not innovate

significantly more, we conclude that our analysis illustrates the need

for better-designed local content or alternative industrial policies.

In the following, Section 2sketches out a verbal theoretical frame-

work. Section 3provides contextual information. Section 4presents

the causal identification strategy and the data. Section 5presents the

results. Section 6discusses the results. Section 7concludes and provides

policy implications.

2. Theoretical framework: Local content and firm innovation

Section 2theorizes how LCR in public electricity auctions may pro-

mote domestic firms’ innovation and production. We keep the frame-

work general so that it could apply to other latecomers countries.

The starting point of the framework (see Fig. 1) is the organization

of public auctions. In these auctions, bidders offering to build and oper-

ate a solar power plant at the lowest price will win a power purchasing

agreement (PPA). The PPA guarantees to the bidder that the govern-

ment, as in the state utility, will buy the electricity from the bidder at

specified tariffs for the next 25 years. The public auctions effectively

create a domestic market for solar energy and generate predictable

demand for the components necessary to build these plants (Mazzucato,

2018;Edler and Yeow,2016).

Bidders must decide whether to import the components necessary

to build the solar power plants or buy them from local manufacturers.

Given foreign components, especially solar cells and modules, are

cheaper at the outset as domestic manufacturers have less experience

and do not benefit from economies of scale yet, there is a strong

price incentive for bidders in open auctions to import these models.

This is depicted on the right-hand side of the framework in Figure

1. Otherwise, bidders would either have to increase their bidding

price, which would reduce their chances to win, or reduce their profit

margins. To counteract this price incentive, the government introduces

LCR, which specifically target the components in which the government

aims to develop local industry, e.g., solar cells and modules.

LCR affect domestic and international firms’ decisions in the fol-

lowing way. Bidders that do not manufacture modules are forced to

source from domestic manufacturers or set up their own manufactur-

ing facility. Domestic manufacturers either participate directly in the

auctions or may receive orders from local or international project de-

velopers. International firms, in contrast, are forced to order from local

manufacturers or establish their own domestic production facilities.

LCR should expose local manufacturers to additional demand at the

margin, which would not have occurred without LCR, resulting in the

following two key impact mechanisms. First, additional demand should

translate into additional production and opportunities for learning-

by-doing (Arrow,1962;Andreoni and Chang,2016;Lucas,1993).

Learning-by-doing, or learning by experience, can take shape of effi-

ciency improvements leading to productivity increases (Arrow,1962;

Lucas,1993), learning from client requirements (Atkin et al.,2017)

or development of productive capabilities, e.g., through innovation to

solve problems encountered in the production process (Andreoni and

Chang,2016;Chang and Andreoni,2020;Arrow,1962). In contrast,

in open auctions, components are imported, and thus no domestic

production would occur. In summary, there are positive knowledge

externalities associated with production.

A second related hypothetical impact mechanism is the additional

revenue and profit generated through orders from LCR auctions. The

additional revenue and profit could help firms finance investments into

R&D as many firms rely on internal resources to finance R&D (Hall

and Lerner,2010;Aghion et al.,2005;Schumpeter,1942). The orders

local content creates generate revenue and profit for domestic firms,

which can, in principle, be reinvested into R&D, e.g., to solve problems

occurring in the production process, to test cheaper processes or materi-

als, and improve quality, e.g., module performance or longevity. Given

firms compete for market shares in the new market organized through

public auctions, firms have an incentive to patent particularly innova-

tive solutions identified in the production process to avoid competitors

imitating them.

We also depict how open auctions may impact domestic innovation

and production on the right side of the graph. In the absence of LCR,

modules will be imported, as they still make up 40% of installation

costs (IRENA,2022) and hence allow significantly lower bidding prices,

which is also found empirically (Probst et al.,2020). Due to the absence

of demand for locally-produced modules, innovation remains restricted

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Fig. 1. Verbal theoretical framework: Hypothetical impact mechanism for local content in auctions and innovation in patenting.

to R&D financed activities in this scenario. Furthermore, in both cases

(LCR and open), post-production innovations may occur related to the

construction and operation of solar plants. Fig. 1 demonstrates that the

primary difference between both pathways is that LCR auction enables

local companies to gain production-related experience. Furthermore,

local companies are receiving revenues that would otherwise be out

of the market, which also provides additional sources for R&D at the

margin.

There are several assumptions that underline this theoretical frame-

work. Firstly, we assume that the local prices of the products targeted

by the LCR are higher than the imported version of the good. This is the

case in India (IRENA,2022), and it is usually the key reason why gov-

ernments introduce LCR. Secondly, we assume there is still a sufficient

level of competition in LCR auctions. Absent sufficient competition,

there is no incentive to propel firms to invest in R&D and learning-by-

doing as they are certain to win anyways (Aghion et al.,2005;Rodrik,

2008;Wade,2018;Hufbauer et al.,2013;Panagariya,2011;Krueger,

1997). Thirdly, we assume that the demand from LCR is sufficiently

large to enable learning-by-doing and re-investment into R&D. Finally,

LCR supporters have often been quick to assume LCR create ‘‘additional

demand’’ for local firms. It is unclear whether demand from LCR creates

additional production or simply leads firms to substitute production

capacity or even reduce production to maximize profits (Grossman,

1981;Hufbauer et al.,2013). Production substitution could occur if

a local company has limited production capacity and decides to sell

to the government rather than to other clients. If the government pays

above world market prices, which is the purpose of LCR auctions, firms

also have the incentive to produce less if that maximizes profits or

stop exporting, as local governments may require lower quality and pay

higher prices than foreign clients.

In Section 5, we empirically assess whether the two central mecha-

nisms and the underlying assumptions have been fulfilled in the case of

the LCR introduced in India’s National Solar Mission, which we present

next.

3. Context of the empirical case study

3.1. The Jawaharlal Nehru National Solar Mission

In January 2010, the Indian government inaugurated the Jawaharlal

Nehru National Solar Mission, NSM (see figure Fig. 2 for a visual

overview). In line with the policy objectives of other countries adopting

LCR (Scheifele et al.,2022), the NSM aimed to achieve a global

leadership role in solar manufacturing using leading-edge solar tech-

nologies across the value chain (MnRE,2009). The NSM also explicitly

mentions technical innovation and efficiency improvements as policy

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Fig. 2. Evolution of solar patents filed by the 113 companies participating in solar

auctions in India.

targets (MnRE,2009). The key policy tool to achieve this objective

was the incorporation of LCR as a mandatory requirement in public

solar auctions. The LCR mandated that the solar PV modules and solar

PV cells used by the bidders had to be manufactured in India. While

the NSM also included an R&D program, current beneficiaries are only

universities and research institutes (MnRE,2022).

Before describing the NSM’s and LCR’s design in detail, we examine

whether the conditions for LCR effectiveness were fulfilled prior to the

policy (Hansen et al.,2020). The first factor is market size and stability.

The NSM created effectively a vast market, auctioning off as much

capacity in ten years as Germany had developed over two decades and

transforming India into the fifth largest solar market worldwide (IEA,

2021). As the NSM pre-defined clear and transparent milestones and or-

ganized regular auctions (Sahoo and Shrimali,2013;Singh and Pandey,

2021), stability was also provided. The second success factor is policy

design and coherence and the third one is restrictiveness. Several

studies have documented that there were loopholes in the LCR policy in

the first two years (Sahoo and Shrimali,2013), which incentivized the

use of (imported) thin-film modules, but the Indian government fixed

the loopholes in 2013, which is the year when our analysis starts. From

2013 onward, the LCR policy was coherent, clear and specific in its

focus on solar cells and modules and effective, as reflected in higher

bidding prices (Probst et al.,2020). The focus on solar cells and solar

modules was certainly ambitious, which has a restrictive element, but

LCR did not result in auction cancellation nor did LCR jeopardize the

overall auction schemes cost-effectiveness. Hansen et al. (2020)’s final

criterion, an industrial base, has been in place as India already had

a small, export-oriented PV module manufacturing industry (Johnson,

2016) and several large industrial conglomerates, such as Tata, Adani

or Mahindra. Hence, ex-ante, the Indian solar LCR fulfilled most of the

policy-level success factors.

The NSM was designed in three phases depicted in Fig. 4. Phase I

(2011–2013) auctions included LCR for polysilicon PV, but as India did

not have a strong manufacturing base in thin-film technologies, thin-

film was exempted from the LCR. As an unintended result, 70% of the

auctioned capacity in Phase I was based on mostly imported thin-film

modules (Sahoo and Shrimali,2013).

The Indian government amended the LCR policy for phase II (2013–

2017) of the NSM to encompass both PV cells and modules of both

polysilicon and thin-film technologies. Open auctions without LCR were

Table 1

Balance table: Characteristics of auctions with and without local content.

Variable (1) (2) T-test

auction w/o LCR LCR auction 𝑃-value

Mean (SD) Mean (SD) (1)–(2)

Number of bidders 8.87 4.82 0.07*

(7.17) (5.76)

BOO+PPA vs. EPC+O&M 1.10 1.36 0.11

(0.31) (0.50)

Total MW auctioned 750.43 49.73 0.00***

(1184.14) (109.31)

International bidders invited 1.33 1.09 0.06*

(0.48) (0.30)

Projects in solar park 0.30 0.09 0.10*

(0.47) (0.30)

Climate zone of plant location 2.23 1.91 0.27

(0.82) (0.83)

Technology neutral 1.90 1.64 0.11

(0.31) (0.50)

Max. plant size 255.10 20.18 0.00***

(347.81) (20.18)

Final bid price, INR/kwh 1.41 3.18 0.04**

(2.00) (2.56)

Length of contract 23.40 18.64 0.10*

(4.90) (8.97)

International quality standards 12.77 14.82 0.18

(2.60) (4.87)

Viability-gap-funding, INR 23080464.16 55392127.27 0.41

(53525686.01) (127339657.93)

N 30 11

Notes: The value displayed for t-tests are p-values. Standard deviations are robust.

All missing values in balance variables are treated as zero.

***, **, and * indicate significance at the 1, 5, and 10 percent critical level.

Abbreviations for two types of legal contracts:

BOO =Build, Own, Operate. PPA =Power Purchasing Agreement.

EPC =Engineering, Procurement, Construction. O&M =Operation & Maintenance.

now held in parallel. Having parallel auctions with and without LCR

during Phase II provides the quasi-experimental setting that we exploit

to identify the causal effect of LCR on innovation. In 2017, India

withdrew LCR due to a complaint by the United States of America at

the World Trade Organization (2018). Therefore, Phase III (2017–2022)

of the NSM no longer includes LCR auctions. The Indian Government

revised the targets for Phase III from 20 GW to 100 GW, to be installed

by 2022.

