Policy Research Working Paper 10556 Masters of Disasters The Heterogeneous Effects of a Crisis on Micro-Sized Firms Arlan Brucal Arti Grover International Finance Corporation & South Asia Region August 2023 Policy Research Working Paper 10556 Abstract Most crises have a disproportionately larger negative effect Within the two groups of micro-sized firms, resilience is cor- on micro-sized firms. Yet, the heterogeneity of impact related with their liquidity position, managerial attitudes within micro-sized firms is lesser known. Using five waves as well as their abilities. Using discriminant analysis, this of the World Bank’s Business Pulse Survey data, this paper paper confirms that a significant proportion of micro-sized finds that firms with zero to four employees have a much firms mimic the behavior of larger firms in terms of their larger drop in sales and slower recovery rate compared to resilience to shocks and could potentially be “misclassified” micro-sized firms with five to nine employees. The over- as micro-sized. These findings have important implications all differences in the resilience between the two groups of for targeting and tailoring support for enhancing businesses’ micro-sized firms could potentially be due to a uniformly resilience to shocks. lower productivity level of firms with zero to four employees. This paper is a product of the International Finance Corporation and the Office of the Chief Economist, South Asia Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http:// www.worldbank.org/prwp. The authors may be contacted at agrover1@ifc.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Masters of Disasters: The Heterogeneous Effects of a Crisis on Micro-Sized Firms Arlan Brucal∗ Arti Grover† JEL classification: D22, L20, L25, O10 Keywords: COVID-19, crisis, micro and small firms, informality, resilience Acknowledgements: The authors are grateful to the South Asia Regional Trade Facilitation Program (SARTFP) for funding support and Denis Medvedev for his strategic guidance during the inception of the project. The team thanks Maurizio Bussolo and Siddarth Sharma for their helpful feedback. Disclaimer: The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. It does not reflect the views of the Governments covered by this study. The findings of the study, thus, would not be binding on the countries covered by the study. ∗ Land, Environment, Economics and Policy Institute, University of Exeter and Gratham Research Institute and Grantham Research Institute on Climate Change and the Environment, London School of Economics and Political Science E-mail: a.brucal@exeter.ac.uk; a.z.brucal@lse.ac.uk. † Corresponding author; International Finance Corporation, 2121 Pennsylvania Avenue, Washington DC, USA, E-mail: agrover1@ifc.org 1. Introduction It is widely known that smaller businesses are ’masters of disasters’ and particularly vulnerable to external shocks. This has been evidenced not only in the context of the recent COVID-19 pandemic (Bartik et al., 2020; Apedo-Amah et al., 2020) but also in past crises (e.g., Sahin et al., 2011; Bricongne et al., 2012; Cowling et al., 2015). Smaller firms face larger market contractions, higher uncertainty (Cirera et al., 2021) and greater financial and credit restrictions during a recession (Gertler and Gilchrist, 1994; Laderman et al., 2013). However, most studies assume smaller firms to be a monolithic and homogeneous entity which raises issues in policies targeting such firms because the group is highly heterogeneous in performance and characteristics.1 By using a unique firm-level data set comprising 185,804 firm-year observations in 86 countries, collected in five waves between May 2020 and August 2022, we illustrate the heterogeneous effects of a crisis within the group of micro-sized firms (that is, those with fewer than 10 employees). Our work emphasizes the need to clearly distinguish typology within this group of firms for a targeted approach to support recovery of the private sector. Smaller businesses are critical for driving economic growth and job creation, (see, Manzoor et al., 2021 for Pakistan, Fiseha and Oyelana, 2015 for South Africa, and Tambunan, 2019 for Indonesia) and are expected to play a vital role in overall recovery efforts (UNCTAD, 2021). While there is extensive documentation of the heterogeneity of impacts and responses of the crisis among firms in general, and smaller firms, in particular (Cirera et al., 2021; Constantinescu et al., 2022; Fernández- Cerezo et al., 2022), little is known about the asymmetric effects and responses within the class of smaller firms that bunches together micro and small firms. An understanding of the heterogeneity within smaller firms is critical for informing the design of recovery policies that can stimulate entrepreneurship and a healthy firm dynamics (Meyer et al., 2021). This paper aims to investigate the heterogeneity of micro-sized firms with respect to the impact of the COVID-19 pandemic on their operations, sales, financial fragility, and recovery pathways. It also compares the responses of such firms to that of other larger-sized firms and identifies the attributes that allow them to mimic resilience comparable with categories of larger firms. Using five waves of survey data collected by the World Bank Group during the various phases of the COVID-19 crisis, we show that: First, although micro-sized firms are disproportionately affected by the crisis, the impact within this category of firms is not homogeneous. Splitting this group of firms into (0-4) and (5-9) employees suggests that the two types of micro-sized firms are quite distinct from each other in terms of their impact on sales and recovery from the crisis. Firms with (0-4) employees have much larger drop in sales and slower recovery rate, implying that combining the two groups, as is often the case in many 1 Smaller or micro-sized firms in developing countries often belong to the informal or unorganized sector. The heterogeneity of the informal sector, for instance, has been categorized into upper and lower tiers based on factors such as skill level, income generation, and economic activities. Lower-tier informal activities include street vending, waste picking, and domestic and day labor. Upper tiers may involve higher setup costs and limited entry opportunities, such as skilled artisan and small-scale manufacturing. The boundaries between upper and lower tiers are fluid, however (Cunningham and Maloney, 2001). 1 developing countries, may treat two very different types of firms in the same way. Second, comparing the impact of the crisis within the two groups of micro-sized firms, the results are contingent on the chosen yardstick. Size variation within the two groups is not associated with resilience to the crisis, while differences in liquidity position, and managerial attitudes matter. For instance, one-person firms within the 0-4 worker group have an impact comparable to that of a firm with 4 employees. By comparison, firms with higher cash, and optimistic managerial outlook within each of the two groups weather the crisis better. Thus, firms can be resilient based on these characteristics which may supersede the size criterion. Third, comparing the impact of the crisis across the two groups of micro-sized firms based on initial productivity and managerial ability it appears that the (0-4) employees group suffers uniformly from the crisis compared to the (5-9) group based on productivity. Likewise, firms in (5-9) employees group are invariant to the managerial ability to reasonably predict expected future sales. The overall difference in the resilience between the two groups of micro-sized firms could potentially be emerging from a lack of variation in productivity within micro-sized firms with (0-4) employees. This suggests that there could be other levers within the (0-4) group of firms such as liquidity, managerial outlook which can still aid their resilience to shocks. Fourth, discriminant analysis pertaining to the misclassification of firms,2 , suggests that a significant proportion of micro-sized firms, particularly those on the right tail of the firm size distribution, are misclassified. This is not surprising, given the wide differences in (0-4) and (5-9) employees firms observed in regressions. The discriminant analysis confirms that some of these firms have the ability to recover from the shock that is different from, and at times mimic that of the larger firms, which are more resilient to shocks. The probability of micro-sized firms to be misclassified as non-micro-sized depends on their liquidity constraints and managerial outlook. We contribute to the growing body of literature on the significant impact of the recent pandemic on smaller businesses and their adaptive responses to recessionary conditions in advanced and devel- oping economies (Apedo-Amah et al., 2020; Bartik et al., 2020; Bloom et al., 2020; Fernández-Cerezo et al., 2022). The overall findings coincide with previous research suggesting that smaller firms are disproportionately more affected by the crisis and that they respond differently to periods of uncertainty relative to larger firms (Cowling et al., 2012; Smallbone et al., 2012). For example, the credit crunch during the 2008 Great Recession brought larger employment cuts in smaller firms (Chodorow-Reich, 2014), not only because of the acute decline in small business loans (Greenstone et al., 2020), but also because small businesses are relatively more dependent on depository in- stitutions, and especially their relationships with commercial banks for credit and other financial services (Kennickell et al., 2016). Our work differs from previous research by, first, highlighting the heterogeneity within the group of micro-sized firms, and second by examining the likelihood of such firms to mimic the pattern of more resilient larger firms in coping with the challenges of a crisis. 2 Misclassification refers to the fact that in certain firms the predicted size category based on the chosen statistical model is different from the observed category (Doumpos and Zopounidis, 2002; Doukas et al., 2021). 2 Our work has significant implications for policies to support firms. First, While public support to firms is usually targeted by their size and the choice of firm size threshold in defining the unorganized or "informal" sector is rather arbitrary – ranging from fewer than 4 employees, to 5 or 6 employees in certain countries (e.g., Burkina Faso, Colombia) and up to 10 in others (e.g., Honduras, India, Pakistan). Size-dependent policies harbor misallocation of resources and limit the ability of firms to grow (Garicano et al., 2016; Martin et al., 2017; Rotemberg, 2019) and respond to the crisis. By comparison, policies that provide general support to all small firms, such as tax breaks or subsidies, may not be efficient in addressing the specific needs of different types of firms(Fontainha and Lazzaro, 2019). Instead, policies that target specific subgroups of small firms based on their characteristics, such as their export potential or managerial capabilities, may be more effective (Atkin et al., 2017; McKenzie and Woodruff, 2017; McKenzie, 2021; Amin and Islam, 2022). Our work suggests that an effective policy targeting firm resilience and recovery even within the micro-sized firms will separate them by their attributes rather than size. Second, the crisis illustrates that micro-sized firms in better liquidity position were more resilient even when the COVID-19 pandemic induced crisis did not originate in the financial sector. While the existence of a target cash holding level to which small firms attempt to converge can help the financially constrained firms (Almeida et al., 2004; Berger and Udell, 1998), such precautionary motive for larger cash holdings can curb growth opportunities due to the low rate of return of liquid assets (Opler et al., 1999; Lockhart, 2014). This underscores the crucial role of credit and access to finance in leveling the playing field for both small and large firms during a crisis. Financial institutions can potentially cushion the disproportionate impact that crises have on smaller and micro-sized businesses and promote economic stability. The paper is structured as follows. Section 2 outlines the data sources and empirical strategy, building on previous methodological contributions by Cunningham and Maloney (2001) and Bruhn (2013). Section 3 presents the results of our analysis, beginning with regression framework, followed by the discriminant analysis that formalizes the methodology for inferring misclassification of firms. We then present the results of logistic regression to identify the predominant factors that influence the extent to which micro-sized firms mirror the behavior of other firms, particularly those in larger size categories. Finally, section 4, summarizes our findings and discusses the implications for policy. 