Economic Policy Uncertainty, Cost of Capital, and Corporate Innovation

Transcription

1 Economic Policy Uncertainty, Cost of Capital, and Corporate Innovation Zhaoxia Xu Abstract We examine the cost-of-capital transmission channel through which government economic policy uncertainty (GEPU) can affect corporate innovation activities. We find that GEPU increases firms cost of capital, which in turn translates into lower innovation activities. As economic policy uncertainty rises, firms with more exposure to such uncertainty face a higher weighted average cost of capital and reduce investment more. Moreover, GEPU s impact is stronger on R&D than on fixed-asset investments. Our study provides novel evidence that more uncertainty about economic policy hinders innovation not only through the investment irreversibility channel, but also through the cost-of-capital channel. Key Words: Economic Policy Uncertainty, Implied Cost of Equity, Cost of Debt, Innovation. JEL No: G31, G32, O30, E60. I gratefully acknowledge the support of the National Bureau of Economic Research Innovation Policy Research Grant. I thank Viral Acharya, Chang-Mo Kang, Qiang Kang, Ron Masulis, and the seminar participants at Florida International University, New York University, University of New South Wales, the Rising Stars Conference Financial Market Workshop for the helpful comments. Department of Finance and Risk Engineering, New York University, 6 MetroTech Center, New York, USA, zhaoxiaxu@nyu.edu. Phone: (646)

2 1 Introduction The intensified uncertainty about government policies in the wake of partisan conflicts in the recent years highlights the importance of understanding their impact on the real economy and the underlying transmission channels. Most of the theoretical and empirical studies analyze the influence of uncertainty on investment in the framework of irreversible investment. 1 With costly reversibility, uncertainty changes the optimal timing of investment because of the real-option feature of investment. In this study, we investigate another crucial transmission mechanism, cost of capital, through which government economic policy uncertainty (GEPU) affects corporate innovation activities. The cost of capital is important in transmitting economic policy uncertainty shocks to investment. Recent studies show that policy uncertainty commands an equity risk premium because of the undiversifiable political risk (Pastor and Veronesi (2012, 2013)). Since equity risk premium is a building block for cost of equity, economic policy uncertainty should affect firms cost of capital. The impact could vary across firms because of their varying exposure to GEPU. The cost of capital is widely used as a discount rate to evaluate a project when firms make capital budgeting decisions (Graham and Harvey (2001), Association for Financial Professionals (2013)). An increase in the cost of capital can turn a positive net present value project into a negative one. Consequently, firms investment decision is affected by the cost of capital. An increase in the cost of capital may also reduce firms incentives to raise 1 See, for example, Bernanke (1983), McDonald and Siegel (1986), Abel and Eberly (1994), Dixit and Pindyck (1994), and many others for the theoretical framework. Bloom et al. (2007), Bloom (2009), Julio and Yook (2012), and Gulen and Ion (2016) among others provide empirical evidence supporting the investment irreversibility channel. 1

3 external capital. Firms with less capital available may invest less in innovative activities. Despite theoretical soundness, no study has empirically linked economic policy uncertainty to individual firms financing costs and their subsequent innovation activities. 2 We focus on innovation for several reasons. First, although previous studies show that uncertainty influences firms investment and hiring decisions, little is known about its impact on corporate innovation. Second, increasing evidence shows that economic factors affect innovation and other investments differently, owing to the unique features of research and development (R&D). 3 Third, R&D investment is more susceptible to uncertainty because it often takes a long time to yield fruitful results and involves a substantial amount of investment in human capital, which is hard to recoup. Although the cost of capital affects investment in physical assets as well, financing costs are especially important for investment in innovation. Studies have shown that financing is essential for developing, implementing, and commercializing new technologies (Rajan (2012) and Kerr and Nanda (2015)). Financing innovation tends to be costlier because of the high uncertainty of investment outcomes (Hall and Lerner (2010)). Moreover, the uncertainty of innovation projects is likely to be exacerbated by policy uncertainty. To analyze the real effects of uncertainty related to the U.S. economic policy, we measure 2 Pastor and Veronesi (2012, 2013) develop general equilibrium models featuring government policy uncertainty and use S&P 500 index return data and the Baker et al. (2013) index to show that policy uncertainty commands an equity risk premium at the aggregate level. Based on the time-series analysis of Fama-French size-momentum portfolio returns, Brogaard and Detzel (2015) show that economic policy uncertainty is an important risk factor. Kelly et al. (2016) use equity index option prices around national elections and global summits to show that political uncertainty is priced. However, none of these studies investigate the impact of economic policy uncertainty on firm-level cost of capital and subsequent investments. 3 Market factors, such as stock liquidity, short selling, coverage by financial analysts, and market sentiment, are found to have varying effects on R&D and capital expenditures (Becker-Blease and Paul (2006), Fang et al. (2014), Grullon et al. (2015), He and Tian (2013, 2014), Derrien and Kecskes (2013) and Dang and Xu (2015)). Unlike capital expenditures that produce tangible assets, R&D investments generate intangible assets that are more difficult to serve as collateral. Furthermore, the outcome of innovation investments is more uncertain. 2

