Policy Uncertainty, Investment, and the Cost of Capital

Transcription

1 Policy Uncertainty, Investment, and the Cost of Capital Wolfgang Drobetz a, Sadok El Ghoul b, Omrane Guedhami c, and Malte Janzen d,e Abstract We examine the effect of economic policy uncertainty on the relation between investment and the cost of capital. Using a novel news-based index developed by Baker, Bloom, and Davis (2016) for 18 countries, we find that the strength of the negative relation between investment and the cost of capital decreases in times of high economic policy uncertainty. For the subsample of U.S. firms, we find similar effects for CPI and tax-related policy uncertainty. Our results further suggest that an increase in policy uncertainty reduces the sensitivity of investment to the cost of capital the most for small firms and for industries that depend more strongly on government subsidies and government consumption. We conclude that economic policy uncertainty distorts the fundamental relation between investment and the cost of capital. Keywords: Economic uncertainty, policy uncertainty, investment, cost of capital JEL Classification Codes: P16, G32 a University of Hamburg, Hamburg Business School, Moorweidenstrasse 18, Hamburg, Germany. wolfgang.drobetz@uni-hamburg.de. b University of Alberta, 8406, Rue Marie-Anne-Gaboury (91 Street), Edmonton, AB, T6C 4G9, Canada. elghoul@ualberta.ca. c University of South Carolina, 1014 Greene Street, Columbia, SC 29208, USA. omrane.guedhami@moore.sc.edu. d University of Hamburg, Hamburg Business School, Moorweidenstrasse 18, Hamburg, Germany. malte.janzen@uni-hamburg.de. e We thank Narjess Boubakri, Joseph Clougherty, Ralf Elsas, Michael Halling, Florian Heider, Emanuel Moench, and Christoph Schneider for helpful comments. We appreciate generous financial support from Canada s Social Sciences and Humanities Research Council.

2 1. Introduction The relation between a firm s cost of capital and its level of investment appears to be unambiguous. Standard corporate finance theory dictates that capital budgeting decisions be made using the net present value (NPV) approach. Under this approach, the decision of whether to invest in a project is based on a forecast of the project s discounted cash flows, where the firm s cost of capital is the discount rate. An increase in the firm s weighted average cost of capital (WACC) leads to lower investment rates. This approach is widely used by financial decision makers in practice. For instance, in their survey of U.S. CFOs, Graham and Harvey (2001) document that almost three-quarters of the survey participants indicate that they use the NPV criterion to evaluate investment opportunities. Despite the theoretical importance of the relation between WACC and investment and the widespread use of the NPV criterion, empirical evidence is scarce. In a recent study, Frank and Shen (2016) address this gap. Building on the theoretical model of Abel and Blanchard (1986), they relate a firm s investment to its cash flow and cost of capital. When they take the model to the data, they find that the inverse relation between corporate investment and the cost of capital holds when the cost of equity capital is constructed using the implied cost of capital approach, but when they employ cost of equity capital estimates based on expected returns from the CAPM and other asset pricing models, the relation between the cost of capital and investment becomes positive. 1 The evidence in Frank and Shen 1 Frank and Shen (2016) find that the CAPM-based cost of equity provides empirically relevant information for investment that is different from that provided by the implied cost of equity capital approach. Most important, the impact of the CAPM-based cost of equity on investment is not the cost of capital effect as in Abel and Blanchard s (1986) model. Frank and Shen (2016) conclude that the CAPM measure must reflect forces outside the standard model. However, when they test alternative mechanisms, they do not find conclusive evidence (see Section 4.7 in Frank and Shen, 2016). 2

3 (2016) thus suggests that the relation between investment and the cost of capital depends on the form that the cost of capital measure takes. In this paper, we extend Frank and Shen s (2016) study by using Baker et al. s (2016) index of policy uncertainty to assess how economic policy uncertainty affects the relation between the cost of capital and investment. Our study builds on a growing literature documenting that uncertainty affects economic outcomes. In particular, uncertainty has been shown to affect economic growth (Baker and Bloom, 2013), bank liquidity creation (Berger, Guedhami, Kim, and Li, 2017), business cycles (Basu and Bundick, 2012; Bidder and Smith, 2012; Christiano et al., 2014; Bianchi at al., 2014; Bloom et al., 2012), investment dynamics (Bachmann and Bayer, 2014), and equity prices as well as risk premiums (Pástor and Veronesi, 2012, 2013; Brogaard and Detzel, 2015). 2 Complementing this aggregate-level evidence, several studies find that uncertainty impacts firm-level outcomes such as capital expenditures (Pindyck, 1993; Dixit and Pindyck, 1994; Bloom, 2009; Julio and Yook, 2012; Gulen and Ion, 2016), R&D expenditures and hiring (Stein and Stone, 2014), mergers and acquisitions (Bonaime et al., 2017; Nguyen and Phan, 2017), equity issuance (Colak et al., 2016; Jens, 2017), firm risk taking (Akey and Lewellen, 2017), cost of corporate debt (Waisman et al., 2015), investment in accruals (Arif et al., 2016), and earnings management (Stein and Wang, 2016). Overall, the empirical evidence suggests that uncertainty amplifies business cycles by affecting corporate decisions. Focusing largely on real options, the theoretical literature on the effects of economic uncertainty predicts a negative relation between uncertainty and investment: the option value 2 See Bloom (2014) for a survey of this strand of literature. 3

