Economic Policy Uncertainty: A - PDF document

1 Economic Policy Uncertainty: A gl
1 Economic Policy Uncertainty: A global view from the firms Matías Braun, Gabriel Fernández, Tiago Ferreira, and Claudio Raddatz1 December 2019 Abstract We construct a novel firm-level economic policy uncertainty (FEPU) index for companies headquartered in more than 50 countries during the last 15 years by applying text analysis to regulatory reports filed by foreign firms listed in the US. The index combines the frequency with which firms report on issues related to both economic policy (FEP) and the uncertainty surrounding it (FU). FEP is not necessarily viewed as negative but is strongly correlated with FU, especially in countries with poor institutions. While FEPU correlates well with existing country level measures of economic policy uncertainty and country risk, it also captures unique features of firms’ assessment of the economic policy environment that are important for their decisions. Country-aggregates of FEPU are negatively correlated with future GDP and investment growth. At the firm-level, higher FEPU is associated to lower capital expenditure and employment, higher cash balances and lower leverage. The effect is not simply related to general uncertainty, nor comes entirely from the fact that FEPU is a good predictor of future fundamentals and is independent of the expected outcome of policy changes. Moreover, the relation between the firms’ decisions and FEPU is stronger for those more financially constrained and when investment is more irreversible. 1 Braun is with ESE Business School, Universidad de los Andes (email: mbraun.ese@uandes.cl). Fernández

is with Asociación Chilena de Seguridad
is with Asociación Chilena de Seguridad (email: jgfernandez.ese@uandes.cl). Ferreira is with the Faculty of Economy and Business, Universidad Alberto Hurtado (email: talves@uahurtado.cl). Raddatz is with the International Monetary Fund (email: craddatzkiefer@imf.org). We gratefully acknowledge comments from seminar participants at IFABS-Medellin 2019 conference, Universidad Adolfo Ibanez, Sociedad Económica de Chile, and ESE Business School, Universidad de los Andes. 2 1. Introduction Economic policy determines the landscape where firms operate, and firms’ views about the direction of future economic policies are a critical determinant of their decisions and economic outcomes (Baker et al., 2016). Faced with heightened uncertainty about the direction of economic policy, firms may find it optimal to postpone business plans until uncertainty resolves, with adverse consequences for their current performance and for macroeconomic activity (Dixit and Pindick, 1994). Yet, the last decade has been a period of significant uncertainty around economic policy. The global financial crisis that started in 2008 pushed the boundaries of monetary policy into formerly uncharted territory, triggered a large global regulatory reform agenda for the financial sector, and resulted in important fiscal expansions in many countries, among many other consequences. Even policies that enjoyed widespread support from academics, policymakers and the public for decades, such as free trade, have come into question lately. Starting with the work of Baker et al. (2016), who constructed a country-level index of economic p

olicy uncertainty derived from textual
olicy uncertainty derived from textual analysis of news articles, a growing literature has studied the macroeconomic consequences of aggregate economic policy uncertainty. Baker et al. (2016), Bordo et al. (2016), Julio and Yook (2016), Berger et al. (2017), Valencia (2017), and Bloom et al. (2018), document an important negative impact on several macroeconomic aggregates (GDP growth, bank lending and liquidity creation, and capital flows, among others). A nascent literature has also explored the consequences for firm behavior, shedding much needed light on the mechanism through which it impacts the economy. Gulen and Ion (2016), Bonaime et al. (2018), Nguyen, Kim and Papanastassiou (2018), Tian and Ye (2017), Nguyen and Phan (2017), Kelly et al. (2016), Kang et al. (2014), Pástor and Veronesi (2012), and Waisman et al. (2015) are some examples that show how economic policy uncertainty reduces corporate investment, M&A activity, and affects stock and option prices, among other effects. 3 This paper builds a new firm-level measure of economic policy uncertainty for a large sample of firms domiciled across 52 countries and jurisdictions to study the drivers of economic policy uncertainty and its consequences for corporate actions. The measure is based on textual analysis of annual SEC 20F filings of foreign firms listed in the US. Following Baker et al. (2016), we measure for each firm and in each year the frequency of sentences containing key terms related to economic policy (FEP) and uncertainty (FU). Our firm-level measure of economic policy uncertainty, which we label FEPU, is the product of thes

e two. Since both the degree of un
e two. Since both the degree of uncertainty and the importance of policies vary for each firm over time, FEPU is a time-varying, firm-level measure. To our knowledge, this is the first firm-level measure of economic policy uncertainty available for a broad sample of countries. Our measure has several advantages over existing ones. A significant part of economic policy uncertainty is likely driven by country-wide policies and regulatory changes that affect all firms, are hard to diversify, and can be adequately captured in country-level measures. However, the consequences and concern about economic policy uncertainty can vary importantly across firms. The literature has already documented that firms in different sectors are diversely affected by various policies and regulations and the degree of uncertainty around policy will also probably exhibit sector-specific time variation. Furthermore, even across firms of the same sector, economic policy uncertainty may vary because of differences in size, geographical location, type of costumers, etc. Indeed, we will document that the idiosyncratic, firm level component of FEPU is quite large and matters for firms’ decisions. Another, more practical advantage of our measure it that is solely based on an official standardized regulatory report. The report must touch on the main factors explaining past firm performance and the risks facing the future. Firms devote significant effort to produce these reports and they reflect the firms’ own views on these issues. An additional advantage of this approach is that all reports are filled in English, regardless of the jur

isdiction of origin of the firm. On the
isdiction of origin of the firm. On the other hand, a shortcoming of our measure relative to news-based ones is that it can only be constructed at annual frequency. 4 We validate our measure in several ways. We show that FEPU is significantly and positively correlated with the country-level EPU measure of Baker et al. (2016) as well as with their US and global EPU measure, consistently with our firms being listed in the US and tilted to firms with transnational operations. The measure is also correlated with the firm-level measures of political risk (PRisk) and overall risk (Risk) of Hassan et al. (2019). We further show that FEPU rises in anticipation of tax increases, especially those of large magnitude. Across sectors, we find that FEPU is higher among firms in the financial sector, which report a heightened concern about overall uncertainty and about monetary policy. Our analysis of the dynamics of economic policy uncertainty reveals that firms have been giving increasing importance to it and express greater uncertainty over time. Nonetheless, on top of this general time trend, FEPU is relatively higher in the years immediately following the global financial crisis (2009 to 2013), which provides further validation to the measure. Furthermore, the dispersion of FEPU across all firms and within a country has also increased importantly during the last decade. This is mainly due to a large increase in FU shortly after the global financial crisis, as the dispersion of FEP does not exhibit a clear trend. FU has a strong global time component, while firm level variation is relatively more important for FEP. As a result,

there is ample within-firm variation
there is ample within-firm variation in FEPU that is not explained by either the country of origin of the firm or the sector to which it belongs. We also find that while uncertainty is positively correlated with a negative tone in the text, policy is not. This suggests that it is the uncertainty surrounding policy changes what matters most to the firm, and not necessarily the policies themselves. This is something that more strongly affects firms in countries with poor institutions. At the country level, an increase in FEPU forecasts lower GDP and investment growth, extending the results previously found for the US for a broad sample of countries. At the firm-level, higher FEPU is associated to lower capital expenditure and employment, higher cash balances and lower leverage. This is consistent with a view where uncertainty increases the incentives to delay investment and makes the firm more conservative by hoarding cash, for precautionary motives, and 5 strengthening their balance sheet. The effect is not simply related to general uncertainty, nor comes entirely from the fact that FEPU is a good predictor of future fundamentals. It is also independent of the expected outcome of policy changes. Moreover, the relation between the firms’ decisions and FEPU is stronger for those more financially constrained and when investment is more irreversible. This paper contributes to the literature in several ways. To the best of our knowledge ours is the first systematic study of economic policy uncertainty across a large sample of countries, including both developed and developing ones. Existing research on the consequences of economic policy uncerta