4 innovation by its inputs (R&D) and outcomes (patents) and estimate each firm s annual cost of equity, cost of debt, and weighted average cost of capital (WACC). The cost-ofcapital transmission channel is identified by exploring cross-sectional variation across firms with varying exposure to economic policy uncertainty. If cost of capital is the underlying transmission mechanism through which GEPU affects innovation, we expect that the cost of capital and investments of firms with more exposure to GEPU would be affected more. One of the main challenges in our analysis is to measure GEPU. Previous studies use proxies such as volatility in stock returns or the dispersion in analyst forecasts to measure firm-specific uncertainty. 4 However, these measures cannot capture uncertainty attributable to the government economic policy. Applying a cross-country framework, Julio and Yook (2012) use national election years to represent periods of political uncertainty. Unfortunately, this proxy does not reflect policy uncertainty in non-election years. Therefore, we adopt a continuous economic policy uncertainty index recently developed by Baker et al. (2013) to measure time variation in uncertainty. 5 This GEPU index is a weighted average of uncertainty related to taxation, government spending, and inflation, as well as the frequency of major newspaper articles discussing uncertainty in economic policy. The index exhibits substantial variation over time and tends to spike during high economic policy uncertainty periods, such as debt-ceiling crises, budgetary disputes, and tight presidential elections. Furthermore, it captures uncertainty about the government s future policy beyond electoral uncertainty. Another challenge is to estimate individual firms cost of capital, because the cost of 4 See, for example, Bloom et al. (2007) and Bond and Cummins (2004). 5 The economic policy uncertainty index of Baker et al. (2013) has been used in numerous academic studies (e.g., Pastor and Veronesi (2012, 2013), Brogaard and Detzel (2015), and Gulen and Ion (2016), among many others), as well as by practitioners and policy makers. 3

5 equity and the cost of debt are not observable. Previous studies use average realized returns to estimate the cost of equity capital (Campbell (1987)). This approach uses ex post realized stock returns as a proxy for ex ante expected returns. However, realized returns are shown to be a poor measure of expected returns (Elton (1999)). In addition, the capital asset-pricing model (CAPM) and Fama and French (1993) three-factor model are found to be imprecise in measuring the cost of equity (Fama and French (1997), Pastor and Stambaugh (1999)). Thus, we estimate the cost of equity capital by using an ex ante proxy developed by recent accounting and asset-pricing research, namely the implied cost of equity. 6 The implied cost of equity is defined as the discount rate that equates a stock s present value of expected cash flows to its current price. This internal rate of return is viewed as the market s expected rate of return on the stock. Recent studies show that the implied cost of equity capital is a good proxy for a stock s conditional expected return (Pastor et al. (2008), Lee et al. (2009), Chava and Purnanandam (2010), and Li et al. (2013)). We derive the implied cost of equity from Gebhardt et al. (2001) and three alternative models and estimate a firm s cost of debt as the actual yield on the debt carried by the firm and two alternative approaches. The cost of capital is then measured as the weighted average cost of capital, which comprises a firm s cost of equity and after-tax cost of debt, with its market leverage ratio as the weight. 7 The third challenge is identification, because confounding factors such as overall economic conditions or investment opportunities may cause a spurious relationship. We adopt 6 See Easton (2007) for a review of the implied cost of equity estimations in the accounting literature. In the asset-pricing literature, implied cost of capital is used as a proxy for expected stock returns to test the risk return relationship (Pastor et al. (2008), Lee et al. (2009), Chava and Purnanandam (2010)). 7 Frank and Shen (2015) estimate the cost of equity by using the Gordon growth model and measure the WACC through a similar approach. Their study evaluates the impact of WACC on firms capital expenditures, while this study investigates how economic policy uncertainty affects firms financing costs. 4