2. Data and Methodology 2.1 Data During the COVID-19 pandemic, the World Bank’s Business Pulse Surveys (BPS) and the Enterprise Surveys (WBES) collected data from businesses to fill the knowledge gap on the impact of the crisis on the private sector. The first wave of the BPS was conducted between May and September 2020 in 82 countries, followed by a second wave between October 2020 and March 2021 in 49 countries, 3 and the following three waves were conducted between April 2021 and August 2022 covering 49 countries. These surveys cover firm characteristics such as size, age, debt, and pre-COVID-19 sales, as well as changes since the pandemic began, such as in employment, sales, expectations, digitalization, and public support received.3 The surveys target firms of all sectors and sizes. Our sample includes firms in agriculture (3.6%), manufacturing (31.6%), retail (23.2%), and other service activities (41.6%). We include all size groups in the sample: micro (firms with 0-4 employees, 36.8%; 20.4% with 5-9 employees), small (10- 49 employees, 35.4%), medium (50-249 employees, 3.3%), and large (at least 250 employees, 4.0%).4 In most countries, the sampling frame is based on censuses from Statistics Agencies, Ministries of Finance or Economy, or business listings from Business Associations, and typically only includes businesses that can be found in some registers or listings. The WBES COVID-19 follow-up surveys cover only formal firms by design. Table A1 presents descriptive statistics. Of the observations, 49% (or 42% of the total number of firms) are micro-enterprises with employment size below 10. For the outcomes of interest. the five rounds of data show that, on average, and without controlling for any other factor, we observe a firm to be open 98% of the time. Firms also exhibit a decrease of about 10 percentage points against their pre-pandemic level. The decline is much larger at 44 percentage points in the first wave, as some firms will have been recovered in succeeding periods. Recovery rate, as measured by ∆Change in salest,t−1 , stood at 13 percentage points, on the average. Notwithstanding, the median value of the recovery rate is zero, implying that half of the firms have not recovered, while some have as indicated by the high standard deviation of about three times that of the mean. To measure the shock suffered by firms, we use data from Google mobility reports around transit stations (Mathieu et al., 2020). For countries without available data, we impute data based on the Oxford Government Response Tracker index (Hale et al., 2021). We construct an indicator of the severity of the crisis that is a weighted average of 30-day periods since the start of the pandemic until the date of the survey. Specifically, the 30-day period average just before the survey has a weight of 1, the average from day 31-60 has a weight of 1/2, the average from day 61 to 90 has a weight of 1/3, and so on until the start of the pandemic. We also use data from Google mobility reports around transit stations (Mathieu et al., 2020) to measure the size of the shock suffered by firms. For countries without available data, we impute data based on the Oxford Government Response Tracker index (Hale et al., 2021). We construct an indicator of the severity of the crisis that is a weighted average of 30-day periods since the start of the pandemic until the date of the survey. Specifically, the 30-day average just before the survey has a weight of 1, the average from day 31-60 has a weight of 1/2, the average from day 61 to 90 has a weight of 1/3, and so on until the start of the pandemic. 3 For more information, see Apedo-Amah et al. (2020) and Cirera et al. (2021). 4 Classification of firms by size varies across countries, with different metrics used to distinguish smaller firms from larger ones. In this study, we follow the classification of the OECD based number of people employed: micro enterprises (fewer than 10 employees); small enterprises (10 to 49 employees); medium-sized enterprises (50 to 249 employees); large enterprises (250 or more people). 4 To use the harmonized BPS and WBES follow-up data for assessing the impact of COVID-19 on firm performance and recovery, we are concerned about the heterogeneity related to the differences in country samples, implementation strategy, and the timing of the surveys, also noted in Apedo-Amah et al. (2020). These concerns are addressed in the following methodology subsection. Our analysis may be affected by attrition in the sample. Although each wave was specifically targeted follow-up surveys of firms from wave 1, some firms were not reachable or declined participation. The challenges associated with attrition, typical of longitudinal studies, are exacerbated in the context of the COVID-19 crisis due to mobility restrictions, in times of higher likelihood of business closures, and potentially larger opportunity costs of responding to a survey rather than working when restrictions are lifted. Such idiosyncrasies may systematically influence the availability of certain types of firms to respond to the follow up surveys. To avoid a reduction in sample size, firms that were no longer observed in the data were replaced with other firms with similar characteristics, as per the original sample design, such as stratification by size and sector. Thus, we observe a balance between waves in terms of general characteristics of the firms, such as size and sector. Using a variety of techniques to address firm attrition, Cirera et al. (2021) show that the results using BPS data remain broadly robust. 2.2 Methodology To assess the heterogeneity in firm responses to the COVID-19 pandemic and the varying impacts of lockdowns on firms we use the following approaches Regression analysis: To account for the differences in country samples, implementation strategies, and timing of surveys, we run probit and simple linear regressions on binary and continuous outcome variables, controlling for differences in sample composition by including sub-sector dummies (10 groups) and country-specific effects. We estimate the following logistic equation for discrete outcomes pertaining to operations and financial fragility: P r(yisc = 1) = α + βXi + θc + γs + λw + Shockic + ϵis (1) where P r(yisc ) is the probability of a binary outcome of firm i in industry sector s in country c. Binary outcomes considered in this paper include indicators of firm performance such as operating status and firms falling into arrears. Estimations also control for 12 sector-, survey wave- and country- specific effects, γs , λw and θc , respectively. To control for variation in the severity of the shock within the period the survey was implemented, we include Shockic which is a cubic spline function of the severity of the crisis measure explained in the data subsection5 , to allow the probabilities to move flexibly across different intensity of the crisis. Our variable of interest here is Xi , which denote firm-level size categories based on number of employees (e.g., micro-sized, small, medium or large). 5 For details, see Apedo-Amah et al. (2020). 5 We run the following estimations for continuous variables on change in sales compared to 2019 levels, and recovery as measured by the percentage difference between changes in sales6 relative to 2019 levels and the previous period: yis = α + βXi + θc + γs + λw + Shockic + ϵis (2) An important aspect of our regression framework examines the differences within micro-sized firms by separating them into two groups based on their employment size: (1) those employing 0-4 employees and (2) those with 5-9 employees. Regression equations 1 and 2 are re-estimated using this categorization as Xi to quantify the impact of the pandemic within the group of micro- sized firms. Later on, this will be expanded to include other firm-level attributes such as liquidity constraints, pre-pandemic per capita sales as a measure of labor productivity, gender of leadership, the firm’s future outlook, and managerial ability as proxied by prediction error in estimating future sales. Our analysis employs a pooled cross-section approach including the full sample of firms in all survey waves. To avoid any confounding effect associated with the potential non-random differences in the implementation of the survey in different time periods, countries and sectors, we control for the survey wave-, country- and sector-specific effects. We do this for both binary and continuous firm performance outcomes. All presented results are estimated conditional averages from regressions, unless specified. We also measure the marginal effect of each firm size dummies on the selected outcomes, allowing us to measure the heterogeneous effects after netting out the effect of these potential confounders. To further explore the diverse effects of the pandemic on micro-sized firms, we extend our analysis by incorporating additional factors. These factors include liquidity constraints, pre-pandemic per capita sales as a measure of labor productivity, gender of leadership, the firm’s future outlook and managerial ability. In order to capture these dimensions, we conducted probit and OLS regressions, introducing dummy variables for liquidity constraints, per capita sales terciles, gender of leadership, managerial style and quality (measured by outlook and prediction error) in each regression run. Discriminant analysis: By maximizing the ratio of between-group variance to within-group variance, we determine the optimal linear combination of independent variables that effectively discriminates between the pre-defined groups, in this case, firm size groups. This enables us to predict which firms belong to a specific size category and which ones belong to different categories based on the observed impacts. In sum, discriminant analysis allows us to effectively distinguish firms across size categories and assess the misclassification imposed by rigid size thresholds. 6 Change in total sales is the reported percentage change in the 30 days prior to the survey relative to the previous period. 