Policy Research Working Paper 10898 Conflict and Firms’ Performance A Global View Emanuele Brancati Michele Di Maio Roberta Gatti Asif M. Islam Middle East and North Africa Region Office of the Chief Economist September 2024 Policy Research Working Paper 10898 Abstract This study provides a global analysis of the effect of conflict conflict-induced reduction in the availability of production exposure on firms’ performance, combining geolocalized inputs and the increase in informal competition. Firms react longitudinal firm-level data and information on political to lower sales by reducing labor costs and expenditure on violence events from 91 countries between 2006 and 2019. other production inputs. The effect of conflict is more det- Higher conflict exposure does not affect firm profits, as it rimental for firms in countries with low-quality bureaucracy reduces both sales and total costs. Sales decline due to the and that are initially at peace. This paper is a product of the Office of the Chief Economist, Middle East and North Africa 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 aislam@worldbank.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 Conflict and Firms’ Performance: A Global View * Emanuele Brancatit Michele Di Maio:j Roberta Gatti§ Asif M. Islam1 Keywords: firm, conflict, political violence, intermediate inputs, labor market, mar- ket competition JEL Classification: C23 D22 D74 L20 O12 * This study is exempt from pre-registration because it only uses secondary data. The findings in- terpretations, and conclusions expressed in this paper do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. All errors are our own. t emanuele.brancati@uniroma1.it Sapienza University of Rome (Italy) and IZA :j michele.dimaio@uniroma1.it, Sapienza University of Rome (Italy) and IZA §rgatti@worldbank.org, The World Bank 1 aislam@worldbank.org, The World Bank 1 Introduction Conflict and political violence seem to have become the new normal in many developing and developed countries, affecting the everyday lives of more than 1.7 billion people world- wide. In 2022, 139,000 political violence events were reported, resulting in over 147,000 fatalities across 167 countries (ACLED, 2023).1 Despite the dramatic suffering and challenges caused by conflict and political violence, economic activity continues, even under extreme conditions. Economists and policy makers are increasingly interested in understanding how this occurs across different contexts. While earlier research primarily adopted a macroeconomic perspective, the current focus has shifted to country-level analysis of the microeconomic effects of conflict and violence on firms (Verwimp et al., 2019; Couttenier et al., 2022; Premand and Rohner, 2024). Although these studies provide rigorous estimates of the effect of conflict on firms and detailed analyses of the mechanisms, their analyses are specific to individual countries and often dependent on the unique characteristics of each economy. Our paper is the first to provide a global view of the effect of conflict on firms, allowing for cross-country comparisons of the impact of conflict exposure on various firm outcomes and the mechanisms behind these effects. We use a confidential version of the global World Bank Enterprise Survey (WBES), which, beyond the rich information in the standard dataset, includes the geolocation of each firm. The original sample comprises longitudinal firm-level data for more than 153,000 firms in 91 countries from 2006 to 2019 (to exclude the COVID-19 pandemic). By combining this information with geolocalized data on conflict and political violence events worldwide, we build a firm-specific, time-varying measure of conflict exposure. This measure is used in a firm-level fixed-effects model to estimate the impact of conflict events on several outcomes. Our analysis documents a number of empirical findings. Results indicate that higher conflict exposure does not significantly impact a firm’s profits. This null effect is due to conflict-induced reductions in both sales and expenditure on production inputs, suggest- ing a decrease in the firm’s output due to the exposure to conflict events. This implies that, although profits remain stable, the overall economic activity is negatively affected by conflict. These findings are not driven by reverse causality or selection bias based on ob- servable factors, and they withstand several robustness checks. Regarding the underlying mechanisms, we show that conflict diminishes a firm’s access to intermediate production inputs, particularly imported ones, and reduces the availability of electricity, leading to decreased output. Furthermore, conflict exacerbates informal competition faced by the firm. In response, firms reduce costs by substituting skilled workers with unskilled ones. violence is the deliberate use of power and force to achieve political goals. Political violence 1Political includes war and related violent conflicts, state violence, and similar acts carried out by larger groups WHO (2002). Throughout the paper, we will interchangeably use the terms exposure to political violence events and exposure to conflict-related events. 1 In sum, while sales decline due to reduced output and increased informal competition, profits remain unchanged due to the firm’s adjustments in production input expenditure. Other mechanisms, such as the conflict-induced decrease in local demand or increase in the severity of various obstacles to a firm’s activity, do not appear to play a role in explaining our findings. A key advantage of using a harmonized global dataset of firms is that it allows for comparisons of the effect of conflict events on firms across countries with different char- acteristics. Our results show no significant differences in the impact of conflict exposure between firms located in high-income versus low-income countries, or in fragile versus non- fragile countries. Yet, results exhibit significant heterogeneity across two other country characteristics, namely the quality of bureaucracy and the conflict status of the coun- try at the beginning and during the period of analysis. We find that the adverse effects of conflict exposure are more pronounced for firms located in countries with low-quality bureaucracy, resulting in decreased sales and profits, while the effects are negligible in countries with high-quality bureaucracy. These results align with the idea that ceteris paribus, a more efficient government can better support firms in coping with negative shocks. This is confirmed by evidence that in countries with low-quality bureaucracy, the negative effect of conflict exposure on firms’ profits is largely driven by reduced access to domestic and imported production inputs and decreased electricity availability. Finally, results show that for firms located in countries that are at peace at the be- ginning of the analysis period, an increase in conflict exposure has a negative and sizable effect on all outcomes, including profits. In contrast, for firms in countries already in conflict at the beginning of the period, conflict exposure has a small negative effect on sales and no significant impact on other outcomes. We obtain similar results when comparing firms in countries that are not in conflict during the entire period of analysis to those in countries that are always in conflict in the sample of analysis, with the former suffering more negative effects. These findings are consistent with the possibility that firms in countries where conflict is business as usual have already adjusted their production strategies to cope with its effects on the overall economy. On the other hand, firms in countries that are initially at peace or not always in conflict in the sample of analysis may not have yet undertaken such costly adjustments. In line with this interpretation, results indicate that in these economies, an increase in conflict exposure leads to a larger reduction in the use of raw and intermediate inputs and in the availability of electricity, and to a greater substitution of skilled workers with unskilled ones, resulting in diminished output and, consequently, reduced sales. Moreover, higher conflict intensity increases informal competition, further decreasing sales and profits. Our paper contributes to the literature on the micro-economic consequences of con- flict. This strand of research has largely focused on the effect of conflict on outcomes such as education and health (for reviews, see Verwimp et al., 2019 and Rohner, 2023). The literature on the economic effects is instead relatively smaller, partly because of the 2 difficulty in accessing economic data in the context of conflict-affected countries.2 Early papers estimated the macroeconomic effects of conflict (see e.g., Abadie and Gardeazabal, 2003; Martin et al., 2008; Glick and Taylor, 2010). More recently, empirical analyses have started looking at the impact on firms.3 Previous studies have focused on various forms of conflict and political violence and their influence on different firm outcomes in specific economies, including Afghanistan (Blumenstock et al., 2024), Angola (Guidolin and La Ferrara, 2007), Colombia (Camacho and Rodriguez, 2013), India (Couttenier et al., 2022), Kenya (Ksoll et al., 2023), Libya (Del Prete et al., 2023), Mozambique (Custo- dio et al., 2024), Ukraine (Korovkin and Makarin, 2020, 2023), Sierra Leone (Collier and Duponchel, 2013), and the West Bank (Amodio and Di Maio, 2018).4 However, all these stud- ies are single-country analyses, examining different outcomes, using various types of data, and employing different methodologies, making their results not directly comparable. Our paper contributes to the literature along three main dimensions. First, it represents the first global study on the effects of conflict on firms. By leveraging comparable firm- level data from both developed and developing countries, we document several empirical regularities that hold across different settings and contexts. Second, we compare the effects of conflict exposure across a broad range of firm and country-level characteristics, revealing that country-level factors are more influential than firm-level heterogeneities. This suggests that the impact of conflict-related events varies more between countries than within them. Third, we test several mechanisms to explain our main results. While some of our findings align with previous studies, we also show how these mechanisms may operate differently depending on a country’s characteristics. In terms of policy implications, our results suggest that interventions in conflict and post-conflict contexts should be tailored to the specific characteristics of each country. The paper proceeds as follows. Section 2 describes the data. Section 3 outlines the empirical strategy and Section 4 presents the results. Section 5 concludes the paper. 2 Data 2.1 Firm data Our main source of data is the World Bank Enterprise Survey (WBES), a global dataset providing information on firms in the manufacturing, retail, and service sectors across 2Acompanion literature looks at the economic determinants of conflict, including price shocks (Dube and Vargas, 2013), natural resources (Berman et al., 2017), and import restrictions (Amodio et al., 2021). 3A few papers document the effect of crime-related violence on economic activity (Pinotti 2015; Rozo 2018; Utar 2024; Piemontese 2023). 4Two papers look at the effect of the return to peace in Colombia on entrepreneurship (Bernal et al., 2024) and investments (De Roux and Martinez, 2023). 3 148 countries.5 The original sample consists of approximately 180,000 firms interviewed between 2006 and 2019 and is designed to be representative at the country level. It covers privately owned firms with at least 5 employees operating in the formal (non-agriculture) sector.6 The original dataset is a repeated cross-section, but several firms are interviewed in multiple waves.7 To implement our identification strategy, we consider the panel compo- nent of the WBES, which includes 91 countries and 43,700 firm-year observations.8 Table A1 provides a detailed list of the countries and waves included in our analysis, while Fig- ure 1 shows their location. We discuss possible issues related to selection and attrition in Section 2.1, where we show that the probability of belonging to the panel is not correlated with our main variables of interest, and we discuss the determinants of exiting the panel. We use a confidential version of the WBES global dataset, that also provides firms’ geo-localization. This information allows us to match each firm with conflict and political violence events occurring in its geographical neighborhood and it is thus essential to construct the firm-specific measure of conflict exposure we use in our analysis. Our initial sample includes firms that belong to the panel and for which we have geolo- calized information, totaling 42,100 firm-year observations. The sample is then reduced to about 34,200 due to missing values in outcome and control variables. Importantly, we show below that missing values are not correlated with the intensity of conflict exposure. A typical firm in our sample is domestically-owned and operates in the private sector. As shown in Table 1, our median firm has 20 employees and has been operating for 16 years. It mainly serves the domestic market (16% of the firms are exporters) and generates $333,000 and $268,000 in annual sales and profits (in 2000 constant USD), respectively. All variables are defined in Table C1. 