inty has mostly focused on the US. Whi
inty has mostly focused on the US. While there is an increasing number of countries for which the Baker et al. (2016) measure of economic policy uncertainty has been constructed, this number is still relatively small and, with a few exceptions, relegated to similar, developed economies. Our paper is also related to Hassan et al. (2019), who build a time-varying firm-level measure and document several firm-level results. Their study differs from ours in its focus on political risk as opposed to economic policy uncertainty, and in covering only US firms. The source used in building the measure is also different, as they analyze conference calls’ transcripts which may not only measure what the firm thinks about the relevance of policy and the uncertainty surrounding it, but also capture the opinion of analysts and journalists. Still, it is reassuring that similar results can also be found in our international sample. The rest of the paper is organized as follows: section 2 describes the methodology and the data used, section 3 presents and discusses the main patterns of the measure and decomposes it across several dimensions, section 4 presents econometric results on the relationship between FEPU and macroeconomic aggregates and institutional environments. Section 5 presents the main analysis on the relationship of FEPU and its various components to firm level decisions, and section 6 concludes and highlights areas of future research. 6 2. Building a Firm-level Measure of Economic Policy Uncertainty Our measure of economic policy uncertainty is based on the appro

ach of Baker et al. (2016), who deve
ach of Baker et al. (2016), who developed a method to measure economic policy uncertainty based on textual analysis of newspapers. Their methodology looks for news articles published in 10 major US news outlets that contain at the same time terms related to economic issues (e.g. “economic” or “economy”), uncertainty (“uncertain”, “uncertainty”) and policy and policymakers (“legislation”, “Congress”) as a share of total news articles published in each outlet every month between 1985 and 2015. Baker et al. (2016) also measure the importance of policy issues by computing the share of sentences containing policy terms in U.S. firms’ 10-K filings. The average of this measure across years for each firm is used as policy risk exposure. The product of this and their country EPU index is then used as an explanatory variable in firm-level regressions. Our measure is also based on text analysis using a dictionary built around the concepts of uncertainty and economic policy. However, instead of using news articles we rely on regulatory reports prepared by firms: the 20-F form that foreign firms listed in the US are required to file with the Security and Exchange Commission (SEC). Under the SEC rules, a foreign private issuer must file its annual 20-F report within the four months after the end of the fiscal year covered by the report. A foreign company will qualify as a foreign private issuer if 50% or less of its outstanding voting securities are held by US residents, a minority of its executive officers or directors are US citizens or residents, less than 50% of

the issuer’s assets are located in th
the issuer’s assets are located in the United States, or the issuer’s business is administered principally outside the United States.2 Foreign firms that choose to list their securities in the U.S. are not a random sample. This needs to be considered when interpreting our results. First, they tend to be larger than the typical firm. The way this may affect the results depends on the specific question. For instance, larger firms may be 2 https://www.sec.gov/divisions/corpfin/internatl/foreign-private-issuers-overview.shtml 7 more affected by policy changes but also better able to affect the chosen policies and carve out exceptions (see Hassan et al. (2019), and Shang et al. (2018)). Second, these firms are also likely to be less financially constrained and less affected by local financial conditions. Third, they are also more internationally oriented, which is important when relating their FEPU to local macroeconomic and political and regulatory circumstances. The main advantage of relying on firms’ sources is that the measure likely reflects more precisely the firms’ views on the importance of economic policies and on the degree of uncertainty surrounding them rather than the perceptions of the press or analysts. Of course, these perceptions are not unrelated because companies are an important source for newspapers. Moreover, they are related indirectly because newspapers also report on the performance and decisions of the firms, and may even cover their regulatory filings, which are publicly available. Another advantage of our approach is that the SEC requires firms to report in a strict s

tandardized way. Firms must address 1
tandardized way. Firms must address 19 different items. Three items correspond to the firm’s financial statements, while the rest contained detailed commentaries on all relevant aspects of the company and the offering. Importantly, these are not free-form commentaries but are instead well organized, splitting each item into several (up to 9) sub-categories. The reports are also in the same language (English), which eases concerns that differences in grammar or the use of words might introduce noise. Moreover, in almost all cases, they are reviewed by professional editors.3 Comparability is a critical issue for cross-country analysis. Indeed, Baker et al. (2016) put a lot of effort in resolving language differences when building their index for some countries outside the U.S. and the authors that have replicated it for other countries adapt the same procedure to achieve greater comparability. Still, differences in the characteristics of the newspaper industry, language, style, and ideology, to name a few, make this a challenging task. 3 Of course, there are always some variations in style (see, for instance, Lundholm et al. (2014). 8 There are also some disadvantages of our approach. One is the paucity of data since the reports are filled only once a year, unlike news pieces and conference calls. Still, this also means that the opinions contained in the reports are well thought and argued. Another potential weakness is that the reports contain opinions and explanations about past performance and projections about the future. One cannot rule out that firms might be tempted to manipulate the perception of inve

stors by shifting responsibility for
stors by shifting responsibility for poor results to changes in policy. If this were the case, one would find that uncertainty and economic policy matters play a larger role when the firms do poorly than when they do well. However, the potential bias is of greater concern when looking at results than when looking at investment and financing choices, as we do. For this to matter, one would need that firms anticipating reduced investment prospects start preparing the road by depicting an uncertain economic policy environment. This is unlikely because forecastable factors that may lead management to shift responsibility, such as poor performance, can be controlled for in the analysis. The firm’s actions and opinions are co-determined, and therefore one cannot as readily interpret our results as causal. In our analysis, we consider firm (and therefore country) and year fixed effects and control for the traditional determinants of investment and financing to ease endogeneity concerns arising from omitted variable bias and confounding factors. Another mitigating factor is the temporal distance between the opinions and actions. Firms are required to file their 20-F reports within four months after their fiscal year-end, based on the state of affairs at fiscal year-end—including their audited financial statements—while we relate FEPU to firms’ actions at the end of the reporting year. Furthermore, not all the variation of FEPU is firm-specific, as an important part comes from industry and country developments. Finally, even if one could not fully separate the causal effect of economic policy uncertainty, the question of whe

ther the firm assigns it a role on the
ther the firm assigns it a role on the investment decisions and the way it is financed is interesting on its own. We measure FEPU as the product of the share of sentences on a firm’s 20-F report that contain concepts related to economic policy (FEP) and the share that is related to uncertainty (FU): 9 ܨܧܷܲ�,�=ܨܧܲ�,�×ܨܷ�,� ×100 (1) ܨܧܲ�,�=ܨܧܲ�,�ி�ௌ��௅+ܨܧܲ�,�ெைோ்�ோ�+ܨܧܲ�,�ோாீ௎௅�்ைோ� (2) where ܨܧܷܲ�,� is the firm-level economic policy uncertainty reported by firm � at time �, ܨܧܲ is our measure of the importance assigned to economic policy by the firm, and ܨܷ is the measure of the overall degree of uncertainty reported by the firm. Thus, our unit of observation is a firm-report, and we apply a dictionary-based method to separately measure the three components of FEP and FU as the share of sentences that contain at least one of the words in the corresponding dictionary. The resulting economic policy uncertainty measure, FEPU, is high when concepts related to both economic policies and uncertainty capture a large share of sentences in the report. The measure is truly a time-varying firm-level one since both the degree of uncertainty and the importance of policies vary for each firm over time. We use dictionaries of words that

captures each of these concepts. A
captures each of these concepts. As is standard, we take the list of synonyms and derivations of “risk” and “uncertainty” from the Oxford English Dictionary. We expand this to include similar terms used by Hassan et al. (2019) in building their firm-level uncertainty measure. For the economic policy terms, we begin with Baker et al. (2016)’s list and adapt it to our larger set of countries. In particular, we exclude terms related to issues that are peculiar or very relevant in the U.S. (entitlement programs, health, and national security) but not so much elsewhere. This leaves out terms such as “tort reform”, “Obamacare”, “earned income tax credit”, and so on. We expand the list to include the equivalent government and regulatory bodies in both their English and local denominations like “Banco Central de Chile”, “Australian Securities and Investments Commission”, “Conseil du marché financier”, etc. In implementing our approach, we split our dictionary of economic policies across different dimensions (fiscal, monetary, and regulatory) and separately compute the share of sentences discussing each dimension before adding them into the overall economic policy relevance measure FEP, as described in Equation (2). We also 10 measure the extent to which the overall tone of the text is negative as the ratio of sentences with negative words net of those with positive ones following Loughran and Mcdonald (2014). An alternative approach would be to identify parts of the text (e.g. sentences or paragraphs) that contain both uncertainty and policy-related words at the same time, as in the approach of