6 The linear combinations in discriminant analysis are derived from the following: R Z= wr Xr (3) r=1 where Z is the discriminant score, wr are discriminant weights and Xr are the independent variables that relate to firm-level resilience such as indicator variables for remaining open and falling into arrears, and change in sales relative to 2019. In order to account for systematic non-random differences associated with the differences in survey waves, subsector and countries, as well as firm-specific attributes such as age and pre-pandemic sales level, we need to net out the effect of these variables from both the dependent variable and the independent variables consistent with the Frisch-Waugh-Lovell theorem (Frisch and Waugh, 1933; Lovell, 2008). We do this by estimating the equation below for each of the variables in equation 3 against age, subsector, country and survey wave-specific effects and sales level in 2019 and then compute the residuals: Xi = agei + sales2009,i + θc + γs + +λw + zi (4) ˆi , are then used to determine the discriminant score for each firm, Zi , in The estimated residuals, z equation 3. The estimated discriminant score represents the firm’s position relative to the decision boundary that separates the pre-defined groups. Hence, firms can be "misclassified" if their grouping based on this methodology does not match their true (observed) grouping. For instance, an observed micro-sized firm can be classified as a non-micro firm, depending on its estimated discriminant score’s position relative to the decision boundary. The decision boundary, which is estimated using within-group and between-group variance through the computation of the Fisher’s linear discriminant function (Ghojogh and Crowley, 2019; Doumpos and Zopounidis, 2002), maximizes the separation between different classes or groups, allowing for classification of observations. Discriminant analysis allows us to uncover the characteristics of micro-sized firms that allows them to mimic the behavior of larger and presumably more resilient firms. To do this, we extend our empirical strategy and conduct a probit regression analysis using the estimated probability of misclassification as the dependent variable to gain insights into the factors that may influence a firm’s resilience to shocks. Building on the predicted discriminant score for each firm, Zi , we will predict the likelihood of a firm falling into a different firm-size classification in terms of variables used in the discriminant analysis. The variable of interest is the misclassification indicator, yi , which has a binomial distribution: yi ∼ B (ni , pi ) (5) 7 where ni and pi are the trials and success probability parameters, respectively. We then analyze the variation in yi using the proposed Probit regression specification: yi = P r(misclassfication = 1) = Φ(AT β ) (6) where P r(misclassif ication = 1) is the probability of a firm being misclassified into a different firm size group and Φ is the cumulative density function of the standard normal distribution. A is a vector of standard normal distribution, which includes cash liquidity constraints, productivity, gender, and outlook dummies. Here, when the A’s are all zero, the interpretation from estimating the equation (through maximum likelihood procedure) will be for a baseline firm, which has no cash availability to pay for its immediate need, at the bottom of the labor productivity tercile, led by a male entrepreneur or manager, and the firm is highly pessimistic in its outlook towards the crisis. Similar to the above specification, the estimated parameters will then be translated into the marginal effect of a unit change in any of the X , in this case the change in any of the categorical or indicator variables, falling into different firm categories relative to the baseline firm. 3. Results This section presents results on the heterogeneity in the impact and recovery of micro-sized firms. 3.1 Master of Disasters: The Impact of the Pandemic on Micro-Sized Firms While large shocks impact all firms indiscriminately, the resulting economic downturn of the recession has a heterogeneous effect (Brucal et al., 2021; Apedo-Amah et al., 2020; Cirera et al., 2021).7 For example, an unanticipated increase in production costs results in a fall in profit margin of all firms. However, the magnitude of effect can be heightened for micro-sized firms due to differences in access to capital, market power, bargaining power and managerial ability. Yet, smaller or micro-sized firms may have greater operational flexibility allowing them to quickly adapt their production processes, explore alternative inputs, or target new markets to alleviate the effects of higher costs. Our data confirms the significant heterogeneity across firm size in the full sample – firms that are small (10-49 employees), medium (50-249 employees) and large (250+ employees) have significantly higher probability of remaining operational (Column 1 in panel (a) of Table 1) relative to micro-sized firms. Predicted margins suggest that micro-sized firms employing less than 10 workers had the lowest likelihood of remaining open during the pandemic with 0.86 probability, compared to 0.93 and 0.94 for the medium and large firms, respectively. 7 (For further reading, see Paunov, 2012 for the 2008 global financial crisis; Haykir et al., 2022 for the explosion of a crude oil bubble; and Liao, 2023 for the Russian Federation-Ukraine crisis with emphasis on energy firms. 8 Table 1: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery, by firm size. Size (a) All Sample (b) Micro-sized firms (no. of employees) Open Change in sales Arrears Recovery Open Change in sales Arrears Recovery (1) (2) (3) (4) (5) (6) (7) (8) Coefficients Coefficients 10-49 0.228*** 4.034*** 0.017 0.109 (0.014) (0.317) (0.012) (0.440) 50-249 0.489*** 10.613*** -0.177*** 0.549 (0.039) (0.697) (0.027) (0.941) 250+ 0.589*** 12.980*** -0.278*** -0.655 (0.038) (0.664) (0.026) (0.909) 5-9 0.120*** 2.485*** 0.061*** -0.292 (0.018) (0.439) (0.018) (0.628) Constant 0.860*** 19.181*** 1.021*** -17.222*** 0.715*** 52.598*** 1.256*** -7.402 9 (0.082) (4.518) (0.099) (4.102) (0.107) (8.687) (0.153) (5.683) Country-effects Yes Yes Yes Yes Yes Yes Yes Yes Survey-effects Yes Yes Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Observations 82412 57643 63189 29543 44564 32211 29808 15326 Predicted Margins Predicted Margins 0-9 0.86 -26.64 0.37 10.18 10-49 0.90 -22.60 0.38 10.29 50-249 0.93 -16.02 0.32 10.73 250+ 0.94 -13.66 0.29 9.53 0-4 0.85 -29.45 0.36 10.75 5-9 0.87 -26.96 0.38 10.46 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on firm size or no. of employees as the main predictor. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. The reference groups for all sample and micro-sized firms are those with 0-9 and 5-9 employees, respectively. Robust standard errors are in parentheses. *,**, *** denote significance level at 10, 5, and 1 percent, respectively. The trend in other outcomes tell a similar story. For example, change in sales is an increasing function of size column 2 in panel (a) of Table 1), implying that as firm size increases, the decline in sales becomes less severe. Specifically, relative to micro-sized firms (with 0-9 employees) the decline in sales for small, medium and large firms was respectively 4, 11 and 13 percentage point lower. Predicted margins show that micro-sized firms experienced the biggest decline in sales at approximately 27%, followed by small firms (10-49 employees) at 23%, medium-sized firms (50-249 employees) at 16%, and large firms (250+ employees) at 14%. The probability of falling into arrears exhibits a similar pattern with micro-sized and small firms being most disadvantaged (column 3 of Table 1). In contrast, we do not find significance difference in the recovery rate of firms across different size categories.8 While we included a number of controls such as country-, sector- and wave-specific effects, it is likely that there are unobserved time-varying factors that can affect both the within- and between- firm dynamics such as changes in management strategies and in inter-industry linkages associated with firm entry and exit. Robustness checks limiting the analysis to the first wave presented in Table A2 confirms the broad finding relating to the heterogeneity of effects by firm size. As expected, the estimated effects as illustrated in the predicted margins are monotonically higher than when we consider all waves, consistent with our hypothesis that the unobserved recovery efforts and adjustments had not took effect yet in this period. Notwithstanding, we still find the heterogeneity between firm groups that is comparable to that of the full sample. 3.2 Characteristics of the Real Masters of Disasters While our results in Table 1 confirm the findings from literature on crises that micro-sized firms are more impacted, it remains unclear whether there are significant differences within this broad size category. In this section, we unpack the impact of the crisis on micro-sized firms. 3.2.1 Firm Size While firm size matters across broad categories, results in panel (b) of Table 1 show that within micro-sized firms, the impact of the crisis varies distinctly across two groups – firms with 0-4 employees behave significantly different from those that are 5-9 employees. Such differences are particularly salient in firm’s drop in sales, and probability of remaining operational, and falling into arrears. In contrast, we do not find strong evidence that the recovery rates, are distinct between these two classes.9 Within each size grouping (0-4) and (5-9), we further split firms into quintiles of workers. Most outcomes are not statistically different within each of the two groups. One exception here are the single person firms that are far less likely to remain open relative to other quintiles within 0-4 employees group (See Table A3). 8 For the p-values on differences in the statistical significance of differences in coefficients, see Table B1. 9 See Table B1 for p-values of the pairwise difference in the predicted margins between size categories. 10 3.2.2 Liquidity The significant financial constraints faced by smaller or micro-sized firms contributes to their vulnerability during a crisis(Bartik et al., 2020). Without access to financial assistance or external credit, liquidity constraints can render such firms more susceptible to external shocks, increasing the likelihood of industry exit (Cefis et al., 2022). Maintaining adequate liquidity is crucial for these firms, particularly during times of crisis when their revenue streams may be disrupted. The BPS collected information on the number of days that allow firms to use their available cash to pay for its immediate financial obligations. Using this variable, we categorize firms into four groups of financial fragility: (1) firms with no days of coverage; (2) firms with some days but less than 30 days; (3) 30-120 days; and (4) firms with coverage greater than 120 days.10 Using the liquidity constraint categorical variable in the regression model specified in equations 1 and 2, we find that, first, firm’s resilience, as measured by our four outcome variables, is positively correlated with cash availability. For instance, relative to firms with zero days of cash, firms with some liquidity have a higher probability of being in operations, have a lower drop in sales, and lower probability of falling into arrears and a faster recovery rate (panel (a) of Table 2). The difference is statistically significant across various categories of liquidity constraints, even after controlling for firm sizes.Second, micro-sized firms behave in a way similar to the full sample of firms across the various categories of liquidity constraints (see panel (b) of Table 2. We observe a similar pattern even within the two categories of micro-sized firms, (0-4) and (5-9) employees (See Table A5). 3.2.3 Firm Productivity Lower productivity of smaller firms implies that such firms may have larger gaps in business information (Hartarska and Gonzalez-Vega, 2006; Bloom et al., 2014; De Loecker et al., 2022), lower investments in technologies or the ability to adapt their business models (Bartoloni et al., 2021), and hence may be at a competitive disadvantage during shocks (Hopenhayn, 1992; Melitz and Ottaviano, 2008; Bartoloni et al., 2021). Using sales per worker of firms categorized in terciles as proxy for firm productivity, results from estimating equations 1 and 2 illustrated in Table 3 suggest that firms with upper tercile of productivity distribution prior to the crisis are more likely to continue their operational status, experience lower drop in sales, and have a lower likelihood of falling into arrears. For instance, firms in upper tercile of productivity distribution are likely to register 5.4 percentage point lower drop in sales compare to the bottom tercile with the differences in the magnitude of impact being statistically significant (see Table B6). Micro-sized firms also display a similar pattern although the differences across productivity terciles are less sharper and insignificant for the probability of falling into arrears and recovery rate. 10 25% of micro-sized firms have 0 days of coverage, 40% have available cash that can last from some days but fewer than 30 days. Table A4 presents the distribution of liquidity constraints across micro-sized firms. 11 Table 2: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery, by liquidity constraint. Liquidity (a) All Sample (b) Microfirms constraint Open Change in sales Arrears Recovery Open Change in sales Arrears Recovery (no of days) (1) (2) (3) (4) (5) (6) (7) (8) Coefficients Coefficients 0120 days 1.081*** 18.104*** -0.775*** 1.702* 1.132*** 16.386*** -0.717*** 3.668*** (0.039) (0.650) (0.026) (0.953) (0.050) (0.810) (0.036) (1.195) Constant 0.254*** -35.503*** 1.422*** -20.231*** 0.184 -26.885*** 1.597*** -10.533* (0.096) (2.193) (0.105) (4.216) (0.125) (2.859) (0.165) (5.821) Country-effects Yes Yes Yes Yes Yes Yes Yes Yes 12 Survey-effects Yes Yes Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Yes Yes Size-effects Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Observations 61455 40316 50595 23376 32575 22165 23858 12446 Predicted Margins Predicted Margins 0 days 0.81 -35.28 0.47 5.41 0.92 -17.82 0.32 14.13 0120 days 0.96 -17.17 0.22 7.12 0.87 -26.62 0.40 9.39 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on liquidity constraint as the main predictor. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. The reference group comprises firms with zero days of cash (or those with no cash on hand). Robust standard errors are in parentheses. *,**, *** denote significance, Calculated p-values for the difference in predicted margins for each pairwise comparison categories are available upon request. Table 3: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery, by productivity tercile. Productivity (a) All Sample (b) Microfirms (tercile) Open Change in sales Arrears Recovery Open Change in sales Arrears Recovery (1) (2) (3) (4) (5) (6) (7) (8) Coefficients Coefficients middle 0.201*** 3.338*** -0.063*** -1.326** 0.193*** 2.932*** -0.028 -1.449 (0.020) (0.429) (0.016) (0.564) (0.028) (0.683) (0.025) (0.923) upper 0.320*** 5.439*** -0.123*** -0.038 0.287*** 3.762*** -0.025 -0.459 (0.021) (0.439) (0.016) (0.577) (0.030) (0.712) (0.026) (0.958) Constant 0.862*** 7.799 1.064*** -21.766*** 1.001*** 21.845** 1.376*** -20.064** (0.115) (5.074) (0.130) (5.543) (0.167) (9.634) (0.217) (8.622) Country-effects Yes Yes Yes Yes Yes Yes Yes Yes Survey-effects Yes Yes Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Yes Yes 13 Size-effects Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Observations 58666 41626 45511 21714 28802 21867 18946 10210 Predicted Margins Predicted Margins bottom 0.87 -26.62 0.40 9.39 0.85 -30.95 0.39 8.91 middle 0.90 -23.29 0.38 8.06 0.88 -28.02 0.38 7.46 upper 0.92 -21.19 0.36 9.35 0.90 -27.19 0.38 8.45 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on labor productivity (tercile) as the main predictor. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. The reference group comprises firms at the bottom tercile of the labor productivity distribution. Robust standard errors are in parentheses. *,**, *** denote significance, Calculated p-values for the difference in predicted margins for each pairwise comparison categories are available upon request. When distinguishing between the two groups of micro-sized firms, (0-4) and (5-9) employees, the impact of the crisis in terms of the firm’s resilience does not vary widely by firm productivity in the (0-4) category (except in the probability of remaining open), while firms within (5-9) employees continue to follow the patterns of the full sample (see Table A6). 3.2.4 Managerial Attitudes and Outlook While crises pose challenges to firms, they also present opportunities for innovation, and strategic decision-making (e.g., pivoting to new products and markets) that can enhance performance in the long term. Thus, managerial style in terms of attitudes and outlook, encompassing aspects such as long-term planning, risk preferences, and decision-making (Lin et al., 2005; Bloom and Van Reenen, 2007; Bandiera et al., 2018), are important factors in shaping how firms respond to crises.For instance, evidence from firms in the United Kingdom during economic downturns suggests the presence of significant heterogeneity in performance across firms emanating from managerial attitude and outlook – those with optimistic attitude are less likely to have abandoned or postponed investments that have more positive long-run repercussions on innovations and competitiveness (Ayyagari et al., 2011). Recent evidence on COVID-19 pandemic also points to the varied importance of the effect of firm’s outlook on performance, such that the risk of closure was found to be negatively associated with the expected length of the crisis Bartik et al. (2020). Given that BPS captures expectations on sales and employment under a regular scenario, we define outlook as "very pessimistic" if expected sales under this scenario is less than -50%. Taking the full distribution of expected sales within industries above -50%, we define outlook as "pessimistic" if expected sales are within the 1st quartile; "neutral" if expected sales is in the 2nd quartile; "optimistic" if expected sales is in the 3rd quartile; and "overly pessimistic" if expected sales is in the 4th quartile. The results from estimating equations 1 and 2 using firm outlook as the explanatory variable of interest suggest that firms in the left tail of sales expectations have remarkably the lowest probability of remaining open, the largest decline in sales, the highest probability of falling into arrears and the least recovery from the drop in sales (panel (a) of Table 4). Similar pattern is observed within the restricted sample of micro-sized firms (panel (b) of Table 4). The difference between very pessimistic firms and those that have neutral to positive outlook are remarkably significant, especially in terms of operational status, financial fragility and recovery rate. When splitting micro-sized firms into two groups (0-4) and (5-9) employees, we find that relative to the very pessimistic firms, other firms outperformed in all outcomes with the pattern being comparable across the two groups (See Table A7). 14 Table 4: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery, by managerial outlook. (a) All Sample (b) Microfirms Outlook Open Change in sales Arrears Recovery Open Change in sales Arrears Recovery (1) (2) (3) (4) (5) (6) (7) (8) Coefficients Coefficients pessimistic 0.590*** 9.968*** -0.442*** 9.678*** 0.549*** 4.966*** -0.352*** 9.818*** (0.035) (0.958) (0.034) (1.672) (0.046) (1.305) (0.045) (2.180) neutral 0.830*** 18.783*** -0.574*** 11.740*** 0.746*** 11.768*** -0.442*** 12.245*** (0.050) (1.035) (0.042) (2.059) (0.069) (1.457) (0.061) (2.897) optimistic 0.827*** 20.941*** -0.591*** 12.920*** 0.784*** 15.556*** -0.520*** 13.801*** (0.042) (1.033) (0.037) (1.757) (0.056) (1.425) (0.049) (2.329) very optimistic 0.751*** 24.518*** -0.586*** 14.481*** 0.689*** 19.327*** -0.424*** 15.395*** (0.039) (1.045) (0.036) (1.783) (0.051) (1.438) (0.048) (2.343) Constant 0.289*** -2.743 1.500*** -20.614*** 0.252* 9.288 1.582*** -12.247** 15 (0.104) (7.670) (0.107) (4.319) (0.140) (11.916) (0.165) (6.044) Country-effects Yes Yes Yes Yes Yes Yes Yes Yes Survey-effects Yes Yes Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Yes Yes Size-effects Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Observations 42845 28471 40659 16798 23013 16202 21708 8912 Predicted Margins Predicted Margins very pessimistic 0.80 -33.78 0.53 1.78 0.78 -29.59 0.47 1.88 pessimistic 0.91 -23.81 0.38 11.46 0.89 -24.62 0.35 11.70 neutral 0.93 -14.99 0.34 13.52 0.92 -17.82 0.32 14.13 optimistic 0.93 -12.84 0.33 14.70 0.92 -14.03 0.30 15.68 very optimistic 0.93 -9.26 0.33 16.26 0.91 -10.26 0.33 17.28 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on labor productivity (tercile) as the main predictor. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. The reference group comprises firms with very pessimistic manager’s outlook. Robust standard errors are in parentheses. *,**, *** denote significance, Calculated p-values for the difference in predicted margins for each pairwise comparison categories are available upon request. 3.2.5 Managerial Ability In times of crisis, managerial ability to predict reasonably well can be critical in guiding firm strategies and decision-making. Inaccurate forecasts, managerial overconfidence and optimism bias, for example, can result in misallocation of resources (Flyvbjerg and Bester, 2021), inadequate cost planning (Love et al., 2019a,b), and ill-informed investment decisions (Ben-David et al., 2013; Kim et al., 2016; Mohamed et al., 2020), ultimately leading to financial losses (Lee et al., 2019; Mohamed et al., 2020). We construct a measure of prediction error by comparing the expected change in sales collected by the BPS with the actual change in sales in the subsequent round for firms responding to at least two rounds of BPS. We group firms into three categories based on their position in the distribution of prediction error defined as the absolute value of the difference between expected change in sales under normal condition and the realized or observed change, within each sector-wave bin. The three categories are: "low" if prediction error is below the 25th percentile; "median" if within the inter-quartile range; and "high" if above the 75th percentile. Augmenting our regression analysis in equations 1 and 2 with prediction error measures, estimations presented in Table 5 suggest that firms with higher margins of prediction error in sales forecasts are correlated with negative outcomes on all front – including the lowest probability of remaining operational, heightened drop in sales, higher probability of falling into arrears and slower recovery. Firms with the lowest prediction errors were more resilient relative to the median baseline category and experienced a higher rate of recovery perhaps because firms with better managerial ability have a clearer understanding of market dynamics and are able to navigate through the crisis (see related work by Grover and Karplus, 2021). The pattern observed within the restricted sample of micro-sized firms is much sharper (panel (b) of Table 5). The difference between firms with low prediction error and those with high error is wider and significant in all outcomes, except for recovery rates. Within micro-sized firms, the impact of the crisis in terms of the firm’s resilience in the (0-4) employee category has only modest variation by managerial ability across different outcomes, while firms in (5-9) employees do not show much variation (see Table A8). 16 Table 5: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery, by managerial ability (as measured by prediction error). Prediction (a) All Sample (b) Microfirms error Open Change in sales Arrears Recovery Open Change in sales Arrears Recovery (1) (2) (3) (4) (5) (6) (7) (8) Coefficients Coefficients low 0.133** 7.387*** -0.170*** -1.664 0.166** 6.640*** -0.198*** -0.532 (0.056) (1.025) (0.036) (1.290) (0.074) (1.516) (0.058) (2.178) high -0.158*** -2.993*** 0.093*** -3.357** -0.118** -1.465 0.096* -3.933 (0.045) (1.056) (0.036) (1.609) (0.058) (1.444) (0.052) (2.432) Constant 2.282** -59.146*** -0.371 -39.275 2.891** 5.055 -1.771* 30.107 (1.055) (10.433) (0.638) (27.122) (1.437) (17.630) (0.934) (39.137) Country-effects Yes Yes Yes Yes Yes Yes Yes Yes 17 Survey-effects Yes Yes Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Yes Yes Size-effects Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Observations 10089 6817 9142 3528 4697 3694 3946 1471 Predicted Margins Predicted Margins median 0.89 -37.61 0.45 9.63 0.84 -39.54 0.47 9.07 low 0.91 -30.22 0.40 7.97 0.87 -32.90 0.41 8.53 high 0.87 -40.60 0.49 6.28 0.82 -41.01 0.51 5.13 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on prediction error as the main predictor. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. The reference group comprises firms at the median tercile of the prediction error distribution. Robust standard errors are in parentheses. *,**, *** denote significance, Calculated p-values for the difference in predicted margins for each pairwise comparison categories are available upon request. 