6 several approaches to isolate the effect of economic policy uncertainty. First, we follow the recommendation by Bernanke et al. (1996) and investigate cross-sectional differences in the cost of capital and innovation in response to economic policy uncertainty. Second, we exploit close congressional elections and close presidential elections as two plausible exogenous shocks to policy uncertainty. Since the timing of elections is independent from economic conditions and the outcomes of close elections are difficult to predict, close elections provide a natural experiment framework to isolate the impact of uncertainty related to general government policies. Third, we use political polarization as an instrumental variable to ease the endogeneity concern. Fourth, we control for the time-varying first-moment factors that capture aggregate-level growth opportunities and control for observable and unobservable factors that reflect firm-level investment opportunities. This helps disentangle the impact of second-moment shocks from first-moment effects associated with business cycles and investment opportunities. Fifth, we perform a fixed-effects estimation to explore the effects of economic policy uncertainty within each firm over time. Finally, we extract the exogenous component from the GEPU index by purging the endogenous component of the index. We find that investments in innovation and the quantity and quality of innovation outcomes are negatively associated with GEPU. Economic policy uncertainty has an immediate impact on innovation inputs, while the impact on innovation outcomes becomes significant in the third year and the influence strengthens in the fourth year. Our robustness analyses show that investment opportunities or macroeconomic conditions are less likely to be the main factor driving the relationship between economic policy uncertainty and innovation. In addition, we find that uncertainty about economic policy has a weaker impact on fixed-asset investments than on R&D investments, which is consistent with the view that innovation is 5

7 more sensitive to uncertainty. We observe that the cost of capital rises as GEPU intensifies. During , the weighted average cost of capital increased by approximately 129 basis points from the lowest to the highest levels of economic policy uncertainty. We further show that firms have a lower propensity to issue equity and debt and raise less capital when economic policy uncertainty increases. Firms invest less in innovation when the cost of capital rises. The results establish an explicit link between economic policy uncertainty and financing costs: uncertainty makes financing innovation costlier for firms. To identify the cost-of-capital channel, we construct a measure of firms exposure to economic policy uncertainty that captures the sensitivity of a firm s stock returns to changes in the GEPU index. Using this measure, we observe that the increase in the WACC of firms with higher sensitivity to economic policy uncertainty is more than that of the firms with lower sensitivity when economic policy uncertainty moves from low to high periods. Economic policy uncertainty has a stronger impact on the cost of capital and investments for firms with more exposure to uncertainty than those with less exposure. These results lend support to the cost-of-capital transmission channel perspective. In addition, we find that the cost-of-capital channel still plays an important role in shaping firms investment under uncertainty after we control for the investment irreversibility mechanism. This study contributes to the debate on the relation between uncertainty and the real economy. The theoretical literature offers contradictory predictions on the uncertaintyinvestment relationship. On the one hand, the literature suggests that it is optimal for firms facing uncertainty to defer investment because reversing investment is costly and uncertainty increases the value of the option to wait (Bernanke (1983), McDonald and Siegel (1986), Abel 6

8 and Eberly (1994), and Dixit and Pindyck (1994)). On the other hand, Carballero (1991), Grenadier (2002), and Weeds (2002) argue that the option to wait is less valuable when firms face competition or when investments can create valuable growth opportunities. Uncertainty increases investments if firms have an option to resell the assets later (Abel et al. (1996)) or if the marginal revenue product of capital is convex (Hartman (1972) and Abel (1983)). Further, the empirical evidence on the real effects of uncertainty is mixed. 8 In terms of investment, Gulen and Ion (2016) and Jens (2016) find that firms delay spending on fixed assets when facing policy uncertainty or electoral uncertainty. Using implied volatility from equity options to capture firm-specific uncertainty, Stein and Stone (2013) document that implied volatility reduces capital investment but increases R&D investment. A contemporaneous paper, Atanassov et al. (2015), find that firms spend more on R&D during gubernatorial election years. Differing from Stein and Stone (2013) and Atanassov et al. (2015) s focus on uncertainty about individual firms or outcomes of gubernatorial elections, we investigate uncertainty about government economic policies that affects the general economic environment within which businesses operate. Such uncertainty goes beyond the potential changes in local leadership in the midst of elections. 9 We find that such aggregate policy risk, which 8 Bernanke (1983), Bloom et al. (2007), Baker et al. (2013), Julio and Yook (2012), Gulen and Ion (2016), and Jens (2016) document that uncertainty has a negative impact on fixed-asset investments and employment. However, Ghosal and Lougani (1996) find no impact of uncertainty on investment in U.S. manufacturing industries, except for competitive industries. Driver et al. (2008) show that the negative effect of uncertainty on investment is offset by the first-mover advantage. 9 For example, events such as debt ceiling crises, budgetary disputes, and debates over the stimulus package dramatically elevated uncertainty about government policies, but they were unrelated to elections. Gubernatorial elections take place every four years. Election years do not capture the variation in policy uncertainty that may occur between elections. An election dummy variable also does not take into account the fact that elections at different points in time may have different implications for the level of policy uncertainty in the economy. Moreover, Gubernatorial elections may affect firms that operate within the state. Given that most firms operate and compete in the national or international market, the impacts of Gubernatorial elections might be limited. This is especially the case when firms often have R&D centers in locations different from the headquarters. 7

