Does it pay to be green? Environmental performance and �rm �nancing during COVID-19 outbreaks

This paper examines the effect of �rm environmental performance on �rm �nancing during the COVID-19 outbreak. Crises in multiple forms curtail Micro, Small and Medium Enterprises (MSMEs) stability and the livelihood of hundreds of millions of people who derive their living from these activities. The way in which MSMEs deal with crises and the extent to which environmental performance is bene�cial when the market suffers a negative shock is relatively unexplored in the literature. We consider three aspects of �nancing -- �rm level liquidity, bank credit and bankruptcy probabilities -- and argue that it pays for �rms to show commitment to environmental responsibilities in a global pandemic. Through an examination of 3,356 MSMEs, we �nd that �rms with better environmental performance reduce their probability of bankruptcies and their liquidities decreasing during the COVID-19 pandemic. Furthermore, analysis shows that the impact of a �rm’s environmental performance is more pronounced in sensitive industries (hospitality and retail). The results are robust based on a series of robustness checks, including propensity score matching and the Heckman two-stage sample selection model. Our study suggests that the trust between a �rm and its stakeholders, if it is grounded on environmental performance, pays off when the overall level of trust in markets suffers a negative shock


Introduction
The coronavirus (COVID-19) outbreak was declared a global pandemic by the World Health Organisation (WHO) on March 11, 2020.Across the world, the number of con rmed cases exceeded 17 million, and the number of deaths exceeded 186 thousand people as of July 04, 2020. [1]To prevent the spread of the virus, several countries have taken strict measures which include regional and national lockdown.This has slowed down the economies and resulted in severe crises, such as a drastic decline in production and demand, cash ow problems, unemployment, and bankruptcies in most industries across the world (Papadopoulos et al. 2020).
Various countries have witnessed a signi cant decline in their gross domestic product (GDP); for example, the GDP of Eurozone has declined by 3.8% and the GDP of the United States has declined by 1.2% in the rst quarter. [2] The impact of the COVID-19 pandemic on Micro, Small and Medium Enterprises (MSMEs) [3] is more severe than on larger companies (Bartik et al. 2020).In particular, MSMEs in the tourism and hospitality industries are severely affected. [4]Recent studies provide evidence that 50% of some small rms have temporarily stopped operating and 60% of MSMEs are at risk of running out of cash reserves across the world (Bartik et al. 2020;Cowling et al. 2018).The International Trade Centre conducted a survey of MSMEs in 132 countries and found that the pandemic has strongly affected two-thirds of businesses, and one-fth of businesses are likely to shut down in the next three months (International Trade Centre 2020).In the United States alone, the pandemic has led to the closure of 43% of MSMEs (Bartik et al. 2020).
Even in normal economic conditions, banks tend to consider MSMEs as high-risk borrowers because MSMEs are more prone to liquidity shortfalls and bankruptcies (Berger and Udell 1998).MSMEs are more inclined to asymmetric information and agency problems, as they often lack information and credit history (Beck and Demirguc-Kunt 2006).Weak property rights and ine cient insolvency practices further increase the risk of lending to MSMEs.Relationship banking prevents type II errors in MSME lending and in evaluating the creditworthiness of rms (Baas and Schrooten 2006).However, shocks such as the COVID-19 pandemic can severely disrupt this form of relationship lending (Lerner et al. 2020).The literature on the impact of COVID-19 on MSMEs is still sparse (Brown and Rocha 2020), although there is evidence to suggest that access to bank nance is more problematic for small rms during a global nancial crisis (Demirguc-Kunt et al. 2020; N. Lee et al. 2015).
The environmental performance of rms has been a subject of interest for researchers and policymakers (Banerjee et al. 2019;Bragdon and Marlin 1972;Muhammad et al. 2015b;Porter and Van der Linde 1995;Williamson et al. 2006).Proactive environmental practices are exposed to a low degree of risk (Godfrey et al. 2009;Muhammad et al. 2015a), low cost of debt (Bauer and Hann 2010), easier access to the nancial market (Jo and Na 2012), higher leverage (Sharfman and Fernando 2008) and better conditions on loans (Goss and Roberts 2011;Magnanelli and Izzo 2017).Nevertheless, some studies report that the nancial risk incurred by socially and environmentally responsible rms is higher than other rms (Kiernan 2007;Seeger and Hipfel 2007).
Overall, empirical studies relating to environmental performance and rm nancing appear inconclusive (D. Lee and Faff 2009) and most of these investigate the relationship within normal economic conditions.How environmental performance impacts on rm nancing when the market suffers a negative shock is relatively unexplored in the literature [5] .By examining SMEs in seven countries in the context of the COVID-19 pandemic, our study aims to bridge this knowledge gap and focuses on three different aspects of nancing: liquidity; access to bank nancing; and bankruptcy.