The Indian solar sector first expanded and later consolidated as

the size of auctions and solar power plants increased substantially,

and early adventurous bidders exited. Consequently, the number of

auction participants and the bidding price decreased (see Fig. 10 in

the Appendix). A similar trajectory can be observed in the patent

applications (see Fig. 2). As of January 2022, India has installed 49.3

GW of solar power, which puts it in fifth place after China, USA,

Japan, and Germany (IEA,2021). The NSM is regarded as successful

thanks to the large-scale deployment of solar energy and its long time

horizon (Singh and Pandey,2021;Sahoo and Shrimali,2013).

3.2. The design of auctions and local content requirements

India’s LCR policy required the manufacturing of solar modules

and cells to take place in India, contrary to other countries like South

Africa or China, which mandated minimum shares of local content in

their auctions (Hansen et al.,2020). Fig. 3 illustrates the steps in the

manufacturing process of solar PV modules that the Indian LCR targeted

(black frame).

Table 1 provides an overview of the characteristics and differences

between LCR and open auctions. LCR auctions are less competitive

as only 4.82 bidders, on average, are participating compared to 8.9

bidders in open auctions. LCR auctions are much smaller in auctioned

capacity, with an average of 50 MW per auction compared to 750 MW

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Fig. 3. Steps in the manufacturing process of solar PV modules. Black frame refers to components targeted by LCR. Own depiction based on Şahin and Okumuş (2016).

Fig. 4. Nurturing a domestic solar sector from scratch to 50 GW solar energy: India’s Jawaharlal Nehru National Solar Mission. Sources: Own elaboration based on MnRE (2009)

and Johnson (2016), SECI online archives & data from Indian patent office. Note: The capacity objective for Phase III was revised from originally 20 GW to 100 GW. Objectives

may not necessarily match with the allocated capacity ex-post.

in the open auctions. The size difference is also reflected in the higher

share of solar park projects and the average plant size, which is more

than tenfold in open auctions (255 MW) compared to LCR auctions (20

MW). A fourth difference is the higher average bid price in the LCR

auctions, which re-confirms the findings by Probst et al. (2020). We

explore in Sections 5.4 and 5.5 how the different auction characteristics

may have affected the impact of LCR and open auctions.

4. Data and methods

4.1. Causal identification strategy

Our central identification strategy explores that the Indian govern-

ment ran simultaneous auctions with and without LCR between 2013

and 2017 (Probst et al.,2021), which provides an otherwise unobserved

counterfactual or control group. In contrast to Probst et al. (2021),

we focus on the firm rather than the bid level to identify whether

participation in LCR auctions has had an impact on firms patenting

activities and sales. The LCR or treatment group consists of 33 firms

that either exclusively participated in LCR auctions or participated in

both auction types. The non-LCR group or control group comprises 80

firms that only participated in open auctions.

Since we dispose of annual sales and patent data, we can compare

post and pre-treatment outcomes of the firms in the LCR (‘‘treat-

ment’’) and non-LCR group (‘‘control’’). The causal identification as-

sumption is that the pre-treatment trend in solar patenting activity

among matched non-LCR (control) group firms is not statistically dif-

ferent from matched LCR (treatment) group firms and is assumed

would have developed similarly in absence of the treatment (LCR). We

examine this assumption through an event study design in Fig. 6. There

was no significant difference in the outcome variable before the LCR

introduction between the LCR group and the control group.

Despite equal pre-trends, firms could select in which auction type

to participate since 2013. Hence, LCR participation may be correlated

with (un-) observable differences in firm characteristics that may also

affect patenting and sales. For this reason, we combine the difference-

in-difference estimation with propensity score matching (PSM). The

intuition of PSM is to identify a single or a weighted combination of

several firms in the control group that constitute a control unit as simi-

lar as possible to a firm or several firms in the treatment group. Table 2

provides results from a balance test on the firm characteristics that

predict participation in LCR auctions (see Section 5.1) before (columns

1 and 2) and after matching (columns 3 and 4). The PSM procedure

eliminates statistically significant differences between LCR and open

auction participants in the variables that predict LCR participation. This

implies that conditional on matching, LCR participation should not be

correlated with any other firm characteristics that predict participation

in LCR auctions anymore.

One potential concern is that our definition of the treatment vari-

able – firms that won at least one LCR auction – may not correctly

assign treatment status. To assess that, we calculate the share of won

LCR auctions for all firms that won at least one between 2013 and 2017.

Fig. 12 in the Appendix illustrates that for 13 out of the 16 firms that

won LCR auctions made up 50% or more of the total won auctions

between 2013 and 2017.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Table 2

Balance table: Firm characteristics predicting participation in auctions with & without local content before & after matching.

Variable Unmatched Matched

No LCR participation LCR participation T-Test No LCR participation LCR participation T-test

Mean (SD) Mean (SD) (1)–(2) P-Value Mean (SD) Mean (SD) (3)–(4) P-Value

log_total_employees 4.77

(1.95)

4.75

(2.18)

0.95 4.93

(2.87)

4.61

(2.00)

0.51

ihs transf. pre-LCR sales 15.62

(8.29)

18.35

(7.27)

0.06* 18.28

(7.87)

17.93

(7.27)

0.82

Solar patents 2001–2010 0.03

(0.16)

0.58

(2.80)

0.26 0.11

(0.74)

0.06

(0.36)

0.67

Indian company 0.71

(0.46)

0.91

(0.29)

0.01*** 0.86

(0.40)

0.90

(0.30)

0.56

Manufacturing company 0.07

(0.27)

0.33

(0.48)

0.00*** 0.18

(0.76)

0.29

(0.46)

0.34

Part 1 NSM 0.03

(0.16)

0.09

(0.29)

0.22 0.12

(0.72)

0.10

(0.30)

0.81

N 80 33 79 31

Notes: The value displayed for t-tests are p-values. Standard deviations are robust.

All missing values in balance variables are treated as zero. ***, **, and *, indicate significance at the 1, 5, and 10 percent critical level.

Balance is evaluated for all variables that predicted participation in LCR.

Sample size is reduced after matching given firms outside the caliper are excluded.

4.2. Matching procedure

We follow the steps outlined in Caliendo and Kopeinig (2008) for

the PSM analysis. We opt for a logit model over a probit model to

estimate a propensity score (PS) as the former has more density mass

in the bounds, which fits better to our data (for details, see 5.1).

We select variables used for matching in the following way. First,

we select variables that were either empirically predictive of or theo-

retically important for predicting treatment status or the main outcome

variable (Caliendo and Kopeinig,2008). Table 5 in the Appendix

illustrates which variables we tested. Second, we used the value of

these variables before the start of the treatment (Caliendo and Kopeinig,

2008). The final logit estimation of firms’ propensity to participate in

LCR auctions is shown in Eq. (1):

𝑃(𝑌𝑖= 1) =𝛽0+𝛽1𝑠𝑜𝑙𝑎𝑟𝑝𝑎𝑡𝑒𝑛𝑡𝑠𝑖+𝛽2𝑒𝑚𝑝𝑙𝑖+𝛽3𝑖𝑛𝑑𝑖𝑎𝑛𝑖+

𝛽4𝑚𝑎𝑛𝑢𝑓𝑖+𝛽5𝑛𝑠𝑚𝑖+𝛽6𝑠𝑎𝑙𝑒𝑠𝑖+𝜀𝑖

(1)

where P(Yi=1) is a dummy variable that equals 1 if a firm participated

in or won at least one auction with local content, solarpatents is the

number of solar patents filed before the initiation of the NSM in 2011,

empl is the log-transformed number of employees, indian is a dummy

whether a company is considered Indian, manuf is a dummy whether

the company is a manufacturer, nsm is a dummy whether the company

has already participated in the first phase of the NSM auctions, and

sales is the inverse hyperbolic sine transformation of the average sales

of the company in the years before LCR in 2013 or the first year of

available data if the company was founded after 2013. Besides being

empirically significant predictors of LCR participation, the inclusion of

the Indian origin and manufacturer dummy variables is also in line

with the theoretical expectation from our framework that shows that

domestic producers are primary beneficiaries of the policy. As firms’

size, both in revenues and employees, influences firms’ R&D capacity,

they may influence patenting outcomes and are therefore included

in the matching. The results of the PS estimation are discussed in

Section 5.1.

We employ a caliper radius matching (Dehejia and Wahba,2002)

but also provide results from nearest neighbor matching with replace-

ment as a robustness check. Caliper radius matching reduces bias

by limiting matches to control observations within a radius of the

propensity score, called the ‘‘caliper’’, of the treated observation (Lunt,

2014;Cochran and Rubin,1973). We adopt a conventional, rather strict

caliper radius of 0.2 standard deviations of the propensity score (0.05

and 0.1 PS units) that should eliminate 93 to 98% of the remaining

bias (Austin,2011;Cochran and Rubin,1973). We use caliper matching

with equal weights rather than distance-weighted radius matching as

we have several observations with identical PS (Huber et al.,2013).

Observations outside the caliper are dropped, which is how common

support is imposed. Fig. 18 in the Appendix illustrates substantial

common support at the lower levels of the PS, while common support

is weaker at higher levels of the PS. There are fewer firms that share

the characteristics predicting LCR participation that did not participate

in LCR auctions. We lose two firms (Bharat and Photon) in the LCR

group with the highest PS when imposing a narrow caliper of 0.05

PS units. Firms in the control group with a high propensity score,

such as ACME and Mahindra, play an important role, given they serve

as counterfactuals for more than one similar firm in the treatment

group. Overall, as we lose at most 2 companies in the strictest caliper

matching, common support holds reasonably.

4.3. Model specification

We estimate two main difference-in-difference specifications. The

first specification is a staggered difference-in-difference as defined

by Callaway and Sant’Anna (2021), and the second is a canonical

2×2 difference-in-difference specification as in Munch and Schaur

(2018). The staggered difference-in-difference takes into account that

firms won (LCR) auctions in different years between 2013 and 2017

and estimates and aggregates the effect for cohorts of firms that won

an LCR auction for the first time in the same year between 2013

and 2017. The advantage of this specification is that it can capture

if the treatment effect on solar patents and revenue only materializes

in the year or years after the firm is treated (and may also vanish

earlier if it is short-lived). The canonical 2 ×2 difference-in-difference

has the advantage that all annual treatment effects are aggregated,

which may increase power if many small effects add up over years.