9 is hard to diversify, has a negative impact on firms innovation activities. The literature has emphasized that investment irreversibility shapes the relation between uncertainty and investments. Our study provides novel evidence on another important mechanism cost of capital through which economic policy uncertainty is transmitted to firms innovation activities. We contribute to the literature by explicitly showing that the firm-level cost of capital is affected by economic policy uncertainty and the impact varies across firms. To the best of our knowledge, this study is the first to empirically demonstrate that high uncertainty about government economic policy has a detrimental effect on innovation by driving up firms cost of capital. Our study also adds to the nascent literature on finance and innovation. The literature shows that innovation is affected by various economic factors such as the development of financial markets (Hsu et al. (2014)), law and legal systems (Acharya and Subramanian (2009), Acharya et al. (2014), Brown et al. (2013)), institutional ownership (Aghion et al. (2013)), credit supply (Amore et al. (2013), Chava et al. (2013), and Cornaggia et al. (2015)), and access to stock markets (Acharya and Xu (2015)), among others. In a cross-country analysis, Bhattacharya et al. (2015) show that overall uncertainty resulting from national elections rather than a country s policy itself affects technological innovation. Unlike their focus on comparing the relative importance of policy and policy uncertainty, we are interested in understanding the real effects of time-varying uncertainty related to U.S. economic policy and identifying the underlying transmission mechanism. We find that the influence of economic policy uncertainty on innovation through firms financing costs is stronger for firms with higher exposure to such uncertainty. 8

10 2 Data and Measures 2.1 Data To measure innovation activities, we use patent and citation data from Kogan et al. (2012) and technology class data from the United States Patent and Trademark Office. 10 The stock price data are from the Center for Research in Security Prices (CRSP) monthly stock database and the firm-level financial data are from the merged CRSP-Compustat database. The sample period is from January 1, 1986 to December 31, The sample ends in 2007 because the average time lag between the patent application date and grant date is 2 3 years (Hall et al. (2001)). An advantage of focusing on innovation before the financial crisis is that it minimizes the influence of deteriorating macroeconomic conditions on firms innovation performance. Firms in financial and utilities industries (SIC code and ) are excluded. We require firms to have complete data on total assets and a positive value on sales and cash. Firm years with total assets less than 1 million are excluded. Further, we focus on R&D firms that invest in R&D during the sample period. Variables are winsorized at 1 percent and 99 percent to avoid the effect of outliers. We merge patent data with our financial data. Following the innovation literature, the patent and citation counts are set to zero when no patent and citation information is available. Including firm-year observations with no patent alleviates the sample selection concern. The final sample consists of 6,167 firms and 54,050 firm-year observations. 10 The patent data by Kogan et al. (2012) contain patent and citation data for U.S. firms up to We find similar results using the National Bureau of Economic Research (NBER) Patent Citation database that covers patent data until

11 2.2 Innovation Measure To gauge a firm s innovation activities, we follow the literature and measure innovation input by using R&D spending and innovation output by using patent-based metrics (Hall et al. (2001, 2005)). The quantity of innovation output is measured by the number of patents applied by a firm in a given year that are eventually granted. The patent application year rather than the grant year is used because the former is closer to the time of the actual innovation (Griliches et al. (1987)). We measure the quality of patents on the basis of the citation count that each patent receives in the subsequent years. Patent citations are subject to truncation biases because patents created in later years have less time to accumulate citations than those established in earlier ones. Additionally, the citation intensities of patents might vary across industries. To adjust for truncation bias, following Hall et al. (2001, 2005), we scale the raw patent citation counts by the average citation counts of all patents applied in the same year and technology class. 2.3 Government Economic Policy Uncertainty Index The time-varying economic policy uncertainty index developed by Baker et al. (2013) is a weighted average of a news-based policy uncertainty index, tax expirations, inflation forecast disagreement, and government purchase disagreement. The news-based policy index is a normalized monthly count of the number of articles that contain terms related to economic policy uncertainty in 10 of the largest newspapers. The tax expirations measure the discounted dollar-weighted sum of expiring federal tax code provisions. Forecast disagreement 10

12 about inflation and government purchases captures uncertainty associated with monetary policy and government spending through forecast dispersion by participants of the Federal Reserve Bank of Philadelphia s Survey of Professional Forecasters. Figure 1 plots the monthly GEPU index and NBER recession periods between January 1986 and December Uncertainty spikes during the debt-ceiling crisis of 2011, budgetary disputes, tight presidential elections, the 9/11 terror attack, and recessions. We match the monthly government economic policy uncertainty index to the annual financial data by the fiscal year and month. 11 Since the GEPU index is likely to be correlated with macroeconomic conditions and/or investment opportunities, we adopt several identification strategies to minimize the influence of the confounding factors. Further, it is possible that the GEPU index contains uncertainty unrelated to the economic policy. Baker et al. (2013) employ various approaches, such as human audit and comparison with alternative measures of economic uncertainty, to ensure accuracy and reliability of the index. A potential measurement error in the GEPU index would result in attenuation bias, which is against finding a statistically significant effect of economic policy uncertainty and underestimates the magnitude of the effect. We use the instrumental variable approach to ease the concern about measurement error. 2.4 Cost of Capital The cost of capital, defined as the weighted average cost of equity and cost of debt, is an important component in firms capital budgeting decisions. We measure the cost of equity by using the implied cost of equity approach, which estimates the ex ante expected return 11 We find similar results by using the annual average economic policy uncertainty index. 11

