Firm Performance Under Financial Constraints: Evidence from Sub-Saharan African Firms

Business environment in which a rm operates has an important impact on rm performance. This study examined the impact of credit constraint and power outages on the rm’s investment decision using World Bank Enterprise Survey data collected from rms operating in 13 SSA countries. The study employed a two-part model and the Heckman selection model to estimate the impact of lack of access to nance and poor power supply on a rm’s decision to invest in self-generation. The result obtained suggest that there is a negative correlation between credit constraint and a rm’s decision to invest in self-generation. This indicates that credit constraint negatively affects a rm’s decision to invest in self-generation and rms that are credit constrained have less incentive to invest in self-generation compared to those that are not credit constrained. To test the robustness of the result obtained, alternative denitions of credit constraints were used. Results from alternative regressions using different denitions of credit constraints show that credit constraint affects a rm’s decision to invest in self-generation but not the volume of investment.


Introduction
It was established in the literature that business environment-encompassing features of legal and regulatory services, infrastructure, nancial and institutional systems of the country affect rm performance and the entrepreneur's willingness to invest (Abdisa, 2019). According to Essmui et al.
(2014), a good business environment makes a country an attractive destination for foreign investment and a place in which domestic entrepreneurs of all sizes and across industries are willing to invest. Crosscountry empirical studies also show that strong evidence that the underdeveloped business environment is associated with a poor investment, employment, and economic growths (Escribano et  Firm performance is strongly linked to the availability and access to nance, which is a main component of the business environment in which rms operate. Empirical studies showed that the degree to which rms face nancial constraints mainly depends on rm size-small rms face bigger challenges in obtaining nance as compared to larger rms (Schiffer and Weder, 2001;Beck et al., 2002;Abdisa, 2018). This further magni es the relative impact of the nancial constraints on the rm's investment decision. In this regard, a study by OECD (2006) documented that access to nance allows rms to expand their business activities and grow faster.
However, the problem of nancial constraint and its effect on rm performance signi cantly varies across regions and countries. For example, Fowowe (2017) showed that nancial constraint is the main problem for African countries than in other developing countries, posing a signi cant challenge to rm growth and further investment decisions. The author, based on the survey data of 26 African countries, found that lack of access to nance was a major constraint among rms operating in SSA countries. The author also noted that within SAA rms, those that have better access to nance have better growth experience, growth being measured by the number of permanent full-time workers. In this regard, about 60% of the sample rms used in this study are reported to be nancially constrained, suggesting that nancial constraint is the main obstacle to rm performance in SSA countries.
In addition to lack of access to nance, the poor power supply is also the main obstacle to rms' doing business in SSA. The World Bank Enterprise Survey (WBES) report in 2007 shows that the average Sub-Saharan African rm suffered a loss of economic activities for around 77 hours per month due to power outages. The situation is even worse in some countries and particularly when compared with other developing regions of the world. The WBES report relating to 2010/2011 shows that about 22% of business managers consider electricity as the most serious obstacle to doing their business (Word Bank, 2015). Many empirical studies have been devoted to examining the impact of poor supply on rm performance and the strategies that rms adopt to cope with the poor power supply (Steinbuks and Abdisa, 2018 and Abdisa, 2020). In this regard, empirical studies by (Beenstock et al., 1997, Oseni and Pollitt, 2015and Abdisa, 2020 found that rms that invested in self-generation continue to face higher unmitigated loss which shows that rms make only partial investments which cannot fully backup back up their electricity load. Our contribution complements the above empirical evidence. Speci cally, the study provides answer to the question "why do rms that invested in self-generation continue to face outage loss?" However, unlike the studies cited above, our study contributes to the existing literature in three important ways. First, the existing empirical investigation by Beenstock et al. (1997); Oseni and Pollitt (2015) and Abdisa (2020) predicted that investment in self-generation of electricity does not guarantee complete mitigation of power outages and a rm that invested in self-generation may continue to face outage loss. However, it is not clear from these studies that why do rms that invested in self-generation continuous to face outage loss? Second, we deviate from many existing literature by exploring factors behind the rm's sub-optimal investment in self-generation using rm-level data for SSA countries and hence we offer new insights in understanding the performances of rms operating in SSA countries. Finally, examining the impact of access to nance and power outages pose a signi cant identi cation challenge due to the potential reverse causality bias, as rms with poor investment opportunities are expected to have a higher probability of being credit constrained (Fowowe, 2017). To tackle this challenge, several identi cation strategies were employed in this study using the two-part model and Heckman selection model (1979).
In nutshell, we explored the joint effect of the lack of access to nance and the poor supply of electricity on a rm's incentive to invest in self-generation. The result obtained suggest that there is a negative correlation between credit constraint and a rm's decision to invest in self-generation. This indicates that rms that are credit constrained have less incentive to invest in self-generation compared to those that are not credit constrained. Results from alternative regressions using different de nitions of credit constraints show that credit constraint negatively affects a rm's decision to invest in self-generation. In particular, credit constraint affects a rm's decision to invest but not the volume of investment. This shows result obtained is insensitive to the alternative de nitions of credit-constrained used indicating the robustness of the result obtained Page 4/20 The remaining part of the paper is organized as follows: Data source and descriptions, estimation strategies, and the empirical models are discussed in Sect. 2. Section 3 presents empirical results, while conclusions and policy implications drawn from the study are presented in Sect. 4.