4 of a delay in investment is high when uncertainty is high, and firms may prefer to wait and avoid a costly mistake. Put differently, uncertainty increases adjustment costs, making investments expensive to reverse (Bernanke, 1983; Brennan and Schwartz, 1985; McDonald and Siegel, 1986; Ingersoll and Ross, 1992; Dixit and Pindyck, 1994; Schwartz and Trigeorgis, 2004; Grenadier and Malenko, 2010) and inducing managers to be cautious when making investment decisions. The empirical evidence supports this prediction. For example, using Baker et al. s (2016) index of policy uncertainty, Gulen and Ion (2016) examine the effect of policy uncertainty on the level of investment and report that high uncertainty leads to lower investment rates. 3 The negative effect is observed for up to five quarters, after which investment starts to increase and eventually makes up for the delayed investment caused by the elevated level of policy uncertainty. This evidence is consistent with the idea that investment decisions contain an option to wait component. While uncertainty is widely understood to distort the level of investment through the real option channel, Bloom (2014) highlights, in addition to this level effect, a sensitivity effect that has received little attention in the extant literature. Bloom (2014) argues that the real option argument not only suggests that uncertainty reduces the level of economic outcomes such as investment, hiring, or consumption, but also makes economic agents less sensitive to changes in business conditions. For example, when uncertainty is high, the sensitivity of investment to changes in interest rates should be lower; thus a policy intervention in the form of lower interest 3 For earlier empirical evidence on the negative relation between uncertainty and corporate investment, see Pindyck and Solimano (1993), Leahy and Whited (1996), Guiso and Parigi (1999), Bond and Cummins (2004), Eisdorfer (2008), and Badertscher et al. (2013). 4

5 rates will be less effective. 4 This sensitivity effect dampens the productivity-enhancing reallocation of resources across firms and may in turn reduce economic growth. We empirically test this sensitivity effect by examining the degree to which economic policy uncertainty distorts the relation between investment and the cost of capital. We expect an increase in policy uncertainty to reduce the sensitivity of investment to changes in the cost of capital. Our prediction is rooted in prior literature emphasizing the role of the information content of stock prices in managerial decisions. Speculators produce private information and trade on it, implying that their information is incorporated in stock prices. Although managers are better informed than outsiders, they receive information about their company from the company s stock price that would otherwise not be available to them, and they use this information in their investment decisions (Dow and Gorton, 1997; Chen et al., 2007; Bakke and Whited, 2010; Edmans et al., 2012). Because both the volatility of individual stocks and the correlation between stocks increases during times of higher uncertainty (Pástor and Veronesi, 2012, 2013), stock prices are likely to deviate from revelatory price efficiency and include less private information that managers can use to make their investment decisions. 5 This effect should attenuate the negative link between investment and cost of capital during times of elevated uncertainty. 4 Eickmeier et al. (2016) find that monetary policy is more effective in stimulating economic activity in periods of lower volatility. A comparable effect is observed for the sensitivity of private consumption of durable goods to demand and price signals (Foote et al., 2000; Bertola et al., 2005). 5 Bond et al. (2012) refer to the extent to which prices reveal the information necessary for real efficiency as revelatory price efficiency (as opposed to the traditional forecasting price efficiency). Real efficiency implies that the price reveals information necessary for decision makers to take value-maximizing actions. 5

6 Our data set comprises 64,158 observations for 11,220 firms from 18 countries over the time period 1989 to Given the evidence in Frank and Shen (2016), we focus on the exante cost of capital and use the implied cost of equity capital (ICC) measures of Claus and Thomas (2001), Ohlson and Juettner-Nauroth (2005), Easton (2004), and Gebhardt et al. (2001) to construct a firm s WACC. Our cross-country evidence corroborates earlier results based on U.S. data. In particular, we find that an increase in policy uncertainty leads to lower investment, in line with Gulen and Ion (2016), and we find that an increase in the cost of capital decreases investment, consistent with Frank and Shen (2016). More importantly, we find that the link between the cost of capital and investment is distorted by policy uncertainty: the sensitivity of investment to WACC is significantly lower when policy uncertainty is higher, i.e., in high policy uncertainty years the negative relation between investment and the cost of capital becomes less pronounced. When we decompose WACC into its cost of debt and equity components, we find that the direct (level) effects of the cost of debt and the cost of equity on investment are comparable. Interestingly, the sensitivity effects for the cost of debt and the cost of equity are also similar in magnitude. This result is in line with prior findings that stock and bond markets are similar in terms of informational efficiency if institutional trade dominance and other specific bond trading features are accounted for (Downing et al., 2009; Ronen and Zhou, 2013; Tsai, 2014). To shed additional light on our main results, we examine cross-sectional differences across firms and industries. We find that firms are not equally exposed to the effect of policy uncertainty. The moderating effect of policy uncertainty on the link between investment and 6