We combine both difference-in-difference estimators with propensity

score matching (which is built-in in the case of Callaway and Sant’Anna

(2021)). This has the advantage that the firms in both groups are

more homogeneous in terms of the variables that predict selection into

LCR auctions, which makes the identification assumption of equal pre-

trends more convincing. The staggered difference-in-difference follows

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

the specification defined by (Callaway and Sant’Anna,2021):

𝐴𝑇 𝑇 (𝑔, 𝑡, 𝛿) = E

⎡

⎢

⎣

⎛

⎜

⎝

𝐺𝑔

E[𝐺𝑔]−

𝑝𝑔,𝑡+𝛿(𝑋)(1−𝐷𝑡+𝛿)

1−𝑝𝑔,𝑡+𝛿(𝑋)

E[𝑝𝑔,𝑡+𝛿(𝑋)(1−𝐷𝑡+𝛿)

1−𝑝𝑔,𝑡+𝛿(𝑋)]

⎞

⎟

⎠

(𝑌𝑡−𝑌𝑔−𝛿−1 −𝑚𝑔,𝑡,𝛿 (𝑋))

⎤

⎥

⎦

(2)

is a regression for the evolution of the outcome Yon covariates. In

other words, we deduct from the outcome Yin time tthe variation in

Yexplained by covariates. 𝑝𝑔,𝑡+𝛿(𝑋)(1 − 𝐷𝑡+𝛿)is the propensity score

estimate that a firm wins an LCR auction for the first time in a specific

treatment year, which turns on if the firm is treated in this year and

is zero otherwise. Otherwise, grefers to the cohort or group of firms

that won an LCR auction in the same year, trefers to a year between

2004 and 2020, 𝛿=0 is an assumption of no anticipation effects, pis

the propensity score that a firm is part of a specific cohort, Drefers to

the cohort of firms treatment status (=1 if firms won LCR auction, and

=0 if firms won an open auction), and Xis a vector of the following

covariates: sales in the year preceding or the earliest available before

LCR, a dummy for a firm having manufacturing capacity, and a dummy

for a firm being majority owned by Indian capital. As long as firms have

never won an LCR auction, they are part of the control group. Given

we use covariates, we use the doubly robust difference-in-difference

estimator as suggested by Sant’Anna and Zhao (2020). All specifications

use robust, firm-level clustered standard errors.

The second difference-in-difference specification is equivalent to

what Goodman-Bacon (2021) calls a canonical 2 ×2 design. In this

case, we simply aggregate all solar patents filed by the firms in the

LCR and in the open auction group before and after the policy. Com-

bining the PSM with the 2 ×2 difference-in-difference approach, as

in Munch and Schaur (2018), leads to the following specification for

the estimation of the treatment effect on the treated:

𝛿𝐴𝑇 𝑇 =1

𝑁𝛴𝑁

𝑖=1(𝛥𝑦𝑇

𝑖−𝑤𝑐∗𝛥𝑦𝐶

𝑖)(3)

where 𝛥𝑌 represents the difference in solar patents or revenue for firm

iin the treatment group Tor control group Csummed over the pre-

treatment period 2005–2012 and post-treatment period 2013–2020. W

refers to the weight that each of the control observations receives. Note

that wdoes not vary across firms ibut only across calipers c, as it is

the same for all firms within the same caliper c.

We also run several other specifications to examine the robustness

of the results, which we report in the results section.

4.4. Variables and data collection

Our sample is composed of 113 firms that were bidding in a public

auction at the federal level between 2011 and 2019. We further de-

scribe firms’ characteristics in Section 5and focus in the following on

the description and documentation of the data.

Table 3 provides an overview of the dependent treatment variable as

well as other firm-level variables that were used in the PSM. The main

outcome variable is the number of solar patent applications per com-

pany filed at the Indian patent office, either annually or cumulative,

before and after the introduction of the LCR policy in the second phase

of the NSM. We also examine the number of cells and module-related

patents. We follow an established approach in the literature by using

patents as an indicator of innovation output (Griliches,1990). While

companies may use informal methods of intellectual property protec-

tion, such as trade secrecy, patents present a quantifiable measure

of inventive innovation and have been increasingly filed in emerging

markets in the solar industry (Luan et al.,2021).

We use firms’ revenue as a secondary outcome variable to ap-

proximate their production. Revenues are a more immediate outcome

than patents and may directly capture a potential demand effect from

LCR auctions. We use the revenue from the latest year available in

Table 3

Descriptive statistics.

Variable Mean SD Min Max

Solar patents 2011–2020 0.696 2.850 0 18

Solar patents 2001–2010 0.188 1.528 0 16

Participated (or not) in LCR auction 0.295 0.458 0 1

Indian company 0.768 0.424 0 1

Manufacturing company 0.152 0.360 0 1

ihs transf. pre-LCR sales 16.40 8.106 0 27.71

ihs transf. post-LCR sales 21.35 3.085 10.17 28.36

log_total_employees 4.763 2.020 0 10.65

Part 1 NSM 0.0446 0.207 0 1

Observations 112

Note: There are 112 observations as employees is missing for one company.

the data to allow for the longest possible time span between auction

participation and outcome.

The main explanatory or treatment variable is a dummy variable

equal to one (the firm is treated) if a company ever participated in an

LCR auction during the second phase of the NSM or zero otherwise (the

firm is in the control group). Given LCR participation may have been

infrequent or unsuccessful, we create a continuous treatment variable

that refers to the number of times a company participated and a dummy

of whether a firm has ever won an LCR auction. Control variables used

in the PSM are listed in Table 5.

The data flowchart in Fig. 9 in the Appendix illustrates how

different data sources have been combined. Firstly, data about auctions,

technical and financial eligibility criteria, and the firms bidding have

been collected from the Solar Energy Corporation India (SECI) online

archives as outlined in Münch and Marian (2022).1

Secondly, we searched the Indian Patent Office’s online portal using

the bidding firms’ name as the applicant name.2After refinement of the

firm name, leaving out common words and terms relating to the legal

form of the company, we could identify and semi-automatically web-

scrape 8845 patents filed by these firms at the Indian patent office. We

used the list of international patent classification (IPC) codes relating

to solar PV components by Shubbak (2019) to identify all solar-related

patents among the 8845 patents.

Thirdly, we collected data about the firm characteristics such as

origin and number of employees in Mergent Intellect’s database as

in Probst et al. (2021). Revenue data was obtained through the Indian

data provider Tofler, which digitally compiles the financial data that

firms file to the Indian Ministry of Corporate Affairs. To avoid the

inclusion of revenues that may have been unaffected by the auctions,

we only used revenue data from the legal entity mentioned in the

auction documents rather than its Indian or foreign mother company.

5. Results

5.1. What predicts winning local content auctions and innovation?

In the following, we analyze which firm-level characteristics predict

whether firms participate in LCR auctions and whether they file a solar

patent or not.

While market entrants come from different sectors (see Fig. 11 in

Appendix), one can classify them broadly into three categories. The

largest group is Indian and international firms that focus on solar plant

development (engineering, procurement, and construction (EPC) and

1Link to SECI’s online archives: https://www.seci.co.in/archives/data_

archives, last accessed May 15th 2022.

2Link to Indian Patents Office’s online search portal: https:

//ipindiaservices.gov.in/publicsearch, last accessed May 15th 2022.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

operations and maintenance (O&M) and sell electricity commercially

to government utilities. Some of these partly Indian and primarily

international firms have been active in developing renewable or con-

ventional power projects before (e.g., NTPC, Larsen & Toubro, Enel,

EDF, SunEdison). Some were newly founded and turned into publicly

listed companies (e.g., Azure), and others came from the civil or real-

estate infrastructure construction more broadly. A second category is

medium-sized (e.g., ACME) and large Indian (e.g., Tata, Mahindra,

Adani), and international conglomerates (e.g., Softbank, Bosch). They

differ from the previous category through their in-house manufacturing

capacity in solar or other related industries, particularly in electronics

(e.g., Bharat). A third category is Indian solar manufacturers that had

mostly assembled modules for export to the EU or the US prior to the

NSM (e.g., Waree, Surana, Vikram Solar). Note that one cannot clearly

distinguish project developers and manufacturers, as, in general, there

is a tendency for the firms that started out with one to diversify and

add additional activities from the other. For example, Vikram Solar and

ACME started as manufacturers of solar modules but have since added

substantial EPC and O&M business. Inversely, Adani, Azure, Tata, and

Alfanar began as project developers but later started to develop their

own solar module production.

Table 10 examines the firm-level determinants of having at least

filed one solar patent. The results suggest that company size, proxied

by its sales and employees, and manufacturing capacity, are positively

and statistically significantly associated with solar patenting. Having

said this, Fig. 14 and Fig. 15 show that EPC companies also filed a

substantial share of solar patents. In addition, an analysis of solar patent

abstracts suggests that manufacturers do not exclusively or at all patent

in regards to a solar module or cell. Tata, for example, while having so-

lar module manufacturing capabilities, exclusively filed patents related

to EPC activities. Being an Indian rather than an international company

has a marginally negative significant effect, which disappears, however,

once one controls for the manufacturing status.

The predictive characteristics of companies participating in LCR

auctions align with the theoretical framework presented in Section 2.

Being an Indian (solar) manufacturer, an electronics component man-

ufacturer, the number of solar patents filed between 2001 and 2010,

and average sales before LCR positively predict a firm’s participation

in LCR auctions (see Table 5 in the Appendix). This translated into

the following difference in these characteristics between LCR and open

auction participants (see Table 12). Firstly, LCR auction participants

were almost exclusively of Indian ownership. Only 3 out of 33 or 9%

of the firms in LCR auctions were international, while 23 of 80 or

28.75% of open auction participants were international. Secondly, 11

of the 17 manufacturing companies participated in the LCR auctions.

Finally, even after controlling for outliers, firms in LCR auctions were,

on average, still 5 years older than firms in open auctions. This suggests

that LCR auctions primarily attracted the intended target group: local,

incumbent manufacturers of solar modules or related products.

5.2. What is the impact of local content on firm innovation?

We first examine visually the evolution of solar patents filed by the

firms that participated in the Indian solar auctions (Fig. 5). While the

absolute amount of solar patents filed by the 113 participating firms is

rather low (100 since 1982, thereof 99 since 2001), the dynamic has

increased since the beginning of the NSM as the firms filed in the last

decade twice as many solar patents as in the previous three decades

from 1982 to 2010. The increase relative to the decade before the

NSM is stronger among open auction participants (+1500%) vs. LCR

participants (+142%).

Fig. 6 visualizes the results from the staggered difference-in-

difference model (Callaway and Sant’Anna,2021) defined in Section 4.