Data
The study employed the World Bank Enterprise Survey (WBES) which is collected from business enterprises operating in 13 SSA countries. The WBES was collected from manufacturing and service in every region of the world including SSA countries. Even though the WBES covers different themes related business environment, the data utilized in this study relates to rms' perceptions related to doing their business, the relative signi cance of various constraints to rms' business operations which are mainly under the infrastructure and services theme of the survey.
The WBES provides an array of economic data on more than 140,000 rms in more than 141 countries worldwide. The data used in this study is, however, restricted to selected rms operating in 13 SSA countries .These countries were selected based on the number of rms included in the survey and the year the survey was conducted. Accordingly, this study considered only countries for which the survey was conducted after the year 2010 and countries for which data on at least 100 rms are available after cleaning for missing information.
Combining rm data for 13 SSA countries selected for this study yields 5129 observations. However, data analysis was made with 3594 observations after cleaning the dataset for missing values and outliers.
The main advantage of using the WBES is that the survey uses standardized survey instruments and the same sampling methodologies across countries. This minimizes measurement error and yields data that are comparable across different economies. This is important to capture cross-country variation in the business climate and its impact on rm performance.

Variables and Descriptive Statistics
Alternative de nitions of credit constraints are used and discussed in this section.

a) Perception approach
In the perception approach to credit constraint, rms are asked to rate the degree to which lack of access to nance is an obstacle to doing their business (Beck and Demirguc-Kunt, 2006;Asiedu et al., 2013). In the WBES, rms are given a categorized choice from no obstacle to a very severe obstacle. Following the approach in Hansen and Rand (2014) and Asiedu et al (2013), two versions of credit constraint variables are constructed from a rm's response to this question. The rst is a categorical variable-constraintwhich takes a value ranging from 0 to 4 in which higher value implies that the rm is more credit constraint. The second is a dummy variable-constraint a -which equals 1 if the rm has reported access to nance is a moderate, major and very severe constraint to doing its business and zero otherwise (details are reported in Table A.1 in the annex). The variable constraint is the rm's response to the question "to what degree lack of access to nance is an obstacle to doing your business". This a categorical variable taking a value ranging from 0 to 4. The variable "constraint_a" is dummy variable version of the variable "constraint" in which rms are classi ed as credit constrained if they have responded to the above question as a moderate, major, and severe constraint. While variables constraint_1 and constraint_2 are the alternative de nitions of credit constraint de ned in alternative b and c, respectively.

b) Credit application information
Based on the credit application information, rms are classi ed as credit constrained or not based on whether they have applied for a loan and the stated reasons for not applying. In the spirit of Bigsten et al (2003), and Hansen and Rand (2014), a rm is classi ed as credit constrained-constraint 1 -if: (i) the rm has applied for a loan and was denied, (ii) did not apply for a loan due to reasons such as ''application procedures was complex'', ''collateral requirements were too high'', or ''possible loan size and maturity were insu cient''. If a rm did not apply for a loan because it does not need one or applied for a loan and were approved, the rm is classi ed as unconstrained (see Table A.2 for details).