7 the cost of capital is stronger for small firms and for firms operating in industries that depend more on government subsidies and direct government consumption. In additional robustness tests, we use alternative policy uncertainty indices that also come from Baker et al. (2016) and are available only for the U.S. We find that our main results persist when we use an alternative policy uncertainty index that is based on the dispersion in CPI forecasts, an index based on the expiration of tax provisions, and an aggregate index, but we find no significant effect when policy uncertainty is captured by an index based on future government purchases. Next, we extend our measure of investment to include R&D expenses. The results remain similar. We further include market-to-book (a proxy for Tobin s q), countrylevel controls for the level of stock and debt market development, and proxies for expectations about future investment opportunities to mitigate concerns about omitted factors that may explain corporate investment behavior. Again, our main results continue to hold. We further find that our results are robust to addressing potential biases that could arise if our estimates were driven by one particular model of ICC and its underlying assumptions as well as to controlling for the unbalanced distribution of firm-year observations across countries by using a weighted regression. We contribute to the growing literature on the relation between economic policy uncertainty and investment behavior in four ways. First, we extend Frank and Shen (2016) and Gulen and Ion (2016) to a cross-country setting. We show that the inverse relation between the cost of capital and investment as well as the negative link between policy uncertainty and investment obtain in our international sample of 18 countries. Second, we provide evidence on the prediction in Bloom (2014) that policy uncertainty affects not only the level of investment, as 7

8 documented by Gulen and Ion (2016), but also the (negative) sensitivity of investment to the cost of capital. Third, we find that exposure to policy uncertainty is not equal across firm and industry characteristics. Finally, for the subsample of U.S. firms, we find that the effect of policy uncertainty is not uniform across the different components of the policy uncertainty index. While the effect of policy uncertainly on the sensitivity of investment to the cost of capital using the news-based index persists when we use measures based on CPI forecast dispersion or the expiration of tax provisions, disagreement about future government purchases does not affect the sensitivity of investment to the cost of capital. The remainder of the paper is structured as follows: Section 2 develops our main hypotheses. Section 3 describes the variables used in our analysis, the sample, and our research design. Section 4 presents our main results, and Section 5 contains the results of additional robustness tests. In Section 6 we conclude and discuss policy implications. 2. Motivation and hypotheses The extant literature shows that, at the macro level, uncertainty is strongly countercyclical. Bloom (2009) links the majority of high uncertainty periods in recent years to negative political, social, and economic shocks. Bloom (2014) and Baker et al. (2016) show that uncertainty about future stock prices, as measured by implied volatility, is negatively related to stock market returns. Bachmann et al. (2013), Scotti (2016), Bloom (2014), and Jurado et al. (2015) document that macroeconomic forecasts tend to be noisier, forecasters are less confident about their forecasts, and their forecasts are more optimistic than actual outcomes during recession periods. 8

9 Similar patterns arise at the micro level. For example, volatility in industry- and firmlevel stock returns increases (Campbell et al., 2001), and output growth dispersion across industries widens (Bloom, 2014) during recessions. Using consumer product prices used by the Bureau of Labor Statistics to calculate the Consumer Price Index, Vavra (2014) finds that the volatility of price changes across products increases significantly during recession periods. The above evidence suggests a causal link, rather than mere correlation, between uncertainty and the state of the economy. Prior literature suggests several arguments for how recessions can increase uncertainty. One such argument is based on the idea that firms production, hiring, and investment activities transmit information to the markets. During recessions, these activities slow down. As a result, less new information reaches markets, thereby increasing uncertainty (Fajgelbaum et al., 2014). Another argument holds that uncertainty rises during recessions because they are rarely observed states of the economy, which works against forecasters ability to generate accurate forecasts (Orlik and Veldkamp, 2014). Yet another argument suggests that while policy makers prefer to maintain the status quo when the economy is in a good state, they have incentives to make policy changes during a bad state, which results in higher (policy) uncertainty (Pástor and Veronesi, 2012). It may be the case, however, that a causal link between uncertainty and economic activity runs in the other direction. Bloom (2009) develops a model in which high uncertainty reduces economic growth by inducing firms to optimally postpone investment and hiring. Bloom (2014) shows that uncertainty may further depress economic growth by decreasing optimal resource allocation across firms. Baker and Bloom (2013) provide empirical evidence for a causal relation in which uncertainty drives the state of the economy. 9