Overall, the results suggest that LCR still had no significant effect

on the number of annual solar patents firms filed. However, taking

into account when cohorts of firms won an LCR auction for the first

Fig. 5. Solar patents filed by firms participating in solar auctions before and after the

use of LCR.

Fig. 6. Staggered difference-in-difference based on Callaway and Sant’Anna (2021),

which takes the timing into account when firms won LCR vs. open auctions. The sample

is restricted to firms that won either one LCR or one open auction. Years to treatment

are normalized and are relative to the year when firms won an LCR auction for the

first time. Not-yet-treated firms are part of the control group until they are treated. The

box in the center-top of the figure provides the aggregated average treatment effect

on the treated (ATT), the associated confidence interval (CI), and the z-value of the

coefficient.

time reveals an interesting, suggestive pattern. Albeit insignificant, the

visualization suggests that LCR may have had a small short-term effect

as the coefficient for the first year post-treatment is the only positive

one with a confidence interval ranging from −0.28 to +.92 additional

solar patents. The treatment effect remains insignificant throughout the

years, but the central tendency turns negative.

Table 4 reports the results from the canonical 2 ×2 differences-

in-differences estimates, which pool the outcome variables before and

after LCR introduction. Panel A in Table 4 shows that the only signifi-

cant estimate is a simple, unmatched post-LCR comparison of the means

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Table 4

Results of matching combined with difference in differences.

All firms Winner firms All w/o outliers

(1) (2) (3) (4) (5) (6) (7) (8)

Simple post difference Unmatched DiD caliper =0.1 caliper =0.05 caliper =0.1 caliper =0.05 caliper =0.1 caliper =0.05

Panel A: Solar PV patents

Participated in LCR 0.99* 0.44 −0.06 −0.03 −0.93 −0.80 −0.85 −0.78

(0.58) (0.57) (0.80) (0.80) (1.19) (0.69) (0.75) (0.63)

Constant 0.40 0.37* 0.89 0.87 1.45 0.80 1.00 0.82

(0.31) (0.19) (0.57) (0.56) (1.14) (0.69) (0.74) (0.62)

Observations 113 113 110 109 66 60 106 104

Panel B: PV module & PV cell patents only

Participated in LCR 0.61 0.25 −0.33 −0.36 −0.69 −0.65 −0.47 −0.54

(0.38) (0.27) (0.48) (0.48) (1.00) (0.57) (0.66) (0.53)

Constant 0.21 0.20 0.65 0.66 1.21 0.65 0.72 0.58

(0.21) (0.14) (0.44) (0.44) (0.95) (0.57) (0.63) (0.53)

Observations 113 113 110 109 66 60 106 104

Panel C: Post-LCR solar patent (binary)

Participated in LCR 0.06 −0.00 −0.05 −0.07 −0.12 −0.17 −0.13 −0.16

(0.06) (0.07) (0.10) (0.10) (0.20) (0.11) (0.14) (0.12)

Constant 0.09** 0.06* 0.18** 0.17** 0.27 0.17 0.23* 0.20*

(0.03) (0.03) (0.08) (0.08) (0.19) (0.11) (0.13) (0.11)

Observations 113 113 110 109 66 60 106 104

Panel D: Revenues (in INR)

Participated in LCR −2.00e+10 −2.00e+10 −1.29e+10 −1.04e+10 −3.07e+10* −1.97e+10 −2.81e+09 −6.60e+09

(1.33e+10) (1.24e+10) (8.09e+09) (6.83e+09) (1.73e+10) (1.22e+10) (4.05e+09) (4.82e+09)

Constant 1.49e+10** 1.49e+10* 1.76e+10** 1.49e+10** 1.91e+10** 2.13e+10* 7.47e+09** 9.95e+09**

(7.16e+09) (7.57e+09) (7.94e+09) (6.66e+09) (8.41e+09) (1.22e+10) (3.76e+09) (4.56e+09)

Observations 113 113 110 109 66 60 108 104

Results in columns (1) & (2) use unmatched counterfactuals and columns (3)–(8) use matched counterfactuals based on the specified parameters.

Robust Standard errors in parentheses.

**𝑝 < 0.05, *𝑝 < 0.1.

(Column 1), which reflects Fig. 5. A simple difference-in-difference

(Column 2), difference-in-difference with caliper matching (Columns 3

and 4), and restriction to LCR vs. open auction winner firms (Columns 5

and 6) as well as controlling for removing outliers (Columns 7 and 8) all

suggest LCR had no significant impact on firms’ solar patenting activity.

Panels B and C illustrate that this result is also robust to restricting

the scope of patents to module and cell solar patents (Panel B), which

are the components that were targeted by the LCR policy, and the

probability of filling a solar PV patent at all (Panel C).

What is more, the point estimates are small. Even if they were

significant, the estimates suggest that the extensive margin of LCR

participation has a close to zero effect, the magnitude of the point

estimates increase to a third to almost a whole solar patent as well

as a 12 to 17% reduction in the probability of filing a solar patent

when narrowing the scope to cell and module patents and to winning

rather than participating. We undertake an additional robustness test

by using the times firms participated in LCR auctions rather than a

binary treatment indicator (see Table 9 in the Appendix). Considering

firms’ intensity of LCR auction participation reduces the negative point

estimate, but results remain insignificant. Finally, results are also robust

to an alternative matching algorithm (see Table 8 in Appendix).

Overall, the results suggest that LCR auctions had no significant effect

on firms’ solar patenting activity.

5.3. What is the impact of local content on firm revenue?

Fig. 7 visualizes the dynamic effect of LCR on firms’ annual revenue.

The results resonate with the picture that emerged from the analysis of

the effect of LCR on firms’ annual solar patents. Overall, LCR had no

statistically significant effect on firms’ annual revenue as documented

by the ATT reported in the center top of the figure and the overlap of

the 95% confidence intervals with zero. Yet, there is some suggestive

evidence that LCR may have had a positive, short-term effect, partic-

ularly in the year right after winning an LCR auction. Thereafter the

point estimates turn negative and increase in their magnitude from the

second year onward but remain insignificant. Note that the increase

in the confidence intervals with the distance to the relative years to

treatment is due to the shrinking in the sample size. For example, one

cohort of LCR winners has only been treated in 2015 and another in

2016. For these firms, we logically only observe four and five years

post-treatment.

In Table 4 Panel D and Fig. 16, we report the results for revenues

from the canonical 2 ×2 difference-in-difference specification. The

pooled estimates are, in line with the dynamic annual results, also neg-

ative but insignificant. Columns 1 and 2 present the results for a naive

post difference and a difference-in-difference specification. columns

3 and 4 provide the estimates for 0.1 and 0.05 caliper matching

difference-in-difference. We consider only winning bidders (Columns 5

and 6) and exclude three outliers (NTPC, Bharat, and Larsen & Toubro),

which have very high revenues (Columns 7 and 8). Finally, we run

the same robustness tests with a continuous treatment indicator and

alternative matching algorithm as for the patent analysis (see Panel D

of Tables 9 and 8in the Appendix). The null result is robust to all

mentioned specifications.

In conclusion, the results suggest LCR did not have a significant

impact on either firms’ solar patent innovation or their revenue. The

results are robust to various difference-in-difference specifications, in-

cluding or excluding matching on covariates predicting LCR participa-

tion, definitions of the dependent variable, and sample or treatment

assignment specifications. The dynamics in the insignificant annual

point estimates suggest that LCR may have had a small, positive short-

term effect on innovation and revenues, which turned negative over

time.

5.4. Did local content provide a sufficient carrot to incentive innovation?

Rodrik (2008) suggests that ‘‘carrots & sticks’’, providing sufficient

incentives or business opportunities while simultaneously demand-

ing efficiency and quality improvements to catch up with the global

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Fig. 7. Staggered difference-in-difference based on Callaway and Sant’Anna (2021),

which takes the timing into account when firms won LCR vs. open auctions. The sample

is restricted to firms that won either one LCR or one open auction. Years to treatment

are normalized and are relative to the year when firms won an LCR auction for the

first time. Not-yet-treated firms are part of the control group until they are treated. The

box in the center-top of the figure provides the aggregated average treatment effect

on the treated (ATT), the associated confidence interval (CI), and the z-value of the

coefficient.

frontier, are crucial for successful industrial policy. Following this

argument, we conduct a falsification exercise and assess whether LCR

have created a sufficiently big incentive or demand shock.

We assess this potential explanation in three ways. First, we ex-

amine whether the firms won sufficient capacity (in MW) to result in

learning-by-doing from producing the solar modules necessary to gener-

ate this capacity (see Fig. 1 impact mechanism 1). Fig. 17 illustrates the

accumulated MW each firm won in LCR and in open auctions between

2013–2017, conditional on having at least won one LCR auction. On

average, firms won only 23 MW over the four years, and even the best

performer – Azure – received only 67 MW in total. To put this into

context, large contemporary solar module manufacturing plants have

an annual capacity of 100 MW+. No single firm won more than 67

MW or 14.4% of the total capacity allocated in LCR auctions, and the

larger conglomerates won around five to eight times as much capacity

in open auctions. Smaller domestic module producers, such as Vikram,

Waaree, Swelect, and Surana, won as much MW in LCR as in open

auctions. However, the absolute MW amount they each won is less than

30 MW and thus so low that one could not expect any major learning

effects or inventive activities. This is also reflected in the analysis of

the patents that LCR and non-LCR firms filed. Given the LCR policy

specifically targeted solar modules and cells, we analyzed, based on

IPC patent codes, whether LCR participants were more likely or file

more cells about the targeted components. LCR participants’ patents fall

more or less in the same categories as those filed by firms that never

participated in LCR auctions. We conclude that the demand from local

content created only little production, resulting in few opportunities for

learning by doing over the four years.

Secondly, we assess whether LCR generated sufficient revenue for

reinvestment into R&D required for patenting (see Fig. 1 impact mech-

anism 2). Firms supported by the EU’s SME R&D grant program and

the US Department of Energies Small Business Innovation Research

program file on average 0.7–1.3 patents per million USD of R&D

expenditure per company (Clancy,2021;Santoleri et al.,2020;Howell,

2017). This is similar to the average patent return to R&D expendi-

ture in the US, which is around 0.5 patents per million USD R&D

expenditure (Clancy,2021). Companies that participated in LCR auc-

tions gained on average contracts for modules worth approximately

10 million USD per firm (see Figs. 8)3. Accordingly, firms would have

3To estimate the USD amount per company, we used the allocated MW

amount in auctions and multiplied it with the average international module

had to reinvest approximately 10% of their LCR-induced revenues into

R&D to generate one additional patent on average (assuming a similar

patent return to private R&D in India as in the EU and the US). Indian

firms reinvest on average only 0.9% of their sales and even Tata, the

Indian firm with the highest R&D expenditure, only reinvested 7.1%

of its sales in 2020 (Kumar,2020). Even the firms that receive large,

above-average LCR demand stimuli like Azure (42 million USD), Adani

(37 million USD), Tata (34 million USD), and Waree (31 million USD),

would have still only generated 1.5–2 additional patents per company

assuming 5% re-investment.