From a stakeholder perspective, theoretical identi cation of the effect of environmental performance on rm nancing during the COVID-19 pandemic can be explained in terms of risk mitigation.Higher environmental performance by rms may decrease the likelihood of negative events at the rm level (Bouslah et al. 2018) and at the economic level (i.e., during nancial crises and economic recessions) because such rms with established reputations as environmentally responsible are able to gain access to crucial resources (Branco and Rodrigues 2006;Zeidan et al. 2015).Moreover, rms with high environmental performance are regarded as more trustworthy by stakeholders who are likely to place a valuation premium on such rms even if overall institutional and market trust is low (Lins et al. 2019).Thus, relationship-based intangible assets (i.e.trust and customer loyalty) gained from corporate investment into environmental activities are particularly valuable in negative economic conditions (Godfrey 2005).It was observed that, during the COVID-19 pandemic, rms experienced liquidity decreases regardless of their size (Juergensen et al. 2020).Our study sets out to ascertain whether the environmental performance of rms affect the rates of decline in the liquidity of rms.If nancial institutions do not recognise environmentally responsible practices as a means for rms to lower their level of idiosyncratic risk, then rms may suffer from a competitive disadvantage because investments into environmentally responsible practices lead to costly diversions of the rm's resources (Goss and Roberts 2011).
Our ndings suggest that it pays for rms to be green.In particular, the probit marginal effects show that one unit increase in environmental performance index decreases the probability of liquidity shortfall by 0.0172 and bankruptcy by 0.0077 during the pandemic.Furthermore, this impact of environmental performance on rm nancing is more pronounced in rms from the hospitality and retail industries.The results are robust based on a series of robustness checks, including propensity score matching and the Heckman two-stage sample selection model.Our study proposes that the trust between a rm and its stakeholders, if it is grounded on environmental performance, holds rm when the overall level of trust in markets suffers a negative shock.
This study makes signi cant contributions to the literature in several respects.First, to the best of our knowledge, this is the rst study to empirically investigate the impact of environmental performance on rm nancing (liquidity, access to bank nancing, and bankruptcy) during the COVID-19 pandemic.
Empirical research on the effects of crisis events for MSMEs is scant, despite the fact that MSMEs tend to be the rms most deprived in a crisis event (Doshi et al., 2018).The limited research that exists on MSME nancing in crisis periods (Brown and Rocha 2020;Cowling et al. 2018) has not touched on the environmental performance of rms in terms of MSME resilience and as a crisis management technique, which is what our study does.
Our paper is one of the few to test how environmental performance activities impact on MSME nancing.The majority of research investigating this topic have examined rms in a developed region (Goss and Roberts 2011;Muhammad et al. 2015aMuhammad et al. , 2015b) ) [6] , whereas our study sample dominated by rms in developing economies, which have weak institutional environments.Obviously, nancing is a challenge for rms of all sizes.However, in a context of weak institutional structures and high levels of information opaqueness (Berger and Udell 1998), MSMEs in weak institutional environment face more constraints in relation to liquidity and bank nancing (Baas and Schrooten 2006;Coricelli and Frigerio 2019).Our multi-national study provides valuable policy recommendations for the developing economies investigated and, potentially, for other countries as well.
Third, we use a multidimensional construct of a proxy for rm environmental performance.This proxy provides a broader perspective of environmental activities in unlisted rms.This differs from earlier works measuring environmental performance via Kinder, Lydenberg, and Domini's (KLD's) index (Goss and Roberts 2011;Nandy and Lodh 2012) or the ASEET4 index (Cheng et al. 2014;Cheung 2016), both of which use only one variable of corporate environmental performance.Given that environmental performance is multidimensional (Dragomir 2012), studies using these approaches, we believe, fail to capture the full picture In contrast, our 10-factor environmental performance index considers many aspects of corporate environmental activities, which increases the robustness of our ndings.
The paper is organised as follows.Section 2 describes the impact of the COVID-19 pandemic on MSMEs.Section 3 describes the data and variables, and Section 4 explains the model speci cations.Section 5 discusses our empirical ndings, and Section 6 reports on our robustness studies, while Section 7 provides a conclusion. [1]https://covid19.who.int/ [2]https://ec.europa.eu/eurostat/documents/2995521/10294864/2-15052020-AP-EN.pdf/5a7ea909-e708-f3d3-8375-e2510298e1b8 [3]The de nition of MSME varies across countries.The three common denominators are the number of employees, turnover and assets.In this paper, we adopt the SME de nition used in the World Bank Enterprise surveys.The employment criterion and categories of rms are as follows: micro constitutes fewer than 5 employees; small constitutes 5-19 employees; medium-sized constitutes 20-99 employees; and large constitutes more than 100 employees. [4]http://www.oecd.org/coronavirus/policy-responses/coronavirus-covid-19-sme-policy-responses-04440101/ [5]The literature provides scant evidence of the impact of the nancial crisis on the relationship between rm social performance and rm risk (Bouslah et al. 2018) and between social capital and rm performance (Lins et al. 2017(Lins et al. , 2019). [6]Recent literature has started to report on developing economies (Féres and Reynaud 2012;García-Rodríguez et al. 2013).