10 In this paper, we focus on a specific type of uncertainty, namely, economic policy uncertainty. Politicians and regulatory institutions frequently take decisions that change the environment in which firms operate. Policy uncertainty captures the portion of the overall economic uncertainty that is attributable to the political and regulatory system. Gulen and Ion (2016) show that an increase in economic policy uncertainty has a negative effect on the level of investment: in response to high policy uncertainty, firms reduce investment for up to five quarters, after which investment starts to make up for the delay. This result relates to a strand of literature that links uncertainty and investment via real options. Real option theory predicts a negative relation between uncertainty and investment; it may be optimal for firms to delay investment so they can wait and see when facing high uncertainty (Bernanke, 1983; Brennan and Schwartz, 1985; McDonald and Siegel, 1986). Bloom (2014) argues that the real option channel affects not only the level of investment but also the sensitivity of investment to factors that drive investment behavior. We test this latter theoretical prediction for a particularly important relation, the sensitivity of firm investment to the cost of capital. Abel and Blanchard (1986) develop a theoretical model in which investment is a function of marginal q, the expected present value of the investment s profit. As marginal q cannot be observed directly, they construct a measure of marginal q that relies on marginal profit and a discount factor, the ex-ante cost of capital. They show that variation in their measure of marginal q is due mainly to the discount factor. Given the proposed relation between investment and marginal q, they predict a negative relation between the cost of capital the denominator in their measure of marginal q and investment. 10

11 Frank and Shen (2016) test Abel and Blanchard s (1986) model in a sample of U.S. firms. When they capture the cost of equity using expected returns from conventional factor models like the CAPM, they find that WACC is positively related to investment. Separating the cost of capital into its cost of debt and cost of equity components, they find that while the cost of debt has the expected negative effect on investment, the cost of equity has a positive effect. Therefore, Frank and Shen (2016) conduct an ex-ante estimation of the cost of equity capital and derive the ICC from several versions of the Gordon growth model and the residual income model (Gebhardt et al., 2001) as implemented in Chava and Purnanandam (2010). When they estimate the cost of equity capital using the ICC measures, they find that the expected negative investment sensitivity is observed for both components of the cost of capital. We build on Frank and Shen (2016) to examine the effect of policy uncertainty on the negative relation between investment and the cost of capital. Based on Bloom s (2014) notion that uncertainty not only reduces the level of economic outcomes but also makes investors less sensitive to changes in business conditions, the strength of the relation between investment and the cost of capital should be decreasing in policy uncertainty. Our prediction is consistent with the strand of literature that emphasizes the role of stock markets in transmitting incremental information to managers who do not have perfect information about every decision-relevant factor. Dow and Gorton (1997) study a model where managers learn from prices. Speculators produce private information and trade on it, and thus their information is incorporated in the company s stock price. Managers, albeit better informed than corporate outsiders, receive investors private information about their company from the stock price, which would otherwise 11

12 not be available to them, and they exploit it in making their investment decisions. Ultimately, financial markets have a real effect due to the transmission of information. Several studies provide evidence supporting this feedback effect emanating from prices that are efficient in a revelatory sense (Bond et al., 2012). For example, Chen et al. (2007) show that the sensitivity of investment to price is stronger when more private information is injected into the price through the trading process. In a related study, Bakke and Whited (2010) document that information, not mispricing, guides corporate investment. Edmans et al. (2012) provide additional evidence on the real effects of secondary financial markets that stem from the informational role of market prices by demonstrating the effect of stock price informativeness on firms takeover activity. 6 Policy uncertainty increases uncertainty with respect to firms expected future cash flows, decreasing the quality of the information contained in stock prices. To the extent that high policy uncertainty makes investors less informed and stock prices contain less private information, managers will be less willing to base their investment decisions on the information revealed by stock prices during times of higher policy uncertainty. With higher policy uncertainty, stock prices become noisier or managers become more informed relative to outside shareholders (or both). Therefore, the sensitivity of investment to stock prices will likely be lower during periods of high policy uncertainty. Pástor and Veronesi (2013) find that policy uncertainty increases the volatility of individual stocks and the correlations between stocks. Morck et al. (2000) claim that if a firm s 6 In related studies, Boot and Thakor (1997) and Subrahmanyam and Titman (1999) use the feedback effect of security prices to explain a firm s choice to issue publicly traded securities in the primary markets, rather than receiving private financing (e.g., bank loans). 12

13 stock returns are highly correlated with market returns, then the firm s stock price is likely to contain less firm-specific (or idiosyncratic) information. In contrast, if a firm s stock returns move asynchronously with market returns, more firm-specific information is likely to be incorporated into the stock price. When stocks move in the same direction during higher policy uncertainty periods, individual stock prices contain less private information that managers can exploit in making their investment decisions, and the investment-price sensitivity should be lower under high policy uncertainty. Durnev (2011) finds that the decrease in investment-price sensitivity during U.S. election years is due to stock prices becoming noisier signals for managers. In particular, during election years (with higher political uncertainty) managers appear to pay less attention to stock prices not because they become more informed relative to outside investors but rather because stocks become less informative (as captured by Morck et al. s (2000) stock return synchronicity measure). In an accounting context, Stein and Wang (2016) document that firms facing relatively higher levels of economic uncertainty report more negative discretionary accruals. Most importantly, they find that market prices are less (more) sensitive to earnings surprises during times of higher (lower) economic uncertainty. 7 A similar implication may be that managers do not pay attention to stock prices (or firm fundamentals) in their investment decisions during periods of high policy uncertainty because the market believes that fundamentals are based on 7 The underlying economic mechanism is as follows: a firm with uncertain prospects manages earnings downward, since markets are more likely to attribute good performance to luck or to expect performance at such times to be transient. Conversely, at times when a firm s value is more certain, it manages earnings upward since markets are more likely to attribute good performance to skill or expect performance during such periods to be persistent. 13