c) Use of nancial services
Some studies (Aterido et al, 2013;Muravyev et al, 2009) use the rm's use of formal nancial services as an indicator of credit constraint. According to this approach, rms that use formal nancial services are classi ed as credit unconstrained while rms that do not use formal nancial intuitions are classi ed as credit constrained. Following the same logic, this study also classi es rms that use formal nancial institutions as credit unconstrained and others as credit constrained. Table (2) classi es rms in the sample as credit constraint or not according to the three de nitions the credit constraint given above. Using the rst and third de nitions, about 59% of rms are credit constrained while 47% of rms are credit-constrained based on the direct credit application information.
The credit application information criterion resulted in a relatively less percentage of credit constrained rms compared to the other two.
The classi cation of rms as credit-constrained and unconstrained by rm size shows that a relatively higher percentage of large rms are credit unconstrained while a large share of small rms were found to be credit constrained. This shows that large rms are more likely to have access to external funds to nance their operations and hence less credit constrained than small rms. Outage time ( lnH ) The variable outage time utilized in the study is computed from the reported frequency and duration of power interruptions that a rm faces in a month. A monthly outage time is obtained by multiplying the frequency of power outages with its duration and then it is converted into yearly data assuming the same outage frequencies and duration throughout the year. The outage time-the number of days a rm is without power supply from the public grid-also measures the reliability of power supply.
Furthermore, a correlation between different de nitions of credit constraint and the rm's decision to invest in self-generation is examined and the result is reported in Table 3. The correlation matrix shows a meaningful result in which all measures of credit constraint are negatively correlated with both rm's decision to invest and the volume of investment a rm wishes to invest. On the other hand, a power outage is positively correlated with both rm's decision to invest and volume of investment which implies that unreliable power supply induces rms to invest in private substitutes. Moreover, the Table shows that positive and signi cant correlation between the alternative de nitions of credit constraints which imply the consistency of the alternative measures of credit constraint used. Constraintis the perception approach to credit constraint de nition and takes value from 0 to 4 with higher value implies more credit constraint, constraint a is the binary version of the variable "Constraint" and takes the value of one if a rm reported access to nance is moderate, major and severe constraints to doing business.Constraint 1 is the credit application information de nition of credit constraint and takes 1 if the rm is credit-constrained and 0 otherwise. Outages are the total power interruption in days a rm faces in a year.

Model speci cation
The methodology used in this paper is based on a theoretical model of a rm's investment decision by Abdisa (2020) where a similar approach was used in estimating the rm's investment decision. According to the approach in Abdisa (2020), all costs of investment in self-generation are weighted against the expected future bene ts. This is based on the Net Present Value (NPV) approach to investment decisions and a rm undertakes an investment with a positive NPV.
In order to examine the role of access to nance in a rm's investment decision, we included nancial constraints in the cost component of the rm's NPV computation. The implication is that a high nancial barrier increases a rm's borrowing cost which worsens the NPV of the investment. Based on the NPV of the investment, a rm decides whether to invest in self-generation; and how much to invest. The rst question is a binary outcome which can be modeled by a standard probit model. The second question is the volume of investment which is left-censored at zero. To address this, two-part and Heckman selection models are employed. More formally, the models are stated below.
A rm invests in self-generation if the NPV of the investment is positive. However, we observe whether the rm has invested in self-generation or not. Assuming unobserved latent variable y * that establishes the following linear relationship between the relevant variables. y * = αx i + u i (2) where x i is a vector of explanatory variables, α is the associated parameters to be estimated, u i is a normally distributed error term with mean zero and variance σ ui 2 . The observed variable y, is related to the latent variable y * as follows: Determinants of a rm's incentive to invest in self-generation are estimated by probit model as indicated above. In the second part, linear regression model is used only for estimating a positive value. Thus, the two-part model for y i following the approach stated in Cameron and Trivedi (2005) is given by: Where y denotes the volume of investment, d is a binary indicator such that d = 1 if y > 0 and d = 0 if y = 0. When y = 0 we observe only Pr(d = 0). For those with y > 0, let f(y ⁄ d = 1) be the conditional density of y.
The two-part model has some exibility and computational simplicity by assuming that the two partsthe decision to invest and the volume of investment-are independent. But rms with positive investments are not randomly selected from the population. This may result in second stage regression to suffer from selection bias (Cameron and Trivedi (2005). To allow for the possible dependency between the equations, the selection model of Heckman (1979) is also used.
The main interest in equations (5a) and (5b) is to identify the causal effect of credit constraints on investment decisions. However, there is a potential reverse causality in the model because rms with poor investment opportunities are more likely to be credit constrained. Following the approach in (Petersen and Rajan, 1994 and Garcia-Posada, M. 2018), we implemented different strategies to tackle this identi cation challenge. First, traditional determinants of rm investment opportunities such as rm size and rm age are included as control variables. Second, country and industry dummies are included to control for the country and industry speci c investment opportunities. Third, the perceived nancial obstacles, rather than actual nancing constraints are used as an alternative de nition of credit constraint as a robustness check for the result obtained.
However, including these variables may not perfectly control for a rm's investment opportunities. Thus, as a nal strategy to tackle the potential reverse causality in the model, the study uses an instrumental variable to isolate the exogenous part of credit constraints. Following the logic of Beck and Demirguc-Kunt (2006) and Fowowe (2017), banking regulatory and supervisory structure are used as IV for the credit constraint variable in this study . Speci cally, the average tenure of bank supervisors and an index of overall supervisory independence from both banks and politicians are used as an instrument for credit constraint. It is expected that bank regulation and supervision will in uence a rm's access to nance but do not have a direct impact on rm performance.