Baker et al. (2016) or the one fol
Baker et al. (2016) or the one followed by Hassan et al. (2019). However, the length of the documents we analyze and their structure pose difficulties for these approaches: the average 20-F report has more than a hundred thousand words, deals with an arrange of issues, and some of its parts are not structured in clearly defined paragraphs but in one-sentence bullet points.4 In addition to these practical difficulties, our approach has the advantage of allowing us to explore independently each component of the FEPU. We downloaded machine-readable 20-F filings from the EDGAR database using a Python script and followed the parsing approach from Loughran and Mcdonald (2014) to exclude markup tags, ASCII-encoded graphics, and tables. The reports are available since 1999, but it is not until 2002 that the number of firms reaches a reasonable number. Therefore, we restrict our sample to the 2002-2016 period. We match our data to Compustat, our source for financial and accounting data for firms. Our final sample comprises 7,519 observations, corresponding to 578 non-financial firms and 40 financial firms, headquartered in 53 different countries and jurisdictions outside the U.S. (see Table A1 in the Appendix for the sample countries). Country-level variables come from standard sources: World Development Indicators, International Financial Statistics, Datastream, and World Governance Institutions. 4 Baker et al. (2016) method computes the share of news articles simultaneously containing words in the two types of dictionaries (policies and uncertainty). Applying this method to our text would require substituting the

articles by either paragraphs or sentenc
articles by either paragraphs or sentences as unit of analysis. As mentioned, various parts of the text are not structured as paragraphs, which complicates their use as unit of analysis. On the other hand, sentences are usually short pieces text and finding sentences simultaneously including both types of concepts would be challenging and would likely result in large Type I error. Hassan et al. (2019) method of finding specific types of bigrams around words related to uncertainty is impractical for documents of the length of 20-F reports. 11 3. The Patterns of FEPU across Firms, Countries, Sectors, and Time 3.1 Summary Statistics Table 1 presents summary statistics for the main variables for both non-financial and financial firms. As denoted by the fraction of sentences that contain terms related to economic policy, these matters are quite relevant for firms. On average, non-financial firms devote 6.4% of the sentences in the text to writing about economic policies. The most prevalent policy dimension seems to be fiscal policy, which is mentioned on about 5% of sentences on average. Uncertainty-related concepts account for 3.4%. There is significantly positive correlation between the relevance of policy and uncertainty, meaning that they tend to be mentioned together. Uncertainty is also positively correlated with a negative tone, which is consistent with uncertainty being negative in the view of the firms. Conversely, policy is negatively correlated with a negative tone. All these suggest that, more than the likely outcome of policy, which is not always negative, it is the uncertainty surrounding it that matters mo

st to non-financial firms. Wit
st to non-financial firms. With the caveat of the significantly smaller sample size, the overall patterns are similar across financial firms with a few distinctions. For instance, monetary policy related issues are more frequently discussed among financial firms, taking almost 2% of sentences on average. The correlation between uncertainty and negative tone is also much larger for these firms—0.3 for financial firms versus 0.04 for non-financial firms, resulting in an overall positive correlation between negative tone and FEPU. 3.2 Variance decomposition Table 2 shows the fraction of the overall variance of the policy and uncertainty measures that can be explained by the different dimensions of the data, as captured by the �2 of a linear regression of each measure against various set of fixed-effects that absorb the corresponding dimensions. Across all firms, a very large fraction of the variation in the relevance of economic policy (FEP) (about three quarters) is related to time-invariant characteristics of the firm. Some firms are simply 12 systematically more prone to discuss issues related to economic policy than others. This is due to a large extent to policies being more relevant in some countries and industries. This is not entirely surprising given the differences in the way governments relate to the business sector, and that the application and impact of policies is oftentimes specific to a sector. These patterns largely remain across dimensions of economic policy, although firm specific characteristics explain a larger share of the variation in the importance assigned to monetary policy (93 pe

rcent across of firms and 83 percent am
rcent across of firms and 83 percent among non-financial firms) than to fiscal and regulatory policies. The sources of variation in our uncertainty variable are somewhat different. While time-invariant firm characteristics remain highly relevant, this seems to be more closely associated to sectoral differences than to country variation. Across non-financial firms, overall patterns are similar, but sectoral variation is much less important than across all firms. This indicates that the relevance of this dimension across all firms is largely driven by differences in the coverage of uncertainty in reports of financial firms relative to the rest, consistently with the discussion of Table 1. The relevance of uncertainty and economic policy vary over time. Global cycles explain a small share of variation and are relatively more important for FU than for FEP (10 percent of variance across all firms and 22 across non-financial firms compared to 5 percent in both cases for economic policy). Nonetheless, country and sectoral cycles are by far the most important reason behind movements in both indexes. Here we again find some interesting differences across sectors, with sectoral cycles being relatively more important than country cycles across all firms and the opposite across non-financial firms. This speaks to the presence of a global, financial-sector specific cycle, which is likely related to our sample embedding the recent global financial crisis. This is also consistent with the importance of sector specific cycles in explaining the variance of the monetary policy index across all firms than among non-financial firms (41

and 17 percent, respectively). Puttin
and 17 percent, respectively). Putting policy and uncertainty together in our FEPU measure, we underscore that about two thirds of the variation in FEPU is firm-specific, one third of it coming from the country and industry to which the company belongs. Thus, exposure to changes in economic policy are very much 13 determined by persistent differences across sectors and countries. Time- varying factors, both at the country and sector level also explain an important share of variation and global cycles explain a smaller but non-negligible part. These characteristics validate the approach of Baker et al. (2016) and others that construct a firm-level measure of EPU as the product of a firm, time-invariant exposure to economic policy and a time-varying country-level EPU variable. However, the analysis also indicates that there is a relevant fraction of idiosyncratic firm-level time-varying component in FEPU. 3.3 Time trends Figure 1 shows that there is, on average, an important upward trend in the share of the reports that is devoted to economic policy issues, having this increased by around 30% in the last 15 years (Panel A). The trend is even stronger in the case of uncertainty, which grew by 70% (Panel B). The figure also shows that, while the dispersion of FEP across countries does not exhibit a clear trend (Panel C), the upward trend in uncertainty has been more heterogeneous across countries (Panel D). These traits are inherited to the resulting FEPU measure, which also exhibits an increasing trend as shown in Figure 2. Nonetheless, there is some meaningful variation around that trend, with average