13 by using market prices and accounting data (Gebhardt et al. (2001), Pastor et al. (2008), Lee et al. (2009)). Specifically, the implied cost of equity is the discount rate that equates the current stock price to the present value of expected future cash flows. According to the discounted cash flow model, the stock price of a firm at time t is P t = n=1 E t (F CF E t+n ) (1 + r e ) n, (1) where P t is the market value of the stock at time t, E t (F CF E t+n ) is the expected free cash flow to equity at time t + n, and r e is the implied cost of equity capital. To estimate the cost of equity, we follow the Gebhardt et al. (2001) (GLS) approach to derive the implied cost of equity. The GLS approach uses the residual income valuation model under the assumption of clean-surplus accounting. 12 This approach has been applied in several asset-pricing studies, for example, Pastor et al. (2008), Lee et al. (2009), and Chava and Purnanandam (2010). The stock price in equation (1) can be written as the sum of the book value and the discounted residual income (economic profits) earned by a firm: P t = B t + n=1 E t (NI t+n r e B t+n 1 ) (1 + r e ) n = B t + n=1 E t [(ROE t+n r e )B t+n 1 ] (1 + r e ) n, (2) where B t is the book value of equity at time t, NI t+n is the net income for period t + n, NI t+n r e B t+n 1 is the residual income (economic profit) for period t + n, ROE t+n is the after-tax return on equity for period t + n, and r e is the cost of equity capital. We use the Institutional Brokers Estimate System (I/B/E/S) consensus analyst forecasts to predict future earnings per share (EPS). The residual income model is estimated by forecasting earnings explicitly for the first three years and then implicitly thereafter, assuming 12 Under the clean-surplus assumption, earnings include all gains and losses that affect book value. Thus, the change in the equity book value at t + 1 equals net income during t + 1 minus net dividends paid during t + 1 (b t+1 = b t + NI t D t ). 12

14 that the ROE of each firm would revert to the industry median ROE by the terminal period. The split-adjusted stock price is obtained from CRSP. We solve for the discount rate r e in equation (2). As a robustness check, we estimate the implied cost of equity capital by using alternative models: Gode and Mohanram (2003) (GM) and Easton (2004) (PEG). These models rely on analyst forecasts for future EPS, which are not available for all firms. To circumvent this disadvantage, Hou et al. (2012) (HVZ) estimate future earnings from cross-sectional regressions by using total assets, dividends, earnings, and accruals as the independent variables. Although the HVZ model can provide earnings forecasts for more firms, Gerakos and Gramacy (2013) and Li and Mohanram (2014) show that the HVZ model underperforms a naive random walk (RW) model that sets future earnings to past earnings. Li and Mohanram (2014) propose two other cross-sectional models: the earning persistence (EP) and residual income (RI) models, showing that the RI model outperforms the HVZ, RW, and EP models in forecasting future EPS. Therefore, we use the Li and Mohanram (2014) RI model approach to forecast future EPS and estimate the implied cost of equity from the Claus and Thomas (2001) model. See Appendix for details on estimating the implied cost of equity. We estimate the WACC as follows: W ACC i,t = Debt i,t MV A i,t CoD i,t (1 T axrate) + (1 Debt i,t MV A i,t )CoE i,t, (3) where W ACC i,t is the WACC for firm i in year t. Debt it MV A it is the market leverage ratio. CoD i,t is the cost of debt for firm i in year t, measured as the actual yield on the debt carried by the firm, as in Frank and Shen (2015). The actual yield is defined as the ratio of interest expenses to the sum of long-term debt and debt in current liabilities. Alternatively, the cost 13

15 of debt is measured using BofA Merrill Lynch US Corporate Effective Yields for different rating classes and firms credit ratings. 13 Further, we use the aggregate bond yield from the Barclays Capital Aggregate Bond Index as a proxy for the cost of debt, following Hann et al. (2013). The results using this alternative measure are reported in Table 10. T axrate is the effective tax rate defined as a ratio of total income tax to taxable income. CoE i,t is the cost of equity for firm i in year t. 3 Economic Policy Uncertainty and Innovation Table 1 reports the summary statistics of characteristics and innovation activities by the sample R&D firms. The average R&D spending as a percentage of total assets at the beginning of the year is 10.37%. The mean of the natural logarithm of one plus the number of patents is , that is, 1.41 number of patents on an average. The mean of the natural logarithm of one plus truncation bias adjusted citations is To test the effect of GEPU on innovation activities of individual firms, we estimate the following panel data model: Y i,t+n = α + βuncertainty t + γx i,t + η i + ε i,t, (4) where Y i,t+n is the measure of innovation activities including the R&D ratio, natural logarithm of one plus the number of patents, and natural logarithm of one plus truncation bias-adjusted citations for firm i in year t + n. Since it takes time to generate patents from R&D investments, we estimate the baseline model from t + 3 to t + 4. U ncertainty is the GEPU index, which is the uncertainty index by Baker et al. (2013) scaled by 100. The 13 Since we are interested in variation in the cost of debt over time within each firm and firms do not issue bonds every year, the actual bond issue data are not suitable for our analysis. 14