5The WBES dataset is the most comprehensive dataset to conduct studies on firms at the world level, see for instance, Fisman et al. (2024) and Armangu´e-Jubert et al. (2024). 6Firms are selected using random sampling techniques with three stratification levels to ensure repre- sentativeness across firm size (5-19 employees; 20-99 employees; and 100+ employees) and sector (manufacturing, retail, and other services). Unlike the other main outcomes, Profits is not directly provided by the WBES dataset. We define Profits as the Sales minus Total costs, with the latter being the sum of Labor costs, and Raw materials and intermediate inputs costs (also including energy costs). For the latter group of costs, the WBES questionnaire administered to service firms only inquires about electricity expenditure. Therefore, it is possible that for service firms Total costs underestimate overall production expenditures (e.g., excluding expenses for equipment such as computers, telecommunication services, etc.). This may introduce issues in the comparability of Profits across firms in different sectors. To address this concern, we exclude from our sample all service firms for which the ratio between Total costs and Sales is excessively low (below the fifth percentile of the distribution). This restriction enhances the reliability of comparing economic performance across firms in different sectors. Nonetheless, our results are virtually identical if we focus on the manufacturing sector only (see Section 4.1). 7For additional details on the sampling methodology of the WBES, see https://www.worldbank. org/content/dam/enterprisesurveys/documents/methodology/Enterprise%20Surveys_Manual% 20and%20Guide.pdf. 8More specifically, our sample is made of 44 countries with two waves available, 45 with three waves, and 2 with four waves. In terms of firms, this implies coverage of about 7,421, 11,894, and 336 firms, followed for two, three, and four waves, respectively. 4 Selection, attrition, and missing values Our estimating strategy relies on the subset of firms with multiple observations over time, i.e. those included in the panel component of the WBES (see Section 3). This may lead to potential selection issues affecting our estimating sample. Specifically, one may be concerned about the self-selection of firms that are re-interviewed in subsequent survey waves, vis a ` vis firms that, for various reasons, are interviewed only once. If such selection is correlated with our main regressors of interest and our dependent variables, it could bias our results. In Table A3, we tackle this issue by focusing on the sample of firms included in the first survey wave for each country and test the correlation between the firm’s likelihood of being interviewed a second time —i.e., being in the panel and thus in our sample— and the variables employed in the analysis. Our results allay concerns about systematic sample selection bias by showing no correlation between the firm’s probability of belonging to the panel and Conflict exposure, or any measure of firms’ economic performance (Sales and Profits). In addition, we explore possible issues related to attrition. In Table A4, we test the correlation between our main variables and a dummy taking the value of one for firms exiting the panel. As expected, results show that firms exposed to a higher number of conflict events are more likely to exit the sample. This suggests that, if anything, our findings might be affected by attenuation bias, potentially leading to an underestimation of the negative impact of conflict on overall economic activity. One additional concern with our data is that conflict may make missing values non- random.9 Results reported in Table A5 suggest that this is not the case. Column 1 indicates that in the full sample, there is no correlation between conflict exposure and firms not reporting sales. Column 2 shows that this also holds for the sample of firms for which we have information on the controls included in the baseline specification. Finally, in the sample of firms for which we have information on sales, conflict exposure is not correlated with the non-reporting of any cost component (column 3), raw material and intermediate inputs costs (including energy costs, column 4), or labor costs (column 5). Taken together, these results indicate that, for the firm’s outcome variables collected in the WBES, non- reporting is uncorrelated with conflict exposure. 2.2 Conflict data Our main source of data on political violence events is the Integrated Crisis Early Warning System (ICEWS, Shilliday and Lautenschlager, 2012). For each event, the ICEWS dataset provides information on the date, source actors, target actors, and location (geospatial coordinates). It covers all types of interactions —whether cooperative or hostile— among 9A related concern may be that firms strategically choose not to report in order to hide economic performance to avoid being targeted by fighters. To begin, we note that the inclusion in the WBES is voluntary and firms are not obliged in any way to answer the questionnaire. Moreover, the number of missing values is higher for Total Costs than for Sales (see Table 2), suggesting that it is unlikely that non-reporting is a strategic choice of the firm trying to hide its economic performance. 5 sociopolitical actors, including individuals, groups, and nation-states. These events are categorized using the Conflict and Mediation Event Observations (CAMEO) classification (Schrodt and Yilmaz, 2012). Each one of these categories is assigned an intensity variable in the interval (−10; +10) that ranks events from the most hostile to the most cooperative. We construct our panel dataset of political violence events by selecting all hostile events from the ICEWS dataset (i.e., those with negative intensity values) that are classified as violent according to Amodio et al. (2021).10 This dataset encompasses a wide range of political violence and conflict-related events, including those that have not resulted in fatalities. Our final dataset includes a total of 1.1 million political violence events that occurred between 2006 and 2019. A detailed breakdown of these event types and their frequency during the analysis period is provided in Table A2, while Figure 2 shows their geographical distribution. Finally, we construct a firm-specific measure of conflict exposure, labeled Conflict Exposure. This measure is defined as the logarithm (log of 1+) of the number of political violence events occurring within a 20 km radius of the firm’s location during the 12 months preceding the end of the last fiscal year, as reported by the WBES survey. In absolute terms, the mean is 13, the standard deviation is 7.3, while the median and the third quartile are 13 and 79, respectively. As a robustness check, we construct an alternative measure of conflict exposure using data from the PRIO/Uppsala Armed Conflict and Location Event (ACLED) dataset (Raleigh et al., 2010). ACLED provides the date and the geolocation of different types of conflict-related events.11 However, the country coverage of ACLED for our period of analysis is much smaller than ICEWS.12 The bottom panel of Table A2 lists the number of conflict events by type for our period of analysis. Our measures of conflict exposure may be subject to measurement error, as the ICEWS dataset shares common weaknesses of large datasets compiled from multiple sources. One such issue is the presence of duplicated events; we address this by excluding events with identical characteristics (date, location, actors, etc.). Another potential source of mea- surement error is related to possible biases in event reporting, which may favor certain areas. To mitigate this concern, our analysis accounts for possible (time-invariant) ge- ographical differences in reporting through the inclusion of firm and area-fixed effects. Additionally, we cross-validate ICEWS conflict event data with that from ACLED. The correlation between the number of conflict events reported by the two datasets is high (0.77) in the same sample. Our robustness checks also confirm that our main results are 10See Table A2 in the Online Appendix for details on this classification. 11Conflict events in the ACLED dataset are categorized as follows: i) battle (government regains terri- tory); ii) battle (no change of territory); iii) battle (nonstate actor overtakes territory); iv) headquarters or base established; v) nonviolent transfer of territory; vi) remote violence; vii) riots and protests; viii) strategic development; and ix) violence against civilians. 12Until 2016 ACLED collected data on conflict events only for African countries and a few Asian ones. 6 consistent when using the ACLED dataset as an alternative source for conflict events. 2.3 Other data In our heterogeneity analysis, we categorize countries using different datasets. First, we group countries by their level of economic development, following the World Bank classification.13 We define a country as low–middle income if it falls into the low or lower- middle income categories, while a country is classified as high–middle income if it belongs to the upper-middle or high-income categories. Second, we categorize countries based on their fragility. According to the World Bank, a fragile country is one experiencing deteriorating governance, prolonged political crises, post-conflict transition, or gradual but still fragile reform processes.14 In our sample, 16% of countries are classified as fragile. Third, we explore heterogeneities based on the quality of a country’s bureaucracy using an indicator from the International Country Risk Guide (ICRG).15 We classify a country as having low-quality bureaucracy if its index is below the cross-sectional median (65% of firms in our sample are located in such countries). In additional analyses, we also use an ICRG indicator that assesses the role of corruption within the country’s political system. Finally, we group countries based on their conflict status at the beginning and dur- ing the whole period of analysis. We classify a country as initially at peace if at the time of the first survey in the country there are no major episodes of political violence (such as civil conflicts, ethnic violence, riots, popular protests, or repression of dissidents). Countries not fitting this criterion are classified as initially in conflict. The list of coun- tries with major episodes of political violence is sourced from the Systemic Peace War dataset.16 We expand this list to also include countries involved in Operation Juniper Shield, a large counter-terrorism international operation initiated in 2007 against armed groups in the Saharan and Sahel regions of Africa.17 The final list of economies initially in conflict includes: Afghanistan, Angola, Cameroon, Chad, Colombia, the Democratic Republic of Congo, the Arab Republic of Egypt, Ethiopia, Kenya, Morocco, Myanmar, Niger, Nigeria, Pakistan, the 13The World Bank classifies countries into four income categories: low, lower-middle, upper-middle, and high income, based on per capita income. The full list of countries and their income categories can be found at: https://datahelpdesk.worldbank.org/knowledgebase/articles/906519. 14The list of fragile countries is available at: https://thedocs.worldbank.org/en/doc/ 9b8fbdb62f7183cef819729cc9073671-0090082022/original/FCSList-FY06toFY22.pdf. 15The ICRG rating includes 22 variables across three risk subcategories: political, financial, and economic. For additional details, see https://www.prsgroup.com/wp-content/uploads/2014/08/ icrgmethodology.pdf. 16The list of all major episodes of political violence from 1945 to 2019 is available at: https:// www.systemicpeace.org/warlist/warlist.htm. Major episodes involve at least 500 “directly related” fatalities and reach a level of intensity where the violence is systematic and sustained (with a base rate of 100 “directly related” deaths per year). Only countries directly affected by the violence are listed; countries intervening in these episodes are not included, as the violence does not occur on their territory. 17The countries involved in Operation Juniper Shield are Algeria, Burkina Faso, Cameroon, Chad, Mali, Mauritania, Morocco, Niger, Nigeria, Senegal, and Tunisia (OECD/SWAC, 2020). 