14 luck and will be transient. In contrast, during times of low policy uncertainty, firm fundamentals are believed to be based on managerial skills and will be more persistent. Taken together, we propose three testable hypotheses. Based on the real option argument, there should be a negative link between policy uncertainty and investment. In addition, we expect an inverse relation between investment and the cost of capital, as proposed in Abel and Blanchard s (1986) theoretical model. Therefore, our first and second hypotheses are: Hypothesis 1: There is a negative relation between policy uncertainty and investment. Hypothesis 2: There is a negative relation between the cost of capital and investment. While these level effects have already been documented for U.S. firms by Frank and Shen (2016) and Gulen and Ion (2016), we extend their findings to a large cross-section of countries. Most importantly, we test Bloom s (2014) proposition that uncertainty not only affects the level but also the sensitivity of investment to changes in the cost of capital. Our third hypothesis is as follows: Hypothesis 3: Policy uncertainty weakens the negative relation between a firm s cost of capital and corporate investment. 3. Research design To examine the effect of economic policy uncertainty on the relation between the cost of capital and investment, we use data from Baker et al. (2016), Compustat, and I/B/E/S to 14

15 construct a cross-country data set. 8 Below we begin by discussing our measures of economic policy uncertainty and the cost of capital. We then discuss our sample construction and empirical specifications Economic policy uncertainty Due to its qualitative nature, uncertainty is difficult to measure. Studies typically rely on second moments of firm-level outcomes (e.g., stock returns, profits, and forecasts) to capture economic uncertainty (Jurado et al., 2015). However, because such measures are based on firm-level data, they are not good proxies for economic (aggregate) policy uncertainty. In this paper, we employ a more direct measure of economic policy uncertainty, namely, the time-varying news-based index of economic policy uncertainty (PU index) proposed by Baker et al. (2016). This index is based on the frequency of articles in a country s major newspapers that focus on uncertainty about future economic policy. For the U.S., Baker et al. (2016) count articles in the top 10 newspapers in the U.S. that include keywords in three categories: economy, policy related, and uncertainty. After adjusting the raw article counts to reflect the total number of articles in each newspaper, they aggregate the article counts across the 10 newspapers and then normalize the counts to obtain a mean of 100 for each month over the period 1985 to Baker et al. (2016) repeat this procedure for 17 additional countries for which text-based analysis can be conducted for the country s most relevant newspapers. They find that the PU index rises ahead of elections and during times of war and economic crisis and falls when macroeconomic indicators such as industrial production and employment are low. 8 We use Compustat North America for firms from the U.S. and Canada and Compustat Global for firms from all other sample countries. 15

16 They also find that the PU index properly captures economic policy uncertainty rather than more general economic and political effects. 9 The PU index has been used in several studies to evaluate the effects of policy uncertainty on corporate financing and investment decisions (Francis et al., 2014; Gulen and Ion, 2016) and asset pricing (Pástor and Veronesi, 2013; Da et al., 2015). For the U.S., Baker et al. (2016) also create three alternative economic policy uncertainty indices. The first is based on the degree of disagreement about future purchases by the federal, state, or local government, where the government purchase forecasts come from a quarterly survey of macroeconomic forecasts published by the Federal Reserve Bank of Philadelphia. The second index uses forecast dispersion about expected CPI, which comes from the same survey. The third alternative index is given as the present dollar value of expiring tax provisions over the next 10 years. Baker et al. (2016) provide these alternative indices together with an aggregate index, which is the weighted average of the three alternative indices and the news-based PU index. 10 We examine whether our main results using the news-based index continue to hold using these alternative indices The cost of capital To compute a firm s cost of capital, or WACC, we follow convention and estimate 1 1, (1) 9 Specifically, Baker et al. (2016) show that the PU index is correlated with established measures of economic and policy uncertainty and that the PU index is not biased by the political views of any of the newspapers included in the index. When they control for the quality of their text-search algorithm via a manual audit of a subsample of the articles, they find that results from the manual audit and the computer algorithm are highly correlated, with the deviation between them independent of macroeconomic states or the level of the PU index. 10 The data are available at 16