16 coefficient β captures the effect of GEPU on innovation. The impact of economic policy uncertainty may potentially be confounded by the difference in the first-moment effects of investment opportunities. Particularly, firms with higher growth opportunities may invest more in R&D and innovate. Therefore, we use Tobin s Q to capture variations in investment opportunities and also control for the distinctness in firm attributes. The control variable set, X, includes ln(sales), P P E, CF, Capex, ROA, and T obin s Q. Further, equation (4) includes GDP Growth to control for macroeconomic conditions and a firm s fixed effect η i, which controls for unobserved differences across firms. The firm s fixed-effect estimation exploits the time-series variation within each firm. As shown in Table 2, the coefficients of the uncertainty index in all specifications are negative. Firms spend less on R&D and generate fewer patents as economic policies become more uncertain. Further, the number of citations, a proxy for patent quality, drops as uncertainty intensifies. The coefficient β in the patent specification is in year t + 3 and in year t + 4. The coefficient β in the citation specification goes from in year t+3 to in year t+4. Performing the z-test from Paternoster et al. (1998), we find that the differences in coefficients are statistically significant. The results indicate that the impact of economic policy uncertainty on the quantity and quality of patents strengthens with time. In terms of the magnitude of the impact, a doubling of the level of the uncertainty index is associated with an approximately 1.55% decline in R&D as a percentage of the beginning-of-the-year total assets in year t + 1. Given an average R&D ratio of 10.37%, the impact is also economically significant. 15

17 4 Cost-of-Capital Channel 4.1 Cost of Capital and Economic Policy Uncertainty To understand the cost-of-capital channel, whereby economic policy uncertainty affects innovation, we first investigate how the cost of raising external capital varies with economic policy uncertainty. To this end, we estimate the cost of equity, cost of debt, and WACC for each firm in each year. In specifications (1) (3) of Table 3, we separately regress the cost of equity, cost of debt, and WACC on the economic policy uncertainty index while controlling for firm-fixed effects. The reported results are based on the implied cost of equity estimated using the GLS model and the cost of debt as actual yield. The estimates of the implied cost of equity from the alternative models and bond yields provide similar results (Section 7.7). The coefficients of the uncertainty index are positive and significant. A firm s cost of capital may be affected by other factors, such as its default risk. In specifications (4) (6), we estimate the sensitivity of the cost of capital with respect to economic policy uncertainty while controlling for several factors that directly influence the cost of capital: C i,t = α + βuncertainty t + γx i,t 1 + η i + ε i,t, (5) where C denotes the cost of equity, cost of debt, and WACC. Uncertainty is the GEPU index. X i,t 1 is a set of variables that affects the cost of capital, including ln(sales), P P E, CF, Capex, ROA, T obin s Q, and Z-Score. The Altman Z-score (Z-Score) is defined as 1.2(W C/T A) + 1.4(RE/T A) + 3.3(EBIT/T A) + 0.6(M E/T L) (Sales/T A), where 16

18 W C is working capital, T A represents total assets, RE denotes retained earnings, EBIT denotes earnings before interest and taxes, ME is market value of equity, T L is total book value of liabilities, and Sales is sales. The Altman Z-score measures the financial distress status of a firm. A higher Altman Z-score indicates a lower probability of default. The impact of GDP Growth is also controlled. η i controls for firm-fixed effects. The coefficients of the GEPU index remain positive and significant while controlling for other determinants of the cost of capital. The cost of equity, cost of debt, and WACC increase with rising uncertainty. Given that the uncertainty index has a minimum value of 58 and a maximum of 188 during , the weighted average cost of capital increased 129 basis points. As for the other factors, the negative coefficients for the Z-Score indicate higher costs of equity and debt for firms with a higher probability of default. Tobin s Q is negatively associated with the cost of equity, indicating that firms with high market valuation face a lower cost of equity. 4.2 Financing Decisions and Economic Policy Uncertainty As a further investigation, we test whether economic policy uncertainty affects firms financing decisions. Since external capital becomes costlier as uncertainty increases, the cost-of-capital channel predicts that firms are less likely to issue equity and debt when uncertainty is high. We test this conjecture by examining whether the probability of equity and/or debt issuance is negatively associated with economic policy uncertainty. We estimate a probit model with an equity (debt) issuance dummy as the dependent variable. A firm is considered as issuing equity (debt) if the amount of equity (debt) issuance as a percentage of total assets at the beginning of the fiscal year is over 5%. Equity issuance 17