7 Philippines, the Russian Federation, Tunisia, Türkiye, Uganda, the West Bank and Gaza, and the Republic of Yemen. We also group countries based on their conflict status during the whole period of analysis. A country is classified as always in conflict in the sample if major episodes of political violence begin before the first survey wave and persist throughout all the following waves. Countries not fitting this criterion are classified as not always in conflict in the sample.18 3 Empirical Strategy To estimate the impact of conflict exposure on economic performance, we use longitudinal firm-level data combined with geolocalized information on conflict events. Our specifica- tion reads as follows: Yit = α + β Conflict exposure i + γ′Xit + µi + θt + εit (1) where Yit represents the outcome of interest for firm i in fiscal year t. In our analysis, we consider several different dependent variables including Sales, Total Costs, and Profits. Conflict exposureit denotes the firm-specific measure of conflict exposure. In our base- line specification, it is computed as the log number of conflict events, from the ICEWS dataset, that occurred within a 20 km radius around the firm’s location during the 12 months preceding the closure of the last fiscal year (as reported by the WBES survey).19 The vector Xit represents a set of time-varying firm characteristics, including size, age, and export status. µi indicates firm fixed effects, accounting for all time-invariant observed and unobserved characteristics of the firm that could potentially influence the outcome of interest. θt represents year fixed effects, controlling for overall trends in economic activity and common shocks. As a robustness check, we also include the full set of area-year in- teracted fixed effects to account for all time-varying location-specific determinants of firm performance, such as infrastructure availability and local labor market characteristics.20 Finally, εit is the error term. In all specifications, we report robust standard errors clus- tered at the firm level, which is the unit of observation used to measure the intensity of conflict exposure. All variables are defined in Table C1. Our identification strategy relies on the assumption that, conditional on our full set of fixed effects, conflict events occurring near the firm’s location are uncorrelated with any latent determinant of its economic performance. Under this assumption, β captures the reduced-form effect of the firm-specific conflict exposure on economic performance. list of economies always in conflict in the sample includes Afghanistan, Cameroon, Chad, 18The Colombia, the Democratic Republic of Congo, Egypt, Ethiopia, Morocco, Myanmar, Niger, Nigeria, Pakistan, the Philippines, Russia, Tunisia, Türkiye, the West Bank and Gaza, and Yemen. 19We also employ alternative measures of conflict exposure constructed using: i) a smaller buffer distance from the firm’s location; ii) a different time period; and iii) the ACLED dataset as an alternative source for conflict events (see Section 2.2). 20Area is the most granular geographical unit of observation in the WBES (314 areas in our dataset). 8 4 Results 4.1 Conflict exposure, sales, total costs, and profits We begin our analysis by examining the impact of conflict exposure on a firm’s sales. Results reported in Table 2, columns 1 and 2, show a negative and highly significant effect across both the baseline model and the one with firm-level controls. Based on our estimates, a one-standard-deviation increase in the number of conflict events (about seven) occurring within a 20 km radius around the firm implies a reduction in median sales by 2.8%. We next investigate the impact of conflict on a firm’s total costs, measured by expen- diture on raw materials, intermediate inputs (including electricity), and labor. As shown in Table 2, columns 3 and 4, increased conflict exposure leads to a significant reduction in total production costs. Note that, in principle, this result could indicate either a decrease in input prices or a reduction in the quantity of inputs purchased. Although the WBES dataset does not allow us to directly test for price changes, it is unlikely that input prices decrease during conflict.21 Consequently, these findings suggest that the observed reduction in production costs is more likely due to the diminished availability of inputs. Evidence supporting this hypothesis is provided in Section 4.2.1. Finally, we analyze profits. Results in Table 2, columns 5 and 6, show that the coefficient for Conflict exposure is centered around zero. This null effect on firms’ profits is consistent with the fact that conflict exposure affects sales and total costs in the same direction. This suggests that firms respond to reduced sales by cutting total production costs. In Section 4.2, we discuss the various mechanisms that contribute to explaining this result. Threats to identification The firm-level fixed effects in our baseline specification account for all time-invariant firm, sector, and geographical characteristics. However, our estimates may still capture some factors that vary over time at the local level and are simultaneously correlated with the number of conflict events as well as the firm’s economic performance. To address this concern, we augment model 1 with a full set of area-time interacted fixed effects to also control for all time-varying, location-specific determinants of firm performance, such as the availability of infrastructure or local labor market characteristics. Table A6, columns 1 and 2, shows that the coefficient for sales is still negative, sizable, and significant (although somewhat smaller than in the baseline) and the one for profits remains null. These results confirm that higher conflict exposure does not impact profits but does lead to a decrease in a firm’s sales.22 21Previous studies document that conflict typically raises input prices (Amodio and Di Maio, 2018; Utar, 2024; Del Prete et al., 2023; Couttenier et al., 2022). 22Table A7 shows that the inclusion of area-year fixed effects does not alter the baseline results regarding the effect of conflict exposure on total costs (column 1) and its two components, namely the cost of raw 9 Another potential threat to our identification is reverse causality. One possible concern is that the location and intensity of conflict events may be driven by the presence of firms with better economic performance, making them valuable targets for conflict fighters. To address this, we add the one-year and two-year lags and leads in the number of conflict events to our baseline regression for sales and profits (see Table A6, columns 3-4 and 5-6, respectively). Reassuringly, the results are consistent with the baseline, and the coefficients for the lead and lagged variables are consistently centered around zero. As an additional indirect test for reverse causality, we re-estimate our model by excluding large firms and those located in capital cities. The results, reported in Table A8, are virtually identical to those obtained using the full sample. This suggests that our findings are not driven by the location and intensity of conflict events being influenced by the location of high-value firms or by the activity of firms located in political centers. A further concern may be that the negative relationship between sales and conflict might be due to low- performing firms being targeted because they cannot afford protection. However, this does not appear to be the case: conflict exposure does not increase firms’ losses from theft or the firm’s security expenditures (see Table A15, columns 5 and 6). Finally, we conduct a more formal check for selection on unobservables using the test proposed by Altonji et al. (2005) and further refined by Oster (2019). Table A9 presents the results for the estimated impact of conflict across varying degrees of impor- tance attributed to unobservable factors relative to the included controls, as measured by a parameter 0 < δ ≤ 1.23 Column 1 reports the estimated baseline effects of conflict exposure on sales and total costs (the same as those in Table 2, columns 4 and 6, respec- tively). Results in columns 2-11 of Table A9 show that the estimate of conflict exposure consistently retains its negative sign for any δ ∈ (0, 1). Furthermore, column 12 reports the hypothetical value of δˆ required to make our estimates statistically insignificant. In both scenarios, δˆ is negative and large in absolute value, indicating that the selection on unobservables would need to be substantial and of the opposite sign to the selection on the observables for the effect to be nullified by the bias. This makes our main findings unlikely to be driven by selection bias from unobservable factors. Robustness Our results are robust to a number of checks. First, we examine the es- timation sample. As discussed in Section 2.1, the number of observations differs across outcomes due to missing values. All results presented so far are obtained by exploiting the maximum number of available observations for each outcome in each regression. Table A10 shows that all the results for our main outcomes (Sales, Total costs, and Profits) remain unchanged with respect to our baseline when we restrict the sample to firms for which information is available for all outcomes. As an additional check, we re-run the main materials and intermediate inputs and cost of labor (columns 2 and 3, respectively). 23When δ = 1, the model assumes that the effect of unobservables is directly proportional to the influence of the included observable factors. 10 analysis considering only manufacturing firms, to test whether our findings are driven by differences in the construction of Total cost between sectors (see Section 2.1). Table A11 shows that results remain unchanged with respect to the baseline. Next, we focus on the definition of the conflict exposure measure. Table A12 shows that the effects of conflict on our main outcomes remain qualitatively unchanged with respect to the baseline when: 1) we consider a smaller radius (10 km) for the buffer around the firm to measure conflict exposure (column 1); 2) we apply the hyperbolic sine transformation to the number of conflict events (column 2); and 3) we count the number of conflict events considering the last two years (column 3). Column 4 shows that results for both sales and total costs are also confirmed when we use the ACLED dataset to measure conflict exposure, while the effect on profits is negative and marginally significant. These results are reassuring given that ACLED covers a very different set of countries and conflict-related events with respect to ICEWS. Finally, since our dataset spans 91 economies, we investigate the robustness to the set of countries included in the analysis. To do so, we re-run our baseline regression 91 times, sequentially excluding from the sample one country at a time. Figure A1 indicates that the negative effect on sales and total costs and the null effect on profits do not depend on the inclusion or exclusion of any specific country. Summing up Higher conflict exposure does not have a significant impact on a firm’s profits. This null result is the net effect of a conflict-induced reduction in both sales and expenditure on production inputs, suggesting that conflict reduces the level of production for surviving firms. This implies that, even though firms’ profitability does not decrease, overall economic activity is adversely affected by conflict-related violent events. 4.2 Mechanisms Our results show that conflict exposure reduces a firm’s sales and expenditure on produc- tion inputs. These effects may stem from different mechanisms at play. Higher conflict exposure could reduce the availability of production factors, forcing the firm to use a sub- optimal input mix, and mechanically lead to a reduction in the firm’s expenditure on raw materials, intermediate inputs, and labor. The conflict-induced reduction in the quantity and quality of inputs then lowers the firm’s total output and thus its sales. At the same time, conflict may reduce sales by changing the level and type of competition faced by firms, such as increasing (unfair) competition from informal firms. Finally, conflict may lower sales by decreasing local demand or increasing the severity of the obstacles to a firm’s activity. In the following, we take these conjectures to the data to provide evidence on the mechanisms behind the negative effect of conflict on firms’ sales and expenditure on production inputs. 11 4.2.1 Raw materials and intermediate inputs Table 3, column A1, shows that conflict exposure leads to a decrease in a firm’s expenditure on raw materials and intermediate inputs. This effect could result from either lower input prices or reduced quantities of inputs purchased by the firm. As the WBES does not provide information on the price or amount of intermediate inputs used, we cannot directly test this. Nevertheless, since it is unlikely for conflict to lower the unit cost of inputs, we regard this finding as indicative of a reduction in the availability and use of raw materials and intermediate inputs by the firm.24 In columns A2-A4, we provide evidence supporting this argument. Conflict reduces the availability of electricity, a key production input. Firms more exposed to conflict events have lower expenditure on electricity (column A2) and a higher probability of power outages (column A3), consistent with these firms being forced to reduce their electricity usage. Moreover, conflict decreases the value share of imported intermediate inputs used in production (column A4). This finding aligns with previous single-country studies indicating that conflict hampers firms’ access to imported inputs used by the firm due to increased trade costs or uncertainty.25 Given that imported inputs are key to firm productivity (see e.g., Topalova and Khandelwal, 2011; Halpern et al., 2015), a negative shock to their availability likely leads to a significant reduction in firms’ economic activity (Amodio and Di Maio, 2018; Boehm et al., 2019; Carvalho et al., 2021). Taken together, these findings suggest that conflict reduces the availability of both raw materials and intermediate inputs, especially those that are imported, thereby resulting in a decline in the firm’s overall output. 