Impact Of The Covid-19 Pandemic On The Msme Sector
MSMEs account for approximately 90% of businesses across the world and are responsible for about 50% of employment [7] .MSMEs therefore play a signi cant role in the economy and are considered as engines of economic growth (Wellalage and Fernandez 2019).COVID-19 has, however, resulted in dramatic changes in the economic and political environment in most countries, causing economic shocks and placing stress on healthcare systems across the world (Kuckertz et al. 2020).In a bid to contain the spread of the virus, a number of countries have imposed national or regional lockdowns have had devastating repercussion for MSMEs.In turn, based on the responses from over 1,200 MSMEs in 109 countries, International Trade Centre survey indicates that pandemic has affected the 2/3 of the MSMEs interviewed.While MSMEs from all regions have been affected by the COVID-19 pandemic, African MSMEs have been hit hardest (Borino and Rollo 2020).Also, recent study shows that in the United States alone, 3.3 million small businesses closed during the February-March period (Fairlie 2020).
MSMEs are nancially fragile and generally have liquidity and pro tability problems (Juergensen et al. 2020).MSMEs are more vulnerable because of their ownership structure (Martin et al. 2019), and they generally have lower resilience to external shocks (Juergensen et al., 2020) because of their small size and limited human, nancial and technical resources.The nancial vulnerability of MSMEs was witnessed during the global nancial crisis (GFC) in 2008, which caused a signi cant decline in demand and nancial distress (Cowling et al., 2018).Likewise, the COVID-19 pandemic has severely affected MSMEs but to an even greater extent than the GFC (Baker et al. 2020).The negative impact of COVID-19 on the entrepreneurial nancial market is forecasted to have signi cant and long-lasting damage on small rms (Lerner et al. 2020).In particular, MSMEs must grapple with the effects of the pandemic on supply-related issues, such as logistical challenges owing to labour shortages and reductions in transportation, and on demand-related issues, including a dramatic reduction in demand and revenue. [8] Researchers emphatically speculated that small rm nance may be signi cantly affected by the COVID-19 pandemic.Also, these shocks have negative impact on all types of entrepreneurial and small business nance including debt nance, venture capital and business angels (Demirguc-Kunt et al. 2020).In particular, Brown and Rocha (2020) suggest that start-ups and small businesses in China witnessed a 60% decline in equity investments in the rst quarter of 2020.In a similar vein, a study examining the UK reports that entrepreneurial nance deals in the rst quarter of 2020 are fewer by 30% than the same quarter in 2019 due to a decrease in seed and early-stage nancing (Brown et al. 2020).Although the impacts of the COVID-19 pandemic on small businesses are undoubtedly negative, it has been argued that the small size of such businesses also enable them to explore opportunities during the crisis (Williams and Shepherd 2018).Papadopoulos et al. (2020), for example, suggest that the adoption of digital technology can help SMEs to secure continuity during the crisis.
Using the data of 250 SMEs from Greece, Kottika et al. (2020) suggest that, despite unfavourable conditions, 85% of SMEs survived the pandemic by focusing on productivity, e ciency, downsizing, and new partnerships with suppliers.Within the existing body of research on the impact of the pandemic on small businesses, the question of how corporate environmental performance informs the survival of small businesses during the crisis has been neglected. [7]https://www.worldbank.org/en/topic/smenance [8] http://www.oecd.org/coronavirus/policy-responses/coronavirus-covid-19-sme-policy-responses-04440101/ 3. Data And Variables