17 where LEV is the firm s leverage, which is equal to total debt as a fraction of total assets; KDebt is the firm s average cost of debt, which is equal to interest expenses divided by total debt; TAX is the firm s corporate tax rate, which is equal to income tax expenses divided by net income; and KEquity is our main measure of the firm s ex-ante cost of equity capital, which we estimate following the ICC approach. 11 More specifically, following prior literature (Hail and Leuz, 2006; Chen et al., 2011; Hou et al., 2012; Barth et al., 2013), we calculate KEquity as the arithmetic average of the ICC measures of Claus and Thomas (2001; KCT), Ohlson and Juettner- Nauroth (2005; KOJN), Easton (2004; KMPEG), and Gebhardt et al. (2001; KGLS). 12 To mitigate concerns that our results are biased by the availability of any individual ICC measure, we also estimate our main specification below using the four individual ICC measures. 13 We refer to the left term on the right-hand side of equation (1) as a firm s cost of debt, COD, and the right term on right-hand side of (1) as its cost of equity, COE Sample Our initial sample comprises firms in the merged Compustat I/B/E/S data sets. We first restrict the sample to country-years for which the PU index of Baker et al. (2016) is available. Following Frank and Shen (2016), we next exclude firms with negative average cash flow over the sample period. 14 We also require nonmissing data on investment, cash flow, and the 11 This approach, which Pástor et al. (2008) show is well suited to explain time-variation in expected returns, entails using analyst forecast data and stock price data to compute the internal rate of return at which the market price of equity equals future residual income or abnormal earnings. 12 Appendix A provides a detailed explanation of the construction of the four ICC measures. 13 In additional robustness tests, we examine whether our main results hold when we consider alternative assumptions in the ICC models. 14 Frank and Shen (2016) impose this restriction given the expected signs for marginal q and the WACC derived from Abel and Blanchard s (1986) model of optimal investment. 17

18 WACC. Finally, we exclude utilities and financial firms. Our final sample consists of 64,158 observations representing 11,220 firms from 18 countries over the period 1989 to All variables are winsorized at the 1 st and 99 th percentiles. Table 1 presents summary statistics for our full sample by country. The table reports the number of firm-year observations, the number of distinct firms, the period, and means of the key variables used in our empirical analysis. We find that the number of observations and the number of distinct firms vary by country depending on the availability of accounting and stock market data from Compustat and the PU index of Baker et al. (2016), with the U.S. accounting for almost half of both the number of observations and the number of firms. Following Frank and Shen (2016), we calculate investment, INV, as capital expenditures net of asset sales divided by beginning-of-year fixed assets and cash flow, CF, as operating income before depreciation deflated by beginning-of-year fixed assets. The mean investment ratio is 27%, with notably higher rates in Brazil, France, Germany, and India, and substantially lower rates in Chile, Japan, and Ireland. Average cash flow ranges from 34% in Chile to 171% in Sweden, with a sample mean of 107% of beginning-of-year fixed assets. The sample mean WACC is 10.8%, which is close to the 10.9% for U.S. firms. Firms from Japan have the lowest mean WACC (7.8%), while firms from Russia face the highest mean WACC (15.7%). The average cost of equity (KEquity) is 13.1%, and the average cost of debt (KDebt) is 9.9%. Japanese firms enjoy the lowest costs of equity and debt, while firms from Russia have the highest cost of equity (19.5%), and firms from Brazil have the highest cost of debt (19%). The average leverage ratio in our sample is 24%. Insert Table 1 here 18

19 Table 2 reports detailed descriptive statistics for the main model variables for the full sample of firms. The means are identical to the last row of Table 1. The median is smaller than the mean for all variables of interest. The difference is particularly pronounced for cash flow (0.516 vs ) and total assets (787 million USD vs. 4,777 million USD). The median investment ratio is 0.19, compared to a mean of 0.27; these numbers are close to those reported by Frank and Shen (2016) for their U.S. sample. Insert Table 2 here 3.4. Empirical specification Our main empirical specification builds on Abel and Blanchard (1986) and Frank and Shen (2016), who model investment as a function of cash flow and the cost of capital. Since we are interested in how policy uncertainty affects the relation between the cost of capital and corporate investment, we include the natural logarithm of the 12-month average (over the fiscal year) of the country-level PU index developed by Baker et al. (2016) as well as the interaction between the cost of capital measures and PU. Because the PU index is a time-varying countrylevel variable, it may capture differences in the level of investment and in the sensitivity of investment to the cost of capital between countries and over time. Following Edmans et al. (2017), we include country (Country) and year (Year) fixed effects, which control for the base level of investment in a given country and for time trends in investment, respectively, and the interactions between the cost of capital and country (WACC Country) and year (WACC Year) fixed effects to control for differences in the sensitivity of investment to the cost of capital 19

20 across countries and over time. 15 Finally, we include for (two-digit SIC) industry (Industry) fixed effects to control for differences in investment across industries. Our main specification is thus given by:,,,,,,,,,,,,,,,,,,,,,,,, (2) where i indexes firms, c indexes countries, k indexes industries, and t indexes years. We cluster standard errors by country. To examine whether the effect of the PU index varies across the cost of debt and cost of equity components of the cost of capital, we estimate a specification in which we include these two components, together with their respective interaction terms with PU:,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,. (3) While Baker et al. (2016) attempt to mitigate concerns that their PU index merely reflects macroeconomic conditions, it may nonetheless capture some other macroeconomic information related to investment opportunities. Therefore, we follow Edmans et al. (2017) and re-estimate equations (2) and (3) after adding a set of country-level economic indicators that may influence investment decisions. These controls are the annual gross domestic product in current USD (GDP), the annual percentage change in GDP (GDP growth), the annual inflation 15 We do not include firm fixed effects with country fixed effects because are perfectly correlated to the extent that firms do not change their country of origin. We choose to include country fixed effects rather than firm fixed effects since policy uncertainty varies at the country and not the firm level. As robustness, we report results using firm fixed effects instead of country fixed effects. 20