FEPU increasing relatively more around
FEPU increasing relatively more around the global financial crisis, as shown in the bottom panel of the figure.5 The presence of this increasing trend in the average behavior of our measures during 2002-2016 is also present in other higher-frequency measures of economic policy uncertainty (such as Baker et al., 2016) and likely points to a fundamental pattern of the data which may merit further analysis. Nonetheless, the econometric analysis that follows will remove this trend by including the corresponding fixed effects. 5 This pattern is present across almost all components of FEPU, with the exception of regulatory policy. 14 3.4 Validation FEPU correlates well with existing measures of economic policy uncertainty and political risk and forecasts major policy changes (Table 3). We start comparing our measures based on firm-level reports with the country-level EPU variables that have been constructed for a smaller set of countries based on Baker et al. (2016)’s methodology.6 Among non-financial firms, regressions of both FEPU and FU on next year average EPU yield positive and significant coefficients (row (1)) controlling for country and year fixed effects, showing that they capture similar yet not identical phenomena.7 When controlling for firm and year fixed effects the coefficient drops and significance disappears, confirming that the relationship comes from the within-country, time series variation in average FEPU rather than from the within-firm, time series variation. Results are weaker for all firms, which is consistent with the variance decomposition showing that country-year variation

is less important for firms in the fi
is less important for firms in the financial sector. Figure 3, which show the correlation between the country-year average of FEPU and EPU, and time series of these two measures for the U.K. confirm these findings. Firms in our sample are non-US companies filing 20-F forms. These firms are listed in the US and usually have operations beyond their home country, so they may be concerned not only about economic policy uncertainty in their country of origin but also in the US and globally. Regressions in rows (2) and (3) confirm that both FEPU and FU are strongly correlated with US indicators of EPU.8 However, the relationship with the global indicator is positive but rarely significant, and 6 These measures have been built on a monthly basis for 23 countries, around one third of those in our sample. 7 We regress against next year average of FEPU for two reasons. 20-F forms are filed in March, based on balance sheet information and analysis as of December 31 of the previous year. Thus, they are both trying to assess economic policy uncertainty on the year of the filing—which in our dating corresponds to t+1—and may also be influenced by events transpired in the first quarter of that year. Second, since Baker et al. (2016) EPU is available monthly, one could match FEPU with EPU at the month of the filing or at December. However, the monthly series of EPU are volatile, and the 20-F reports have an annual perspective that better fits with the yearly average. Nonetheless, contemporaneous relationships yield qualitatively similar results. 8 These regressions replace the year fixed effects for a time tren

d since both the US and global EPU measu
d since both the US and global EPU measures vary only across time. 15 much smaller in magnitude. Thus, beyond the time trend present across measures, the common time series variation in FEPU across all firms seems to come mainly from events in the listing country rather than global events.9 At the firm level, FEPU and FU also correlate well with the firm-level indicators of political risk (PRisk) and overall risk (Risk) built by Hassan et al. (2019) using conference call transcripts. With the caveat that only a fraction of our sample of firms is also covered in Hassan et al. (2019), regressions reported in rows (4) and (5) of Table 3 show that this relationship is particularly strong with FU, our firm-level measure of uncertainty, and especially between FU and Risk. This shows that while both approaches similarly capture firm level uncertainty, concerns about economic policy expressed in 20-F and coverage of political risk in conference calls are distinct. Thus, our FEP measure captures firm’s concerns about economic policy uncertainty in a complementary manner to existing measures. The results also show that the relationship between the measures largely disappears after accounting for firm fixed effects. This indicates that the correlation between firm-level values of the indexes are behind the significant results and that the correlation of their within-firm, time series variation is much weaker. If our measure properly captures firm’s perceptions of economic policy uncertainty, it should reflect firms’ concerns about maĽor policy developments in their home countries. We test this by estimating the relationship between FEPU and

FU and tax reforms in regressions rep
FU and tax reforms in regressions reported in rows (6)-(8) of Table 3. Data on the timing, type, and magnitude of tax reforms come from the IMF Tax Policy Reform Database. The results show that there is no significant relationship between either FEPU or FU and future tax measures (row (6)). However, tax measures included in the database may not necessarily be large and may encompass both tax increases and reductions. In fact, the relationship 9 When excluding the time trend, the coefficients are always positive and significant for both US and global EPU. Indeed, in this case the association with global EPU is somewhat stronger. Thus, both FEPU and global EPU share a common time trend. This could be a meaningful relationship but could also be a spurious result. For this reason, we focus on the de-trended relationships. 16 turns significantly positive when focusing on future tax increases (row (7)), and especially so in anticipation of major tax increases (row (8)). This clear relationship between our measures and future major policy changes, of the sort that should be captured in our indicators, provide further reassurance that the measure is not capturing spurious information. 4. Economic policy uncertainty and country-level performance and institutions In this section we explore whether FEPU relates to macroeconomic outcomes and institutional settings.10 We first aggregate FEPU across firms in a given country to obtain a country-year measure and estimate its dynamic relationship with GDP and investment using local projections (Jorda, 2005). The goal is not to establish causality, since policy and macroecon

omic outcomes are likely jointly deter
omic outcomes are likely jointly determined, but to explore whether reduced form shocks to FEPU forecast changes in macroeconomic variables at different horizons.11 We focus on GDP and investment as the main variables to maintain a parsimonious specification and include maximum 4-year ahead forecasting horizon because of the short time series available. The estimated local projection specification corresponds to Δ��,�+ℎ=ߚℎ′Δ��,�−1+ߛℎΔܨܧܷܲ+��,�+ℎ where ��,�=(ܩܦܲ,�) is a vector of macroeconomic variables for country � at year � consisting of (log) GDP and (log) investment, ߚℎ is a vector of coefficients capturing the relationship between 10 Baker et al. (2016) document that their news-based country EPU is related to macroeconomic aggregates such as industrial production, and to the value of the stock market. 11 As demonstrated by Jorda (2005), the paths for the dynamic relationships obtained through local projections are equivalent to the reduced-form impulse responses obtained in a VAR analysis, with the advantage of a simpler estimation and the possibility of more easily accommodating non-linear features, such as interaction effects. 17 lagged changes in the macroeconomic variables and future cumulative changes in macroeconomic variables ℎ periods ahead.12 As can be seen in Figure 4, the cumulative impulse responses for shocks to the lagged value of FEPU on b

oth GDP and investment are negative
oth GDP and investment are negative and significant at most forecasting horizons, meaning that there is at least some effect of economic policy uncertainty that feeds back into the determination of aggregate outcomes. The effect is material in economic terms: a one standard deviation shock to (changes in) FEPU today predicts a cumulative average decline of about 20 bps of GDP growth three years ahead (total cumulative decline of 60 bps) and to about 80 bps of lower investment growth 2 years ahead (total of 160 bp over the 2 year period). Across countries, firms headquartered in richer and institutionally developed countries tend to refer less often to economic policy in their 20-F filings. On the other hand, economic policy appears to be more frequently reported by firms in countries with a larger government sector. Furthermore, as seen in Table 1, there is a significant positive relation between the share of filings devoted to economic policy (FEP) and the references to uncertainty (FU) across non-financial firms. One can investigate whether there are differences across countries on this relation to have an idea of the circumstances under which firms’ perceptions of uncertainty are more closely related to economic policy issues. To do this, in Table 5 we regress FEP on FU and the interaction between the latter and different country characteristics. The signs of the interaction coefficients show that the tendency to attach uncertainty to matters related to economic policy is attenuated in wealthier countries (per GDP per capita) and in those where policy is likely o

f better quality; that is, where go
f better quality; that is, where government is more effective, regulations are of high quality, and the law rules. Furthermore, greater accountability seems also to reduce the correlation with uncertainty, probably because bad 12 The specification implicitly considers country and year fixed effects by first partialling out these fixed effects from the variables before the estimation. Results are similar if the fixed effects are included into each estimation, but this method allows such fixed effects to change with the forecasting horizon, changing therefore their interpretation. 18 policies aren’t as likely to last in these environments. All these characteristics are highly correlated across countries and, therefore, is difficult to disentangle how much of the effect of the institutions is direct and how much works through increasing the country’s income, although the results reported in the last two columns, which simultaneously include all interaction terms suggest that the quality of institutions is more relevant than the level of GDP per capita in attenuating the relationship between uncertainty and economic policy for firms. These results are consistent with a view where economic policy has a major impact on the firms. But, if strong institutions are in place, firms do not pay much attention to changes in policies, and when they do, these are not as strongly associated to uncertainty. 5. Economic policy uncertainty and firm-level behavior 5.1 Baseline results Standard investment models predict that economic policy uncertainty should affect firm’s decisions. Increased economic policy uncertainty relates to higher vola