19 is defined as a change in stockholders equity minus change in retained earnings divided by the beginning-of-the-year total assets. Debt issuance is defined as a change in long-term debt scaled by the beginning-of-the-year total assets. The independent variables are the GEPU index and factors affecting a firm s financing decision. Columns (1) and (2) of Table 4 show that the coefficients of the GEPU index are negative and significant, which is consistent with the hypothesis that firms are less likely to raise external capital in the case of higher uncertainty. Next, we examine whether economic policy uncertainty also affects the amount of securities issues. Our empirical specification uses equity or debt issuance as the dependent variable and tests the influence of economic policy uncertainty while controlling for the confounding factors. Columns (3) and (4) show that the coefficients of the GEPU index are negative and statistically significant. In other words, as uncertainty increases the cost of capital, we observe that firms reduce their equity and debt issuances. Overall, the results are consistent with the financing cost view that firms have a lower propensity to raise external capital when capital becomes more expensive owing to higher uncertainty. 4.3 Cost of Capital and R&D Investments The results thus far show that higher GEPU is associated with a higher cost of capital and fewer innovations. In this subsection, we investigate how cost of capital affects investments in innovation. Table 5 Columns (1) (3) report the estimation results of regressing R&D ratio on cost of equity (CoE), cost of debt (CoD), and WACC (W ACC), while controlling for variables that affect firms R&D investments, GDP growth, and firm-fixed effects. The results show 18

20 that the coefficients on CoE, CoD, and W ACC are all negative and significant, indicating that firms invest less in R&D when the cost of capital increases. A potential concern about this test is that more innovative firms may have higher cost of capital. Although fixed-effects estimations control for time-invariant omitted variables that affect firms cost of capital, it is likely that firms that invest more in innovative projects in the past tend to face higher financing costs. Past R&D investments is a time-varying confounding variable that cannot be subsumed in the fixed effects. To ease this concern, we control for firms past R&D expenditures. 14 When including the lagged R&D ratio to the regression, fixed-effect estimation yields biased estimation because lagged R&D is correlated with the error term (Hsiao (2003) and Baltagi (2005)). To address the endogeneity problem in the dynamic panel data model, we estimate the model by using system Generalized Method of Moments (GMM) developed by Arellano and Bover (1995) and Blundell and Bond (1998). System GMM jointly estimates the dynamic R&D model in differences and in levels, using lagged R&D variables as instruments for differenced equations and lagged first differences in R&D as instruments for the level equations. System GMM provides unbiased estimates for the dynamic panel data model (Blundell and Dias (2000), Bond (2002), and Baltagi (2005)). The results from system GMM estimation are reported on Columns (4) (6). The coefficients on cost of capital remain negative and significant. Given that an average R&D as a percentage of total assets is 10.37% and an average total assets is $ million, a onepercentage increase in WACC could lead to $12.22 million decrease in R&D investment. The magnitude of the impact is economically significant. To the extent that the change in cost 14 See Angrist and Pischke (2008) for differences between fixed effects and lagged dependent variables in establishing causal inferences. 19

21 of capital is owing to economic policy uncertainty, the results provide evidence that increasing uncertainty about government economic policy lowers firms investments in innovation through driving up the cost of capital. 5 Heterogeneity in Exposure to Economic Policy Uncertainty To provide further evidence for the cost-of-capital transmission mechanism of the economic policy uncertainty, we employ the strategy of exploiting the cross-sectional heterogeneity in firms exposure to GEPU. If economic policy uncertainty affects innovation through its influence on firms financing costs, then the cost of capital and investments of firms with higher exposure to such uncertainty should be more affected. 5.1 Exposure to Economic Policy Uncertainty We estimate each firm s exposure to economic policy uncertainty as in Brogaard and Detzel (2015) and examine whether firms with higher uncertainty exposure face a higher cost of capital. We conduct the test by using the monthly stock returns from CRSP s monthly return file adjusted for the CRSP delisting return when available. For firms with more than one class of shares, we maintain the stock returns of class A shares. For each firm, we estimate the following model by using stock returns from the previous 60 months: r i,t r f,t = α + β MKT i (r M,t r f,t ) + βi HML HML t + βi SMB SMB t (6) +βi UMD UMD t + βi GEP U log(gep U) t + ε i,t, 20