Finally, we examine the costs of the policy to the government. Was

the government’s objective of nurturing a domestic, globally compet-

itive solar industry aligned with the resources it invested? In total,

the Indian government allocated approximately 6 GW that generated

a demand for solar PV modules worth an estimated 8.65 billion USD

from 2013–2017. Only around 10% of the capacity or 329 million USD

worth of solar PV modules were auctioned in LCR auctions. Hence,

even after assuming slightly higher module prices, the overall demand

that firms received from LCR auctions between 2013–2017 was small,

only around 12.5%–20% of the total demand. We also calculate the

additional costs of the LCR policy to the Indian government compared

to a scenario where the Indian government had just auctioned the same

capacity in open auctions. The results suggest the LCR policy has cost

the Indian government an additional 18.5 million USD compared to a

scenario where they auctioned off all the capacity in open auctions.

Needless to say that these are small investment costs relative to the

objective of nurturing a globally competitive and innovative solar

industry for a country of the size of India.

As a result, we conclude that the size of the demand shock from the

LCR policy was too small to induce learning-by-doing from production

or to generate sufficient revenue for reinvestment into R&D. Accord-

ingly, the two main impact mechanisms identified in Section 2could

not materialize.

5.5. Did local content provide a sufficient stick to incentivize innovation?

A second potential explanation relates to the common critique that

industrial policies provide too much protection and thus no incentive

for beneficiaries to improve, given they are spared from competing with

the globally most productive firms. If true, LCR may unintentionally

and unexpectedly reduce firms’ incentive to innovate.

The fact that competition in LCR auctions was considerably lower

than in open auctions lends meaning to this explanation. Table 11 illus-

trates that firms had, on average, a 63.5% chance that their bid would

be successful in LCR auctions, which was 22 or 15, after controlling for

covariates, percentage points higher than in open auctions. In Table 1,

we have already shown that the number of bidders in LCR auctions

(4.82) was only about half the number in open auctions (8.9) and that

bidders were less international. Moreover, Münch and Marian (2022)

showed that LCRs were not combined with any other performance

requirement, such as international quality standards, which could have

functioned as performance requirements in the absence of competition

from international (and many national) firms.

Finally, Fig. 13 in the Appendix shows that only very few firms

managed to file a single solar patent since the beginning of the LCR

auctions in 2013. Most of the firms that filed solar patents are ei-

ther large Indian or international conglomerates. This illustrates that

consolidation and concentration of the major share of business among

a few potent players have followed the initial entry euphoria in the

Indian solar auctions. These ‘‘infants’’ may not necessarily only require

protection but may, in the absence of foreign competition, benefit from

alternative performance requirements that push them to innovate.

price between 2013–2017 (source: Our World in Data global average solar

(PV) module price) Given previous research (Probst et al.,2020) had shown

prices in Indian LCR auctions were 6% higher, we adjusted the value for LCR

auctions accordingly.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Fig. 8. Est. USD value of module demand of LCR vs. open auctions (2013–2017).

6. Discussion

One alternative explanation could be that we simply lack statistical

power and therefore measure a null effect (type-II error). To assess

the plausibility of the null effect, we estimate how many patents and

revenue LCR would have needed to create to measure a statistically

significant effect. We conduct ex-post power calculations using the esti-

mated standard errors from Table 4.4We find that the LCR policy would

have had to induce, on average, 2.1 [1.59-3.33] additional patents

based on the standard errors in Table 4 in the LCR group relative to

the open auction group to detect statistically significant effects given

the sample size. As revenue is an outcome with much higher variance

than patents, precise estimation ideally requires a larger sample size

and, as we showed above, a more substantial treatment. For example,

given the estimate standards errors in Table 4, the smallest statistically

significant change in revenue we could have detected was 188 million

USD. As we outlined, the average demand shock was only around 10

million USD, which illustrates that we lack statistical power to measure

a precise effect and therefore implies that it is difficult to tell apart a

null effect and type-2 error in the case of the revenue analysis.

Are 2.1 [1.59–3.33] patents per firm in response to an LCR policy

realistic or an implausible effect? While there is no existing estimate

of the specific impact of any LCR policy on firm performance, there

are studies of other government support programs, which provide a

good benchmark in so far as the Indian government could have opted

4To have an 80% chance of drawing an estimator that is 1.96 standard

errors away from zero, we multiply the estimated standard errors of the LCR

coefficient in Table 4 with 2.8 as the inverse normal of 80% is 0.84 and 1.96

+0.84 =2.8. See, for example, Ioannidis et al. (2017) for a more detailed

explanation.

for different policy options, such as R&D grants, to promote domestic

solar component production and innovation (Clancy,2021;Howell,

2017;Santoleri et al.,2020). Accordingly, to generate 2.1 patents per

firm, LCR firms would have needed to receive an average demand

shock worth 60 million USD, assuming an average 5% reinvestment

of revenue into R&D5and assuming every LCR dollar has the same

effect as a dollar from an R&D subsidy in the EU or the US public

or private sector.6This is six times as much as firms won on average

in LCR auctions. At the same time, it is not an unrealistic amount,

in particular given the governments’ objective to develop a globally

competitive solar manufacturing sector (MnRE,2009). Assuming the 33

LCR participants had won 60 rather than 10 million USD, on average,

the total expenditure on PV modules would have been 1.98 billion

instead of 329 million USD or 22% of the total demand allocated in

the solar auctions in this period (8.65 billion).

Finally, there remains the question of what can be concluded regard-

ing the overall effectiveness of LCR. While we show that LCR did not

induce more innovation nor increase revenues, an overall judgment of

the policy’s effectiveness ultimately depends on what is being defined

as success. While we argue, based on the industrial policy literature,

that global competitiveness should be the ultimate success criterion for

LCR policies, as it is required in order to terminate the protection, one

may include alternative measures. Those can be but are not limited to,

employment creation (Hansen et al.,2020), firm creation, or survival

5For comparison Tata, the Indian firm with the highest R&D expenditure,

reinvested 7.1% of its sales in 2020 (Kumar,2020).

660 ∗0.05 =3. 3 ∗0.7 =2.1 patents. Note that we do not account for

learning-by-doing in this simplified calculation. If the input–output efficiency

of the LCR policy is lower (higher) than the EU or US private and public R&D,

the per-company demand shock would need to be even higher (lower).

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

and may lead to alternative conclusions. For example, one benefit

of the LCR policy may have been that Indian manufacturers ‘‘sur-

vived’’ (Johnson,2016), compared to manufacturers in other countries,

such as Germany or the United States. Future studies should assess these

impacts by collecting, if available, employee data to study whether LCR

create jobs, even if industries may not be internationally competitive.

Furthermore, LCR may promote technological capabilities that are not

necessarily new to the frontier inventions and, therefore, may not be

captured well by patents or which may occur in down-or upstream

companies through backward and forward linkages. Future research

should explore this by employing firm innovation surveys or analyzing

firm administrative records to capture innovative activity that does

not lead to a patent. What is more, it may take even more time than

the observed 8 years (even 10 years if one counts the first two years

without parallel open auctions) until increases in firms’ technological

capabilities, which are potentially unobserved in this study, translate

into inventions protected via patents. While we cannot ultimately ex-

clude this option, it seems very unlikely that future patents are related

to the LCR auctions observed in this period as, as discussed in detail

in Section 5.4, the firm-specific production capacity related to these

auctions was too small to enable learning-by-doing from production or

incentivize investment into new production facilities.

7. Conclusion & policy implications

While there has been an increase in patent applications with the

onset of the National Solar Mission in 2011 in India, we do not find

evidence that winning an auction with rather than without LCR led to

more solar patents or revenue. This result is robust to various specifica-

tions and robustness checks. The analysis underlines the predicament

governments face as open auction winners, despite having won 9

times as much capacity, do not patent much (more). Overall, patenting

activity remains limited to a small number of auction participants (14

out of 113).

We derive two primary explanations for why the LCR auction

scheme did not have a significant innovation impact. The first expla-

nation is that the demand stimulus provided by the LCR auction was

insufficient. We conduct a falsification exercise and show that, given

the average demand stimulus each company received in terms of capac-

ity (23 MW) and revenue ($ 10 million) over the four-year treatment

period, the two theoretical impact channels – learning-by-doing and

revenue re-investment into R&D – could not materialize. More specifi-

cally, we illustrate that the average demand shock translated only into

a fifth of the annual capacity of a contemporary module manufacturing

plant and that companies would have had a revenue-R&D investment

rate twice as high as India’s firm with the highest corporate R&D

reinvestment rate. The dynamic evolution in the insignificant point esti-

mates (switch from positive in the first two to negative in the following

years) resonates with this analysis. Manufacturing occurs within the

first 24 months after an auction is won, while revenues from operating

and selling electricity, which is the main impact channel identified for

open auctions, start to kick in once the plant has been commissioned

(approximately 18–24 months after the auction was won). Finally, 16

out of the 23 LCR-winning firms won only one LCR auction, and LCR

auctions remained small in size, which implies that the LCR scheme

failed to provide firms with regular and recurring, increasingly large

demand shocks.

A second explanation is that local content reduced competition in

auctions and, as no alternative performance requirements were defined,

reduced firms’ incentive to innovate. There are two supporting pieces

of evidence. Firstly, the competition in LCR auctions is much lower

than in open ones, as the number of bidders is almost half and less

international. In addition, the winning probability in LCR auctions is 15

percentage points higher than in open auctions, controlling for several

covariates.

The findings point towards the following policy implications. Firstly,

the size of local content auctions needs to be more substantial or

concentrated among a few companies, increase gradually in size, and

occur regularly. For many countries, in particular small countries, this

implies that their auction-related component demand, relative to what

is required to enable learning-by-doing and revenue re-investment into

R&D, is simply not large enough to propel domestic manufacturers

to catch up with the global technology frontier. Secondly, while LCR

were successful in attracting its target group, Indian module and cell

manufacturers, they significantly reduced competition. This implies it

is imperative for any government using LCR in the future to combine

LCR with alternative performance requirements, such as international

quality standards (Münch and Marian,2022) or innovation require-

ments, such as gradually increasing improvements in cell efficiency

levels, as done for example in the Chinese top runner program (IEA,

2022). In sum, we define, similar as Hansen et al. (2020) on the

sector-level, two firm-level success factors for LCR effectiveness. First,

substantial, firm-specific, regular, and recurring demand needs to be

built up gradually as firms and the solar sector grow and alternative

performance incentives need to be in place and gradually adjusted.