3.3 Misclassification of Micro-Sized Firms Many developing countries define their unorganized and informal sector based on the cut-off number of employees. In countries such as Burkina Faso, Colombia, Costa Rica, El Savlador, Niger, South Africa, Uruguay, and Zambia firms with up to 5 employees can qualify to remain out of the tax net and hence informal (OECD, 2019), while other countries such as Honduras (OECD, 2019), India, and Pakistan use the 10 employee cut-off (Williams et al., 2016). Other countries such as Chile, Ghana, Madagascar, and Paraguay define informality with a threshold of 6 workers (OECD, 2019). Given the wide range of size cut-offs that are not necessarily related to income per capita or the firm size distribution in the country, there could be potential concerns of misclassification of firms in "wrong" size groupings that do not reflect their innate capabilities to grow or to cope with the crisis. This is critical because it affects not only the government’s ability to appropriately target firms for taxation but also for public support policies that affects the incentives for firms to grow.11 Building upon the work of Bruhn (2013), we use discriminant analysis for assessing the misclassi- fication of micro-sized firms. We examine the differences between certain firms within a specific group in terms of their resilience, while also considering their similarities to firms in other groups. We deploy the available impact metrics considered earlier in this paper to group firms based on similarities in these metrics, prioritizing an analysis of firm-level resilience over firm-level attributes. This is different from Bruhn (2013) whose classifying of firms into specific groups is based on an optimal combination of detailed firm-level characteristics. Using the method described in Section 2, we calculate discriminant scores, Z , specified in equation 3 using residualized variables as described in equation 4 to account for potential confounders such as age-, wave-, country- and sector-specific effects, and the pre-pandemic sales level. This data intensive exercise left us with a total sample of up to 38,657, and about 68% have complete observations for the factors included in the analysis. Inclusion of recovery rates further reduces our sample as it necessitates a firm to be observed at least twice. Given that we are interested in potential misclassification of micro-sized firms, we limit our sample to firms with less than 49 employees (over 90% of full sample) because misclassification can reasonably be only considered across the following categories (0-4), (5-9) and (10- 49) due to the observed proximity of impacts and recovery rates among these firms relative to bigger ones such that the managerial strategies and coping mechanisms could potentially be comparable (Lighthouse, 2021). Despite the data intensive method and limited metrics pertaining to firm-level impact of the crisis, our specification correctly classifies 44%, 25% and 38% of firms in (0-4), (5-9) and (10-49) categories, respectively. For comparison, the richer set of characteristics used by Bruhn (2013) allowed for the classifying of up to 65% percent of the groups correctly. For the micro-sized firms in the (0-4) and (5-9) employees, 35% and 36% respectively have shown resilience that is comparable to small firms 11 Several countries implement size-dependent policies that support or regulate activities of firms based on their size, in particular, their number of employees. Some examples of such policies include labor regulation policies in France, tax enforcement rules in Spain, and small-scale firm protection in India, and small-and-medium enterprises (SMEs) support policies in the Republic of Korea. Evidence suggests that size-dependent policies can potentially accentuate misallocation of resources (Garicano et al., 2016; Dabla-Norris et al., 2018; Bertrand et al., 2021). 18 with (10-49) employees firms (see Table 6). By comparison, nearly 40% of firms with (5-9) and (10-49) employees show resilience comparable to firms with (0-4) employees. These firms are at the margin of either graduating to non-micro-sized firms or reverting back to the micro-sized group.12 Table 6: Results of Discriminant Analysis: Impacts Predicted Category True Category Total 0-4 5-9 10-49 0-4 2,655 1,312 2,113 6,080 43.67 21.58 34.75 100 5-9 2,588 1,587 2,304 6,479 39.94 24.49 35.56 100 10-49 5,188 3,174 5,213 13,575 38.22 23.38 38.4 100 10,431 6,073 9,630 26,134 Total 39.91 23.24 36.85 100 Notes: Figures shown are the number of firms and their respective share to total for each classification using dis- criminant analysis specified in equation 3; Authors’ cal- culation. 3.4 Correlates of Misclassification In this section, we examine the factors that are associated with the misclassification of micro-sized firms in (5-9) employees group by estimating equation 6, with the results presented in Table 7. We focus on micro-sized firms in the (5-9 employees group because they are primarily at the margin of graduating to non-micro-sized firms (10-49) employees or reverting back to the smallest group of (0-4) employees.A positive coefficient implies higher probability of misclassification relative to when all indicator variables is set to zero. Our results suggest that firms in the (5-9) group have a higher probability of being misclassified as (10-49) firms monotonically as liquidity constraints relax, when firm productivity is in the highest tercile and when managerial outlook is anything better than least pessimistic. By comparison, the misclassification in the (0-4) category decreases when firms are in the highest productivity tercile while the relationship with liqiduity is non-monotonic. The probability of misclassification in (0-4) employee group is lower when cash increases between 30 to 120 days, however, the risk of misclassification in (0-4) employees category increases if cash exceeds 120 days reflecting the possible under-utilization of liquid resources.13 . 12 When considering both impact and recovery metrics, our results remain similar (see Annex Table A9). 13 Notice that about 14% of firms in both the (0-4) and (5-9) employees group has cash-days>120, see Annex Table A4. 19 Table 7: Results of estimating Probit regression on the event of a (5-9)-sized firm being misclassified into a different size category. (a) Without Recovery (b) With Recovery Covariates size = (0-4) size = (10-49) size = (0-4) size = (10-49) (1) (2) (3) (4) (0 < cash-days <=30)=1; 0 else -0.073 0.164** -0.478*** 0.305** (0.074) (0.078) (0.121) (0.132) (30< cash-days <=120)=1; 0 else -0.176* 0.241** -0.728*** 0.792*** (0.097) (0.099) (0.169) (0.171) (cash-days >120)=1; 0 else 0.302*** 0.251*** 0.079 0.265* (0.081) (0.085) (0.136) (0.146) (sales/employment = 2nd tercile)=1; 0 else -0.020 -0.024 0.059 -0.001 (0.065) (0.067) (0.103) (0.107) (sales/employment = 3rd tercile)=1; 0 else -0.128* 0.161** -0.249** 0.348*** (0.067) (0.068) (0.110) (0.110) 20 (outlook is pessemistic) = 1; 0 else 0.152 0.706*** -0.146 0.934* (0.140) (0.189) (0.344) (0.539) (outlook is neutral) = 1; 0 else -0.218 0.903*** -0.963* 1.122* (0.165) (0.205) (0.502) (0.635) (outlook is optimistic) = 1; 0 else -0.060 0.947*** -0.543 1.281** (0.150) (0.196) (0.353) (0.543) (outlook is very optimistic) = 1; 0 else -0.185 0.986*** -0.629* 1.363** (0.148) (0.194) (0.360) (0.547) Constant -0.204 -1.428*** 0.535 -1.912*** (0.144) (0.197) (0.356) (0.569) N 2220 2220 883 883 Notes: The figures presented represent the parameter estimates obtained by running the probit model specified in equation 6 on firms belonging to a specific size category in Table 6. The top heading of each regression corresponds to the true size category, while each regression provides the results of estimating the probability of a firm being misclassified into a different size category. 4. Conclusions Using the five waves of the the World Bank Business Pulse survey data covering the period 2020-2022, we assess the impact of COVID-19 on the full sample relative to micro-sized firms. Thereafter, we focus on two groups within the micro-sized firms: (0-4) employees and (5-9) employees. Although the micro-sized firms are most affected by the crisis, our research reveals significant heterogeneity within the micro-sized firms that vary by firm liquidity constraints, productivity, managerial attitudes and managerial ability. Within the two groups of micro-sized firms, liquidity position and managerial attitudes important predictors of resilience to shocks. By comparison, firm size variation within each group of micro-sized firms is not a good predictor of resilience to crisis. Finally, the limited variation in productivity for firms with (0-4) employees also makes the group uniform in response to the crisis, while the resilience on the group with (5-9) employees is invariant to managerial ability to reasonably predict future sales. Given that some micro-sized firms can exhibit patterns of resilience similar to other firms, we explore the possibility of misclassification of firms into different size categories based on resilience and recovery. This is particularly confirmed among (5-9) employee micro-sized firms where more than a third of these firms are prone to being misclassified as firms with (0-4) employees, while a quite similar proportion falls into the (10-49) employees firm category. Factors such as higher liquidity, and better managerial outlook ability can enable smaller firms to exhibit resilience comparable to larger ones. There are three broad policy lessons from our nuanced results on the resilience of micro-sized firms during the crisis. First, size-based policies to support smaller firms especially at the time of the crisis may not be appropriate given the arbitrariness of the size cut-offs and the differences in resilience even within the narrowly defined micro-sized firm groupings. Second, our results call for a targeted support to enhance business resilience. For all micro-sized firms, financial and psychological resilience contribute to a firm’s ability to withstand challenges. In this context, financial institutions can play a critical role in leveling the playing field for both small and large firms during a crisis, while business development agencies can work with firms on building a positive and growth- oriented outlook especially during crisis. Third, given that there are differences within micro-sized firms, for instance, firms with (5-9) employees can potentially be as resilient as those with (10-49) employee firms, the type of support should be more tailored and perhaps focus on efforts to enhance productivity (e.g., management practices, innovation and technology adoption) and those that work with entrepreneur’s’ psychology (e.g., personal initiative skills training). 21 References H. Almeida, M. Campello, and M. S. Weisbach. The cash flow sensitivity of cash. The Journal of Finance, 59(4):1777–1804, 2004. M. Amin and A. M. Islam. Does manager education play a role in the productivity of informal firms in developing economies? evidence from firm-level surveys. Review of Development Economics, 26 (2):962–984, 2022. M. C. Apedo-Amah, B. Avdiu, M. X. Cirera, E. Davies, A. Grover, L. Iacovone, U. Kilinc, D. Medvedev, F. Maduko, S. Poupakis, et al. Businesses through the covid-19 shock: Firm-level evidence from around the world. Policy Research Working Paper, 9434, 2020. D. Atkin, A. K. Khandelwal, and A. Osman. Exporting and firm performance: Evidence from a randomized experiment. The Quarterly Journal of Economics, 132(2):551–615, 2017. M. Ayyagari, A. Demirgüç-Kunt, and V. Maksimovic. Firm innovation in emerging markets: The role of finance, governance, and competition. Journal of Financial and Quantitative Analysis, 46(6): 1545–1580, 2011. O. Bandiera, R. Lemos, A. Prat, and R. Sadun. Managing the family firm: Evidence from ceos at work. The Review of Financial Studies, 31(5):1605–1653, 2018. A. W. Bartik, M. Bertrand, Z. Cullen, E. L. Glaeser, M. Luca, and C. Stanton. The impact of covid-19 on small business outcomes and expectations. Proceedings of the National Academy of Sciences, 117 (30):17656–17666, 2020. E. Bartoloni, A. Arrighetti, and F. Landini. Recession and firm survival: is selection based on cleansing or skill accumulation? Small Business Economics, 57:1893–1914, 2021. I. Ben-David, J. R. Graham, and C. R. Harvey. Managerial miscalibration. The Quarterly Journal of Economics, 128(4):1547–1584, 2013. A. N. Berger and G. F. Udell. The economics of small business finance: The roles of private equity and debt markets in the financial growth cycle. Journal of banking & finance, 22(6-8):613–673, 1998. M. Bertrand, C.-T. Hsieh, and N. Tsivanidis. Contract labor and firm growth in india. Technical report, National Bureau of Economic Research, 2021. N. Bloom and J. Van Reenen. Measuring and explaining management practices across firms and countries. The Quarterly Journal of Economics, 122(4):1351–1408, 2007. N. Bloom, R. Lemos, R. Sadun, D. Scur, and J. Van Reenen. Jeea-fbbva lecture 2013: The new empirical economics of management. Journal of the European Economic Association, 12(4):835–876, 2014. N. Bloom, P. Bunn, P. Mizen, P. Smietanka, and G. Thwaites. The impact of covid-19 on productivity. Technical report, National Bureau of Economic Research, 2020. 22 J.-C. Bricongne, L. Fontagné, G. Gaulier, D. Taglioni, and V. Vicard. Firms and the global crisis: French exports in the turmoil. Journal of International Economics, 87(1):134–146, 2012. A. Brucal, A. Grover, and S. Reyes Ortega. Damaged by the disaster. 