Data
To explore the relationship between rm level environmental performance and rm nancing during the COVID-19 pandemic, we utilise rm level data from two sources: (i) the Enterprise surveys from the World Bank (see http://www.enterprisesurveys.org);and (ii) COVID-19 follow up surveys from the World Bank (see https://www.enterprisesurveys.org/en/covid-19).Enterprise surveys from the World Bank encompass a representative random sample of rms with data collected across the world by using the same core questionnaire and the same sampling method.Face to face interviews are held with the owner/manager or representative.Surveys provide rm level information about unlisted rms of all sizes, such as rm characteristics and owner/manager demographics.This survey also reports information about the rm's access to nance, corruption and crimes, competition and infrastructure.The 2019 Entreprises Surveys provide a green economy module which contains information about environment-related aspects of the sampled rms.
As part of the overall response of the Development Economics Vice Presidency (DEC) of the World Bank, the Enterprise Analysis Unit has been developing different approaches to measure the impact of coronavirus on the private sector.In addition to conducting Enterprise Surveys, phone interviews were held in seven Eastern European and Central Asian countries.The Enterprise Surveys team has now released data pertaining to Albania, Cyprus, Georgia, Greece, Italy, Moldova and the Russian Federation.Enterprise Surveys follow-up surveys data aimed at measuring the impact of the COVID-19 pandemic on businesses [9] , which includes follow-up surveys that aim to measure the impact of COVID-19 on businesses by collecting information about closures.The Enterprise Surveys team has similar surveys planned or under implementation in 39 countries.
Our study merges two datasets using unique rm identi cations.Our sample includes micro, small and medium rms from 22 sub-industries.From the survey data, our study sample provides information about 3,356 unlisted rms.The number of respondent rms by rm size, the country combination and industry distribution are reported in Table 1. Figure 1 shows the main source that MSMEs have used to deal with cash ow shortages in the outbreak of COVID-19.Figure 1 A (full sample) indicates that the majority of rms used equity nancing (29%) as the main source to deal with cash ow shortages during the pandemic.However, approximately 28% of MSMEs do not have access to any nancial source to deal with cash ow shortages.These results highlight the nancial fragility of many small businesses during the pandemic.About 21% of MSMEs delayed payments to suppliers and workers in response to cash ow shortfalls and only 17% of MSMEs accessed bank loans.As with the trend across the full sample, rms from Georgia, Cyprus and Moldova used equity nancing to address cash ow shortfalls.However, Figure 1 shows that the majority of MSMEs from Greece, Albania and Russia have no access to a nancial source in order to deal with their cash ow shortage.
Table 2 reports the descriptive statistics of the study sample.According to Table 2, about 72% of the sample rms have experienced a decrease in liquidity during the pandemic.This is aligned with recent OECD ndings, which report that the risk of liquidity shortage is high for a signi cant portion of rms in OECD countries and that, if quarantine measures last beyond seven months, more than 50% of rms would face a shortfall of cash (OECD 2020).Studies from developed countries show that approximately half of all small rms have temporarily stopped trading since movement restrictions have been enforced and 60% of SMEs are at risk of running out of cash reserves (Bartik et al. 2020;Cowling et al. 2018).Table 2 shows that only 17% of rms in our sample used banks to deal with cash ow shortages, which means approximately 83% of rms have not borrowed funds from commercial banks.Three percent of rms led for bankruptcy/insolvency during the pandemic and, while this gure may appear small, the high liquidity shortfall in 72% of rms may soon turn into a solvency crisis.An ongoing reduction in liquidity and revenues over a prolonged period, coupled with limited sources to deal with this shortfall, push many SMEs to voluntary closure and bankruptcies in the long run.
The mean environmental performance score for the full sample is 1.4, which ranges from 0 (lowest) to 10 (highest).The low mean value of the environmental performance index (EPI) indicates that SMEs are not highly engaged with environmental performance activities.About 34% of rms have at least one female owner.The average percentage of SMEs that are family-owned is 43%, but it varies from 0% (non-family owned rms) to 100% (fully family-owned rms).The average rm age is 18 years, but it varies from 1 year to 199 years.Approximately 58% of sample rms are micro and small, and 42% are medium.In terms of rm legal ownership, the majority of rms are a company (68%).Just over a quarter of rms are exporters (26%).The majority of rms come from the manufacturing industry (46%), followed by services (32%) and other services (21%).Liquidity_Dec: This variable takes the value of one if rm liquidity decreased during the COVID-19 pandemic.Firm level liquidity is an important factor which determines a rm's ability to pay its creditors and remain solvent in the short run.Liquidity is more of a strategic concept related to MSMEs (Kontuš and Mihanović 2019).A healthy liquidity level is an indicator of the creditworthiness of small rms and improves their borrowing ability.
Bank: This variable takes the value of one if the rm has used commercial banks as the main mechanism for dealing with cash ow shortages during the COVID-19 pandemic.
Bankruptcy: This variable takes the value of one if the rm led for insolvency or bankruptcy during the pandemic.