Finally, demand-side incentives should be accompanied by supply-

side policies. For example, the IEA (2022) documented that the Chinese

government used several supply-side tools, such as grants, subsidies,

and low-cost loans, for more than a decade before it combined them

with demand-side policy tools, which it also linked to performance

requirements in cell manufacturing. These lessons seem crucial for the

success of new programs, such as new auctions linking component

manufacturing with electricity generation as used in Turkey and India,

and India’s recent production-linked incentive scheme (IEA,2022).

Otherwise, government support will remain a condition for India’s (and

other countries’) solar industry continuation.

CRediT authorship contribution statement

Florian Anselm Münch: Conceptualization, Methodology, Formal

analysis, Writing – original draft, Visualization, Validation. Fabian

Scheifele: Conceptualization, Methodology, Formal analysis, Writing

– original draft, Visualization, Validation.

Declaration of competing interest

The authors declare that they have no known competing finan-

cial interests or personal relationships that could have appeared to

influence the work reported in this paper.

Data availability

The code and data to replicate the results can be found here: https:

//doi.org/10.7910/DVN/UW06US .

Acknowledgments

We would like to thank Marco Caliendo, Benedict Probst and Beatriz

Couto Ribeiro for their valuable comments that helped us improve early

versions of this paper and Daniel Kral, Jannik Karl, Siwar Hakim and

Jonathan Muth for excellent research assistance.

Appendix

A.1. Tables

See Tables 5–12.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Table 5

Selection of variables used for PSM.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)

Indian HQ in Delhi pre-LCR patents pre-LCR solar patents pre-LCR sales pre-LCR employees Sector Electronics SOE Age Energy focus Manufacturer Subsidiary Phase 1 All Final

Indian 1.40** 1.39** 1.50** 1.25* 1.45** 1.35* 1.40* 1.33* 1.34* 1.39* 1.61** 1.58* 1.61* 1.80 1.85**

(0.66) (0.65) (0.65) (0.68) (0.69) (0.72) (0.75) (0.75) (0.73) (0.77) (0.75) (0.82) (0.83) (1.08) (0.84)

HQ in Delhi=1−0.08

(0.51)

Not solar patents, ihs transformed 0.07 −0.79

(0.18) (0.54)

Solar patents 2001–2010 1.24 4.27** 1.70*

(0.81) (2.05) (0.96)

ihs transf. pre-LCR sales 0.03 0.04 0.04

(0.03) (0.04) (0.03)

log_total_employees −0.05 −0.22 −0.28**

(0.11) (0.18) (0.15)

Industry 0.66

(1.25)

Construction −0.71

(1.02)

Business services −1.04

(1.10)

Electrical services, EPC −0.46

(0.86)

Electronics, component manufacturers 1.51

(0.98)

Utility 0.00

(.)

Sector 0.19

(0.25)

Electronics sector=1 1.97*** 1.96*** 1.85*** 1.99***

(0.61) (0.61) (0.64) (0.63)

Indian SOE=1 1.09 1.36

(0.95) (1.55)

Age 0.01 0.00

(0.01) (0.01)

Main business is energy=1−0.06 0.00

(0.49) (0.62)

Manufacturing company=1 1.99*** 1.65**

(0.59) (0.80)

Solar manufacturing=1 1.59*** 1.49** 1.58**

(0.58) (0.61) (0.68)

Subsidiary of mother company −0.00

(0.54)

Part 1 NSM=1 1.07 1.09 1.01

(1.13) (1.78) (1.16)

Subsidiary −0.26

(0.67)

Constant −2.04*** −0.87*** −2.07*** −2.22*** −2.49*** −1.83*** −1.81 −2.31*** −2.31*** −2.43*** −2.27*** −2.57*** −2.42** −2.49*** −3.26*** −2.06**

(0.62) (0.23) (0.59) (0.61) (0.65) (0.70) (1.11) (0.73) (0.73) (0.68) (0.87) (0.74) (1.21) (0.84) (1.21) (0.84)

Observations 113 113 113 113 113 112 110 113 113 113 113 113 113 113 112 112

Standard errors in parentheses

All estimates are based on a Logit model with robust standard errors in parentheses.

*𝑝 < 0.1, **𝑝 < 0.05, ***𝑝 < 0.01.

Table 6

Predicting participation in LCR auctions.

(1)

Participated (or not) in LCR auction

Participated (or not) in

LCR auction

log_total_employees −0.28**

(0.15)

ihs transf. pre-LCR sales 0.04

(0.03)

Solar patents 2001–2010 1.70*

(0.96)

Indian company 1.85**

(0.83)

Solar manufacturing 1.58**

(0.68)

Part 1 NSM 1.06

(1.13)

Constant −2.09**

(0.87)

Observations 112

Standard errors in parentheses

Estimates are based on a Logit model with robust standard errors in parentheses.

*𝑝 < 0.1, **𝑝 < 0.05, ***𝑝 < 0.01.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Table 7

Coefficients from staggered difference-in-differences estimation.

(1) (2) (3)

Solar patents Solar patents binary Revenue

Pre_avg 0.032 0.033 −0.605

(0.030) (0.031) (5.147)

Post_avg −0.225 −0.210 −19.452

(0.148) (0.138) (13.422)

T(−10) 0.402 0.201

(0.372) (0.186)

T(−9) −0.201

(0.186)

T(−8) 0.057 −0.003

(0.058) (0.004)

T(−7) 0.242 −0.013

(0.242) (0.014)

T(−6) −0.242 0.013

(0.242) (0.014)

T(−5) 0.035 −0.026

(0.060) (0.026)

T(−4) −0.035 0.026 22.020

(0.060) (0.026) (21.917)

T(−3) 0.120 1.245

(0.119) (4.771)

T(−2) −0.120 −17.114

(0.119) (12.086)

T(−1) 0.057 −8.570

(0.058) (8.688)

T(0) −0.323 −0.437 −1.375

(0.269) (0.352) (2.414)

T(+1) 0.320 −0.361 1.923

(0.310) (0.367) (5.770)

T(+2) −0.413 −0.114 −6.086

(0.255) (0.047) (4.375)

T(+3) −0.316 −0.194 −11.680

(0.223) (0.140) (7.662)

T(+4) −0.233 −0.067 −24.919*

(0.200) (0.084) (13.285)

T(+5) −0.525* −0.080* −40.269*

(0.300) (0.112) (22.296)

T(+6) −0.083 −0.021 −45.188

−28.027

(28.091)

T(+7) -0.083 -0.041 -45.188

(0.081) (0.041) (36.772)

Observations

Standard errors in parentheses

Column (1) and (2) include Indian origin dummy, Manufacturer dummy and pre-LCR revenues as

matching covariates.

Column (3) does not include any covariates.

All estimations include not-yet treated observations as controls.

*𝑝 < 0.1, **𝑝 < 0.05, ***𝑝 < 0.01.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Table 8

Robustness: Results with nearest neighbor matching.

1 nearest neighbor 2 nearest neighbors

(1) (2) (3) (4) (5) (6)

All firms Winner firms All w/o outliers All firms Winner firms All w/o outliers

Panel A: Solar PV Patents

Participated in LCR −1.33 −1.38 −0.97 −0.42 −0.44 −0.41

(1.64) (1.83) (1.15) (1.05) (0.99) (0.61)

Constant 2.15 1.88 1.31 1.24 0.94 0.75

(1.55) (1.80) (1.13) (0.91) (0.94) (0.57)

Observations 51 35 49 65 45 62

Panel B: PV Module & PV Cell Patents only

Participated in LCR −1.36 −0.96 −0.56 −0.61 −0.23 −0.09

(1.30) (1.57) (1.00) (0.77) (0.84) (0.54)

Constant 1.82 1.46 0.94 1.06 0.73 0.47

(1.28) (1.54) (0.97) (0.74) (0.78) (0.49)

Observations 51 35 49 65 45 62

Panel C: Post-LCR solar patent (binary)

Participated in LCR −0.12 −0.21 −0.22 −0.06 −0.02 −0.16

(0.17) (0.30) (0.21) (0.13) (0.18) (0.19)

Constant 0.27* 0.38 0.34* 0.21** 0.19 0.28*

(0.16) (0.29) (0.20) (0.11) (0.16) (0.18)

Observations 51 35 49 65 45 62

Panel D: Revenues (in INR)

Participated in LCR −8.92e+09 −1.78e+10 −2.19e+09 −9.50e+09 −1.42e+10 −9.57e+08

(1.04e+10) (1.58e+10) (4.32e+09) (1.04e+10) (1.41e+10) (3.51e+09)

Constant 3.84e+09 9.12e+09 6.66e+09* 4.43e+09 5.55e+09 5.43e+09*

(3.54e+09) (8.15e+09) (4.05e+09) (3.41e+09) (4.15e+09) (3.18e+09)

Observations 51 35 54 65 45 67

Robust Standard errors in parentheses

**𝑝 < 0.05, *𝑝 < 0.1.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Table 9

Adjusted for treatment intensity: Difference-in-differences combined with propensity score matching.

All firms Winner firms All w/o outliers

(1) (2) (3) (4) (5) (6) (7) (8)

Simple post difference DiD caliper =0.1 caliper =0.05 caliper =0.1 caliper =0.05 caliper =0.1 caliper =0.05

Panel A: Solar PV Patents

No. of times

participated in an LCR

auction

0.51* 0.21 −0.02 −0.01 −0.17 −0.41 −0.19 −0.16

(0.31) (0.20) (0.29) (0.29) (0.55) (0.39) (0.41) (0.37)

Constant 0.45 0.41** 0.88* 0.86* 1.14 0.69 0.72 0.55

(0.30) (0.20) (0.52) (0.52) (1.18) (0.66) (0.70) (0.57)

Observations 113 113 110 109 66 60 106 104

Panel B: PV Module & PV Cell Patents only

No. of times

participated in an LCR

auction

0.37* 0.18 −0.08 −0.08 −0.14 −0.33 −0.09 −0.14

(0.20) (0.15) (0.21) (0.21) (0.44) (0.33) (0.31) (0.27)

Constant 0.22 0.19 0.55 0.54 1.00 0.57 0.56 0.42

(0.20) (0.13) (0.37) (0.37) (0.97) (0.55) (0.58) (0.47)

Observations 113 113 110 109 66 60 106 104

Panel C: Post-LCR solar patent (binary)

No. of times

participated in an LCR

auction

0.07* 0.04 0.02 0.02 0.02 −0.09 0.01 0.00

(0.03) (0.06) (0.08) (0.08) (0.12) (0.07) (0.11) (0.11)

Constant 0.08** 0.04 0.14* 0.12 0.18 0.15 0.16 0.12

(0.03) (0.03) (0.08) (0.08) (0.20) (0.11) (0.13) (0.11)

Observations 113 113 110 109 66 60 106 104

Panel D: Revenues (in INR)

No. of times

participated in an LCR

auction

−9.88e+09 −9.88e+09 −4.38e+09 −3.32e+09 −1.08e+10 −1.00e+10 −5.28e+08 −1.55e+09

(7.00e+09) (7.52e+09) (3.44e+09) (2.89e+09) (8.83e+09) (6.48e+09) (1.86e+09) (2.19e+09)

Constant (6.84e+09) (6.94e+09) (6.41e+09) (5.34e+09) (7.94e+09) (1.05e+10) (3.08e+09) (3.82e+09)

Observations 113 113 110 109 66 60 108 104

Results in columns (1) & (2) use unmatched counterfactuals and columns (3)–(8) use matched counterfactuals based on the specified parameters.