2021. M. Bruhn. A tale of two species: Revisiting the effect of registration reform on informal business owners in mexico. Journal of Development Economics, 103:275–283, 2013. E. Cefis, C. Bettinelli, A. Coad, and O. Marsili. Understanding firm exit: a systematic literature review. Small Business Economics, 59(2):423–446, 2022. G. Chodorow-Reich. The employment effects of credit market disruptions: Firm-level evidence from the 2008–9 financial crisis. The Quarterly Journal of Economics, 129(1):1–59, 2014. X. Cirera, M. Cruz, E. Davies, A. Grover, L. Iacovone, J. E. L. Cordova, D. Medvedev, F. O. Maduko, G. Nayyar, S. Reyes Ortega, et al. Policies to support businesses through the covid-19 shock: A firm level perspective. The World Bank Research Observer, 36(1):41–66, 2021. C. Constantinescu, A. M. Fernandes, S. Poupakis, A. G. Goswami, and S. Reyes. Globally engaged firms in the covid-19 crisis. 2022. M. Cowling, W. Liu, and A. Ledger. Small business financing in the uk before and during the current financial crisis. International Small Business Journal, 30(7):778–800, 2012. M. Cowling, W. Liu, A. Ledger, and N. Zhang. What really happens to small and medium-sized enterprises in a global economic recession? UK evidence on sales and job dynamics. International Small Business Journal, 33(5):488–513, 2015. W. V. Cunningham and W. F. Maloney. Heterogeneity among Mexico’s microenterprises: An application of factor and cluster analysis. Economic Development and Cultural Change, 50(1):131– 156, 2001. M. E. Dabla-Norris, L. Jaramillo, F. Lima, and A. Sollaci. Size dependent policies, informality and misallocation. International Monetary Fund, 2018. J. De Loecker, T. Obermeier, and J. Van Reenen. Firms and inequalities. Institute for Fiscal Studies, 1, 2022. H. Doukas, P. Xidonas, and N. Mastromichalakis. How successful are energy efficiency investments? a comparative analysis for classification & performance prediction. Computational Economics, pages 1–20, 2021. M. Doumpos and C. Zopounidis. Multicriteria decision aid classification methods, volume 73. Springer Science & Business Media, 2002. A. Fernández-Cerezo, B. González, M. Izquierdo Peinado, and E. Moral-Benito. Firm-level hetero- geneity in the impact of the covid-19 pandemic. Applied Economics, pages 1–29, 2022. 23 G. G. Fiseha and A. A. Oyelana. An assessment of the roles of small and medium enterprises (SMEs) in the local economic development (LED) in South Africa. Journal of Economics, 6(3):280–290, 2015. B. Flyvbjerg and D. W. Bester. How (in) accurate is cost-benefit analysis? data, explanations, and suggestions for reform. Data, Explanations, and Suggestions for Reform (January 15, 2022). Infrastructure Economics and Policy: International Perspectives (Cambridge, MA: Lincoln Institute of Land Policy), page 174, 2021. E. Fontainha and E. Lazzaro. Cultural and creative entrepreneurs in financial crises: Sailing against the tide? Scientific Annals of Economics and Business, 66:73–89, 2019. R. Frisch and F. V. Waugh. Partial time regressions as compared with individual trends. Econometrica: Journal of the Econometric Society, pages 387–401, 1933. L. Garicano, C. Lelarge, and J. Van Reenen. Firm size distortions and the productivity distribution: Evidence from france. American Economic Review, 106(11):3439–3479, 2016. M. Gertler and S. Gilchrist. Monetary policy, business cycles, and the behavior of small manufacturing firms. The Quarterly Journal of Economics, 109(2):309–340, 1994. B. Ghojogh and M. Crowley. Linear and quadratic discriminant analysis: Tutorial. arXiv preprint arXiv:1906.02590, 2019. M. Greenstone, A. Mas, and H.-L. Nguyen. Do credit market shocks affect the real economy? quasi- experimental evidence from the great recession and" normal" economic times. American Economic Journal: Economic Policy, 12(1):200–225, 2020. A. Grover and V. J. Karplus. Coping with COVID-19: Does Management Make Firms More Resilient? Policy Research Working Paper No. 9514, 2021. T. Hale, N. Angrist, R. Goldszmidt, B. Kira, A. Petherick, T. Phillips, S. Webster, E. Cameron-Blake, L. Hallas, S. Majumdar, et al. A global panel database of pandemic policies (Oxford COVID-19 Government Response Tracker). Nature human behaviour, 5(4):529–538, 2021. V. Hartarska and C. Gonzalez-Vega. What affects new and established firms’ expansion? Evidence from small firms in Russia. Small Business Economics, 27:195–206, 2006. O. Haykir, I. Yagli, E. D. A. Gok, and H. Budak. Oil price explosivity and stock return: Do sector and firm size matter? Resources Policy, 78:102892, 2022. H. A. Hopenhayn. Entry, exit, and firm dynamics in long run equilibrium. Econometrica: Journal of the Econometric Society, pages 1127–1150, 1992. A. B. Kennickell, M. L. Kwast, and J. Pogach. Small businesses and small business finance during the financial crisis and the great recession: New evidence from the survey of consumer finances. In Measuring Entrepreneurial Businesses: Current Knowledge and Challenges, pages 291–349. University of Chicago Press, 2016. 24 J.-B. Kim, Z. Wang, and L. Zhang. Ceo overconfidence and stock price crash risk. Contemporary Accounting Research, 33(4):1720–1749, 2016. E. Laderman et al. Small businesses hit hard by weak job gains. FRBSF Economic Letter, 26, 2013. Y.-C. Lee, Y.-C. Lu, and Y.-C. Wang. Corporate social irresponsibility, ceo overconfidence, and stock price crash risk. Applied Economics Letters, 26(14):1143–1147, 2019. S. Liao. The Russia–Ukraine outbreak and the value of renewable energy. Economics Letters, 225: 111045, 2023. Lighthouse. Why everything breaks when you reach 25 employees, 2021. URL https://getlighthouse. com/blog/company-growth-25-employees/. Y.-h. Lin, S.-Y. Hu, and M.-s. Chen. Managerial optimism and corporate investment: Some empirical evidence from Taiwan. Pacific-Basin Finance Journal, 13(5):523–546, 2005. G. B. Lockhart. Credit lines and leverage adjustments. Journal of Corporate Finance, 25:274–288, 2014. P. E. Love, L. A. Ika, and M. C. Sing. Does the planning fallacy prevail in social infrastructure projects? empirical evidence and competing explanations. IEEE transactions on engineering management, 69 (6):2588–2602, 2019a. P. E. Love, M. C. Sing, L. A. Ika, and S. Newton. The cost performance of transportation projects: The fallacy of the planning fallacy account. Transportation Research Part A: Policy and Practice, 122: 1–20, 2019b. M. C. Lovell. A simple proof of the fwl theorem. The Journal of Economic Education, 39(1):88–91, 2008. F. Manzoor, L. Wei, and M. Siraj. Small and medium-sized enterprises and economic growth in pakistan: An ardl bounds cointegration approach. Heliyon, 7(2):e06340, 2021. L. A. Martin, S. Nataraj, and A. E. Harrison. In with the big, out with the small: Removing small-scale reservations in india. American Economic Review, 107(2):354–386, 2017. E. Mathieu, H. Ritchie, L. Rodés-Guirao, C. Appel, C. Giattino, J. Hasell, B. Macdonald, S. Dattani, D. Beltekian, E. Ortiz-Ospina, and M. Roser. Coronavirus pandemic (covid-19). Our World in Data, 2020. https://ourworldindata.org/coronavirus. D. McKenzie. Small business training to improve management practices in developing countries: re-assessing the evidence for ‘training doesn’t work’. Oxford Review of Economic Policy, 37(2): 276–301, 2021. D. McKenzie and C. Woodruff. Business practices in small firms in developing countries. Management Science, 63(9):2967–2981, 2017. M. J. Melitz and G. I. Ottaviano. Market size, trade, and productivity. The Review of Economic Studies, 75(1):295–316, 2008. 25 K. E. Meyer, S. Prashantham, and S. Xu. Entrepreneurship and the post-covid-19 recovery in emerging economies. Management and Organization Review, 17(5):1101–1118, 2021. doi: 10.1017/mor.2021.49. E. B. Mohamed, N. Garoui, and K. Naoui. Do optimistic managers destroy firm value? Journal of Behavioral and Experimental Finance, 26:100292, 2020. OECD. Tackling vulnerability in the informal economy, 2019. T. Opler, L. Pinkowitz, R. Stulz, and R. Williamson. The determinants and implications of corporate cash holdings. Journal of financial economics, 52(1):3–46, 1999. C. Paunov. The global crisis and firms’ investments in innovation. Research policy, 41(1):24–35, 2012. M. Rotemberg. Equilibrium effects of firm subsidies. American Economic Review, 109(10):3475–3513, 2019. A. Sahin, S. Kitao, A. Cororaton, and S. Laiu. Why small businesses were hit harder by the recent recession. Current Issues in Economics and Finance, 17(4), 2011. D. Smallbone, D. Deakins, M. Battisti, and J. Kitching. Small business responses to a major economic downturn: Empirical perspectives from New Zealand and the United Kingdom. International Small Business Journal, 30(7):754–777, 2012. T. Tambunan. Recent evidence of the development of micro, small and medium enterprises in indonesia. Journal of Global Entrepreneurship Research, 9(1):18, 2019. UNCTAD. Supporting small businesses is critical for covid-19 recovery. https://unctad.org/news/ supporting-small-businesses-critical-covid-19-recovery, 2021. C. C. Williams, M. S. Shahid, and A. Martínez. Determinants of the level of informality of informal micro-enterprises: Some evidence from the city of Lahore, Pakistan. World Development, 84:312–325, 2016. 26 Appendices .0.1 Description of the survey and the data set The World Bank Group (WBG) has developed a brief firm survey instrument to collect data mea- suring the impact of the COVID-19 pandemic on the private sector. The questionnaire includes the following dimensions: operations of the business, sales, liquidity and insolvency, labor adjustments, firms’ responses, expectations and uncertainty about the future, and preferred mechanisms of public support. In most countries interviews were conducted over the phone, but in a few countries such as Colombia or Türkiye, the questionnaire was administered online. In 31 of these countries, a fresh sample of businesses was collected and the survey was implemented in collaboration with private sector associations, statistical agencies, and other government agencies (mainly Ministries of Finance and Economy). Data for the remaining 20 countries were collected as a follow-up to the World Bank Enterprise Survey, using a questionnaire that excluded some questions from the standard version. The two instruments, the standard pulse survey and the Enterprise Survey follow-up, were implemented in Togo. In Bangladesh, the standard pulse survey was implemented on different samples and at different times of the shock. The survey instrument differed across countries but in most cases the Enterprise Survey COVID-19 follow-up excludes some questions on the adjustment to employment and the channels affecting the operations of the business, the module on expectations, and most questions on the adoption of technology as a response to the crisis. The sampling frame in most countries where the pulse survey was not a follow-up of the Enterprise Survey was based on censuses from Statistics Agencies, Ministries of Finance or Economy, or business listings from Business Associations, and typically only included registered businesses. In the case of the Enterprise Survey, by design the implementation covers only formal firms. Only Cambodia, Gabon, Ghana, Pakistan, the Philippines, Senegal, South Africa, Sudan, and Tunisia include informal firms in their sample. .0.2 Data harmonization and cleaning The analysis excludes observations of businesses contacted but that reported their status as per- manently closed at the time of the interview. We also exclude businesses in Education and Health services. The implementation of the survey in some countries presented the respondent with a different menu of options for the status of the operations of the business and the adjustments to their labor force on the intensive and the extensive margin. We group open and partially open businesses into one category; and temporarily closed by mandate and choice into a second one. Similarly, we group plants that granted leave without pay and with pay into one group. 