Explanatory variables
We created a composite index in order to establish a proxy for rm level environmental performance indicators in our regression analyses.This method was initially developed by Jaggi and Freedman (1992) in the adaptation used by Wagner (2005).Following Wagner and Schaltegger (2004), we measure environmental sustainability activities in terms of an index that assesses the reduction of environmental impacts by rms in several categories, each measured by a separate item variable.For each of the items, the Enterprise Survey questionnaire had asked about the series of environmental activities that were undertaken to reduce the company's environmental impact.The following ten questions were used to create the index value: Over the last three years, did this establishment adopt any of the following measures?
(i) Heating and cooling improvements, (ii)More climate-friendly energy generation on-site, (iii) Machinery and equipment upgrades, (iv) Energy management, (v) Waste minimisation, recycling and waste management, (vi) Air pollution control measures, (vii) Water management, (viii) Upgrades of vehicles, (ix) Improvements to lighting systems, (x) Other pollution control measures.
A high score on the index indicates environmental performance above the mean, i.e. a rm that is highly engaged in environmentally sustainable practices.
There are no universally accepted standards or methodologies for assessing a rm's environmentally sustainable performance (Ameer and Othman 2012).Consequently, there is no quantitative data available for the environmental performance of unlisted rms.As such, we can argue that measurement of environmental performance needs to be included as a component of rm performance proxies because environmental performance is a holistic view of the effects of environmental management.Following prior literature, we use self-assessment by rms as the most suitable approach for establishing a proxy for environmental performance (Sharma 2001;Wagner and Schaltegger 2004).The responses yielded from the ten questions posed about the rm's environmental activities provide the basis for our measurement of environmental performance.
Earlier studies adopted several methods to ascertain corporate environmental performance, including external audits, external awards or accreditation processes, indices, and non-quanti able sustainability initiatives.Such variation leads to inconsistency of environmental performance variables utilised in empirical studies (Braam et al. 2016).Recent nancial studies have used environmental, social, and governance performance scores obtained from the Thomson Reuters ASSET4 index as a reliable proxy for environmental performance for large listed rms (Cheng et al. 2014).ASSET4 includes data about the energy used by a rm, water and waste recycling practices, carbon emissions, and spills and pollution controversies.Due to data unavailability, we cannot apply the ASSET4 index to the unlisted rms we sampled.Moreover, our proxy incorporates ten indicators, which better represent the environmental activities of unlisted rms.It should be noted that our index is output-oriented, which diverges from other studies which have used environmental performance effort as input measures, (such as the amount of environmental management activities a rm records), as proxies for environmental performance activities.Given that all input efforts do not necessarily equate to the production of desired outputs, we contend that our output proxy is more reliable.
One concern with the speci cations in our regression models is that the strong performance of high environmental performance rms during the COVID-19 pandemic may be due to omitted variables that happen to be correlated with EPI, and not due to environmental performance itself.To address this issue, we include several control variables.The control variables chosen for the analysis are widely recognised in the literature as those that affect rm level nancing.
First, we control for a rm's nancial health in the year before the COVID-19 pandemic on the basis that rms with high nancial stability before the pandemic may have a greater ability to withstand a downturn in the economy.Therefore, we include: Credit_Constraint (constraints as value one if the rm does not have (i) an overdraft facility or (ii) a line of credit or loan from a nancial institution and zero otherwise) and Finance_Obstacle (obstacle as value one if the rm reports that access to nance is a major or very severe obstacle to the current operations of the rm and zero otherwise) as a proxy for the nancial health of the rm prior to the outbreak of COVID-19.
Additional rm characteristics may also be important for rm level nancing.As a proxy for rm size, we use two categorical variables based on numbers of employees, such as Micro & Small and Medium.We also included rm age.The following proxies indicate the ownership characteristics of the rm: the legal entity of the rm -sole proprietorship, partnership and company; and the percentage of foreign ownership, family ownership and female ownership.Export and innovation status of the rm, top manager experience in the eld, and rm level bribes payment were also variables included in our regression.
Moreover, we control for the overall nancial situation at country-by using the depth of credit information index (Credit_Index), the ease of doing business index (Ease_Business) and the percentage of rms using banks to nance working capital (Bank_Capital).To address possible valuation differences among industries, we included industry dummies.
Table A1 in the Appendix presents de nitions and metrics for the variables of this study.
[ 9] The Enterprise Survey also provided Armenian data, but this was excluded from our study as it contains only micro rms.