Robust Standard errors in parentheses.

**𝑝 < 0.05, *𝑝 < 0.1.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Table 10

Probability of filing at least one solar patent.

Solar patent (binary) Solar patent (binary, only

post-2013)

(1) (2) (3) (4)

At least one solar patent At least one solar patent At least one solar patent

after LCR introduction

At least one solar patent

after LCR introduction

main

Participated (or not) in LCR auction −0.170 −0.348 −0.051 −0.240

(0.965) (1.006) (1.092) (1.152)

Indian company −2.055* −1.223 −2.582* −1.823

(1.146) (1.166) (1.343) (1.278)

ihs transf. pre-LCR sales 0.302*** 0.216*** 0.267*** 0.173***

(0.098) (0.057) (0.097) (0.047)

Age 0.016 0.006 0.023 0.012

(0.017) (0.013) (0.018) (0.014)

Pre-LCR employees 0.000* 0.000*** 0.000** 0.000***

(0.000) (0.000) (0.000) (0.000)

Real estate 0.000 0.000

(.) (.)

Industry −1.516 −2.108

(1.551) (1.758)

Construction 0.000 0.000

(.) (.)

Business services 0.000 0.000

(.) (.)

Electrical services, EPC 0.108 −0.284

(1.470) (1.533)

Electronics, component manufacturers 1.717 1.203

(1.442) (1.377)

Utility 0.000 0.000

(.) (.)

Manufacturing company 3.192*** 3.287***

(0.976) (1.181)

Observations 86 112 86 112

Standard errors in parentheses

Robust standard errors in parentheses.

Business services are baselevel of sector variable.

*𝑝 < 0.1, **𝑝 < 0.05, ***𝑝 < 0.01.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Table 11

Probability of winning an auction.

(1) (2) (3) (4)

LPM LPM Logit Probit

LCR auction 0.219*** 0.157** 0.155** 0.154**

(0.073) (0.075) (0.077) (0.076)

No. of bidders 0.002 0.002 0.001

(0.003) (0.003) (0.003)

Desired quantity in MW −0.000 −0.000 −0.000

(0.000) (0.000) (0.000)

Bid related to a solar park −0.357*** −0.325*** −0.329***

(0.059) (0.050) (0.050)

Age −0.001 −0.001 −0.001

(0.001) (0.001) (0.001)

Firm of Indian origin 0.032 0.032 0.031

(0.059) (0.058) (0.059)

Build-own-operate 0.000 0.000 0.000

(.) (.) (.)

Engineering, Procurement, Construction −0.386*** −0.309*** −0.316***

(0.100) (0.073) (0.071)

Observations 324 324 324 324

Robust standard errors in parentheses.

For Model (3) and (4) coefficients denote the marginal effects.

*𝑝 < 0.1, **𝑝 < 0.05, ***𝑝 < 0.01.

Table 12

Mean differences of key characteristics between firms that participate in LCR auctions and those who do not.

Variable (1) (2) T-test

did not partici- participated P-value

pate in LCR in LCR (1)–(2)

Mean/SD Mean/SD

Indian company 0.71 0.91 0.01**

(0.46) (0.29)

Filed patent before 2012 0.15 0.12 0.68

(0.36) (0.33)

Non-solar patents 1982–2011 7.96 48.73 0.25

(57.60) (198.73)

Solar patents 1982–2011 0.03 0.58 0.26

(0.16) (2.80)

Solar patents 2012–2021 0.40 1.38 0.21

(1.75) (4.46)

Non-solar patents 2012–2021 16.79 141.30 0.25

(117.29) (611.66)

Indian SOE 0.03 0.09 0.22

(0.16) (0.29)

Age 20.88 30.52 0.10*

(22.36) (30.53)

Main business is energy 0.60 0.64 0.72

(0.49) (0.49)

Manufacturing company 0.07 0.33 0.00***

(0.27) (0.48)

Solar manufacturing 0.07 0.24 0.04**

(0.19) (0.44)

Subsidiary of mother company 1.29 1.21 0.39

(0.46) (0.42)

N 80 33

Notes: The value displayed for t-tests are p-values. Standard deviations are robust. All missing values in

balance variables are treated as zero. ***, **, and * indicate significance at the 1, 5, and 10 percent

critical level.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

A.2. Figures

See Figs. 9–18.

Fig. 9. Data flowchart illustrating the construction and combination of the different data sources.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Fig. 10. Auctioned capacity, patent applications and bid price development.

Fig. 11. Auction participants by main sectors.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Fig. 12. Share of LCR auction wins among LCR firms.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Fig. 13. The firms that filed solar patents and their participation in LCR auctions.

Fig. 14. Solarpatents by main sectors.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Fig. 15. Solarpatents by main manufacturing status and auction type.

Fig. 16. Revenue development of auction-winning firms.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Fig. 17. Demand shock LCR vs. open auctions: MW won 2013–2017.

Fig. 18. Common support for the whole sample. Several bin width are reported for transparency.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

References

Aghion, P., Bloom, N., Blundell, R., Griffith, R., Howitt, P., 2005. Competition and

innovation: an inverted-U relationship. Q. J. Econ. 120 (2), 701–728. http://dx.

doi.org/10.1093/qje/120.2.701.

Altenburg, Tilman, Rodrik, Dani, 2017. Green Industrial Policy: Concept, Policies,

Country Experiences. UN Environment, United Nations Environment Programe,

Geneva.

Andreoni, Antonio, Chang, Ha-Joon, 2016. Industrial policy and the future of manufac-

turing. Econ. Polit. Ind. 43 (4), 491–502. http://dx.doi.org/10.1007/s40812-016-

0057-2.

Arrow, Kenneth J., 1962. The economic implications of learning by doing. Rev. Econom.

Stud. 29 (3), 155. http://dx.doi.org/10.2307/2295952.

Atkin, David, Khandelwal, Amit K., Osman, Adam, 2017. Exporting and firm perfor-

mance: Evidence from a randomized experiment*. Q. J. Econ. 132 (2), 551–615.

http://dx.doi.org/10.1093/qje/qjx002.

Austin, Peter C., 2011. Optimal caliper widths for propensity-score matching when

estimating differences in means and differences in proportions in observational

studies. Pharm. Statist. 10 (2), 150–161. http://dx.doi.org/10.1002/pst.433.

Bayer, Benjamin, 2018. Experience with auctions for wind power in Brazil. Renew.

Sustain. Energy Rev. 81, 2644–2658. http://dx.doi.org/10.1016/j.rser.2017.06.070.

Caliendo, Marco, Kopeinig, Sabine, 2008. Some practical guidance for the implementa-

tion of propensity score matching. J. Econ. Surv. 22 (1), 31–72. http://dx.doi.org/

10.1111/j.1467-6419.2007.00527.x.

Callaway, Brantly, Sant’Anna, Pedro H.C., 2021. Difference-in-differences with multiple

time periods. J. Econometrics 225 (2), 200–230. http://dx.doi.org/10.1016/j.

jeconom.2020.12.001.

Chang, Ha-Joon, 2003. Kicking away the ladder: Infant industry promotion in his-

torical perspective. Oxf. Dev. Stud. 31 (1), 21–32. http://dx.doi.org/10.1080/

1360081032000047168.

Chang, Ha-Joon, Andreoni, Antonio, 2020. Industrial policy in the 21st century. Dev.

Change 51 (2), 324–351. http://dx.doi.org/10.1111/dech.12570.

Clancy, Matt, 2021. An example of high returns to publicly funded R&D. What’s New

under Sun.

Cochran, William G., Rubin, Donald B., 1973. Controlling bias in observational studies:

A review. Sankhy¯

a 35 (4), 417–446.

Dehejia, Rajeev H., Wahba, Sadek, 2002. Propensity score-matching methods for

nonexperimental causal studies. Rev. Econ. Stat. 84 (1), 151–161. http://dx.doi.

org/10.1162/003465302317331982.

Del Río, Pablo, Kiefer, Christoph P., 2022. Which policy instruments promote innova-

tion in renewable electricity technologies? A critical review of the literature with

a focus on auctions. Energy Res. Soc. Sci. 89, 102501. http://dx.doi.org/10.1016/

j.erss.2022.102501.

Del Río, Pablo, Linares, Pedro, 2014. Back to the future? Rethinking auctions for

renewable electricity support. Renew. Sustain. Energy Rev. 35, 42–56. http://dx.

doi.org/10.1016/j.rser.2014.03.039.

Dobrotkova, Z., Surana, K., Audinet, P., 2018. The price of solar energy: Comparing

competitive auctions for utility-scale solar PV in developing countries. Energy

Policy 118, 133–148. http://dx.doi.org/10.1016/j.enpol.2018.03.036.

Edler, Jakob, Uyarra, Elvira, 2013. Public procurement of innovation. In: Hand-

book of Innovation in Public Services. https://www.elgaronline.com/view/edcoll/

9781849809740/9781849809740.00025.xml.

Edler, Jakob, Yeow, Jillian, 2016. Connecting demand and supply: The role of

intermediation in public procurement of innovation. Res. Policy 45 (2), 414–426.

http://dx.doi.org/10.1016/j.respol.2015.10.010.

Goodman-Bacon, Andrew, 2021. Difference-in-differences with variation in treatment

timing. J. Econometrics http://dx.doi.org/10.1016/j.jeconom.2021.03.014.

Griliches, Zvi, 1990. Patent Statistics as Economic Indicators: A Survey. National Bureau

of Economic Research, Cambridge, MA, http://dx.doi.org/10.3386/w3301.