27 Change in sales is only available for businesses open or partially open, or that closed less than 4 weeks prior to the time of the survey (temporarily or permanently). We set change in sales -100 for businesses that have been temporarily closed for more than four weeks at the time of the interview. For comparison purposes, size and sector in each country is obtained from the pulse survey data, even if in some countries these variables are available from the sampling frame. In some countries where the survey was a follow-up from the Enterprise Survey, size excludes part-time workers. To compute the percentage change in employment, we subtract workers laid off from workers hired, but we exclude observations with measurement error in the question on workers hired (number identical or higher than the size of the firm). We trim the top 1% in the number of workers hired and in the percentage change in sales relative to the same period of last year. We also trim the top and bottom 2% in the predicted changes to sales in the three scenarios (pessimistic, regular, optimistic). To study expectations and uncertainty, we only use subjective probability distributions where the probabilities for the three scenarios total 100. 28 A.1. Annex A Figure A.1: Survey Implementation and Google mobility trends around transit stations. AFG BGD 100 100 0 0 80 80 lockdown measures lockdown measures -20 -20 index (baseline = 0) index (baseline = 0) Google mobility Google mobility Stringency of Stringency of 60 60 -40 -40 -60 40 -60 40 -80 20 -80 20 -100 0 -100 0 03/01 04/01 05/01 06/01 07/01 08/01 03/01 04/01 05/01 06/01 07/01 08/01 IND LKA 100 100 0 0 80 80 lockdown measures lockdown measures -20 -20 index (baseline = 0) index (baseline = 0) Google mobility Google mobility Stringency of Stringency of 60 60 -40 -40 -60 40 -60 40 -80 20 -80 20 -100 0 -100 0 03/01 04/01 05/01 06/01 07/01 08/01 03/01 04/01 05/01 06/01 07/01 08/01 NPL PAK 100 100 0 0 80 80 lockdown measures lockdown measures -20 -20 index (baseline = 0) index (baseline = 0) Google mobility Google mobility Stringency of Stringency of 60 60 -40 -40 -60 40 -60 40 -80 20 -80 20 -100 0 -100 0 03/01 04/01 05/01 06/01 07/01 08/01 03/01 04/01 05/01 06/01 07/01 08/01 Stringency of Mobility around Period of interview lockdown transit stations restrictions This figure shows how BPS was implemented at varying points of mobility and lock-down. Gray area represents the time of the survey implementation. Table A1: Summary Statistics Variables N Mean Std. Dev. Minimum Maximum Micro-sized = 1, 0 else 185,802 0.49 0.50 0 1 open =1; 0 else 164,792 0.98 0.14 0 1 Change in sales 105,297 - 9.58 31.35 -100 70 in arrears =1, 0 else 119,626 0.29 0.46 0 1 Delta change in sales_t,t-1 44,907 13.28 32.41 -80 110 Size category 144,573 1.63 0.75 1 4 Observations Unique values Firm ID 144,573 104,706 Sector 139,035 12 Survey wave 185,802 5 Country 185,802 86 Source: Authors’ calculations using data from the World Bank Business Pulse Survey. 29 Table A2: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery, wave 1 only, by size categories Size (a) All Sample (b) Micro-sized firms Category Open Change in sales Arrears Open Change in sales Arrears (no. of employees) (1) (2) (3) (4) (5) (6) Coefficients Coefficients 10-49 0.209*** 3.635*** -0.105*** (0.027) (0.514) (0.027) 50-249 0.566*** 11.889*** -0.387*** (0.077) (1.200) (0.055) 250+ 0.610*** 13.607*** -0.546*** (0.069) (1.075) (0.054) 5-9 0.056 2.585*** 0.009 (0.035) (0.721) (0.052) Constant -0.571 -25.427*** -0.114 -0.232 -18.352 -1.119 (0.455) (7.094) (0.404) (0.709) (12.784) (0.773) Country-effects Yes Yes Yes Yes Yes Yes 30 Survey-effects Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Observations 21483 22277 13077 11589 12476 4581 Predicted Margins Predicted Margins 0-9 0.83 -44.46 0.45 10-49 0.87 -40.83 0.41 50-249 0.93 -32.58 0.32 250+ 0.93 -30.86 0.28 0-4 0.83 -47.40 0.47 5-9 0.85 -44.82 0.47 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on size category as the main predictor. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. The reference groups for all sample and micro-sized firms are those with 0-9 and 0-4 employees, respectively. Robust standard errors are in parentheses. *,**, *** denote significance, Calculated p-values for the difference in predicted margins for each pairwise comparison categories are available upon request. Table A3: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery for micro-sized firms, by size categories (in quintile) Size (a) 0-4 Firms (b) 5-9 Firms (quintile) Open Change in sales Arrears Recovery Open Change in sales Arrears Recovery (1) (2) (3) (4) (5) (6) (7) (8) Coefficients Coefficients 1st (base) 2nd 0.211*** -0.913 0.058 -0.424 0.095** -0.117 -0.021 0.624 (0.033) (0.816) (0.037) (1.200) (0.037) (0.835) (0.033) (1.162) 3rd 0.247*** -0.679 0.075** -1.587 0.112*** -0.491 0.011 0.646 (0.032) (0.812) (0.034) (1.194) (0.041) (0.902) (0.035) (1.244) 4th 0.313*** 0.810 0.111*** -0.487 0.142*** 1.594* 0.010 1.470 (0.034) (0.865) (0.037) (1.268) (0.040) (0.874) (0.034) (1.214) 5th 0.298*** -0.118 0.074* -0.286 0.240*** 1.257 -0.076* 0.626 31 (0.036) (0.883) (0.038) (1.305) (0.052) (1.053) (0.042) (1.464) Constant 0.224* 82.281*** 1.395*** 7.887** 1.219*** 31.277*** 1.149*** -30.320*** (0.136) (15.369) (0.219) (3.380) (0.188) (10.828) (0.217) (10.169) Country-effects Yes Yes Yes Yes Yes Yes Yes Yes Survey-effects Yes Yes Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Observations 27419 20099 16298 9357 16858 12022 13421 5930 Predicted margins Predicted margins 1st 0.80 -30.34 0.34 11.98 0.86 -25.35 0.38 8.76 2nd 0.85 -31.25 0.36 11.56 0.88 -25.47 0.38 9.38 3rd 0.86 -31.02 0.36 10.39 0.88 -25.84 0.39 9.40 4th 0.87 -29.53 0.37 11.49 0.88 -23.76 0.39 10.23 5th 0.87 -30.46 0.36 11.69 0.90 -24.09 0.36 9.38 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on size category (in quintiles) as the main predictor. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. Robust standard errors are in parentheses. *,**, *** denote significance, Calculated p-values for the difference in predicted margins for each pairwise comparison categories are available upon request. Table A4: Distribution of observations according to liquidity constraints and classes of micro-sized firms. Liquidity Constraints Categories Size Total days = 0 0120 0-4 9,546 13,177 4,977 4,510 32,210 29.64 40.91 15.45 14 100 5-9 3,646 9,332 3,605 2,670 19,253 18.94 48.47 18.72 13.87 100 13,192 22,509 8,582 7,180 51,463 Total 25.63 43.74 16. 68 13.95 100 Source: Authors’ calculation using data from the World Bank Business Pulse Survey. 32 Table A5: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery for micro-sized firms, by liquidity constraints Liquidity (a) 0-4 employees (b) 5-9 employees constraints Open Change in sales Arrears Recovery Open Change in sales Arrears Recovery (no. of days) (1) (2) (3) (4) (5) (6) (7) (8) Coefficients Coefficients 0120 days 1.208*** 16.378*** -0.678*** 5.172*** 0.994*** 16.315*** -0.763*** 2.058 (0.063) (1.015) (0.048) (1.557) (0.084) (1.365) (0.055) (1.867) Constant -0.005 -15.533*** 1.558*** 5.894* 0.406* -38.629*** 1.663*** -39.804*** (0.164) (4.268) (0.235) (3.325) (0.213) (5.259) (0.238) (10.058) 33 Country-effects Yes Yes Yes Yes Yes Yes Yes Yes Survey-effects Yes Yes Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Yes Yes Size-effects Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Observations 19565 13416 12801 7531 12623 8676 10981 4882 Predicted Margins Predicted Margins 0 days 0.78 -39.62 0.44 7.38 0.82 -34.57 0.48 3.16 0120 days 0.96 -23.24 0.23 12.55 0.96 -18.25 0.24 5.22 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on liquidity constraint as the main predictor. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. The reference group comprises firms with zero days of cash. Robust standard errors are in parentheses. *,**, *** denote significance, Calculated p-values for the difference in predicted margins for each pairwise comparison categories are available upon request. Table A6: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery for micro-sized firms, by productivity. Productivity (a) 0-4 employees (b) 5-9 employees (tercile) Open Change in sales Arrears Recovery Open Change in sales Arrears Recovery (1) (2) (3) (4) (5) (6) (7) (8) Coefficients Coefficients middle 0.239*** 1.189 0.007 -0.622 0.136*** 4.330*** -0.037 -1.856 (0.041) (1.067) (0.038) (1.443) (0.040) (0.896) (0.033) (1.208) upper 0.238*** 1.964* 0.075* 0.795 0.360*** 5.471*** -0.108*** -1.573 (0.043) (1.089) (0.039) (1.512) (0.045) (0.962) (0.035) (1.261) Constant 0.726*** 25.450 1.527*** 3.250 1.172*** 23.871** 1.196*** -25.249** (0.248) (17.103) (0.324) (3.547) (0.249) (12.144) (0.292) (9.816) Country-effects Yes Yes Yes Yes Yes Yes Yes Yes 34 Survey-effects Yes Yes Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Yes Yes Size-effects Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Observations 16130 12877 8946 5697 12415 8909 9920 4478 Predicted Margins Predicted Margins bottom 0.85 -33.30 0.36 8.26 0.86 -27.50 0.41 9.50 middle 0.89 -32.11 0.36 7.63 0.88 -23.17 0.40 7.65 upper 0.89 -31.33 0.39 9.05 0.91 -22.03 0.38 7.93 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on firm labor productivity as the main predictor. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. The reference group comprises firms at the bottom tercile of the labor productivity distribution. Robust standard errors are in parentheses. *,**, *** denote significance, Calculated p-values for the difference in predicted margins for each pairwise comparison categories are available upon request. Table A7: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery for micro-sized firms, by managerial outlook. (a) 0-4 employees (b) 5-9 employees Outlook Open Change in sales Arrears Recovery Open Change in sales Arrears Recovery (1) (2) (3) (4) (5) (6) (7) (8) Coefficients Coefficients pessimistic 0.571*** 4.541*** -0.287*** 7.566*** 0.518*** 6.004*** -0.465*** 13.422*** (0.057) (1.705) (0.056) (2.593) (0.078) (1.997) (0.074) (3.949) neutral 0.753*** 7.447*** -0.262*** 8.987** 0.754*** 18.169*** -0.636*** 18.415*** (0.101) (1.904) (0.087) (3.488) (0.101) (2.236) (0.089) (5.149) optimistic 0.783*** 15.281*** -0.444*** 11.666*** 0.796*** 16.281*** -0.645*** 17.697*** (0.070) (1.866) (0.062) (2.811) (0.095) (2.186) (0.081) (4.143) very optimistic 0.649*** 18.726*** -0.336*** 13.813*** 0.777*** 20.775*** -0.572*** 17.727*** (0.064) (1.883) (0.060) (2.808) (0.085) (2.195) (0.078) (4.229) Constant -0.079 -31.662 1.440*** 0.880 0.577** 25.831 1.649*** -31.271*** 35 (0.191) (20.645) (0.248) (4.154) (0.226) (16.088) (0.229) (10.728) Country-effects Yes Yes Yes Yes Yes Yes Yes Yes Survey-effects Yes Yes Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Yes Yes Size-effects Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Observations 14550 10153 13702 5790 8399 6048 8005 3121 Predicted Margins Predicted Margins very pessimistic 0.77 -27.43 0.43 4.58 0.80 -33.61 0.54 -2.73 pessimistic 0.89 -22.89 0.34 12.14 0.89 -27.60 0.38 10.69 neutral 0.91 -19.99 0.34 13.56 0.92 -15.44 0.33 15.68 optimistic 0.92 -12.15 0.29 16.24 0.93 -17.33 0.32 14.97 very optimistic 0.90 -8.71 0.32 18.39 0.93 -12.83 0.35 15.00 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on managerial outlook as the main predictor and on micro-sized firms. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. The reference group comprises firms with very pessimistic manager’s outlook. Robust standard errors are in parentheses. *,**, *** denote significance, Calculated p-values for the difference in predicted margins for each pairwise comparison categories are available upon request. Table A8: Estimated coefficients and predicted margins for operation, sales, financial fragility and recovery for micro-sized firms, by sales prediction error. Prediction (a) 0-4 employees (b) 5-9 employees error Open Change in sales Arrears Recovery Open Change in sales Arrears Recovery (1) (2) (3) (4) (5) (6) (7) (8) Coefficients Coefficients low 0.253** 2.557 -0.216** 2.880 0.194 4.541 -0.335*** 1.175 (0.111) (2.232) (0.093) (3.243) (0.144) (2.803) (0.102) (4.266) high -0.116* -6.139*** 0.064 2.746 -0.131 -1.970 -0.153** -1.588 (0.062) (1.620) (0.062) (2.601) (0.081) (1.806) (0.068) (2.919) Constant 3.387*** -6.257 -1.687*** 71.145** 0.416 -18.331*** -0.841* 57.335** (0.678) (6.708) (0.486) (35.554) (0.723) (4.817) (0.434) (23.880) Country-effects Yes Yes Yes Yes Yes Yes Yes Yes 36 Survey-effects Yes Yes Yes Yes Yes Yes Yes Yes Sector-effects Yes Yes Yes Yes Yes Yes Yes Yes Size-effects Yes Yes Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Yes Yes Observations 7672 18677 18336 5673 12729 10467 7098 2204 Predicted Margins Predicted Margins median 0.81 -39.02 0.47 7.86 0.87 -40.61 0.47 10.89 low 0.84 -32.17 0.41 8.46 0.89 -33.40 0.41 10.24 high 0.80 -39.43 0.49 1.23 0.82 -42.58 0.52 10.58 Note: The table presents the estimated coefficients and predicted margins for estimating equations 1 and 2, focusing on prediction error (management quality) as the main predictor and on micro-sized firms. Predicted margins are interpreted as predicted probability of remaining operational or falling into arrears, or predicted change in change in sales or recovery rate. The reference group comprises firms at the median of the prediction error distribution. Robust standard errors are in parentheses. *,**, *** denote significance, Calculated p-values for the difference in predicted margins for each pairwise comparison categories are available upon request. Table A9: Results of Discriminant Analysis: Sales impact & recovery True Predicted Category Total Categorry Classification 0-4 5-9 10-49 0-4 993 602 974 2,569 38.65 23.43 37.91 100 5-9 822 591 985 2,398 34.28 24.65 41.08 100 10-49 1,717 1,251 2,384 5,352 32.08 23.37 44.54 100 Total 3,532 2,444 4,343 10,319 34.23 23.68 42.09 100 Notes: Figures shown are the number of firms and their respective share to total for each classification using discriminant analysis specified in equation 3; Authors’ calculation. 