21 rate (Inflation), and exports plus imports scaled by GDP (Trade). For robustness, we also include proxies for expected future investment opportunities such as the GDP forecasts, a consumer confidence index, and a composite leading indicator. 4. Results In this section, we examine the effect of policy uncertainty on the link between the cost of capital and corporate investment. We first examine the effect of policy uncertainty as captured by the news-based PU index for the full cross-country sample. Second, we analyze whether our main results vary across firms and industries Main analysis Table 3 presents estimation results for equations (2) and (3), with and without the set of country-level economic indicators discussed above. In these specifications, the WACC is calculated using KEquity (the arithmetic average of our ICC measures) and KDebt. Policy uncertainty is captured using the news-based policy uncertainty index of Baker et al. (2016). In column 1, the coefficients of interest are those on PU, WACC, and the interaction term WACC PU. Consistent with Hypothesis 1, we find a significantly negative coefficient on PU, implying that an increase in economic policy uncertainty is associated with lower investment, even after controlling for the cost of capital. This result extends the finding of an investment distortion caused by policy uncertainty documented by Gulen and Ion (2016) to a crosscountry data set. Supporting Hypothesis 2, we observe a negative relation between investment and the cost of capital. The estimate implies that a one standard deviation decrease in WACC 21

22 increases investment by 1.1%, which is equivalent to 3.7% of the mean investment ratio. Although this effect is less pronounced than the effect documented by Frank and Shen (2016) for the U.S., it is economically significant given that we control for country-, year-, and industryfixed effects. 16 Finally, corroborating Hypothesis 3, we find that the coefficient on the interaction term between WACC and PU is significantly positive. In line with Bloom s (2014) sensitivity effect argument, an increase in policy uncertainty decreases the effect of the cost of capital on investment. In particular, when PU is set to its first quartile, the effect of a one standard deviation decrease in WACC results in a 1.70% increase in the investment ratio. The increase is only 0.3% when we set PU to its third quartile. 17 The resulting difference of 1.4% is equal to 4.9% of the mean investment ratio. Turning to cash flow, we find that investment is positively linked to changes in cash flow. This result is not surprising given that investment cash flow sensitivities are well documented (Fazzari et al., 1988; Kaplan and Zingales, 1997; Chen and Chen, 2012; Lewellen and Lewellen, 2016; among others), and cash flow likely captures part of marginal q in the absence of additional control variables and corrections for measurement error in marginal q (Erickson and Whited, 2000, 2012). In column 2, we decompose WACC into its components, COD and COE. We find that COD and COE have significantly negative effects on investment that are comparable in magnitude. Interestingly, the estimated sensitivity effects for both components are also similar in 16 In Frank and Shen (2016), the increase in investment is equivalent to 7.6% of the investment ratio using the one-period Gordon growth model. 17 A one standard deviation decrease of WACC is equal to The average coefficients of the statistically significant interaction terms WACC Country ( ) and WACC Year ( ) are and , respectively. The economic impact of a one standard deviation change in WACC on investment is then calculated as: This is equal to when PU is set to its first quartile (4.373) and when PU is set to its third quartile (4.835). 22

23 magnitude. The difference in the increase in investment in response to a one standard deviation decrease in COD and CEO when PU moves from the first to the third quartile is 1.0% and 1.3%, respectively. Again, this difference constitutes a sizeable fraction of firms mean investment ratio (3.5% for COD and 4.6% for COE). This result is consistent with prior findings that stock and bond markets are similar in terms of informational efficiency once institutional trade dominance and other specific bond trading features are accounted for (Downing et al., 2009; Ronen and Zhou, 2013; Tsai, 2014). Related evidence shows that macroeconomic announcements have a significant impact on the returns of both stock and bond markets (Balduzzi et al., 2001; Flannery and Protopapadakis, 2002), also indicating that the feedback effect from private information is similarly important for equity and debt markets. Insert Table 3 here In columns 3 and 4, we add the four country-level economic indicators as control variables in order to mitigate concerns that our policy uncertainty measure is merely capturing other economic variables that affect firm investment behavior. We find that the effect of PU on the level of investment does not change when we incorporate these country-level controls. More importantly, although the estimates for WACC and WACC PU decrease slightly in magnitude, they remain statistically significant. The estimates for the COD and COE components are either little changed or only marginally smaller, but again remain statistically significant. Finally, in columns 5 and 6 we repeat the regressions in columns 3 and 4 by including firm fixed effects instead of country and industry fixed effects. The results remain qualitatively similar. 23