Grossman, Gene M., 1981. The theory of domestic content protection and content

preference. Q. J. Econ. 96 (4), 583. http://dx.doi.org/10.2307/1880742.

Hall, Bronwyn H., Lerner, Josh, 2010. Chapter 14 - the financing of R&D and

innovation. In: Hall, Bronwyn H., Rosenberg, Nathan (Eds.), Handbook of the

Economics of Innovation : Handbook of the Economics of Innovation, Vol. 1.

North-Holland, pp. 609–639. http://dx.doi.org/10.1016/S0169-7218(10)01014-2,

Vol. 1.

Hansen, U.E., Nygaard, I., Morris, M., Robbins, G., 2020. The effects of local content

requirements in auction schemes for renewable energy in developing countries: A

literature review. Renew. Sustain. Energy Rev. 127, 109843. http://dx.doi.org/10.

1016/j.rser.2020.109843.

Hochberg, Michael, Poudineh, Rahmattalah, 2018. Renewable auction design in theory

and practice: lessons from the experiences of Brazil and Mexico. http://dx.doi.org/

10.26889/9781784671068.

Howell, Sabrina T., 2017. Financing innovation: Evidence from R&D grants. Amer.

Econ. Rev. 107 (4), 1136–1164. http://dx.doi.org/10.1257/aer.20150808.

Huber, Martin, Lechner, Michael, Wunsch, Conny, 2013. The performance of estimators

based on the propensity score. J. Econometrics 175 (1), 1–21. http://dx.doi.org/

10.1016/j.jeconom.2012.11.006.

Hufbauer, Gary Clyde, Schott, Jeffrey J., Cimino-Isaacs, Cathleen, 2013. Local Content

Requirements: A Global Problem. Columbia University Press.

IEA, 2021. Renewable energy market update- outlook for 2021 and 2022.

IEA, 2022. Special report.

Ioannidis, John P.A., Stanley, T.D., Doucouliagos, Hristos, 2017. The power of bias

in economics research. Econ. J. 127 (605), F236–F265. http://dx.doi.org/10.1111/

ecoj.12461.

IRENA, 2019. Renewable energy auctions: Status and trends beyond price.

IRENA, 2022. Renewable Power Generation Costs in 2021. Abu Dhabi.

Irwin, Douglas A., 2021. The rise and fall of import substitution. World Dev. 139,

105306. http://dx.doi.org/10.1016/j.worlddev.2020.105306.

Johnson, Oliver, 2016. Promoting green industrial development through local content

requirements: India’s national solar mission. Clim. Policy 16 (2), 178–195. http:

//dx.doi.org/10.1080/14693062.2014.992296.

Juhász, Réka, 2018. Temporary protection and technology adoption: Evidence from the

napoleonic blockade. Amer. Econ. Rev. 108 (11), 3339–3376. http://dx.doi.org/10.

1257/aer.20151730.

Krueger, Anne, 1990. Government Failures in Development. National Bureau of

Economic Research, Cambridge, MA, http://dx.doi.org/10.3386/w3340.

Krueger, Anne, 1997. Trade policy and economic development: How we learn. http:

//dx.doi.org/10.3386/w5896.

Kruger, Wikus, Eberhard, Anton, 2018. Renewable energy auctions in sub–Saharan

Africa: Comparing the South African, Ugandan, and Zambian Programs. Wiley

Interdiscip. Rev.: Energy Environ. 7 (4), e295. http://dx.doi.org/10.1002/wene.

295.

Kumar, Chitranjan, 2020. At 0.9% of revenues, firms spend dismal rs 36,000 crore on

R&D in FY20. Bus. Today.

Kuntze, Jan-Christoph, Moerenhout, Tom, 2012. Local content requirements and the

renewable energy industry - A good match? SSRN Electron. J. http://dx.doi.org/

10.2139/ssrn.2188607.

Lane, Nathaniel, 2020. The new empirics of industrial policy. J. Ind. Compet. Trade

20 (2), 209–234. http://dx.doi.org/10.1007/s10842-019-00323-2.

Lewis, Joanna I., Wiser, Ryan H., 2007. Fostering a renewable energy technology

industry: An international comparison of wind industry policy support mechanisms.

Energy Policy 35 (3), 1844–1857. http://dx.doi.org/10.1016/j.enpol.2006.06.005.

Luan, Chunjuan, Sun, Xiaoming, Wang, Yalan, 2021. Driving forces of solar energy

technology innovation and evolution. J. Clean. Prod. 287, 125019. http://dx.doi.

org/10.1016/j.jclepro.2020.125019.

Lucas, Robert E., 1993. Making a miracle. Econometrica 61 (2), 251. http://dx.doi.org/

10.2307/2951551.

Lunt, Mark, 2014. Selecting an appropriate caliper can be essential for achieving good

balance with propensity score matching. Am. J. Epidemiol. 179 (2), 226–235.

http://dx.doi.org/10.1093/aje/kwt212.

Matthäus, David, Schwenen, Sebastian, Wozabal, David, 2021. Renewable auctions:

Bidding for real options. European J. Oper. Res. 291 (3), 1091–1105. http://dx.

doi.org/10.1016/j.ejor.2020.09.047.

Mazzucato, Mariana, 2018. The Entrepreneurial State: Debunking Public Vs. Private

Sector Myths, Revised ed. PublicAffairs, New York, NY.

MnRE, 2009. Jawaharlal nehru national solar mission: Towards building solar India.

MnRE, 2022. Research, development and demonstration (Rd&D) in solar energy. https:

//mnre.gov.in/research-and-development/solar.

Münch, Florian Anselm, Marian, Adela, 2022. The design of technical requirements

in public solar auctions: Evidence from India. Renew. Sustain. Energy Rev. 154,

111713. http://dx.doi.org/10.1016/j.rser.2021.111713.

Munch, Jakob, Schaur, Georg, 2018. The effect of export promotion on firm-level

performance. Am. Econ. J.: Econ. Policy 10 (1), 357–387. http://dx.doi.org/10.

1257/pol.20150410.

OECD, 2015. Local-content requirements in the solar- and wind-energy global value

chains. http://dx.doi.org/10.1787/9789264227064-6-en.

Panagariya, Arvind, 2011. A re-examination of the infant industry argument for

protection. Margin: J. Appl. Econ. Res. 5 (1), 7–30. http://dx.doi.org/10.1177/

097380101000500102.

PIIE, 2021. Local content requirements threaten renewable energy uptake.

Probst, Benedict, Anatolitis, Vasilios, Kontoleon, Andreas, Anadón, Laura Díaz, 2020.

The short-term costs of local content requirements in the Indian solar auctions.

Nat. Energy 5 (11), 842–850. http://dx.doi.org/10.1038/s41560-020-0677-7.

Probst, Benedict, Touboul, Simon, Glachant, Matthieu, Dechezleprêtre, Antoine, 2021.

Global trends in the invention and diffusion of climate change mitigation tech-

nologies. Nat. Energy 6 (11), 1077–1086. http://dx.doi.org/10.1038/s41560-021-

00931-5.

Rodrik, Dani, 2008. Normalizing Industrial Policy. World Bank, Washington, DC.

Şahin, Mustafa Ergin, Okumuş, Halil İbrahim, 2016. Physical structure, electrical

design, mathematical modeling and simulation of solar cells and modules. Turk. J.

Electromech. Energy 1 (1).

Sahoo, Anshuman, Shrimali, Gireesh, 2013. The effectiveness of domestic content

criteria in India’s Solar Mission. Energy Policy 62, 1470–1480. http://dx.doi.org/

10.1016/j.enpol.2013.06.090.

Sant’Anna, Pedro H.C., Zhao, Jun, 2020. Doubly robust difference-in-differences estima-

tors. J. Econometrics 219 (1), 101–122. http://dx.doi.org/10.1016/j.jeconom.2020.

06.003.

Santoleri, Pietro, Mina, Andrea, Di Minin, Alberto, Martelli, Irene, 2020. The Causal

Effects of R&D Grants: Evidence from a Regression Discontinuity. LEM Working

Paper Series 18, http://dx.doi.org/10.2139/ssrn.3637867.

Energy Policy 179 (2023) 113574

F.A. Münch and F. Scheifele

Scheifele, Fabian, Bräuning, Moritz, Probst, Benedict, 2022. The impact of local content

requirements on the development of export competitiveness in solar and wind

technologies. Renew. Sustain. Energy Rev. 168, 112831.

Schumpeter, A., 1942. Capitalism, Socialism, and Democracy. Allen Unwin, London.

Shrimali, Gireesh, Konda, Charith, Farooquee, Arsalan Ali, 2016. Designing renewable

energy auctions for India: Managing risks to maximize deployment and cost-

effectiveness. Renew. Energy 97, 656–670. http://dx.doi.org/10.1016/j.renene.

2016.05.079.

Shrimali, Gireesh, Sahoo, Anshuman, 2014. Has India’s Solar Mission increased the

deployment of domestically produced solar modules? Energy Policy 69, 501–509.

Shrimali, Gireesh, Srinivasan, Sandhya, Goel, Shobhit, Nelson, David, 2017. The

effectiveness of federal renewable policies in India. Renew. Sustain. Energy Rev.

70, 538–550. http://dx.doi.org/10.1016/j.rser.2016.10.075.

Shubbak, Mahmood H., 2019. Advances in solar photovoltaics: Technology review

and patent trends. Renew. Sustain. Energy Rev. 115, 109383. http://dx.doi.org/

10.1016/j.rser.2019.109383.

Singh, Ayush Kumar, Pandey, Kheelraj, 2021. A review to the progress of solar utility

scale and solar thermal power in India—Policies, barriers and the way forward.

J. Inst. Eng. (India): Ser. C 102 (2), 525–543. http://dx.doi.org/10.1007/s40032-

021-00664-0.

Veloso, Francisco, 2001. Local Content Requirements and Industrial Development :

Economic Analysis and Cost Modeling of the Automotive Supply Chain (Ph.D.

dissertation). Massachusetts Institute of Technology.

Wade, Robert H., 2018. The developmental state: Dead or alive? Dev. Change 49 (2),

518–546. http://dx.doi.org/10.1111/dech.12381.

Winkler, Jenny, Magosch, Magdalena, Ragwitz, Mario, 2018. Effectiveness and ef-

ficiency of auctions for supporting renewable electricity – What can we learn

from recent experiences? Renew. Energy 119, 473–489. http://dx.doi.org/10.1016/

j.renene.2017.09.071.

World Trade Organization, 2018. India – certain measures relating to solar cells and

solar modules.