Credit Constraint and Investment in Self-generation
The effect of credit constraint and a power outage on a rm's investment decision is reported in Table 4. The Table summarizes the results estimated by the two-part model and the Heckman selection model. In both speci cations, the decision to invest is estimated by the probit model. The coe cient estimates of the two-part model are reported in the rst column of Table 4. As can be seen from the Table,  The coe cient of ρ, which measures a correlation between the error terms in the two equations, is signi cant. Furthermore, the likelihood ratio test also rejects the hypothesis that the correlation between the error terms in the selection and outcome equations are not signi cantly different from zero. This shows that the two equations are not independent and there is evidence of sample selection. The discussion of the result is thus, based on the Heckman selection model and the two-part model is presented here as a robustness check to the result obtained.
In the Heckman selection model, there should be at least one variable in the selection equation which is not included in the outcome equation for a robust identi cation. In this study, a set of industry dummies are included only in the selection equation. The assumed hypothesis is that industry dummies affect the decision to invest in self-generation but not the volume of investment. This is mainly due to the fact that some industries need a continuous supply of electricity in which they are more willing to invest in selfgeneration than in other industries.
The coe cient of outage time is positive and signi cant both in the selection and outcome equations.
This shows higher outage time increases a rm's propensity to invest in self-generation and the volume of investment. The theoretical model used in this study shows that the effect of outage time on a rm's decision to invest in self-generation depends on the rm's degree of vulnerability to a power outage and the expected productivity of the installed generator. If the expected return from investing in selfgeneration is less than the rm's vulnerability to a power outage (outage loss), the rm has no incentive to invest in self-generation and vice-versa. The result obtained shows that the coe cient of outage time is positive and signi cant indicating that the return to a rm from the investment outweighs the cost of doing so.
The variable constraint a is negative both in selection and outcome equations. However, it is signi cant only in the selection equation. The result obtained suggests that credit constraints affect a rm's decision to invest in self-generation negatively. This indicates that a rm that is credit constrained is less likely to invest in self-generation compared to rms that are not credit constrained. Even though it is not signi cant in the outcome equations, a sign of the variable is maintained indicating that being credit constrained discourages a rm's investment in self-generation. This is in line with the theoretical prediction in which rms that are credit constrained are those that do not have easy access to external nance. This, on the other hand, increases rms' borrowing costs and worsens rms' Net Present Value (NPV) which eventually discourages rms' incentive to invest.
The coe cients of size dummies are signi cant, and it is positive for large rms. This indicates that large rms are more likely to invest in self-generation compared to medium rms (base category) while small rms are less likely to invest in self-generation compared to medium rms. This could re ect rms' ability to nance investment in self-generation. Larger rms are more likely to have access to external funds to nance their operations, including self-generation, and hence less credit constrained. This adds to the result obtained in descriptive statistics reported in Table 2 and the ndings of (Abdisa, 2018 and Steinbuks, 2010).

Robustness checks
To test the robustness of the result obtained, alternative de nitions of credit constraint are used, and the result is reported in Table 5 and Table 6. In Table 5, the credit application information is used to classify rms as credit constrained or credit unconstrained. The coe cient estimate of credit constraint is negative and signi cant in the Heckman model while it is negative but insigni cant in the two-part model. In Table 6, a categorical variable generated from the rm's response to the question 'do credit constraint is an obstacle to the operation of your establishment' is utilized. The result indicates that rms that perceived lack of access to nance as a major constraint to their operation are less likely to invest in a self-generation compared to rms that perceived lack of access to nance is only a minor obstacle to their operation. In all speci cations, a lack of access to nance is found to affect a rm's investment decision, not the amount of investment to be made.
Needless to say, results from alternative regressions show that credit constraint affects a rm's decision to invest in self-generation. In particular, credit constraint affects a rm's decision to invest but not the volume of investment. The result is insensitive to the alternative de nitions of credit-constrained used indicating the robustness of the result obtained. Compared to the result reported in Table 3, the same estimation strategy is followed except the alternative de nition of credit constraint is used. The variable credit constraint is the rm's response to a question that "does lack of access to nance is an obstacle to operation of your establishment?". The response is classi ed as minor, moderate, and major obstacle. The minor obstacle is the base category in the estimation