4.2.2 Labor Results in Table 3, column A5, indicate that conflict exposure reduces total labor costs. To explore more in detail the mechanism explaining this effect, we examine the impact of conflict on the number of workers, their characteristics, and the average wage, as shown in columns A6-A8. Conflict exposure does not appear to affect the total number of workers employed by the firm (column A6).26 However, conflict does have a significant impact on the composition of the workforce, resulting in a sizable increase in the share of unskilled workers (column A7). Consistent with this substitution effect, higher conflict exposure 24The conflict-induced reduction in the usage of intermediate inputs confirms the findings from previous studies looking at the West Bank, Libya, Ukraine, and India (Amodio and Di Maio, 2018; Korovkin and Makarin, 2020; Del Prete et al., 2023; Couttenier et al., 2022). 25Firms selling to other companies in conflict-affected countries may react to higher local-level violence by decreasing export volume or increasing prices to cover the export risk (Amodio and Di Maio, 2018; Korovkin and Makarin, 2020). 26The number of workers includes both full-time permanent and (equivalent) temporary workers. Tem- porary workers are redefined as “full-time seasonal or temporary employees”, where full-time temporary workers are those employed for short terms (i.e., less than a year) without a guarantee of employment renewal, working full-time. The variable Number of workers is computed as full-time equivalents, consid- ering temporary workers for their effective months of work. 12 also leads to a reduction in average wage (column A8), explaining the observed decrease in total labor costs.27 Table A13 provides additional results for the labor mechanism. To start, column 1 shows that the null effect of conflict exposure on employment masks significant hetero- geneity. Higher conflict exposure reduces employment for smaller firms, yet this effect decreases with firm size. As discussed in Section 4.3, this difference is also observed in the impact of conflict exposure on sales and total costs, but not on profits. Columns 2 and 3 provide suggestive evidence supporting our explanation for the reduction in total labor costs, specifically the substitution of skilled workers with unskilled ones. Results in column 2 indicate that this substitution effect is more pronounced for firms producing non- differentiated products.28 This is expected because the production of differentiated goods typically requires higher-skilled workers, making it less feasible to substitute high- skill workers with low-skill ones. Additionally, column 3 indicates that the magnitude of the substitution effect varies with a country’s labor legislation. In contexts where firing costs are higher (proxied by severance pay levels), the substitution of skilled workers for unskilled ones due to conflict exposure is significantly smaller.29 These results are con- sistent with the idea that worker substitution is a defensive strategy employed by the firm to reduce labor costs when hit by conflict. The remaining columns of Table A13 provide evidence that the substitution of skilled workers —and the consequent reduction in total labor costs— is a choice of the firm and not due to other mechanisms. Column 4 shows that such substitution is not due to the supply of skilled workers becoming a more serious obstacle to firms’ activity (for instance, because of the conflict-induced decline in the education level of the labor force). At the same time, column 5 excludes that the reduction in labor costs is due to a conflict-induced increase in the share of temporary workers employed by the firm. Finally, results in column 6 rule out the possibility that the substitution of skilled workers for unskilled ones is linked with changes in perceptions of the strictness of labor regulations. Taken together, these findings provide suggestive evidence consistent with the substitution of workers being driven by changes in labor demand due to the conflict-induced reduction in the firm’s production possibilities, while excluding an important role played by changes in the labor supply. 27Individual wage data at the firm level is not reported by the WBES. We compute the average wage as the ratio between the total cost of work and the overall number of employees (full-time equivalent, see Footnote 26). 28Drawing from the Rauch (1999) classification, we define a product as differentiated if its trade is neither occurring on an organized exchange nor based on a reference price. As highlighted by Rauch and Trindade (2002), goods possessing a reference price are sufficiently homogeneous as prices convey all relevant information for international trade; this is not the case for more differentiated goods that lack reference prices. The original classification is available at: https://econweb.ucsd.edu/~jrauch/ rauch_classification.html. 29Data on severance pay for redundancy dismissal (in weeks of salary) are from the ILO database or the World Bank Employing Workers project dataset: https://www.worldbank.org/en/research/ employing-workers/data/redundancy-cost. We define a country to have low firing costs if its average severance pay is below the cross-sectional median of the distribution. 13 4.2.3 Market competition An important factor impacting sales is the extent of market competition faced by a firm. We examine this mechanism in Table 3, columns A9-A11. Column A9 indicates that an increase in conflict intensity does not affect the total number of competitors of the firm. However, there is some evidence related to the level of competition from informal firms.30 While the effect of higher conflict exposure on having informal competitors is positive but not statistically significant (column A10), the conflict situation makes competition from the informal sector a more serious threat to the firm’s activity (column A11). Evidence from previous studies shows that competition from informal firms reduces formal firms’ output (Rozo and Winkler, 2021), employment (Amin, 2023), innovation (Avenyo et al., 2021), and demand for credit, thereby impacting investment (Brancati et al., 2024). More in general, an increase in informality creates obstacles to the proper functioning of the overall economy, as it leads to the misallocation of resources, losses in total factor productivity, and can be associated with welfare losses (Ulyssea, 2018). Taken together, this evidence suggests that the increase in informal competition induced by conflict may help explain the reduction in firms’ sales documented in Table 2. 4.2.4 Demand One alternative explanation for the reduction in sales and expenditure on production inputs is that conflict reduces local consumer demand. Firms may then respond to a contraction in consumer expenditure by reducing output levels and consequently, their purchases of inputs.31 While we cannot directly test this possibility because WBES data do not provide information on consumer demand, Table A14 presents some evidence suggesting that our results are unlikely to be driven by this mechanism. The left panel displays the effect of conflict exposure based on the firm’s main destination market, whether local, national, or international. If the effect of conflict exposure on sales were solely driven by consumer demand, we would expect the impact on firms that mainly sell internationally to be negligible, and, in any case, significantly smaller than for firms operating in domestic markets. However, this is not the case; the effect of conflict exposure on sales does not exhibit statistically significant differences across the three categories of firms. In the right panel of Table A14, we then explicitly consider heterogeneities along firms’ export status. As expected, the negative impact of conflict exposure is somewhat smaller for exporting firms (column 2), and this mitigation increases with the export volume (column 3). Yet, for any given level of exports, the direct effect of conflict on a firm’s sales is negative and highly statistically significant. Our calculations indicate that even for a firm exporting 30A large informal sector is often a defining characteristic of developing countries, with formal and informal firms coexisting within the same sectors and producing similar products (Ulyssea, 2020). 31As noted before, since it is unlikely that conflict reduces prices, output must decrease to lead to a reduction in sales. 14 100% of its product (i.e., a full exporter), a one-standard-deviation increase in nearby conflict events reduces its median sales by about 2.7%. The fact that even firms not dependent on local demand are severely affected by conflict events in their vicinity allows us to rule out changes in demand as the primary explanation for our results. 4.2.5 Other obstacles to firms’ activity Next, we examine whether the reduction in a firm’s sales might stem from a conflict- induced worsening of various obstacles to business activity. Table A15 suggests this is unlikely. Columns 1-4 present results for a range of potential obstacles, including access to finance, corruption, transportation, and availability of land. None of these factors are more likely to be reported as severe obstacles by firms exposed to conflict events, indicating that they do not significantly contribute to explaining our main findings. Additionally, columns 5 and 6 show that the decline in sales is not attributable to a conflict-induced increase in crime (measured by losses due to theft) or increased expenditure on security protection by the firm. Summing up Our results indicate that conflict restricts a firm’s access to intermediate production inputs, particularly imported ones, and to electricity, leading to reduced out- put. In response, firms cut production costs by substituting skilled workers with unskilled ones. Moreover, conflict intensifies the informal competition faced by the firm. As a result, while sales decline due to decreased output and heightened informal competition, profits remain unchanged due to the firm’s adjustments in production input and labor expenditures. Other factors, such as changes in demand and in the intensity of various obstacles to firm activity, do not seem to play a role in explaining our results. 4.3 Heterogeneity In this section, we test the heterogeneous effect of conflict exposure on our main outcomes along several firm-level and country-level characteristics. Firm-level characteristics Table A16 presents heterogeneity tests for the effect of conflict exposure across various firm characteristics, including size, age, government ownership, and foreign ownership. The results indicate that for any of the main outcomes (Sales, Total costs, and Profits) there are no significant differences across these characteristics, with the only exception of firm size. Panels A and B show that conflict exposure reduces sales and total costs for small firms, yet these effects tend to diminish with firm size. As a result, conflict exposure (slightly) reduces profits only for small firms (Panel C). Overall, these findings suggest that larger firms are more resilient to conflict and better equipped 15 to handle its adverse effects, while other firm-level characteristics appear less relevant in mediating the negative effects of conflict. Country-level characteristics Table 4 presents the results from a series of cross-country heterogeneity tests.32 Panel A examines the level of economic development of the country where the firm operates. Results show no significant differences in the impact of conflict exposure between firms located in high-income versus low-income countries. In both cases, increased conflict exposure has no effect on profits while reducing sales and expenditures on production inputs. Results in Panel B also suggest that the distinction between fragile and non-fragile countries, largely used in the literature (see for instance Chami et al., 2021; Verwimp et al., 2019), is not associated with significant differences in the effect of conflict exposure on firm’s performance. One country-level characteristic that does play a role is the quality of the bureaucracy. As shown in Table 4, Panel C, conflict exposure has a more detrimental effect in countries with low-quality bureaucracy. In these economies, the negative impact on sales is so severe that, even though firms reduce their expenditure on production inputs in response to conflict, their profits still decline. Conversely, firms in countries with high-quality bureaucracy, even if still forced to cut back on production, do not experience significant reductions in sales or profits.33 This aligns with the idea that ceteris paribus, a more efficient government can better support firms in coping with negative shocks. Finally, we examine whether the impact of conflict on firms varies with the conflict status of the country at the beginning and during the whole period of analysis. We begin by comparing countries that —at the time of the first survey— are at peace (initially at peace) versus those in which a major episode of political violence is ongoing (initially at conflict ). Table 4, Panel D, highlights substantial heterogeneity related to this country’s characteristic. For firms in countries initially at peace, an increase in conflict exposure leads to a sizable decline in sales and expenditure on production inputs. As a result, the effect of conflict exposure on profits is significantly negative and the magnitude is substantial: a one-standard- deviation increase in the number of nearby political violence events results in a 3.9% reduction in median profits. On the contrary, for firms in countries already in a conflict situation at the beginning of the period, the reduction in sales and production input expenditure is much smaller and not statistically significant, resulting in a null impact on profits. As shown in Table A18, we obtain very similar results when we compare firms in countries not always in conflict in the sample versus countries always in conflict in the sample.34 One 32Definitions for all variables used in the following analyses are provided in Section 2.3. 