tility, which raises the option val
tility, which raises the option value of delaying irreversible investments, such as in physical assets or specialized labor, thereby reducing today’s investment and hiring.13 Uncertainty may also affect the capital structure of the firm because of precautionary motives that lead firms to behave more conservatively by hoarding cash and reducing indebtedness. Regressions in Table 6 investigate the relationship between investment in fixed assets (the ratio of capital expenditures over net property plant and equipment), employment growth, cash holdings (the ratio of cash and equivalents to total assets) and net debt issuance and our firm-level measure of economic policy uncertainty FEPU. The specifications add time and firm-fixed effects to focus on the within-firm variation of the data and control for global trends and shocks. The coefficients thus 13 See, for instance, Dixit and Pindyck (1994). 19 capture the idiosyncratic behavior of a firm when reporting different levels of economic policy uncertainty. The table also reports regressions adding time-varying firm controls, such as (lagged values of) size, profitability (return on assets), the relative importance of fixed assets (property, plant, and equipments over assets), and investment opportunities (Tobin’s Q) in the even columns. All variables are winsorized at the 1% level to avoid outliers driving our results and errors are clustered at the firm-level. The results confirm the existence of a negative and statistically significant relationship between FEPU and investment in fixed assets

and the employment growth rate. The ec
and the employment growth rate. The economic magnitude of the effect is sizeable: a one standard deviation increase in FEPU is associated with a 4 percent increase in the investment ratio and 3 percent higher growth in employment. Also, FEPU is positively associated to cash holdings and negatively with the degree of financial leverage of the firm (the net debt issuance). In all these cases, the impact of a one-standard deviation increase in FEPU is economically meaningful. These results corroborate the findings of Baker et al. (2016) and Gulen and Ion (2016) for the US in a broader set of countries and in a richer empirical setting that controls for common time varying factors and exploits the within-firm, time series variation of the data. 5.2 The sources of FEPU and firm behavior In this section we study what dimensions of variation in FEPU are more relevant for firm behavior. Table 7 disentangles the effect of the three types of economic policies that comprise our broad index, namely fiscal, monetary, and regulatory (see Equation (2)). We multiply each of these components by the measure of uncertainty (FU), so that the their sum adds to FEPU. The results show that there is heterogeneity in the relationship between firms’ decisions and the various components of FEPU. Regulation seems to matter most for financing and capital expenditures. This might be due to changes in regulation being perceived as more permanent than those in fiscal and monetary policies and therefore more relevant for investment and capital structure decisions, which have been shown to be lumpy and anchored to l

ong-term targets. On the other hand,
ong-term targets. On the other hand, the relationship between 20 regulatory policy uncertainty and other decisions that are less lumpy and can be more easily adjusted as precautionary considerations change in time, such as cash holdings, is much weaker or inexistent. Uncertainty regarding fiscal policy matters for all types of firm decisions albeit with various degrees of statistical significance. Interestingly, in several instances the relationship with monetary policy uncertainty has the opposite sign than other policy dimensions and this opposite sign is also statistically significant for cash holding decisions. In contrast to fiscal, regulatory, or overall economic policy uncertainty, firms tend to reduce cash holdings when monetary policy uncertainty increases. A possible explanation is that cash is a nominal variable that is affected by changes in short term rates and inflation resulting from monetary policy uncertainty. Thus, precautionary motives for cash holdings are smaller when the source of uncertainty can result in fluctuations in the value of such holdings. Movements in firm-level economic policy uncertainty can also be decomposed into a time-varying component that affects all firms in a country, and a firm specific component. Results in Table 8 test for their relative importance. This is important because the literature studying the consequences of economic policy uncertainty across countries has only focused on the country component. However, the results show that the firm-specific, time varying component of FEPU is as relevant as the country-time variation in policy uncertaint

y when explaining investment and hiring
y when explaining investment and hiring (columns (1) through (4)). Moreover, when looking at capital structure and cash, it is the firm-level component that drives the results (columns (5) to (8)). This suggests that counting with firm-level information on perceived economic policy uncertainty is key to understand their behavior. Table 9 investigates whether FEPU adds information relative to existing country and firm level measures of economic policy uncertainty and political risk. Regressions in columns (1), (4), (7), and (10) add to our specification the newspaper-based country EPU measure of Bloom et.al (2016). In all cases, the coefficient of FEPU maintains its sign, magnitude, and significance. The remaining columns report the results obtaining when adding Hassan’s firm-based measures of political risk 21 (PRisk) and overall risk (Risk), showing that FEPU generally dominates in the sample of firms where both measures are available. This may be because conference calls not only capture what the firm thinks but also what the analysts perceive as being important, which might not coincide perfectly. It might also result from these calls being relatively less informative for non-US firms listed in US markets than for the local firms that compose the bulk of the sample used by Hassan et al. (2019). 5.3 Alternative explanations After some lag, the decisions of the firm to reduce investment and hiring should translate into lower production. We have already seen, at the aggregate level, that GDP growth is lower following periods of high economic policy uncertainty, as perceived by the averag

e firm. The result is confirmed at the
e firm. The result is confirmed at the firm-level, as the rate of growth of sales the following year is negatively correlated with EPU (see Table 10 columns (1)-(2)). This opens the possibility for a different interpretation of the previous results. This is that EPU does not affect the way the firm acts but rather is a good predictor of future fundamentals and it is to them that the firm reacts. The results in columns (3)-(10) do not support this view, as the relationships between FEPU and firm-level decisions are for the most part unaltered when future sales growth is included as an additional explanatory variable. Of course, these regressions suffer from obvious endogeneity and therefore the result should be interpreted with caution. Our point is simply that, when one considers only the part of FEPU that is uncorrelated with future fundamentals, there is still a strong relation with investment and the capital structure. An alternative explanation of the previous findings would be that firms are responding to expected adverse changes in economic policy rather than to the uncertainty surrounding them. This could be possible since firms adopt a more negative tone when their perception of uncertainty is higher, although not when they give more importance to economic policy (see Table 1). Regressions in Table 11 add to the specification a variable that captures the view regarding policy, constructed as the product of the net negative tone indicator and the relevance of economy policy in the text 22 (ܨܧܲ×ܱܶܰܧ). The coefficient for this new variable has the expected sign: when firms discus

s economic policy in their regulatory
s economic policy in their regulatory filings and adopt a negative tone, companies hold back investment and get more conservative in their financing choices. However, this effect is distinct from the economic policy uncertainty channel, as the coefficients for the FEPU variable are largely unchanged and remain statistically significant. The results reported so far indicate that uncertainty matters for firm behavior. However, economic policy is not the only source of uncertainty faced by firms. Since other sources of uncertainty are likely correlated with FEPU, our estimated coefficients might overstate the effect. In this context, the potential role of overall economic uncertainty is of special concern. Although related, the two concepts are not the same: economic policy uncertainty is directly related to the impact of government action, while overall economic uncertainty is associated with how the economy affects the activities of the firm. In Table 12 we add a firm-level Economic Uncertainty index constructed similarly to FEPU but considering words and bi-grams more related to the state of the economy. The correlation between the two measures of uncertainty is high (0.55) and the relationship between economic uncertainty and investment and capital structure is generally negative, as expected. However, the results clearly show that FEPU is a more robust and statistically significant driver of firm behavior. 5.4 Sectoral differences and the channels linking FEPU and firm behavior The firms in our sample differ in multiple dimensions that can be exploited to explore the channels through which econ

omic policy uncertainty influences fi
omic policy uncertainty influences firm behavior. A first dimension is the irreversibility of the firms’ investment decisions. Although the option to delay is always valuable, it is especially so when investment cannot be fully reversed. If investment were fully reversible, the firm would simply undo it when an adverse scenario materializes and recover the costs, but this is not possible when the investment is irreversible. Thus, one would expect that higher uncertainty would be associated with a larger decline in investment for firms in sectors that require less 23 reversible investments. These firms would also be more interested in hoarding cash and keeping a conservative financial structure. To test these predictions, we add to the benchmark specification the interaction between our FEPU variable and two measures of the degree of irreversibility of investment in each of the 48 Fama-French industry classification, following Almeida and Campello (2007).15 The first is a measure of the liquidation value of each type of asset based on the proceeds from discontinued operations for firms in Compustat (gathered by Berger et al. (1996)), corresponding to Liquidation Value = 0.715 x Receivables + 0.547 x Inventory + 0.535 x Fixed Assets. The second is a classification of assets’ durability, based on the cyclicality of the firm’s operations. The idea behind this is that it would be harder for firms to sell their assets if other firms in the industry are also experiencing hardship, that is, if they are in a highly cyclical sector. Since liquidation value is a proxy fo