Method
To analyse the relationship between the nancing of MSMEs and environmental performance, we employ a probit model, where our outcome variable(s) is binary.That is, we assume: SME nancing is captured by three binary variables.The rst is a Liquidity_Dec which takes a value of one if the rms report that the COVID-19 outbreak decreased rm liquidity/cash ow, and zero otherwise.The second variable is Bank which takes the value of one if the rm received loans from commercial banks to deal with cash ow shortages during the COVID-19 outbreaks, and zero otherwise.The third variable is bankruptcy which takes the value of one if the rm led for insolvency or bankruptcy during the pandemic, and zero otherwise.
The main explanatory variable, EPI, is a continuous variable that ranges from zero to ten, with the value zero being for rms which have no engagement with environmental performance activities.Vector X captures rm characteristics and ownership characteristics, such as rm nancial stability proxies, percentage of family ownership, rm age, rm size, rm legal ownership status, innovations, and export status.Vector Y captures institutional characterises, such as Credit_Info, Doing_Business and Bank_Cap.
We cluster the standard errors by country, where vce (cluster country) accommodates and adjusts for the correlation of observations within values of the country.Observations from the same country are likely to be more similar, therefore, the country-level cluster approach is appropriate.

Firm level environmental performance and nancing in the COVID-19 outbreak
The base results of the effect of environmental performance on liquidity decrease, bank credit approval, and insolvency/bankruptcy, are presented in columns I, II and III in Table 3 respectively.In each column, we report probit regression results and marginal effects.
Our results in column I show that the environmental performance of a rm is signi cantly negatively correlated with liquidity decrease.In particular, the marginal effects show that for every one unit increase in the EPI index, the probability of liquidity shortfall decreases by 0.0172.Although the results in column II show that the environmental performance of a rm is positively correlated with bank nancing access during the pandemic, the relationship is insigni cant.
Interestingly, our results in column III show that the environmental performance of a rm is signi cantly negatively correlated with the probability of MSMEs' insolvency/bankruptcy during the pandemic.In particular, the marginal effects show that for every one unit increase in the EPI index, the probability of bankruptcy/insolvency decreases by 0.0077.Our results suggest that the trust between rms and stakeholders built through investment in rm level environmental activities pays off when markets suffer from a negative shock.In sum, rms with greater environmental performance create value and cooperation from stakeholders that leads to an increase in rm cash ow, access to bank nancing, and solvency.This may be because proactive environmental practices are exposed to a low degree of risk (Godfrey et al. 2009;Muhammad et al. 2015a), and easier access to the nancial market (Jo and Na 2012).It may also be that CSR and environmental activities restore stakeholder trust after the crisis (PricewaterhouseCoopers 2013).Overall, our nding is in line with Lins et al. (2017) and Lins et al. (2019) who nd that the rms with high CSR initiatives experience higher pro tability, growth, sales and access to funds during the nancial crisis of 2008-2009.Firms from all industries have been impacted from COVID-19 outbreak.Nevertheless, the hospitality and retail industries are faced with the most signi cant disruption in recent global pandemic (Orr 2020).In this section, we investigate whether the positive effect of environmental performance is outperforming in sensitive industries.