33Table A17 shows similar results when countries are classified by their level of corruption. 34In practice, we compare firms operating in countries that have been continuously affected by conflict with those in countries that have experienced periods of peace or intermittent conflict. As noted by Custodio et al. (2024), a priori, it is not clear how the impact of additional conflict events might vary based on the frequency of past conflicts and expectations about future ones. In a context of continuous conflict, firms may adapt and develop strategies to manage such negative conditions, potentially improving their 16 possible explanation for these results is that firms in countries where conflict is business as usual may have already adjusted their production strategies to cope with the effects of conflict on the overall economy. Conversely, firms in countries initially at peace or not always in conflict in the sample might not have yet undergone such costly adjustments, thus suffering larger negative effects for any conflict event.35 4.3.1 Mechanisms heterogeneity In the following, we explore heterogeneities in the mechanisms to explain why conflict exposure has a more detrimental effect on profits, sales, and production input expenditures for firms in countries with low-quality bureaucracy and those initially at peace. Table 5 Panel A presents results based on the quality of a country’s bureaucracy. Columns A1-A4 show that conflict exposure reduces expenditure on both intermediate and imported inputs more for firms in countries with low-quality bureaucracy. In these economies, conflict exposure also decreases expenditure on electricity and increases the likelihood of power outages, whereas this is the case in high-quality bureaucracy coun- tries. Columns A5-A8 explore the labor market mechanism. Conflict exposure leads to a shift from skilled to unskilled workers only for firms located in low-quality bureaucracy countries. This recomposition lowers total labor costs due to reduced wages. Conversely, total labor costs remain unchanged in countries with high-quality bureaucracy. Finally, columns A9-A11 show that the effect of conflict exposure on market competition is not different across low or high-quality bureaucracy countries. Taken together, these results indicate that the more negative effect of conflict exposure on the economic performance of firms in low-quality bureaucracy countries is due to the larger conflict-induced reduction in the access to production inputs —both domestic and imported— and in the availability of electricity that occurs in countries with a less efficient government. Table 5, Panel B presents heterogeneities based on whether firms are located in coun- tries initially at peace or initially in conflict.36 Columns B1-B4 show that conflict exposure reduces expenditure on intermediate inputs and electricity, and increases the probability of power outages for both groups. However, except for the value share of imported inputs, the adverse effects are more pronounced for firms in countries initially at peace. Columns B5- B8 look at the labor market mechanism. Firms in countries initially at peace substitute skilled workers with unskilled labor, leading to a reduction in the average wage and, consequently, in the total labor cost. For firms in countries initially in conflict, an increase in conflict exposure instead leads to a reduction in the number of workers. Finally, performance over time. Conversely, it is possible that persistent exposure to conflict might exacerbate the effects of new conflict events, leading to increasingly severe impacts on firms. 35Evidence consistent with this interpretation is provided in Table A4, column 2, showing that higher conflict intensity increases firm exit only in countries not always in conflict during our period of analysis. 36As shown in Table A19, we obtain very similar results if we compare firms in countries always in conflict in the sample versus those in countries not always in conflict in the sample. 17 columns B9-B11 examine the impact on market competition. Our findings suggest that while an increase in conflict intensity does not affect the overall number of competitors (column B9) in countries that are initially at peace, it does increase the probability of facing informal competitors (column B10) and the perceived threat of such competition (column B11). Given the significant negative impact of informality on firms’ economic activities (see Section 4.2), these findings help explain the substantial decline in both sales and profits for firms in these countries. Taken together, these results are in line with the interpretation that the differential effect of conflict exposure across countries likely reflects differences in the firm’s adjustment costs to the conflict situation. While firms in countries initially in conflict (or always in conflict in the sample) are likely to have already adapted to a conflict-ridden environment, this may not be the case for firms in countries initially at peace (or not always in conflict in the sample), making them suffer disproportionally more from an increase in conflict exposure. 5 Conclusions Our analysis, covering 91 countries from 2006 to 2019, offers a unique perspective on the effects of conflict on firm activity around the world. The results of this paper document several key findings. We show that higher exposure to conflict reduces a firm’s sales and production costs. Conflict-induced disruptions in access to both domestic and imported intermediate inputs and raw materials force firms to scale down their output, which in turn decreases sales. Moreover, sales are further diminished by the increased competition from informal firms. Although lower output already leads to reduced production costs, firms further cut costs by substituting skilled workers with unskilled ones to offset the decline in sales. As a result of the conflict-induced reduction in both sales and production costs, the average effect of conflict exposure on profits is null. Our analysis also highlights important cross-country heterogeneities. The negative impact of conflict exposure is large for firms located in countries with low-quality bu- reaucracy, which suffer a decrease in both sales and profits. Conversely, the impact is negligible for firms in high-quality bureaucracy countries. Moreover, we find that the negative effect of conflict exposure is small for firms in countries that are in conflict at the beginning or during the whole period of analysis, whereas it is very negative for firms located in countries initially at peace or that are not always in conflict during the sample period. In this case, firms undergo larger reductions in the use of raw and intermediate inputs, leading to greater decreases in output and consequently in sales, a greater substitution of skilled workers with unskilled ones, and an increase in the level of informal competition faced by the firm. Overall, these effects lead to a reduction in both sales and profits. Our findings provide a first contribution to understanding which country-level char- 18 acteristics make firms more vulnerable to the impact of conflict-related events and the mechanisms through which these effects occur. This is a necessary preliminary step in designing effective policies aimed at mitigating the negative impact of conflict on the economy. Two crucial policy implications emerge from our analysis. First, the quality of the government plays a pivotal role in enabling firms to withstand negative economic shocks, underscoring the importance of investing in enhancing government capabilities, especially in contexts where such shocks are likely to occur. Second, policies aimed at maintaining peace, though challenging to pursue, offer significant returns, as the cost of (a new or more violent) conflict is much higher for firms in countries that are at peace. References Abadie, A. and J. Gardeazabal (2003). The economic costs of conflict: A case study of the Basque country. American Economic Review 93 (1), 113–132. ACLED (2023). Conflict Index – July 2023. Altonji, J. G., T. E. Elder, and C. R. Taber (2005). An evaluation of instrumental variable strategies for estimating the effects of catholic schooling. Journal of Human Resources 40 (4), 791–821. Amin, M. (2023). Does competition from informal firms hurt job creation by formal manufacturing SMEs in developing and emerging countries? Evidence using firm-level survey data. Small Business Economics 60 (4), 1659–1681. Amodio, F., L. Baccini, and M. Di Maio (2021). Security, trade, and political violence. Journal of the European Economic Association 19 (1), 1–37. Amodio, F. and M. Di Maio (2018). Making do with what you have: Conflict, input misallocation and firm performance. Economic Journal 128 (615), 2559–2612. Armangu´e -Jubert, T., N. Guner, and A. Ruggieri (2024). Labor market power and devel- opment. American Economic Review: Insights, forthcoming. Avenyo, E. K., M. Konte, and P. Mohnen (2021). Product innovation and informal market competition in sub-Saharan Africa. Journal of Evolutionary Economics 31 (2), 605–637. Berman, N., M. Couttenier, D. Rohner, and M. Thoenig (2017). This mine is mine! How minerals fuel conflicts in Africa. American Economic Review 107 (6), 1564–1610. Bernal, C., M. Prem, F. Vargas, and M. Ortiz (2024). Peaceful entry: Entrepreneur- ship dynamics during Colombia’s Peace Agreement. Journal of Development Eco- nomics 166 (103119). 19 Blumenstock, J. E., T. Ghani, S. R. Herskowitz, E. Kapstein, T. Scherer, and O. Toomet (2024). Insecurity and firm displacement: Evidence from Afghan corporate phone records. Mimeo. Boehm, C. E., A. Flaaen, and N. Pandalai-Nayar (2019). Input linkages and the trans- mission o hoku earthquake. Review of Economics of shocks: Firm-level evidence from the 2011 T¯ and Statistics 101 (1), 60–75. Brancati, E., M. Di Maio, and A. Rahman (2024). Finance, informal competition, and expectations: A firm-level analysis. World Development 173, , 106408. Camacho, A. and C. Rodriguez (2013). Firm exit and armed conflict in Colombia. Journal of Conflict Resolution 57 (1), 89–116. Carvalho, V. M., M. Nirei, Y. U. Saito, and A. Tahbaz-Salehi (2021). Supply chain disruptions: Evidence from the great East Japan earthquake. Quarterly Journal of Economics 136 (2), 1255–1321. Chami, R., R. Espinoza, and P. Montiel (2021). Macroeconomic Policy in Fragile States. OUP Oxford. Collier, P. and M. Duponchel (2013). The economic legacy of civil war: Firm-level evidence from Sierra Leone. Journal of Conflict Resolution 57 (1), 65–88. Couttenier, M., N. Monnet, and L. Piemontese (2022). The economic costs of conflict: A production network approach. CEPR Discussion Paper 16984. Custodio, C., B. Mendes, and D. Mendes (2024). Inventory decisions under political violence. Mimeo. De Roux, N. and L. Martinez (2023). Uncertainty, peace, and investment: Evidence from credit microdata in Colombia. Mimeo. Del Prete, D., M. Di Maio, and A. Rahman (2023). Firms amid conflict: Performance, production inputs, and market competition. Journal of Development Economics 164, 103143. Dube, O. and J. F. Vargas (2013). Commodity price shocks and civil conflict: Evidence from Colombia. Review of Economic Studies 80 (4), 1384–1421. Fisman, R., S. Guriev, C. Ioramashvili, and A. Plekhanov (2024). Corruption and firm growth: Evidence from around the world. Economic Journal 134 (4), 1494–1516. Glick, R. and A. M. Taylor (2010). Collateral damage: Trade disruption and the economic impact of war. Review of Economics and Statistics 92 (1), 102–127. 20 Guidolin, M. and E. La Ferrara (2007). Diamonds are forever, wars are not: Is conflict bad for private firms? American Economic Review 97 (5), 1978–1993. Halpern, L., M. Koren, and A. Szeidl (2015). Imported inputs and productivity. American Economic Review 105 (12), 3660–3703. Korovkin, V. and A. Makarin (2020). Production networks and war. Mimeo. Korovkin, V. and A. Makarin (2023). Conflict and intergroup trade: Evidence from the 2014 Russia-Ukraine crisis. American Economic Review 113 (1), 34–70. Ksoll, C., R. Macchiavello, and A. Morjaria (2023). Electoral violence and supply chain disruptions in Kenya’s floriculture industry disruptions. Review of Economics and Statistics 105 (6), 1335–1351. Martin, P., T. Mayer, and M. Thoenig (2008). Civil wars and international trade. Journal of the European Economic Association 6 (2-3), 541–550. OECD/SWAC (2020). The Geography of Conflict in North and West Africa. OECD Publishing, Paris. Oster, E. (2019). Unobservable selection and coefficient stability: Theory and evidence. Journal of Business & Economic Statistics 37 (2), 187–204. Piemontese, L. (2023). Uncovering illegal and underground economies: The case of mafia extortion racketeering. Journal of Public Economics 227, 104997. Pinotti, P. (2015). The economic costs of organised crime: Evidence from Southern Italy. Economic Journal 125 (586), F203–F232. Premand, P. and D. Rohner (2024). Cash and conflict: Large-scale experimental evidence from Niger. American Economic Review: Insights 6, 137–153. Raleigh, C., R. Linke, H. Hegre, and J. Karlsen (2010). Introducing ACLED: An armed conflict location and event dataset. Journal of Peace Research 47 (5), 651–660. Rauch, J. E. (1999). Networks versus markets in international trade. Journal of Interna- tional Economics 48 (1), 7–35. Rauch, J. E. and V. Trindade (2002). Ethnic Chinese networks in international trade. Review of Economics and Statistics 84 (1), 116–130. Rohner, D. (2023). Conflict, civil wars, and human development. In K. Zimmermann (Ed.), Handbook of Labor, Human Resources and Population Economics. Springer. 