37 B.2. Annex B 38 Table B1: Calculated p-values, pairwise difference in the predicted margins of size categories. (a) All Sample (b) Micro-sized firms Size Categories 0-9 10-49 50-249 250+ Size Categories 0-4 5-9 (no. of employees) (1) (2) (3) (4) (no. of employees) (5) (6) Open Open 0-9 - 0.000 0.000 0.000 0-4 - 0.000 10-49 0.000 - 0.000 0.000 5-9 0.000 - 50-249 0.000 0.000 - 0.059 250+ 0.000 0.000 0.059 - Change in sales Change in sales 0-9 - 0.000 0.000 0.000 0-4 - 0.000 10-49 0.000 - 0.000 0.000 5-9 0.000 - 50-249 0.000 0.000 - 0.007 250+ 0.000 0.000 0.007 - Arrears Arrears 0-9 - 0.177 0.000 0.000 0-4 - 0.001 10-49 0.177 - 0.000 0.000 5-9 0.001 - 39 50-249 0.000 0.000 - 0.003 250+ 0.000 0.000 0.003 - Recovery Recovery 0-9 - 0.804 0.560 0.471 0-4 - 0.642 10-49 0.804 - 0.629 0.387 5-9 0.642 - 50-249 0.560 0.629 - 0.310 250+ 0.471 0.387 0.310 - Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of size category. Table B2: Calculated p-values, pairwise difference in the predicted margins of size categories, wave 1 only. All Sample Micro-sized firms Size Categories 0-9 10-49 50-249 250+ Size Categories 0-4 5-9 Open Open 0-9 0.000 0.000 0.000 0-4 0.111 10-49 0.000 0.000 0.000 5-9 0.111 50-249 0.000 0.000 0.660 250+ 0.000 0.000 0.660 Change in sales Change in sales 0-9 0.000 0.000 0.000 0-4 0.000 10-49 0.000 0.000 0.000 5-9 0.000 50-249 0.000 0.000 0.243 250+ 0.000 0.000 0.243 Arrears Arrears 0-9 0.000 0.000 0.000 0-4 0.869 10-49 0.000 0.000 0.000 5-9 0.869 50-249 0.000 0.000 0.022 250+ 0.000 0.000 0.022 40 Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of size category. Table B3: Calculated p-values, pairwise difference in the predicted margins of size categories, micro-sized firms. (0-4) employees (5-9) employees Size (quintile) 1st 2nd 3rd 4th 5th 1st 2nd 3rd 4th 5th Open Open 1st - 0.000 0.000 0.000 0.000 - 0.010 0.005 0.000 0.000 2nd 0.000 - 0.244 0.002 0.013 0.010 - 0.680 0.271 0.005 3rd 0.000 0.244 - 0.040 0.130 0.005 0.680 - 0.524 0.019 4th 0.000 0.002 0.040 - 0.664 0.000 0.271 0.524 - 0.070 5th 0.000 0.013 0.130 0.664 - 0.000 0.005 0.019 0.070 - Change in sales Change in sales 1st - 0.263 0.403 0.349 0.894 - 0.888 0.586 0.068 0.232 2nd 0.263 - 0.741 0.026 0.320 0.888 - 0.693 0.061 0.206 3rd 0.403 0.741 - 0.048 0.470 0.586 0.693 - 0.032 0.123 4th 0.349 0.026 0.048 - 0.257 0.068 0.061 0.032 - 0.761 5th 0.894 0.320 0.470 0.257 - 0.232 0.206 0.123 0.761 - Arrears Arrears 1st - 0.118 0.029 0.002 0.051 - 0.526 0.752 0.761 0.069 2nd 0.118 - 0.601 0.136 0.663 0.526 - 0.388 0.387 0.200 3rd 0.029 0.601 - 0.282 0.966 0.752 0.388 - 0.984 0.053 41 4th 0.002 0.136 0.282 - 0.293 0.761 0.387 0.984 - 0.050 5th 0.051 0.663 0.966 0.293 - 0.069 0.200 0.053 0.050 - Recovery Recovery 1st - 0.159 0.023 0.641 0.150 - 0.977 0.203 0.003 0.433 2nd 0.159 - 0.306 0.393 0.780 0.977 - 0.214 0.003 0.450 3rd 0.023 0.306 - 0.082 0.549 0.203 0.214 - 0.106 0.748 4th 0.641 0.393 0.082 - 0.306 0.003 0.003 0.106 - 0.079 5th 0.150 0.780 0.549 0.306 - 0.433 0.450 0.748 0.079 - Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of size category. Table B4: Calculated p-values, pairwise difference in the predicted parameter of liquidity categories. All Sample Micro-sized firms Liquidity constraints 0 days 0120 0 days 0120 Open Open 0 days - 0.000 0.000 0.000 - 0.000 0.000 0.000 0120 0.000 0.000 0.000 - 0.000 0.000 0.000 - Change in sales Change in sales 0 days - 0.000 0.000 0.000 - 0.000 0.000 0.000 0120 0.000 0.000 0.000 - 0.000 0.000 0.000 - Arrears Arrears 0 days - 0.000 0.000 0.000 - 0.000 0.000 0.000 0120 0.000 0.000 0.000 - 0.000 0.000 0.000 - 42 Recovery Recovery 0 days - 0.000 0.000 0.000 - 0.000 0.000 0.000 0120 0.000 0.049 0.000 - 0.000 0.023 0.044 - Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of size category. Table B5: Calculated p-values, pairwise difference in the predicted parameter of liquidity constraints, micro-sized firms. All Sample Micro-sized firms Liquidity constraints 0 days 0120 0 days 0120 Open Open 0 days 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0120 0.000 0.000 0.000 0.000 0.000 0.000 0.015 0.000 Change in sales Change in sales 0 days 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0120 0.000 0.000 0.000 0.000 0.000 0.000 0.056 0.000 Arrears Arrears 0 days 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0120 0.000 0.000 0.020 0.000 0.000 0.000 0.000 0.000 Recovery Recovery 43 0 days 0.000 0.001 0.000 0.001 0.000 0.001 0.000 0.004 0120 0.001 0.057 0.125 0.000 0.004 0.222 0.193 0.000 Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of size category. Table B6: Calculated p-values, pairwise difference in the predicted parameter of productivity categories All Sample Micro-sized firms Productivity (terciles) bottom middle upper bottom middle upper Open Open bottom - 0.000 0.000 - 0.000 0.000 middle 0.000 - 0.000 0.000 - 0.002 upper 0.000 0.000 - 0.000 0.002 - Change in sales Change in sales bottom - 0.000 0.000 - 0.000 0.000 middle 0.000 - 0.000 0.000 - 0.239 upper 0.000 0.000 - 0.000 0.239 - Arrears Arrears bottom - 0.000 0.000 - 0.261 0.326 middle 0.000 - 0.000 0.261 - 0.912 upper 0.000 0.000 - 0.326 0.912 - 44 Recovery Recovery bottom - 0.019 0.947 - 0.116 0.631 middle 0.019 - 0.021 0.116 - 0.294 upper 0.947 0.021 - 0.631 0.294 - Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of size category. Table B7: Calculated p-values, pairwise difference in the predicted parameter of productivity categories, micro-sized firms. Productivity 0-4 employees 5-9 employees (terciles) bottom middle upper bottom middle upper Open Open bottom - 0.000 0.000 - 0.000 0.000 middle 0.000 - 0.987 0.000 - 0.003 upper 0.000 0.987 - 0.000 0.003 - Change in sales Change in sales bottom - 0.265 0.071 - 0.003 0.000 middle 0.265 - 0.470 0.003 - 0.000 upper 0.071 0.470 - 0.000 0.000 - Arrears Arrears bottom - 0.860 0.057 - 0.000 0.000 middle 0.860 - 0.061 0.000 - 0.007 upper 0.057 0.061 - 0.000 0.007 - Recovery Recovery bottom - 0.667 0.599 - 0.001 0.000 45 middle 0.667 - 0.342 0.001 - 0.152 upper 0.599 0.342 - 0.000 0.152 - Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of productivity category. Table B8: Calculated p-values, pairwise difference in the predicted parameter of outlook categories All Sample Micro-sized firms Outlook very pessimistic pessimistic neutral optimistic very optimistic very pessimistic pessimistic neutral optimistic very optimistic Open Open very pessimistic 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 pessimistic 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 neutral 0.000 0.000 0.000 0.954 0.077 0.000 0.000 0.000 0.581 0.344 optimistic 0.000 0.000 0.954 0.000 0.031 0.000 0.000 0.581 0.000 0.033 very optimistic 0.000 0.000 0.077 0.031 0.000 0.000 0.000 0.344 0.033 0.000 Change in sales Change in sales very pessimistic 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 pessimistic 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 neutral 0.000 0.000 0.000 0.003 0.000 0.000 0.000 0.000 0.001 0.000 optimistic 0.000 0.000 0.003 0.000 0.000 0.000 0.000 0.001 0.000 0.000 very optimistic 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Arrears Arrears very pessimistic 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 pessimistic 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.057 0.000 0.003 neutral 0.000 0.000 0.000 0.611 0.705 0.000 0.057 0.000 0.129 0.719 optimistic 0.000 0.000 0.611 0.000 0.822 0.000 0.000 0.129 0.000 0.002 very optimistic 46 0.000 0.000 0.705 0.822 0.000 0.000 0.003 0.719 0.002 0.000 Recovery Recovery very pessimistic 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 pessimistic 0.000 0.000 0.106 0.000 0.000 0.000 0.000 0.226 0.000 0.000 neutral 0.000 0.106 0.000 0.378 0.045 0.000 0.226 0.000 0.459 0.138 optimistic 0.000 0.000 0.378 0.000 0.072 0.000 0.000 0.459 0.000 0.204 very optimistic 0.000 0.000 0.045 0.072 0.000 0.000 0.000 0.138 0.204 0.000 Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of outlook category. Table B9: Calculated p-values, pairwise difference in the predicted parameter of outlook categories, micro-sized firms. 0-4 employees 5-9 employees Outlook very pessimistic pessimistic neutral optimistic very optimistic very pessimistic pessimistic neutral optimistic very optimistic Open Open very pessimistic - 0.000 0.000 0.000 0.000 - 0.000 0.000 0.000 0.000 pessimistic 0.000 0.000 0.030 0.000 0.064 0.000 - 0.003 0.000 0.000 neutral 0.000 0.030 0.000 0.758 0.240 0.000 0.003 - 0.686 0.796 optimistic 0.000 0.000 0.758 - 0.017 0.000 0.000 0.686 - 0.813 very optimistic 0.000 0.064 0.240 0.017 - 0.000 0.000 0.796 0.813 - Change in sales Change in sales very pessimistic - 0.008 0.000 0.000 0.000 - 0.003 0.000 0.000 0.000 pessimistic 0.008 - 0.026 0.000 0.000 0.003 - 0.000 0.000 0.000 neutral 0.000 0.026 - 0.000 0.000 0.000 0.000 - 0.259 0.121 optimistic 0.000 0.000 0.000 - 0.001 0.000 0.000 0.259 - 0.001 very optimistic 0.000 0.000 0.000 0.001 - 0.000 0.000 0.000 0.001 - Arrears Arrears very pessimistic - 0.000 0.002 0.000 0.000 - 0.000 0.000 0.000 0.000 pessimistic 0.000 - 0.727 0.000 0.111 0.000 - 0.007 0.000 0.010 neutral 0.002 0.727 - 0.017 0.310 0.000 0.007 - 0.899 0.336 optimistic 0.000 0.000 0.017 - 0.005 0.000 0.000 0.899 - 0.146 very optimistic 0.000 0.111 0.310 0.005 - 0.000 0.010 0.336 0.146 - Recovery Recovery very pessimistic - 0.004 0.010 0.000 0.000 - 0.001 0.000 0.000 0.000 47 pessimistic 0.004 - 0.562 0.002 0.000 0.001 - 0.152 0.007 0.020 neutral 0.010 0.562 - 0.301 0.062 0.000 0.152 - 0.842 0.853 optimistic 0.000 0.002 0.301 - 0.182 0.000 0.007 0.842 - 0.989 very optimistic 0.000 0.000 0.062 0.182 - 0.000 0.020 0.853 0.989 - Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of outlook category. Table B10: Calculated p-values, pairwise difference in the predicted parameter of prediction error categories. Prediction All Sample Micro-sized firms error median low high median low high Open Open median - 0.000 0.000 - 0.020 0.045 low 0.000 - 0.003 0.020 - 0.000 high 0.000 0.003 - 0.045 0.000 - Change in sales Change in sales median - 0.000 0.005 - 0.000 0.310 low 0.000 - 0.000 0.000 - 0.000 high 0.005 0.000 - 0.310 0.000 - Arrears Arrears median - 0.001 0.064 - 0.001 0.064 low 0.001 - 0.000 0.001 - 0.000 high 0.064 0.000 - 0.064 0.000 - 48 Recovery Recovery median - 0.197 0.037 - 0.807 0.106 low 0.197 - 0.349 0.807 - 0.225 high 0.037 0.349 - 0.106 0.225 - Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of prediction error category. Table B11: Calculated p-values, pairwise difference in the predicted parameter of prediction error categories, micro-sized firms. 0-4 5-9 Sales prediction median low high median low high Operationability Operationability median - 0.014 0.050 - 0.000 0.000 low 0.014 - 0.000 0.000 - 0.000 high 0.050 0.000 - 0.000 0.000 - Change in sales Change in sales median - 0.252 0.000 - 0.105 0.275 low 0.252 - 0.000 0.105 - 0.013 high 0.000 0.000 - 0.275 0.013 - Financial fragility Financial fragility median - 0.020 0.303 - 0.001 0.026 low 0.020 - 0.002 0.001 - 0.062 high 0.303 0.002 - 0.026 0.062 - 49 Recovery rate Recovery rate median - 0.375 0.291 - 0.783 0.586 low 0.375 - 0.965 0.783 - 0.478 high 0.291 0.965 - 0.586 0.478 - Note: The table presents the estimated p-values of the difference in predicted margins for each pairwise comparison of prediction error category.