Table 4 indicates that the rms with high environmental performance in sensitive industries exhibit superior performance during the pandemic.The coe cient on the EPI is negative and signi cant, indicating that regardless of the industry to which the rm belongs, rm level environmental performance negatively impacts on the probability of cash ow decrease (β= -0.0082) and bankruptcy (β=-0.0059).The signi cant and negative relationship is shown in the interaction variable (EPI X i. Sensitive_industry), rm level liquidity decrease (β= -0.0082 + (-0.0154) = -0.0236)and bankruptcy (β= -0.0059 + (-0.0038) = -0.0097).The results show that the environmental performance effect is more pronounced in sensitive industries than in non-sensitive industries.

Robustness
This study runs a series of tests to ensure the robustness of baseline results.
6.1 Propensity score matching method The probit model tested rm level nancing as a function of rm level environmental performance.However, the results will be biased when an endogeneity issue is present.In empirical research, the propensity score matching (PSM) technique heavily draws on causal inferences by using observational data (Wellalage and Locke 2020).PSM controls causal interferences and self-selection biases by placing them into non-random treatment (Rosenbaum and Rubin 1983).Our focus was the comparison between the propensity for nancing during the COVID-19 pandemic in rms that are exposed to no treatment T=0 (EPI index is less than average EPI of the sample) and the propensity for innovation in rms that are exposed to treatment T=1 (EPI index is above average EPI of the sample).We use three matching methods: nearest neighbour matching, kernel matching, and strati cation matching.To further ensure the e ciency of PSM, we conduct a covariate balance analysis on the mean difference in the covariate used in the probit model between the treatment group and the control group.The unreported results of the covariate balance indicate the difference between the treatment group and control groups are small across all variables, con rming that our PSM is appropriate.
Table 5 includes all three matching models and shows that ATT is statistically signi cant and negative for cash ow decrease and bankruptcy.In particular, the estimated negative average effect of EPI on liquidity decrease and liquidation for rms that have high EPI is between 3% to 3.5%, and between 1.4% and 1.5%, respectively.Overall, these ndings are consistent with our main ndings which support the view that, in every sense of the phrase, it "pays to be green".This study adopts the probit model with sample selection method as a robustness check to control the potential sample selection bias (Heckman 1977).In the rst step, we estimate a probit model with a binary dummy as the dependent variable.However, in the sample, a rm is observable when the rm sales decrease.This raises the issue of selection bias.To address selection bias, we use a COVID_Temp_Closed variable in the selection equation which takes the value of one if the rm reports that it temporarily closed as a result of the COVID-19 outbreak, and zero otherwise.
Similarly, when we estimate a probit model with a binary dummy as dependent, a rm is observable when the cash ow/liquidity decreases.To address selection bias, we use proxies for two other types of nance sources, such as government grants (Government _Grant) and credit from non-nancial institutions and micro nance institutions (MF).
When we estimate a probit model with a binary dummy as dependent, a rm is observable if the establishment has overdue obligations to any nancial institution during the COVID-19 pandemic.To address selection bias, we use proxies for two other types of nance sources: Government_Support, which indicates whether or not the establishment received any national or local government support in response to the crisis, and Snr_Mgr_Time, which indicates the time spent by senior management on dealing with regulations.
Table 6 reports the heckprobit regression results.The marginal effects show that a unit increase in the EPI, on average, leads to a decrease of 0.0079 in the probability of cash ow decrease and 0.0020 in the probability of bankruptcy.Overall, we can see that the EPI is still signi cantly and negatively correlated with cash ow decrease and bankruptcy, demonstrating that environmental performance enhances rm nancing during the pandemic.