Instrument variable
So far, the identi cation strategy has relied on the extensive use of country-industry dummies and rmlevel covariates to control for rms' investment opportunities. In addition, the alternative de nitions of credit constraints are used and the result obtained indicates that rms that are credit constrained are less likely to make an investment in self-generation compared to rms that are credit unconstrained under all speci cations. However, if investment opportunities are not perfectly controlled, then the error term will be correlated with the credit constraint variable which leads to potential reverse-causality bias. Hence, in robustness, an instrumental variable is used to tackle the potential reverse causality bias in the model.
The result of an instrumental variable estimation is reported in Table 7. In the rst stage, the credit constraint variable is regressed on a set of rm control variables, industry dummies, and the instruments. This is estimated by a linear probability model. The rst stage statistics are reported in the last rows of the Table and indicate that the instruments are strong predictors of rm credit constraint. The credit constraint variable in equations 5a and 5b are replaced by the predicted residual (ivresid) from the rst stage regression. Replacing credit constraint by the predicted residual from the rst stage regression, the model in equations 5a and 5b are estimated by the Heckman and the two-part models.
The result is in line with the results obtained previously and con rms the previous ndings that rms that have di culty in obtaining credit access are less likely to invest in self-generation compared to rms that are credit-unconstrained. Like the result obtained earlier, the credit constraint variable negatively affects a rm's decision to invest in self-generation in both Heckman and two-part model.

Conclusion And Policy Implications
The study examined the impact of credit constraint and power outages on the rm's investment decision using WBES data collected from rms operating in 13 SSA countries. The study employed a two-part model and Heckman selection model to estimate the impact of lack of access to nance and poor power supply on a rm's decision to invest in self-generation.
The result obtained suggest that there is a negative correlation between credit constraint and a rm's decision to invest in self-generation. This indicates that rms that are credit constrained have less incentive to invest in self-generation compared to those that are not credit constrained. The effect of outage time is found to be positive under all alternative speci cations indicating that a poor supply of electricity induces rms to invest in self-generation. However, rms are constrained by a lack of access to nance to fully backup their electricity load. This implies that rms that invested in self-generation continuous face outage loss.
To test the robustness of the result obtained, alternative de nitions of credit constraints were used.
Results from alternative regressions using different de nitions of credit constraints show that credit constraint affects a rm's decision to invest in self-generation. In particular, credit constraint affects a rm's decision to invest but not the volume of investment. This shows the result obtained is insensitive to the alternative de nitions of credit-constrained used indicating the robustness of the result obtained. To control potential reverse causality bias that arises from a two-way causality between investment opportunities and credit constraints, the study implemented different strategies. These include controlling for traditional determinants of rm investment opportunities such as age and rm size. Furthermore, country and industry dummies were included to control for country and industry speci c investment opportunities and the perceived nancial obstacles, rather than actual nancing constraints are used as an alternative de nition of credit constraint as a robustness check for the result obtained. As a nal strategy to tackle the potential reverse causality in the model, the study used an instrumental variable to isolate the exogenous part of the credit constraints. The results from alternative speci cation and IV estimation is in line with the results obtained from the two-part and Heckman selection models con rming the ndings that rms that have di culty in obtaining credit access are less likely to invest in self-generation compared to rms that are credit unconstrained.
The result of the study implies that for rms to improve their performance, they should overcome the credit constraints. This, however, poses an important challenge for the governments of the SSA countries.
That means, governments and nancial institutions in African countries should make concrete efforts needed to be undertaken to overcome constraints in obtaining nance and boost access to nancial services for rms. This is mainly important for SSA countries as rms are assumed to play a key role in economic growth, employment creation and hence poverty reduction. Thus, in order to solve the problem, it is quite important to approach the problem from both demand and supply side dimensions. On the demand side, the interaction of rms and nancial institutions should be improved. For example, the data used in this study shows that about 42% of rms reported that they did not apply for loan, but are nancially constrained, because of complex nancial procedure in getting the loan such as high/unfavorable interest rate, collateral requirements and small loan size offered by these nancial institutions. Thus, the government should work with the nancial institutions to ease rms' nancial constraints. On the supply side, the government and rms should work closely to gure out the nature of nancial systems in SSA countries and how demand could meet given the supply. In this regard, the survey data used in this study shows that about 2% of rms that are nancially constrained due to the amount loan of offered to them is less than the amount demanded by rms.

Declarations
Compliance with Ethical Standards Data Availability Statement Data used in this manuscript will be available up on request.
Funding: No funding was received for this paper Con ict of interest: Authors declare no con ict of interest.
Ethical approval: This article does not contain any studies with human participants or animals performed by any of the authors.
Authors' Contribution: First Author worked on data analysis and estimation, interpretation of ndings while the second author worked on review of relevant literatures and methodological sections.