21 Rozo, S. V. (2018). Is murder bad for business? Evidence from Colombia. Review of Economics and Statistics 100 (5), 769–782. Rozo, S. V. and H. Winkler (2021). Is informality good for business?: The impacts of inflows of internally displaced persons on formal firms. Journal of Human Re- sources 56 (4), 1141–1186. Schrodt, P. A. and O. Yilmaz (2012). Conflict and mediation event observations code- book. Center for International Political Analysis Institute for Policy and Social Re- search—University of Kansas. Shilliday, A. and J. Lautenschlager (2012). Data for a global ICEWS and ongoing research. Tech. rep., 2nd International Conference on Cross-Cultural Decision Making: Focus 2012 . Topalova, P. and A. Khandelwal (2011). Trade liberalization and firm productivity: The case of India. Review of Economics and Statistics 93 (3), 995–1009. Ulyssea, G. (2018). Firms, informality, and development: Theory and evidence from Brazil. American Economic Review 108 (8), 2015–2047. Ulyssea, G. (2020). Informality: Causes and consequences for development. Annual Review of Economics 12, 525–546. Utar, H. (2024). Firms and labor in times of violence: Evidence from the Mexican drug war. World Bank Economic Review , forthcoming. ¨ ck (2019). The microeconomics of violent conflict. Verwimp, P., P. Justino, and T. Bru Journal of Development Economics 141, 102297. WHO (2002). World report on violence and health. World Health Organization, Geneva. 22 Figures Figure 1: Coverage of the panel component of the WBES dataset. Note: Dark blue countries are those included in our sample. Source: WBES Figure 2: Geolocalization of conflict events in the ICEWS dataset (2006-2019). Note: Red dots indicate political violence events in our sample of countries. Source: ICEWS. 23 Tables Table 1: Descriptive statistics Mean Median SD Min Max Age 2.766 2.833 0.800 0.000 5.352 Size 3.295 3.045 1.386 0.000 12.044 Export 15.814 0.000 36.488 0.000 100.000 Sales 12.543 12.716 2.877 0.000 29.205 Profits 11.613 12.501 5.358 -18.715 25.788 Total cost 11.461 11.437 2.693 -3.830 21.642 Raw mat & interm cost 11.243 11.548 3.305 0.000 23.431 Electric expenditure 8.026 8.050 2.777 0.000 19.576 Power outages 0.557 1.000 0.497 0.000 1.000 Imported input 32.429 20.000 36.220 0.000 100.000 Labor cost 10.563 10.687 2.642 0.000 21.574 Number workers 3.381 3.135 1.319 0.000 8.700 % unskilled workers 29.237 20.000 31.545 0.000 100.000 Unit wage 7.253 7.556 1.985 0.000 15.716 Number compet 2.578 3.258 1.005 0.000 3.434 Informal (dummy) 0.502 1.000 0.500 0.000 1.000 Informal (intensity) 1.593 2.000 1.399 0.000 4.000 Conflict exposure 2.670 2.639 2.125 0.000 7.191 Low-middle income country 0.665 1.000 0.472 0.000 1.000 Fragile country 0.110 0.000 0.313 0.000 1.000 Low quality bureaucracy country 0.759 0.000 1.523 0.000 6.786 Country always in conflict in the sample 0.286 0.000 0.452 0.000 1.000 Notes: Descriptive statistics for the main variables employed. All variables are defined in Table C1. Table 2: Conflict exposure: sales, total costs, and profits Dependent variable: Sales Total cost Profits (1) (2) (3) (4) (5) (6) Conflict exposure -0.163*** -0.168*** -0.118*** -0.136*** -0.0742 -0.0892 [0.0177] [0.0179] [0.0190] [0.0188] [0.0643] [0.0680] Size 0.651*** 0.687*** 0.637*** [0.0240] [0.0249] [0.0793] Age 0.0720** 0.0509 0.188* [0.0327] [0.0331] [0.108] Export 0.000911* 0.00172*** -0.000927 [0.000531] [0.000562] [0.00211] Firm FE yes yes yes yes yes yes Time FE yes yes yes yes yes yes Adj R-squared 0.677 0.703 0.745 0.773 0.219 0.224 Indentifying obs. 33,713 29,168 22,178 19,504 20,764 18,360 Total obs. 37,080 34,196 29,447 27,444 28,390 26,521 Notes: High-dimensional fixed effect estimates. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. 24 Table 3: Mechanisms: raw materials and intermediate inputs, labor, and market competition Raw materials and intermediate inputs Labor Market competition Dependent variable: Raw mat Electricity Power Imported Labor Number % unskilled Unit Number Informal Informal & interm cost expenditure outages input cost workers workers wage compet (dummy) (intensity) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Conflict exposure -0.181*** -0.161*** 0.0146*** -1.797*** -0.177*** -0.00234 1.013** -0.150*** -0.00102 0.00695 0.0229* 25 [0.0324] [0.0201] [0.00436] [0.576] [0.0170] [0.00364] [0.442] [0.0156] [0.0178] [0.00489] [0.0134] Firm controls yes yes yes yes yes yes yes yes yes yes yes Firm FE yes yes yes yes yes yes yes yes yes yes yes Time FE yes yes yes yes yes yes yes yes yes yes yes Adj R-squared 0.662 0.639 0.241 0.535 0.716 0.928 0.220 0.584 0.203 0.203 0.260 Indentifying obs. 12,148 23,948 34,946 4,794 27,344 34,711 14,427 26,944 9,202 32,525 32,525 Total obs. 16,730 30,760 37,940 8,493 32,987 37,756 17,887 32,675 15,898 36,430 36,430 Notes: High-dimensional fixed effect estimates. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. Table 4: Main results: Heterogeneity by country income class, fragility, quality of bureau- cracy, and conflict status PANEL A Sales Total cost Profits (1) (2) (3) Conflict exposure × Low-middle income country -0.172*** -0.153*** -0.0978 [0.0209] [0.0204] [0.0783] Conflict exposure × High-middle income country -0.155*** -0.0658 -0.0548 [0.0327] [0.0453] [0.131] Firm controls yes yes yes Firm FE yes yes yes Time FE yes yes yes Adj R-squared 0.703 0.773 0.224 Indentifying obs. 29,168 19,504 18,360 Total obs. 34,196 27,444 26,521 PANEL B Sales Total cost Profits (1) (2) (3) Conflict exposure × Fragile country -0.111*** -0.173*** -0.0149 [0.0239] [0.0228] [0.110] Conflict exposure × Not fragile country -0.178*** -0.128*** -0.105 [0.0183] [0.0195] [0.0702] Firm controls yes yes yes Firm FE yes yes yes Time FE yes yes yes Adj R-squared 0.703 0.773 0.224 Indentifying obs. 29,168 19,504 18,360 Total obs. 34,196 27,444 26,521 PANEL C Sales Total cost Profits (1) (2) (3) Conflict exposure × High quality bureaucracy country 0.0298 -0.0139 0.0264 [0.0221] [0.0233] [0.0855] Conflict exposure × Low quality bureaucracy country -0.503*** -0.371*** -0.412*** [0.0362] [0.0409] [0.119] Firm controls yes yes yes Firm FE yes yes yes Time FE yes yes yes Adj R-squared 0.736 0.776 0.233 Indentifying obs. 25,190 17,181 16,183 Total obs. 29,589 23,933 23,125 PANEL D Sales Total cost Profits (1) (2) (3) Conflict exposure × Country initially in conflict -0.0341 -0.0161 0.0471 [0.0245] [0.0229] [0.0974] Conflict exposure × Country initially at peace -0.285*** -0.275*** -0.245*** [0.0247] [0.0288] [0.0848] Firm controls yes yes yes Firm FE yes yes yes Time FE yes yes yes Adj R-squared 0.704 0.774 0.224 Indentifying obs. 29,168 19,504 18,360 Total obs. 34,196 27,444 26,521 Notes: High-dimensional fixed effect estimates. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. 26 Table 5: Mechanisms: Heterogeneity by quality of bureaucracy and conflict status of the country Raw materials and intermediate inputs Labor Market competition Dependent variable: Raw mat Electric Power Imported Labor Number % unskilled Unit Number Informal Informal & interm cost expenditure outages input cost workers workers wage compet (dummy) (intensity) Panel A (A1) (A2) (A3) (A4) (A5) (A6) (A7) (A8) (A9) (A10) (A11) Conflict exposure × High-quality bureaucracy country -0.0630* 0.0119 -0.0118* -1.710** 0.0355 -0.00489 0.599 0.0412** -0.0138 -0.00933 -0.0195 [0.0379] [0.0264] [0.00636] [0.747] [0.0227] [0.00479] [0.559] [0.0206] [0.0215] [0.00713] [0.0190] Conflict exposure × Low-quality bureaucracy country -0.484*** -0.411*** 0.0423*** -2.271* -0.544*** -0.00524 2.160*** -0.471*** 0.0520 0.00844 0.0139 [0.0680] [0.0379] [0.00608] [1.217] [0.0351] [0.00609] [0.758] [0.0320] [0.0352] [0.00781] [0.0214] Firm controls & FE yes yes yes yes yes yes yes yes yes yes yes Adj R-squared 0.677 0.644 0.251 0.520 0.727 0.931 0.225 0.602 0.188 0.208 0.269 27 Indentifying obs. 11,055 21,060 30,219 4,298 23,723 30,004 13,319 23,362 7,642 28,007 28,007 Total obs. 14,988 26,812 32,786 7,627 28,581 32,618 16,289 28,303 13,331 31,399 31,399 Panel B (B1) (B2) (B3) (B4) (B5) (B6) (B7) (B8) (B9) (B10) (B11) Conflict exposure × Country initially in conflict -0.0831** -0.0618** 0.00543 -1.953** -0.0147 -0.00899* 0.0989 0.0130 0.0134 -0.00234 -0.0147 [0.0412] [0.0248] [0.00569] [0.822] [0.0207] [0.00504] [0.553] [0.0196] [0.0224] [0.00732] [0.0202] Conflict exposure × Country initially at peace -0.295*** -0.267*** 0.0226*** -1.619** -0.328*** 0.00339 2.251*** -0.300*** -0.0197 0.0142** 0.0521*** [0.0471] [0.0304] [0.00609] [0.722] [0.0257] [0.00488] [0.675] [0.0231] [0.0265] [0.00605] [0.0165] Firm controls & FE yes yes yes yes yes yes yes yes yes yes yes Adj R-squared 0.663 0.640 0.241 0.535 0.718 0.928 0.220 0.588 0.203 0.203 0.260 Indentifying obs. 12,148 23,948 34,946 4,794 27,344 34,711 14,427 26,944 9,202 32,525 32,525 Total obs. 16,730 30,760 37,940 8,493 32,987 37,756 17,887 32,675 15,898 36,430 36,430 Notes: High-dimensional fixed effect estimates. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. A Appendix: Figures Figure A1: Baseline: excluding one country per time. .1 0 -.1 -.2 -.3 0 10 20 30 40 50 60 70 80 90 Profits -.05 -.1 -.15 -.2 -.25 0 10 20 30 40 50 60 70 80 90 Sales -.05 -.1 -.15 -.2 0 10 20 30 40 50 60 70 80 90 Total cost Notes: This figure presents the estimates for our baseline specifications in columns 2, 4, and 6 of Table 2, obtained by sequentially excluding one country at a time from our estimating sample. 95% confidence intervals are reported. 28 B Appendix: Tables Table A1: List of economies and waves included in the analysis Economy Waves Economy Wave Afghanistan 2008-2014 Malawi 2009-2014 Albania 2013-2019 Mali 2007-2010-2016 Angola 2006-2010 Mexico 2006-2010 Argentina 2006-2010-2017 Moldova 2009-2013-2019 Armenia 2009-2013-2020 Mongolia 2009-2013-2019 Azerbaijan 2009-2013-2019 Montenegro 2009-2013-2019 Bangladesh 2007-2013 Morocco 2013-2019 Belarus 2008-2013-2018 Myanmar 2014-2016 Benin 2009-2016 Nepal 2009-2013 Bhutan 2009-2015 Nicaragua 2006-2010-2016 Bolivia 2006-2010-2017 Niger 2009-2017 Bosnia and Herzegovina 2009-2013-2019 Nigeria 2007-2014 Botswana 2006-2010 North Macedonia 2009-2013-2019 Bulgaria 2007-2009-2013-2019 Pakistan 2007-2013 Cambodia 2013-2016 Panama 2006-2010 Cameroon 2009-2016 Paraguay 2006-2010-2017 Chad 2009-2018 Peru 2006-2010-2017 Chile 2006-2010 Philippines 2009-2015 Colombia 2006-2010-2017 Poland 2009-2013-2019 Croatia 2013-2019 Romania 2009-2013-2019 Czechia 2009-2013-2019 Russian Federation 2009-2012-2019 Côte d’Ivoire 2009-2016 Rwanda 2006-2011-2019 Dominican Republic 2010-2016 Senegal 2007-2014 Congo, Dem. Rep. 2006-2010-2013 Serbia 2009-2013-2019 Ecuador 2006-2010-2017 Sierra Leone 2009-2017 Egypt, Arab Rep. 2013-2016-2020 Slovak Republic 2009-2013-2019 El Salvador 2006-2010-2016 Slovenia 2009-2013-2019 Estonia 2009-2013-2019 South Africa 2007-2020 Ethiopia 2011-2015 Suriname 2010-2018 Georgia 2008-2013-2019 Tajikistan 2008-2013-2019 Ghana 2007-2013 Tanzania 2006-2013 Guatemala 2006-2010-2017 Timor-Leste 2009-2015 Honduras 2006-2010-2016 Togo 2009-2016 Hungary 2009-2013-2019 Tunisia 2013-2020 Indonesia 2009-2015 Türkiye 2008-2013-2019 Jordan 2013-2019 Uganda 2006-2013 Kazakhstan 2009-2013-2019 Ukraine 2008-2013-2019 Kenya 2007-2013-2018 Uruguay 2006-2010-2017 Kosovo 2009-2013-2019 Uzbekistan 2008-2013-2019 Kyrgyz Republic 2009-2013-2019 Venezuela, RB 2006-2010 Lao PDR 2009-2012-2016-2018 Viet Nam 2009-2015 Latvia 2009-2013-2019 West Bank And Gaza 2013-2019 Lebanon 2013-2019 Yemen, Rep. 2010-2013 Lesotho 2009-2016 Zambia 2007-2013-2019 Liberia 2009-2017 Zimbabwe 2011-2016 Lithuania 2009-2013-2019 Notes: List of countries with panel and geolocalized information. Source: WBES and ICEWS. 29 Table A2: List of political violence events or conflict-related events 2006-2019 Type of event Number of events ICEWS Arrest, detain, or charge with legal action 442349 Use conventional military force 203460 Use unconventional violence 186438 Abduct, hijack, or take hostage 35723 Protest violently, riot 29332 Physically assault 23750 Expel or deport individuals 21141 Use tactics of violent repression 20798 Mobilize or increase armed forces 18346 Employ aerial weapons 17757 Sexually assault 13919 Conduct suicide, car, or other non-military bombing 12284 Carry out suicide bombing 11587 Coerce 10386 Kill by physical assault 7817 Mobilize or increase police power 6265 Assassinate 5887 Demonstrate military or police power 5229 Torture 4048 Seize or damage property 1623 Destroy property 1388 Engage in mass killings 1203 Attempt to assassinate 551 Expel or withdraw peacekeepers 528 Expel or withdraw 412 Engage in violent protest for leadership change 206 Use chemical, biological, or radiological weapons 202 Carry out car bombing 129 Carry out roadside bombing 41 Engage in ethnic cleansing 40 Engage in mass expulsion 29 ACLED Battles 91683 Violence against civilians 74565 Explosions/Remote violence 43563 Protests 1312 Riots 7932 Strategic developments 915 Notes: political violence and hostile events in the ICEWS and ACLED datasets between 2006 and 2019. 