r reversibility, the coefficients
r reversibility, the coefficients for the interactions between FEPU and liquidation value should have the opposite sign as those for EPU alone, meaning that firm’s decisions should be less correlated with FEPU when assets are more liquid. The opposite is true for the measure of durability, since it is a proxy for irreversibility. This is indeed what the regressions reported in Table 13A show: the relation between FEPU and investment, cash holdings, and capital structure is stronger for firms in industries with more durable assets and weaker when they fetch higher liquidation values. It has long been recognized that financial frictions play a role in the determination of investment, cash holdings and capital structure. Most models imply that the reaction of the firm to shocks is larger when it cannot access freely the capital markets and smooth out these fluctuations.16 This likely also applies to the response to an increase in uncertainty, since the option value of waiting 15 Since the specification includes firm fixed effects, the main value of the irreversibility variables is not identified. 16 See, for instance, Bernanke and Gertler (1996). 24 would be higher for financially constrained firms with limited and sporadic access to financing. Thus, the relationship between economic policy uncertainty and firm decisions should be stronger for firms that are more likely to be financially constrained. We consider two measures of the degree of financial constraints. The first is based on Kaplan and Zingales (1997) measure of a firm’s own view of how constrained it is by access to f

inancing, which they obtain from
inancing, which they obtain from analyzing firms’ annual reports. Lamont et al. (2001) explain this measure with firm characteristics to produce an index of financial constraints: KZ = 1.1001 x Cash flow + 3.139 x Long term debt to total assets – 39.367 x Cash dividends to total assets – 1.314 x Cash holdings to total assets + 0.282 x Tobin’s Q. The second proxy is due to Whited and Wu (2006), and is based on an estimation of the shadow price of equity financing, which is the following linear function of firm characteristics: WW = -0.091 x Cash flow – 0.062 x Dividend payer + 0.021 x Long term debt to total assets – 0.044 x ln(total assets) + 0.102 x Industry sales growth – 0.035 x Sales growth. We compute these at the firm level and then average across the 48 Fama-French industries. Table 13B presents the results obtained when adding the interaction between FEPU and the industry measures of financial constraints. The signs of the coefficients are in general consistent with the view that financial frictions increase the sensibility of investment, cash holdings and capital structure to economic policy uncertainty. 6. Conclusions This paper introduces a novel firm-level measure of economic policy uncertainty across a large set of developed and developing countries for which it did not exist, which we label FEPU. The measure aims to capture the firms’ perceptions of both uncertainty and the importance of economic policy. We corroborate several results documented in the existing literature, both at the country and firm-level, and provide external validation to firm level results that

are almost invariably based on US dat
are almost invariably based on US data. Country-aggregates of FEPU are negatively associated to GDP and investment growth. At the firm-level, higher FEPU is associated to lower capital expenditure and employment, higher cash 25 balances and lower leverage. We also document new facts: while EP is not necessarily viewed as negative, it is strongly correlated with uncertainty, especially in countries with poor institutions. Different sources of economic policy uncertainty elicit heterogeneous responses by firms, with uncertainty about regulatory policy more strongly related to less reversible decisions and uncertainty about monetary policy inversely related to the holdings of cash balances. The relation of EPU with firm-level choices is independent of the expected outcome of policy changes, that have an impact on their own. Also, there is evidence consistent with the idea that changes in policies generate more uncertainty in countries with poor institutions. Finally, our results show that the within-firm across time variation in perceived economic policy uncertainty is a key driver of firms’ real decisions, highlighting the importance of going beyond country level measures to understand the consequences of uncertainty for firms’ behavior. Having built a measure for firms in a large set of countries, a natural avenue for future research is to delve deeper into how these relations vary across them. Countries differ greatly in terms of the magnitude of changes in policy, the process by which these are determined, the ability of firms to affect the regulatory outcome, and their capacity to cope with risks. There is a strong case for t

hinking that all these affect firms in
hinking that all these affect firms in a material way. Our panel allows looking at the role of these institutions while at the same time controlling for many other differences across countries. Here we focused mostly on the average relation between FEPU and firm-level choices and outcomes and just provided evidence that hints at the importance of these diversity. Relatedly, the variety of firms that we have, and the differing context in which they work, present opportunities for further investigating why some firms are more affected or can cope differently with FEPU. Corporate governance, reporting, ownership, and political connections are clear candidates for having an influence. Finally, knowing how the perceptions of analysts, politicians, and the general public differ from those of the firms themselves is of interest. Being able to compare our measure of firms’ view to other indices allows for this. Are they similar? Do some exaggerate? Who captures better the bigger 26 picture? Do firms react to the others’ opinions or is it the other way around? We have advanced on this direction by comparing our measure with other existing ones constructed with a variety of approaches, emphasizing the relevance of firms’ own views of the relevance of the economic policy uncertainty they face for explaining their behavior, but further research is needed. 27 References Almeida, H., and Campello, M. (2007). Financial constraints, asset tangibility, and corporate investment. The Review of Financial Studies, 20(5), 1429-1460. Baker, S. R., Bloom, N. and Davis, S. J. (2016) ‘Measuring Economic Policy Uncertainty’, Working Paper, (March),

pp. 1–79. doi: 10.3386/w21633. B
pp. 1–79. doi: 10.3386/w21633. Bernanke, Ben, Gertler, Mark and Gilchrist, Simon (1996). "The Financial Accelerator and the Flight to Quality. The Review of Economics and Statistics, vol. 78, issue 1, 1-15. Berger, A. N. et al. (2017) ‘Economic Policy Uncertainty and Bank Liquidity Creation’, Ssrn, (October). doi: 10.2139/ssrn.3030489. Bloom, N., Floetotto, M., Jaimovich, N., Saporta‐Eksten, I., and Terry, S. J. (2018). “Really uncertain business cycles”. Econometrica, 86(3), 1031-1065. Bonaime, A., Gulen, H. and Ion, M. (2018) ‘Does policy uncertainty affect mergers and acquisitions?’, Journal of Financial Economics. Elsevier B.V., 129(3), pp. 531–558. doi: 10.1016/j.jfineco.2018.05.007. Bordo, M. D., Duca, J. V. and Koch, C. (2016) ‘Economic policy uncertainty and the credit channel: Aggregate and bank level U.S. evidence over several decades’, Journal of Financial Stability. Elsevier B.V., 26, pp. 90–106. doi: 10.1016/j.jfs.2016.07.002. Dixit, A. K., Dixit, R. K., Pindyck, R. S., and Pindyck, R. (1994). Investment under uncertainty. Princeton university press. El Ghoul, S. et al. (2018) ‘Policy Uncertainty and Earnings Management’, Working Paper. Gulen, H. and Ion, M. (2016) ‘Policy uncertainty and corporate investment’, Review of Financial Studies, 29(3), pp. 523–564. doi: 10.1093/rfs/hhv050. 28 Hassan Stephan Hollander Laurence van Lent Ahmed Tahoun, T. A. (2017) ‘Firm-Level Political Risk’. doi: 10.3386/w24029. Hassan, T. A. et al. (2016) ‘Firm-Level Political Risk: Measurement and Effects’, Ssrn. doi: 10.2139/ssrn.2838644. Jordà, Ò. (2005). Estimation and inference of impulse respon