24 4.2. Subsample analysis Next, as in Gulen and Ion (2016), we test whether the moderating effect of policy uncertainty on the relation between investment and the cost of capital varies with firm and industry characteristics. To analyze heterogeneous effects, we split our sample into subsamples based on (1) firm size as a proxy for financial constraints, (2) industry-level sensitivity to government subsidies, and (3) industry-level government consumption. We expect that the distortive effect of policy uncertainty on the link between investment and the cost of capital will be stronger for financially constrained firms, as corporate policies and, in particular, investment decisions of financially constrained firms, are more sensitive to market conditions (Campello et al., 2010; Drobetz et al., 2017). Moreover, the same level of policy uncertainty should translate into a higher level of demand uncertainty for firms that are more dependent on government subsidies or direct government consumption. We therefore expect that the effect of policy uncertainty on the relation between investment and cost of capital will be stronger for this group of firms. Our measure of financial constraints is firm size following Erickson and Whited (2000). We classify a firm as small (large) if the mean of total assets over the sample period is in the bottom (top) tercile of the distribution of all firms from the same country. We construct sensitivity to government subsidies at the industry level. Specifically, for each firm we first calculate 24

25 the correlation between the firm s sales and government subsidies as a fraction of total government spending, 18 then aggregate these correlations at the two-digit SIC level, and finally categorize industries into high (above-median) and low (below-median) sensitivity groups. Our third measure, government dependency, is a more direct measure of government consumption. Using the 2005 OECD Structural Analysis Input-Output tables, which provide data on purchases of an industry s production by other industries, households, and the government, we estimate how much of an industry s production is consumed by the government. We first calculate an industry s total consumption as the sum of all outputs for the industry in each country. We then calculate the mean fraction of this consumption that is attributed to the government over all countries, 19 which we then use to categorize industries into high and low government dependency groups based on the median of all industries. 20 Table 4 presents the results. The first four columns report results for subsamples based on firm size. As expected, the coefficients on both WACC and WACC PU are significantly higher in absolute terms for small firms compared to large firms. In the subsample of large firms, the loadings of investment on WACC and WACC PU are statistically insignificant. 18 We use the GC.XPN.TRFT.ZS time series from the World Bank database ( This variable includes all unrequited, non-repayable transfers on current account to private and public enterprises; grants to foreign governments, international organizations, and other government units; and social security, social assistance benefits, and employer social benefits in cash and in kind. 19 The output category is general government final consumption (GGFC), which is defined by the World Bank as all government current expenditures for purchases of goods and services (including compensation of employees). It also includes most expenditures on national defense and security, but excludes government military expenditures that are part of government capital formation. 20 Industries in the Input-Output tables are categorized by the UN s International Standard Industrial Classification of All Economic Activities (ISIC Rev. 3). We use the ISIC Rev. 3 to SIC concordance tables to match industries to Compustat. 25

26 When we decompose WACC into its debt and equity components, we find that these differences persist only for COE as well as the corresponding interaction term with PU. Insert Table 4 here Columns 5 to 8 show the results for the government subsidy subsamples. We find that for firms from industries that are highly sensitive to government subsidies, the estimates for WACC PU and COE PU are statistically higher compared to firms from industries with low sensitivity to government subsidies. These results extend evidence in Gulen and Ion (2016) that adjustments in response to policy uncertainty are particularly strong for firms that depend on government spending by showing that the distortionary effect of policy uncertainty is observed in both the level of investment and the sensitivity of investment to the cost of capital. Examining COD and COE rather than WACC, we observe that the coefficients on COD PU are of equal magnitude across the subsamples, while the coefficient on COE PU is significantly larger for the high subsidy sensitivity subsample. In columns 9 to 12, we report results for the government dependency subsamples. We again find that the moderating effect of PU on the link between investment and the cost of capital is driven by firms from industries that depend on consumption by the government. The effects of WACC and WACC PU are statistically significant only in the subsample of government-dependent industries. When we decompose WACC into its components, we confirm this finding. The coefficients of both components, i.e., COD and COD PU as well as COE and COE PU, are statistically significant only in government-dependent industries. 26

27 Overall, the evidence in Table 4 shows that the sensitivity of investment to the cost of capital is not homogeneous across firms and industries. Any distortive effect of policy uncertainty on the link between investment and the cost of capital is stronger for small and for industries that are more dependent on government subsidies or direct government consumption. 5. Robustness We perform several tests to examine the robustness of our main results. First, we test whether our results persist when we use Baker et al. s (2016) alternative uncertainty indices for a subsample of U.S. firms. Second, we test the robustness of our results towards the use of a broader definition of investment. Third, we control for potentially omitted variables that drive investment or the relation between investment and the cost of capital. Fourth, we address concerns regarding the aggregation of our implied cost of equity measure by using the individual ICC measures rather than their mean. We also test the robustness of our results to changes in the assumptions underlying the construction of the ICC measures. Finally, we perform a weighted least squares regression to mitigate concerns that our results can be attributed to a disproportionate number of observations coming from a small number of countries Alternative measures of policy uncertainty In our main analysis above, we use Baker et al. s (2016) news-based policy uncertainty indices for a sample of 18 countries. Next, we employ Baker et al. s (2016) additional indices of policy uncertainty for the U.S. that are based on disagreement about purchases of goods and services by federal, state, and local governments, dispersion in CPI forecasts, and the dollar 27