30 Table A3: Sample selection Firm included in the panel (1) (2) (3) (4) (5) (6) Conflict exposure -0.00102 -0.00106 -0.00142 [0.00167] [0.00167] [0.00167] Profits -0.000590 -0.000686 -0.000582 [0.000432] [0.000528] [0.000526] Sales 0.000201 -0.0000171 -0.00390 [0.00139] [0.00371] [0.00376] Total cost 0.000680 0.00127 -0.00144 [0.00141] [0.00348] [0.00361] Size 0.00928*** [0.00339] Age 0.0188*** [0.00330] Export -0.0000408 [0.0000726] Time FE yes yes yes yes yes yes Country FE yes yes yes yes yes yes Adj R-squared 0.511 0.511 0.511 0.511 0.511 0.512 Indentifying obs. 15,854 15,854 15,854 15,854 15,854 15,854 Total obs. 15,854 15,854 15,854 15,854 15,854 15,854 Notes: High-dimensional fixed effect estimates. The dependent variable is Firm included in the panel, a dummy that equals one if the firm is interviewed in two consecutive years, and zero, otherwise. Ro- bust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Other variables are defined in Table C1. Table A4: Exit Firm exit (1) (2) Conflict exposure 0.00987*** [0.00375] Conflict exposure × Country always in conflict in the sample 0.00862 [0.00609] Conflict exposure × Country not always in conflict in the sample 0.0107** [0.00475] Firm controls yes yes Time FE yes yes Country FE yes yes Adj R-squared 0.076 0.075 Indentifying obs. 6,100 6,100 Total obs. 6,102 6,102 Notes: High-dimensional fixed effect estimates. The dependent variable is Firm exit, a dummy that equals one if the firm exits the panel, and zero, otherwise. The sample is the cross-section of firms interviewed in the first year of a country’s survey. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Other variables are defined in Table C1. 31 Table A5: Conflict and missing values Missing Missing Missing Missing Missing Sales Sales any cost raw mat & interm labor cost (1) (2) (3) (4) (5) Conflict exposure 0.000262 -0.000395 0.00299 0.00244 -0.000176 [0.00197] [0.00211] [0.00210] [0.00204] [0.00166] Size -0.00387** 0.000842 0.000857 0.00138 [0.00170] [0.00188] [0.00184] [0.00165] Age 0.00235 -0.00443 -0.00399 0.00197 [0.00249] [0.00435] [0.00416] [0.00308] Export -0.0000682 -0.0000215 0.00000446 -0.0000669* [0.0000528] [0.0000737] [0.0000734] [0.0000352] Firm controls no yes yes yes yes Region FE yes yes yes yes yes Adj R-squared 0.139 0.133 0.653 0.663 0.136 Indentifying obs. 41,946 38,114 34,182 34,182 34,182 Total obs. 41,956 38,130 34,196 34,196 34,196 Notes: High-dimensional fixed effect estimates. The dependent variable in columns 1-2 is a dummy that equals one if the firm does not declare its sales, and zero otherwise. In columns 3-5, we employ dummies for missing values in one of the components of Total cost, in the cost of raw materials and intermediates (including electricity expenditure), and in labor cost, respectively. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Other variables are defined in Table C1. Table A6: Threats to identification: Interacted time fixed effects, lags, and leads Sales Profits Sales Profits Sales Profits (1) (2) (3) (4) (5) (6) Conflict exposure -0.0512** -0.0814 -0.0724** -0.325* -0.106*** -0.234 [0.0209] [0.0843] [0.0343] [0.171] [0.0352] [0.169] Future conflict exposure 0.0288 0.0559 0.0215 0.0586 [0.0283] [0.0969] [0.0235] [0.0845] Past conflict exposure 0.00358 0.243 0.0463 0.124 [0.0317] [0.159] [0.0302] [0.140] Firm controls yes yes yes yes yes yes Firm FE yes yes yes yes yes yes Area × Time FE yes yes yes yes yes yes Horizon ±1 year ±1 year ±2 years ±2 years Adj R-squared 0.834 0.245 0.834 0.245 0.834 0.245 Indentifying obs. 29,126 18,259 29,126 18,259 29,126 18,259 Total obs. 34,196 26,521 34,196 26,521 34,196 26,521 Notes: High-dimensional fixed effect estimates. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. 32 Table A7: Total costs: Interacted time fixed effects Total cost Raw mat Labor cost & interm cost (1) (2) (3) Conflict exposure -0.0711*** -0.0973** -0.0387* [0.0223] [0.0405] [0.0205] Firm controls yes yes yes Firm FE yes yes yes Area × Time FE yes yes yes Adj R-squared 0.846 0.719 0.831 Indentifying obs. 19,415 12,033 27,278 Total obs. 27,444 16,730 32,987 Notes: High-dimensional fixed effect estimates. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. Table A8: Threats to identification: Excluding large firms and those located in capital cities Panel A Excluding large firms Sales Total cost Profit (1) (2) (3) Conflict exposure -0.215*** -0.156*** -0.107 [0.0301] [0.0308] [0.103] Firm controls yes yes yes Firm FE yes yes yes Time FE yes yes yes Adj R-squared 0.571 0.664 0.201 Indentifying obs. 12,724 8,458 7,881 Total obs. 18,041 14,463 13,925 Panel B Excluding firms in capital cities Sales Total cost Profit (1) (2) (3) Conflict exposure -0.206*** -0.187*** -0.148 [0.0271] [0.0287] [0.0977] Firm controls yes yes yes Firm FE yes yes yes Time FE yes yes yes Adj R-squared 0.722 0.767 0.218 Indentifying obs. 14,880 10,143 9,450 Total obs. 17,593 14,215 13,664 Notes: High-dimensional fixed effect estimates. Panel A restricts the sample to companies below 20 employees. Panel B excludes firms located in capital cities. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. 33 Table A9: Threats to identification: Confounding factors δ= βˆ 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 δˆ for β = 0 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Sales -0.168 -0.174 -0.180 -0.187 -0.193 -0.199 -0.206 -0.212 -0.219 -0.226 -0.233 -3.596 Total cost -0.136 -0.143 -0.150 -0.158 -0.165 -0.173 -0.181 -0.189 -0.197 -0.206 -0.215 -2.447 Notes: In this table, we employ the Oster (2019) approach to test the stability of our estimates with respect to unobservable confounding factors. βˆ, in column 1, is the estimated coefficient of Conflict exposure in the regressions of Table 2, columns 4 and 6 (only significant estimates are reported). In columns 2-11, we report the estimated β for values of δ in the [0; 1] interval. As suggested by Oster (2019) calculations are based on R2max = 1.3R2. δˆ for β = 0 , in column 12, is the estimated value of δ that would reduce the effect of conflict exposure to zero. Table A10: Robustness: Restricted sample Sales Total cost Profits (1) (2) (3) (4) (5) (6) Conflict exposure -0.0980*** -0.114*** -0.125*** -0.137*** -0.0741 -0.0891 [0.0200] [0.0200] [0.0196] [0.0193] [0.0643] [0.0680] Size 0.651*** 0.676*** 0.638*** [0.0257] [0.0252] [0.0793] Age 0.0889*** 0.0630* 0.188* [0.0344] [0.0337] [0.108] Export 0.000917 0.00179*** -0.000927 [0.000574] [0.000575] [0.00211] Firm FE yes yes yes yes yes yes Time FE yes yes yes yes yes yes Adj R-squared 0.737 0.762 0.757 0.783 0.219 0.224 Indentifying obs. 20,758 18,356 20,758 18,356 20,758 18,356 Total obs. 28,386 26,517 28,386 26,517 28,386 26,517 Notes: High-dimensional fixed effect estimates. This table replicates the analysis in Table 2 restricting the sample to firms for which information on all outcomes is available. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. 34 Table A11: Conflict exposure: Manufacturing sector only Dependent variable: Sales Total cost Profits (1) (2) (3) (4) (5) (6) Conflict exposure -0.184*** -0.200*** -0.198*** -0.209*** -0.0854 -0.127 [0.0242] [0.0240] [0.0271] [0.0277] [0.100] [0.105] Size 0.630*** 0.616*** 0.652*** [0.0321] [0.0358] [0.125] Age 0.0606 0.0443 0.262 [0.0418] [0.0458] [0.160] Export 0.00165** 0.00222*** -0.0000252 [0.000643] [0.000724] [0.00279] Firm FE yes yes yes yes yes yes Time FE yes yes yes yes yes yes Adj R-squared 0.721 0.742 0.755 0.775 0.185 0.189 Indentifying obs. 16,881 14,813 12,346 11,014 11,796 10,557 Total obs. 19,785 18,390 16,354 15,365 15,939 15,000 Notes: High-dimensional fixed effect estimates. This table replicates the analysis in Table 2 restricting the sample to firms in the manufacturing sector. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. 35 Table A12: Robustness: Alternative definitions of conflict exposure Sales (1) (2) (3) (4) Conflict exposure -0.101*** -0.173*** -0.137*** -0.538*** [0.0151] [0.0186] [0.0165] [0.0580] Adj R-squared 0.702 0.703 0.702 0.717 Indentifying obs. 29,168 29,168 29,168 8,160 Total obs. 34,196 34,196 34,196 12,777 Total cost (1) (2) (3) (4) Conflict exposure -0.0805*** -0.139*** -0.103*** -0.206*** [0.0163] [0.0194] [0.0173] [0.0464] Adj R-squared 0.772 0.773 0.772 0.775 Indentifying obs. 19,504 19,504 19,504 6,232 Total obs. 27,444 27,444 27,444 10,879 Profits (1) (2) (3) (4) Conflict exposure -0.0195 -0.0898 -0.0313 -0.272* [0.0603] [0.0705] [0.0635] [0.141] Adj R-squared 0.224 0.224 0.224 0.264 Indentifying obs. 18,360 18,360 18,360 5,843 Total obs. 26,521 26,521 26,521 10,575 Conflict 10km (log) 20km (sine) 20km (log) 20km (log) Period Prev year Prev year Prev 2 years Prev year Source ICEWS ICEWS ICEWS ACLED Firm controls yes yes yes yes Firm FE yes yes yes yes Time FE yes yes yes yes Notes: High-dimensional fixed effect estimates. This table shows the baseline results of Table 2 using alternative definitions of conflict exposure: in columns 1-4, we employ a smaller buffer (10 km) for the level of conflict exposure, the hyperbolic sine transformation, count the number of conflict events over the last two years, or use the ACLED database to construct our conflict measure, respectively. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. 36 Table A13: labor mechanism: Additional results Dependent variable: Number % unskilled % temporary Obstacle: Obstacle: workers workers workers supply skilled labor regulation (1) (2) (3) (4) (5) (6) Conflict exposure -0.0191** 0.0633 -0.137 -0.00413 [0.00884] [0.106] [0.374] [0.00304] Conflict exposure × Size 0.00494** [0.00236] Conflict exposure × Non differentiated 1.183** [0.498] Conflict exposure × Differentiated 0.845* [0.469] Conflict exposure × Low severance pay 1.495** [0.644] Conflict exposure × High severance pay 0.657 [0.550] Firm controls yes yes yes yes yes yes Firm FE yes yes yes yes yes yes Time FE yes yes yes yes yes yes Adj R-squared 0.928 0.219 0.219 0.229 0.145 0.168 Indentifying obs. 34,711 14,173 14,427 32,515 34,328 34,676 Total obs. 37,756 17,734 17,887 36,401 37,551 37,782 Notes: High-dimensional fixed effect estimates. This table explores alternative dependent variables. Number workers is the log-number of employees. % temporary workers is the proportion of temporary workers out of all workers. Obstacle: supply skilled and Obstacle: Labor regulation are dummy variables taking the value of one if the firm indicates, respectively, an inadequately educated workforce or labor regulation to be a major or very severe constraint to its business activity. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Other variables are defined in Table C1. 37 Table A14: Mechanisms: Excluding demand channel Dependent variable: Sales Dependent variable: Sales (1) (2) (3) Conflict exposure × Local -0.180*** Conflict exposure -0.178*** -0.181*** [0.0231] [0.0184] [0.0185] Conflict exposure × National -0.171*** Conflict exposure × Z 0.0535** 0.0937** [0.0228] [0.0215] [0.000385] Conflict exposure × International -0.136*** Z -0.0600 -0.0315 [0.0325] [0.0854] [0.00141] Z definition: Export Export % Firm controls yes Firm controls yes yes Firm FE yes Firm FE yes yes Time FE yes Time FE yes yes Adj R-squared 0.757 Adj R-squared 0.703 0.702 Indentifying obs. 17,920 Indentifying obs. 29,168 29,426 Total obs. 23,264 Total obs. 34,196 34,353 Notes: High-dimensional fixed effect estimates. This table tests heterogeneities across firms’ serving markets in the effect of conflict on sales. In column 1, we allow the effect of conflict to vary depending on whether the main destination market for the firm’s products is local, national, or international. In column 2, we interact conflict with Export, a dummy taking the value of one for firms exporting more than 10% of their sales. In column 3, we employ the interaction with Export %, the share of sales from exports. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Other variables are defined in Table C1. Table A15: Mechanisms: Other obstacles Obstacle Losses due Cost of Finance Corruption Transport Land to Theft Security (1) (2) (3) (4) (5) (6) Conflict exposure 0.00273 0.00212 -0.00438* 0.00228 0.0107 0.0428 [0.00256] [0.00343] [0.00235] [0.00220] [0.0328] [0.0427] Firm controls yes yes yes yes yes yes Firm FE yes yes yes yes yes yes Time FE yes yes yes yes yes yes Adj R-squared 0.092 0.152 0.047 0.071 0.077 0.089 Indentifying obs. 35,267 35,267 35,267 35,267 32,948 30,701 Total obs. 38,130 38,130 38,130 38,130 36,677 35,232 Notes: High-dimensional fixed effect estimates. This table explores alternative dependent variables re- flecting the main obstacles to the firm’s activity. Obstacles: Finance, Corruption, Transport, and Land are dummy variables taking the value of one if the firm indicates, respectively, that i) access to finance, ii) corruption, iii) transport, or iv) access to land to be a major or very severe constraint to its business activity. Losses due to thefts is the estimated losses from theft, robbery, vandalism, or arson that occurred on an establishment’s premises (in percentage of annual sales). Security costs is a continuous variable measuring the average security costs as a percentage of total annual sales of the firm.. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Other variables are defined in Table C1. 