ses by local projections. American econ
ses by local projections. American economic review, 95(1), 161-182. Julio, B. and Yook, Y. (2016) ‘Policy uncertainty, irreversibility, and cross-border flows of capital’, Journal of International Economics. Elsevier B.V., 103, pp. 13–26. doi: 10.1016/j.jinteco.2016.08.004. Kaplan, S. N., and Zingales, L. (1997). Do investment-cash flow sensitivities provide useful measures of financing constraints?. The quarterly journal of economics, 112(1), 169-215. Kelly, B., Pástor, Ľ. and Veronesi, P. (2016) ‘The Price of Political Uncertainty: Theory and Evidence from the Option Market’, Journal of Finance, 71(5), pp. 2417–2480. doi: 10.1111/jofi.12406. Kang, W., Lee, K. and Ratti, R. A. (2014) ‘Economic policy uncertainty and firm-level investment’, Journal of Macroeconomics. Elsevier Inc., 39(PA), pp. 42–53. doi: 10.1016/j.jmacro.2013.10.006. Lamont, O., Polk, C., and Saaá-Requejo, J. (2001). Financial constraints and stock returns. The review of financial studies, 14(2), 529-554.. Loughran, T. and Mcdonald, B. (2014) ‘Measuring readability in financial disclosures’, Journal of Finance, 69(4), pp. 1643–1671. doi: 10.1111/jofi.12162. Nguyen, N. H. and Phan, H. V. (2017) Policy Uncertainty and Mergers and Acquisitions, Journal of Financial and Quantitative Analysis. doi: 10.1017/S0022109017000175. 29 Nguyen, Q., Kim, T. and Papanastassiou, M. (2018) Policy uncertainty, derivatives use, and firm-level FDI, Journal of International Business Studies. Palgrave Macmillan UK. doi: 10.1057/s41267-017-0090-z. Pastor, L., and Veronesi, P. (2012). Uncertainty about government policy and stock prices. The journal o

f Finance, 67(4), 1219-1264. Tian,
f Finance, 67(4), 1219-1264. Tian, X. and Ye, K. (2017) ‘How Does Policy Uncertainty Affect Venture Capital?’, Ssrn, (919). doi: 10.2139/ssrn.2910075. Waisman, M., Ye, P. and Zhu, Y. (2015) ‘The effect of political uncertainty on the cost of corporate debt’, Journal of Financial Stability. Elsevier B.V., 16, pp. 106–117. doi: 10.1016/j.jfs.2015.01.002. Whited, T. M., and Wu, G. (2006). Financial constraints risk. The Review of Financial Studies, 19(2), 531-559. 30 Tables and Figures Figure 1: Trends and Dispersion in Firm-Level Economic Policy (FEP) and Firm-Level Uncertainty (FU) Panels A and B show the evolution of the median across countries of the within-country average index of firm-level economic policy (FEP) and firm-level uncertainty (FU), respectively, as well as their interquartile range. Panels C and D show the evolution of the standard deviation of the within-country average indexes. A. Firm-level Economic Policy (FEP) (median and interquartile range) B. Firm-level Uncertainty (FU) (median and interquartile range) C. Firm-level Economic Policy (FEP) (standard deviation) D. Firm-level Uncertainty (FU) (standard deviation) 31 Figure 2: Trends and Dispersion in Firm-Level Economic Policy Uncertainty (FEPU) Panel A shows the evolution of the median across countries of the within-country average index of firm-level economic policy uncertainty index (FEPU), as well as its interquartile range. Panel B shows the deviation of the average cross country index with respect to a linear time trend. A. Firm-level Economic Policy (FEPU) (median and interquartile range)

B. Firm-level Economic Policy (FE
B. Firm-level Economic Policy (FEPU) (deviation from linear trend) 32 Figure 3: Firm-Level Economic Policy Uncertainty (FEPU) and Country-Level Economic Policy Uncertainty from Baker et al. (2016) Panel A shows the predicted value of the within-country average index of firm-level economic policy uncertainty index (FEPU) from a panel (country and year) regression of this index against the average country-level economic policy uncertainty index (EPU) of Baker et al. (2016). Panel B shows the time series of FEPU and EPU for the United Kingdom.. A. Country level EPU and Predicted FEPU B. Country level EPU and FEPU in the UK 33 Figure 4: The response of GDP and investment growth to a shock to FEPU Panel A shows the dynamic response, obtained through local projections (see Jorda, 2000), of average cumulative real GDP growth one to four years ahead to a contemporaneous one-standard-deviation shock to the changes in FEPU, as well as its 90 percent confidence interval. Panel B presents the analogous dynamic response for the growth in real investment (growth rate of the fixed capital formation in constant US dollars). The dynamic specifications include country and year fixed effects, which are partialed-out of the macroeconomic variables to ensure their consistency across horizons. Only countries with a minimum of 3 firms filing 20F statements are included in the estimation. A. Response of GDP growth to a shock to FEPU B. Response of investment growth to a shock to FEPU 34 Table 1. Summary statistics and correlation Summary Statistics and Correlations. Fir

m-level Economic Policy (FEP) and Unce
m-level Economic Policy (FEP) and Uncertainty (FU) denotes the fraction of sentences that contains terms related to economic policy and uncertainty, respectively. Firm-level Economic Policy Uncertainty (FEPU) is the product of the share of sentences that contains uncertainty-related words (FU) and the share that contains economic policy terms (FEP). ܨܧܲ.ி�ௌ��௅, ܨܧܲ.ெைோ்�ோ� and ܨܧܲ.ோாீ௎௅�்ைோ� denote the share of sentences containing economic policy key terms specifically related to fiscal, monetary, and regulatory dimensions. Capital Expenditures (CAPEX), Employment growth (EMP), Cash and equivalents to assets (CASH) and Net change in debt to assets (CHDEBT) were obtained from Compustat. 35 Table 2. Variance decomposition This table shows the R-squared coefficient from a regression of Firm-level economic, policy and uncertainty measures on a number of different fixed effects and on country, sector and global trends. Firm-level Economic Policy (FEP) and Uncertainty (FU) denotes the fraction of sentences that contains terms related to economic policy and uncertainty, respectively. Firm-level Economic Policy Uncertainty (FEPU) is the product of the share of sentences that contains uncertainty-related words (FU) and the share that contains economic policy terms (FEP). ܨܧܲ.ி�ௌ��௅, ܨܧܲ.ெைோ்�ோ� and ܨܧܲ.ோாீ௎௅�்ைோ� denote the share of sentences containing economic policy key te

rms specifically related to fiscal, mone
rms specifically related to fiscal, monetary, and regulatory dimensions. 36 Table 3. Relationship with other measures of policy uncertainty and major reforms This table shows the coefficients of a series of bivariate regressions where the dependent variables are our firm level economic policy uncertainty (FEPU) and uncertainty (FU) measures and the correlates are existing alternative measures of policy uncertainty at the country, global, and firm level (rows (2) to (5)), as well as indicators of future tax reforms (rows (6) to (8)). FEPU is the frequency of sentences containing key terms of economic policy (FEP) multiplied by the frequency of sentences containing key terms of uncertainty (FU). Each entry in the table corresponds to a different regression. All regressions include country fixed effects, and all but those in rows (2) and (3) include year fixed effects. Standard errors, reported in parentheses, are clustered at the country level. *, **, and *** denote statistical significance at the 10, 5, and 1% level respectively. 37 Table 4. Economic Policy, Uncertainty and Institutions This table shows the relation between Firm-level Economic Policy (FEP), Firm-level Uncertainty (FU) and Institutions. The dependent variable is Firm-level Economic Policy (FEP) measured as the frequency of sentences containing key terms of economic policy. Firm-level Uncertainty (FU) is defined as the frequency of sentences containing key terms of uncertainty. The other Independent variables include per capita GDP, indicator of quality of institutions from World Governance Indicators and their interaction with Firm-level Uncertainty (FU)