Excluded micro rms
In the third robustness test, we assess whether the decision to remove micro-rms from our sample affects our results.We excluded rms (77 in total) which belong to the micro category.This was on the basis that micro rms have limited access to external nancing, and they are mainly nanced by the owner/managers, which could outweigh other factors during the pandemic.We re-estimate our baseline model for only small and medium-sized rms and observe that the unreported results align with our baseline model.

Conclusion
This paper examines the effect of rm environmental performance on rm nancing during the COVID-19 pandemic using a sample of unlisted MSMEs.The evidence summarised in the results section indicates that rm level environmental performance reduces liquidity decrease and bankruptcies during the pandemic.Furthermore, our industry analysis indicates that the impact of environmental performance on rm nancing is more pronounced in sensitive industries (hospitality and retail) than in non-sensitive industries.Overall, our results are consistent with our conjecture that a rms' environmental performance can limit the negative effects when markets and institutions suffer a negative shock.
Worldwide, MSMEs make up more than 95% of all rms, when both formal and informal businesses are considered.Therefore, a widespread collapse of MSMEs could have a signi cant negative impact on global economy.In their response to the COVID-19 pandemic, governments should prioritize policies that support MSMEs.Our study leads to several policy implications.First, it highlights the importance of environmental activities as a business strategy that is integrated with core business objectives to manage rm level nancing.We illustrate that rm-speci c environmental performance can be thought of as an insurance policy for rms that, quite literally, pays off when the overall economy experiences a negative shock.We also highlight the importance of favourable monetary policies, such as support programmes and the provision of short-term credit to the private sector, which work to generate a continued ow of credit to businesses and to cushion the nancial distress on solvent private rms.
As with all environmental performance measures used in prior studies, we cannot fully rule out the possibility that our environmental performance index suffers measurement errors.To alleviate this concern, we suggest that a much broader framework should be considered in future studies.The current lack of a widely accepted framework for measuring the proxy for environmental performance may limit opportunities to compare the environmental performance of different rms.Therefore, future studies can consider working towards a universal proxy for environmental performance in order to facilitate comparative studies.

Declarations
Page 18/18 0 = otherwise Credit_Info This index measures rules affecting the scope, accessibility, and quality of credit information available through public or private credit registries.
0=low to 8=high The higher values indicating the availability of more credit information Doing_Business Ease of doing index indicates the regulatory environment quality to starting and operation of a local rm.The Figures main sources MSMEs have used to deal with cash ow shortages in the outbreak of COVID-19

Table 1
Sample rms

Table 2
Descriptive statistics

Table 3
Probit and marginal probit results of rm nancing and environmental performance in COVID-19 outbreak Sensitive industries, environmental performance activities, and nancing during COVID-19 outbreak.

Table 4
Highly sensitive industries, environmental performance activities, and nancing in COVID-19 outbreak Table 4 reports probit results and marginal effects estimated around means points.Robust standards errors are reported in parentheses.± Omitted rm size is Micro & Small.∞ Omitted rm legal status is soleproperitership.*Signi cant at 10% level, **Signi cant at 5% level, ***Signi cant at 1% level.

Table 5
The impact of environmental performance on nancing: propensity score matching (PSM).

Table 6
Heckprobit results of liquidity decrease, bank loan and bankruptcy and environmental performance in COVID-19 outbreakTable 6 reports heckprobit results and marginal effects estimated around means points.Robust standards errors are reported in parentheses.The Need equals one if the rm reports a need for loans, zero otherwise.Apply equals one if the rm applies for a loan, zero otherwise.Loan approval equals one if the rm gets at least one loan, zero otherwise.Collateral equals one if the rm pledge collateral for their approved loan, zero otherwise.Duration equals to one if the rm received long term loan (more than or equal 5 years), zero otherwise.Crime and Sales_Increase are two instrumental variables in "Need" selection equation.Trade_Credit and Informal_Credit are the two instrumental variables in "Apply" selection equation.Internal_funds and Snr_Mgr_Time are the two instrumental variables in "Approved" equation.* Signi cant at 10% level, **Signi cant at 5% level, ***Signi cant at 1% level.