38 Table A16: Main results: Firm-level heterogeneities Sales (1) (2) (3) (4) Conflict exposure -0.264*** -0.146*** -0.155*** -0.157*** [0.0338] [0.0385] [0.0180] [0.0181] Conflict exposure × Z 0.0273*** -0.00803 -0.0251 0.0137 [0.00800] [0.0119] [0.0476] [0.0160] Z Size Age State Foreign owned owned Adj R-squared 0.703 0.703 0.710 0.710 Indentifying obs. 29,168 29,168 28,805 28,772 Total obs. 34,196 34,196 33,936 33,918 Total cost (1) (2) (3) (4) Conflict exposure -0.233*** -0.119*** -0.140*** -0.139*** [0.0345] [0.0395] [0.0192] [0.0193] Conflict exposure × Z 0.0277*** -0.00586 0.0153 -0.00299 [0.00826] [0.0120] [0.0535] [0.0180] Z Size Age State Foreign owned owned Adj R-squared 0.773 0.773 0.772 0.772 Indentifying obs. 19,504 19,504 19,344 19,328 Total obs. 27,444 27,444 27,334 27,320 Profits (1) (2) (3) (4) Conflict exposure -0.214* -0.0629 -0.0745 -0.0809 [0.117] [0.137] [0.0687] [0.0689] Conflict exposure × Z 0.0351 -0.00933 -0.0635 0.0407 [0.0312] [0.0426] [0.137] [0.0543] Z Size Age State Foreign owned owned Adj R-squared 0.224 0.224 0.225 0.225 Indentifying obs. 18,360 18,360 18,210 18,196 Total obs. 26,521 26,521 26,418 26,405 Firm controls yes yes yes yes Firm FE yes yes yes yes Time FE yes yes yes yes Notes: High-dimensional fixed effect estimates. This table explores firm-level heterogeneities in our baseline analysis of Table 2. Large is a dummy taking the value of one if the firm’s number of employees is above the median distribution. Young is a dummy taking the value of one if the firm’s age is below the median distribution. State owned is a dummy taking the value of one for firms with at least 10% of government/state ownership. Foreign owned is a dummy taking the value of one for foreign-owned firms. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. Other variables are defined in Table C1. 39 Table A17: Main results: Heterogeneity by the level of corruption of the country Dependent variable: Sales Total cost Profits (1) (2) (3) Conflict exposure × Low corruption country -0.135*** -0.114*** -0.0414 [0.0202] [0.0212] [0.0771] Conflict exposure × High corruption country -0.276*** -0.199*** -0.280*** [0.0218] [0.0231] [0.0783] Firm controls yes yes yes Firm FE yes yes yes Time FE yes yes yes Adj R-squared 0.733 0.775 0.234 Indentifying obs. 25,190 17,181 16,183 Total obs. 29,589 23,933 23,125 Notes: High-dimensional fixed effect estimates. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. Table A18: Main results: Heterogeneity by conflict status of the country during the period Dependent variable: Sales Total cost Profits (1) (2) (3) Conflict exposure × Country always in conflict in the sample -0.0470* -0.0236 0.0156 [0.0248] [0.0236] [0.101] Conflict exposure × Country not always in conflict in the sample -0.263*** -0.253*** -0.197** [0.0241] [0.0275] [0.0821] Firm controls yes yes yes Firm FE yes yes yes Time FE yes yes yes Adj R-squared 0.704 0.774 0.224 Indentifying obs. 29,168 19,504 18,360 Total obs. 34,196 27,444 26,521 Notes: High-dimensional fixed effect estimates. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. 40 Table A19: Mechanisms: Heterogeneity by conflict status of the country during the period Raw materials and intermediate inputs Labor Market competition Dependent variable: Raw mat Electric Power Imported Labor Number % unskilled Unit Number Informal Informal & interm cost expenditure outages input cost workers workers wage compet (dummy) (intensity) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Conflict exposure × Country always in conflict in the sample -0.0906** -0.0734*** 0.00285 -2.056** -0.0241 -0.00599 -0.176 0.00251 0.0181 -0.00151 -0.0130 41 [0.0421] [0.0254] [0.00589] [0.830] [0.0211] [0.00517] [0.572] [0.0200] [0.0230] [0.00742] [0.0204] Conflict exposure × Country not always in conflict in the sample -0.278*** -0.244*** 0.0238*** -1.515** -0.307*** 0.000489 2.453*** -0.278*** -0.0240 0.0127** 0.0476*** [0.0457] [0.0293] [0.00589] [0.714] [0.0247] [0.00477] [0.650] [0.0222] [0.0260] [0.00601] [0.0165] Firm controls & FE yes yes yes yes yes yes yes yes yes yes yes Adj R-squared 0.663 0.640 0.241 0.535 0.718 0.928 0.220 0.588 0.203 0.203 0.260 Indentifying obs. 12,148 23,948 34,946 4,794 27,344 34,711 14,427 26,944 9,202 32,525 32,525 Total obs. 16,730 30,760 37,940 8,493 32,987 37,756 17,887 32,675 15,898 36,430 36,430 Notes: High-dimensional fixed effect estimates. Robust standard errors in brackets. *, **, *** indicate statistical significance at the 10%, 5%, and 1%, respectively. All variables are defined in Table C1. C Data Appendix Table C1: Variable definition Variable name Variable definition log–age. Question b5 of the WBES questionnaire: “In what year did this establishment begin Age operations?”. Age = ln(1+T–b5), where T is the last fiscal year in the survey wave. Source: WBES. log–employees. Question l2 of the WBES questionnaire: “Looking back, at the end of two Size fiscal years ago, how many permanent, full–time individuals worked in this establishment? Please include all employees and managers”. Size= ln(1+l2). Source: WBES. dummy for exporting firms. Question d3 of the WBES questionnaire: “Coming back to the last fiscal year, what percentage of this establishment’s sales were: national sales [d3a], Export indirect exports (sold domestically to a third party that exports products) [d3b], direct exports [d3c]?”. Export=1 if d3b+d3c ≥ 10%. Source: WBES. log–sales. Question d2 of the WBES questionnaire: “In the last fiscal year, what was this Sales establishment’s total annual sales for all products and services?”. Sales=ln(1+d2). Values are expressed in 2000 constant USD. Source: WBES. log–total labor cost. Question n2a of the WBES questionnaire: “From this establishment’s income statement for the last fiscal year, please provide the following information: a. Total Labor cost annual cost of labor including wages, salaries, bonuses, social security payments”. Labor cost=ln(1+n2a). Values are expressed in 2000 constant USD. Source: WBES. log– expenditure on electricity. Question n2b of the WBES questionnaire: “From this estab- lishment’s income statement for the last fiscal year, please provide the following information: Electric expenditure b. Total annual cost of electricity”. Electric expend=ln(1+n2b). Values are expressed in 2000 constant USD. Source: WBES. log–expenditure on raw materials and intermediates. Question n2e of the WBES question- naire: “From this establishment’s income statement for the last fiscal year, please provide Raw mat & interm cost the following information: e. Total annual cost of raw materials and intermediate goods used in production”. Raw mat & interm=ln(1+n2e). Values are expressed in 2000 constant USD. Source: WBES. log-total cost. Total cost= ln(1+n2a+n2b+n2e) if the firm is in the manufacturing sector, while it is ln(1+n2a+n2b) if it operates in the service sector (for which n2e is not available). Total cost We exclude from the sample unreasonably low values of total costs identified as the very left tail of the total cost-to-sales ratio (below the fifth percentile). Values are expressed in 2000 constant USD. Source: WBES. inverse hyperbolic sine transformation of profits. Profits=sinh−1(Sales − Total cost). Source: Profits WBES. Note: we compute total costs as the sum of Labor cost, Electric expenditure, and Raw mat & interm cost to maximize the size of our estimating sample. dummy for firms experiencing power outages. Question c6 of the WBES questionnaire: “Over Power outages the last fiscal year, did this establishment experience power outages?”. Power outages=1 if c6=Yes. Source: WBES. share of imported inputs. Question d12 of the WBES questionnaire: “In last fiscal year, as a proportion of all material inputs or supplies purchased that year, what percentage of this Imported input establishment’s material inputs or supplies were: d12b. material inputs or supplies of foreign origin?”. Imported input=d12b. Source: WBES. 42 log–number of permanent and temporary workers. The number of temporary workers is adjusted for the number of months of their employment. Question l1 of the WBES ques- tionnaire: “At the end of last fiscal year, how many permanent, full-time individuals worked in this establishment? Please include all employees and managers (Permanent, full-time em- ployees are defined as all employees that are employed for a term of one or more fiscal years Number workers and/or have a guaranteed renewal of their employment and that work a full shift)”. Question l6: “How many full-time seasonal or temporary employees did this establishment employ dur- ing the last fiscal year? (Full-time, temporary workers are all short-term (i.e. for less than a year) employees with no guarantee of renewal of employment and work full-time)”. Question l8: “What was the average length of employment of all full-time temporary employees in the last fiscal year?” N workers=ln(1+permanent+temporary). Source: WBES. share of of unskilled workers. Question l4 of the WBES questionnaire: “At the end of last fiscal year, how many permanent, full-time individuals working in this establishment were: l4a1. workers in highly skilled production jobs, professionals whose tasks require extensive % unskilled workers theoretical and technical knowledge; l4a2. workers in semi-skilled production jobs, techni- cians whose tasks require some level of mechanical or technical knowledge; l4b. workers in unskilled production jobs, whose tasks involve no specialized knowledge?” % unskilled work- ers=l4b/(l4a1+l4a2+l4b). Source: WBES. Unit wage log–average unitary wage. Unit wage=ln[1+ n2a/(permanent+temporary)]. Source: WBES. number of competitors of the firm (quartiles). Question e2 of the WBES questionnaire: “In the last fiscal year, for the main market in which this establishment sold its main product, Number compet how many competitors did this establishment’s main product face?”. We build quartiles considering firms declaring that competitors are “too many to count” to be in the top quartile. Source: WBES. dummy for firms facing competition from informal firms. Question e30 of the WBES ques- tionnaire: “To what degree are practices of competitors in the informal sector an obstacle Informal (dummy) to the current operations of this establishment?”. Available options: i) no obstacle; ii) mi- nor obstacle; iii) moderate obstacle; iv) major obstacle; v) very severe obstacle. Informal (dummy) =1 if e30= ii-v. Source: WBES. measure for the intensity of the perceived obstacles generated by informal competitors. In- Informal (intensity) formal (intensity)∈ (0; 5) and increases linearly with the possible answers of question e30 (whereby i=0 and v=5). Source: WBES. log-number of conflict events (+1) that occurred within a 20 km radius around the firm’s Conflict exposure location during the 12 months preceding the closure of the last fiscal year. Sources: WBES (firms’ geolocalization), ICEWS or ACLED (geolocalized conflict events). dummy for firms located in low or lower-middle (per capita) income countries at the beginning of Low-middle income country the sample period. Source: The World Bank. dummy for firms located in upper-middle and high (per capita) income countries at the High-middle income country beginning of the sample period. Source: The World Bank. dummy for firms located in fragile countries at the beginning of the sample period. Fragile Fragile country states are those experiencing deteriorating governance, prolonged political crises, post-conflict transition, or gradual but still fragile reform processes. Source: The World Bank. dummy for firms located in not fragile countries at the beginning of the sample period. Not fragile country Source: The World Bank. dummy for firms located in countries with low-quality bureaucracy. The International Coun- try Risk Guide (ICRG) provides an index assessing whether the bureaucracy has the strength and expertise to govern without drastic changes in policy or interruptions in government ser- Low-quality bureaucracy country vices. See https://epub.prsgroup.com/products/icrg. Our measure takes the value of one if the country indicator of bureaucracy quality is below the median distribution, and zero otherwise. Source: ICRG. dummy for firms located in countries with high-quality bureaucracy. The International Coun- try Risk Guide (ICRG) provides an index assessing whether the bureaucracy has the strength and expertise to govern without drastic changes in policy or interruptions in government ser- High-quality bureaucracy country vices. See https://epub.prsgroup.com/products/icrg. Our measure takes the value of one if the country indicator of bureaucracy quality is above the median distribution, and zero otherwise. Source: ICRG. 43 dummy for firms located in countries that are in conflict throughout the entire analysis period. We define a country as Always in conflict in the sample if it has experienced one or more major episodes of political violence (such as civil conflicts, ethnic violence, riots, popular protests, or repression of dissidents) continuously throughout the entire period of Country always in conflict in analysis. The list of countries with major episodes of political violence is sourced from the the sample Systemic Peace War dataset. We enrich this list with countries involved in Operation Juniper Shield, a counter- terrorism operation initiated in February 2007 by the United States and partner nations in the Saharan and Sahel regions of Africa. Country not always in conflict dummy for firms located in initially peaceful countries that are not in conflict throughout in the sample the entire analysis period. Sources: Systemic Peace War dataset and Jupiter list. 44