variable. Robust standard errors, re
variable. Robust standard errors, reported in parentheses, are clustered at the firm level. *, **, and *** denote statistical significance at the 10, 5, and 1% level respectively. 38 Table 5. Economic Policy Uncertainty, Investment and Financing This table shows the relation between our measure of Firm-level Economic Policy Uncertainty (FEPU) and capital expenditure over property plant and equipment (Capex), growth of employees (Employment), cash and equivalents over assets (Cash holdings) and new debt issuance over assets (New debt issuance). All these dependent variables were obtained from Compustat. The independent variable, Firm-level Economic Policy Uncertainty (FEPU), is defined as the product of the share of sentences that contains uncertainty related words (FU) and the share that contains economic policy terms (FEP). All specifications include time and firm fixed effects. We also add time-varying firm controls, such as (lagged values of) size, profitability (return on assets), the character of the assets (PP&E over assets) and investment opportunities (Tobin's Q) in the even columns. Robust errors clustered at the firm-level. Significance *10%, **5%, ***1% 39 Table 6. Firm-level Economic Policy Uncertainty decomposed variance and firm fundamentals This table shows the relation between the variance of Firm-Level Economic Policy Uncertainty (FEPU) components (FEPUFISCAL, FEPUMONETARY, FEPUREGULATORY) and firm fundamentals. FEPUFISCAL is the product of the firm-level share of sentences containing terms related to fiscal policy (FEPFISCAL) and firm-level uncertai

nty (FU). FEPUMONETARY and FEPUREGUL
nty (FU). FEPUMONETARY and FEPUREGULATORY are analogously defined for the firm-level share of sentences containing terms related to monetary policy and overall regulation. The dependent variables are capital expenditure over property plant and equipment (Capex), growth of employees (Employment), cash and equivalents over assets (Cash holdings) and new debt issuance over assets (New debt issuance), all obtained from Compustat. Time-varying firm controls include (lagged values of) size, profitability (return on assets), the character of the assets (PP&E over assets) and investment opportunities (Tobins'sQ) in the even columns. Robust errors clustered at the firm-level. Significance *10%, **5%, ***1% 40 Table 7. Firm-Level Economic Policy Uncertainty (FEPU) variability and firm fundamentals This table shows the relation between the Firm-Level Economic Policy Uncertainty (FEPU) variability, decomposed into country-year and across time within-firm variation, and firm fundamentals. FEPU (country-year) is the fitted value of a univariate regression of Firm-Level Economic Policy Uncertainty (FEPU) on country year indicators. FEPU (residual) is the residual of this regression. Firm-Level Economic Policy Uncertainty is defined as the frequency of sentences containing key terms of economic policy (FEP) multiplied by the frequency of sentences containing key terms of uncertainty (FU). The dependent variables are capital expenditure over property plant and equipment (Capex), growth of employees (Employment), cash and equivalents over assets (Cash holdings) and new debt issuance over assets (New debt issuance), all obtaine

d from Compustat. We also add time-v
d from Compustat. We also add time-varying firm controls, such as (lagged values of) size, profitability (return on assets), the character of the assets (PP&E over assets) and investment opportunities (Tobin's Q) in the even columns. Robust errors clustered at the firm-level. Significance *10%, **5%, ***1% 41 Table 8. Performance comparison of Firm-level Economic Policy Uncertainty and other measures This table shows the relation between uncertainty related measures and firm fundamentals. FEPU is defined as the frequency of sentences containing key terms of economic policy (FEP) multiplied by the frequency of sentences containing key terms of uncertainty (FU). Political Risk (PRisk) and Overall Risk (Risk) are Hassan et al.(2019) measures of risk based on conference calls transcripts. Average EPU Bloom- year is the year-average of newspaper based Economic Policy Uncertainty measure made available by Bloom et al. (2016). The dependent variables are capital expenditure over property plant and equipment (Capex), growth of employees (Employment), cash and equivalents over assets (Cash holdings) and new debt issuance over assets (New debt issuance), all obtained from Compustat. We also add time-varying firm controls, such as (lagged values of) size, profitability (return on assets), the character of the assets (PP&E over assets) and investment opportunities (Tobin’s Q) in the bottom panel. Robust errors clustered at the firm-level. Significance *10%, **5%, ***1% 42 Table 10 Economic Policy Uncertainty and Firm fundamentals This table shows the relation between firm-level Economic Policy Uncertainty (FEPU)

and firm fundamentals. The independent v
and firm fundamentals. The independent variable, FEPU is defined as the product of the share of sentences that contains uncertainty-related words (FU) and the share that contains economic policy terms (FEP). The dependent variables are next year's sales growth (also dependent variable on the first two columns), capital expenditures over property plant and equipment (Capex), growth of employees (Employment), cash and equivalents over assets (Cash holdings), and new debt issuance over assets (New debt issuance). All specifications include time and firm fixed effects. We also add time-varying firm controls, such as (lagged values of) size, profitability (return on assets), the character of the assets (PP&E over assets), and investment opportunities (Tobin’s Q) in the even columns. Robust standard errors clustered at the firm-level. Significance * 10%, ** 5%, *** 1%. Table 11 Uncertainty or Expectations of a Negative Outcome? 43 The dependent variable is capital expenditures over property plant and equipment, growth of employees, cash and equivalents over assets, and new debt issuance over assets. The independent variables are the interaction between negative tone and Economic Policy, and firm-level Economic Policy Uncertainty (FEPU), which is the product of the share of sentences that contains uncertainty-related words (FU) and the share that contains economic policy terms (FEP). Negative Tone is defined as the fraction of sentences containing negative tone words net of the amount of sentences with positive words, based on Loughram and McDonalds (2014) sentiment dictionary. All specifications include time and firm fixed effects. We

also add time-varying firm controls,
also add time-varying firm controls, such as (lagged values of) size, profitability (return on assets), the character of the assets (pp&e over assets), and investment opportunities (Tobin’s Q) in the even columns. Robust standard errors clustered at the firm-level. Significance * 10%, ** 5%, *** 1%. Table 12. Economic or Economic Policy Uncertainty ? 44 The dependent variable is capital expenditures over property plant and equipment, growth of employees, cash and equivalents over assets, and new debt issuance over assets. The independent variables are Economic Uncertainty and firm-level Economic Policy Uncertainty (EPU), which is the product of the share of sentences that contains uncertainty-related words (U) and the share that contains economic policy terms (EP). Economic Uncertainty is the product of the fraction of sentences containing terms relating to the economy and the Uncertainty variable. All specifications include time and firm fixed effects. We also add time-varying firm controls, such as (lagged values of) size, profitability (return on assets), the character of the assets (PP&E over assets), and investment opportunities (Tobin’s Q) in the even columns. Robust standard errors clustered at the firm-level. Significance * 10%, ** 5%, *** 1%. 45 Table 13A. Heterogeneity: Irreversibility The dependent variable is capital expenditures over property plant and equipment, growth of employees, cash and equivalents over assets, and new debt issuance over assets. The independent variables are firm-level Economic Policy Uncertainty (FEPU) and the interaction between this and two industry-level variables of irre

versibility based on Almeida and Campell
versibility based on Almeida and Campello (2007). FEPU is the product of the share of sentences that contains uncertainty-related words (FU) and the share that contains economic policy terms (FEP). All specifications include time and firm fixed effects. We also add time-varying firm controls, such as (lagged values of) size, profitability (return on assets), the character of the assets (PP&E over assets), and investment opportunities (Tobin’s Q) in the even columns. Robust standard errors clustered at the firm-level. Significance: * 10%, ** 5%, *** 1%. 46 Table 13B. Heterogeneity: Financial Constraints The dependent variable is capital expenditures over property plant and equipment, growth of employees, cash and equivalents over assets, and new debt issuance over assets. The independent variable are firm-level Economic Policy Uncertainty (EPU) and the interaction between this and two industry-level variables based on Kaplan and Zingales (1977) and Whited and Wu (2006). EPU is the product of the share of sentences that contains uncertainty-related words (U) and the share that contains economic policy terms (EP). All specifications include time and firm fixed effects. We also add time-varying firm controls, such as (lagged values of) size, profitability (return on assets), the character of the assets (pp&e over assets), and investment opportunities (Tobin’s Q) in the even columns. Robust standard errors clustered at the firm-level. Significance: * 10%, ** 5%, *** 1%. 47 Table A1 Economic Policy Uncertainty - Sample countries 48 Table A2 sample countries and and country means of Firm Level Economic Policy U