22 where r i,t is the returns of stock i for month t and r f,t is the risk-free rate measured by the one-month treasury bill rate. r M,t r f,t, HML t, and SMB t are the excess return on the market, difference between the average returns on small market capitalization portfolios and large portfolios, and difference between the average returns on value (high book-to-market) portfolios and growth (low book-to-market) portfolios in the three-factor model (Fama and French (1993)), respectively. UMD t is the difference between the average returns on the winners and losers of the past year in the four-factor model (Carhart (1997)). log(gep U) t is the natural logarithm of the GEPU index. The coefficient, βi GEP U, measures the exposure of firm i to uncertainty about economic policy. According to a firm s exposure to GEPU, we classify each firm into one of the three portfolios in each year. A more negative β GEP U i indicates greater exposure to economic policy uncertainty because it is the least optimal hedge against increases in uncertainty. Firms in the portfolio with higher exposure to economic policy uncertainty have a more negative βi GEP U, while those in the portfolio with lower exposure to uncertainty have a more positive βi GEP U. The portfolio in the middle is omitted. We classify the high (low) uncertainty state as the time when the GEPU index is above (below) the median. 5.2 Cost of Capital and Heterogeneous Exposure In Panel A of Table 6, we compare the costs of capital for firms with low versus high uncertainty exposure in the low versus high uncertainty states. The comparison results show that the average cost of equity for firms with low uncertainty exposure is 8.13%, while that for firms with high uncertainty exposure is 8.23% when uncertainty is low. The difference is 10 basis points but not statistically significant. When uncertainty is high, the difference 21

23 in the cost of equity between firms with high and low exposure increases to 51 basis points and becomes statistically significant. We observe that the differences in the cost of debt and WACC between firms with high and low exposure are also larger in the high uncertainty state. We then examine whether uncertainty about economic policy has a stronger impact on the cost of capital of firms with higher exposure while controlling for the confounding factors. We regress our proxies for the cost of capital on the GEPU index, an uncertainty exposure dummy (Exposure), and their interaction. Exposure equals one for firms in a portfolio with high exposure to GEPU and zero for firms in the portfolio with low exposure. In Columns (1) (3) of Panel B in Table 6, we control only for firm-fixed effects. In Columns (4) (6), we also explicitly control for other factors affecting firms cost of capital. The estimation results show that the coefficients of the interaction between the uncertainty index and uncertainty exposure dummy are positive and significant in Columns (1), (3), (4), and (6). The results indicate that economic policy uncertainty leads to a larger increase in the cost of equity and WACC for firms with higher uncertainty exposure than for firms with lower exposure. The coefficients of the interaction term are positive but not statistically significant in the cost of debt estimations. Overall, these results provide evidence in support of the cost-of-capital transmission channel. 5.3 Investments and Heterogeneous Exposure The above results confirm the heterogeneous impacts of economic policy uncertainty on firms cost of capital. In this subsection, we first exploit differences in sensitivity of fixedasset investments and R&D to economic policy uncertainty and then examine variations 22

24 across firms. Recent studies find that economic factors have differential impacts on R&D and fixed-asset investments owing to the intangibility nature and uncertainty of R&D investment outcome (Brown et al. (2013), Becker-Blease and Paul (2006), Fang et al. (2014), He and Tian (2014), Grullon et al. (2015), and Dang and Xu (2015)). We, therefore, investigate whether economic policy uncertainty affects capital expenditures differently from R&D. In Column (1) of Table 7 Panel A, we report the result of regressing firms capital expenditures as a ratio of total assets at the beginning of the fiscal year on the GEPU index while controlling for other confounding factors and firm-fixed effects. The coefficient for the GEPU index is negative and significant, indicating that uncertainty about government economic policy also has a negative impact on fixed-asset investments. To compare the sensitivities of R&D and capital expenditures to GEPU, we compute the differences between R&D expenses and capital expenditures for each firm in each year and scale the difference by total assets at the beginning of the fiscal year. We regress the difference between R&D and capital expenditures on the GEPU index, while controlling for observable and unobservable factors affecting investments (Column (2)). The coefficient of the GEPU index is negative and statistically significant, indicating that as uncertainty intensifies, firms reduce R&D investments more than they reduce capital expenditures. The result indicates that investments in innovation are more sensitive to economic policy uncertainty than investments in fixed assets. We further explore cross-sectional heterogeneity in the impacts of uncertainty related to economic policy on investments among firms with different exposure to economic policy uncertainty. If the cost-of-capital channel is relevant, the effect of economic policy uncertainty should be stronger for firms with higher exposure to the uncertainty. 23

25 To test this conjecture, we regress the measure of investment on GEPU index, Exposure dummy, interaction with GEPU index and Exposure dummy, and the control variables. Table 7 Panel B shows that the coefficients on the interaction term are negative and significant for the R&D and CAP EX specifications. The results provide evidence that firms with more exposure to economic policy uncertainty reduce their investments in R&D and fixed assets more as uncertainty increases. The coefficient on the interaction is negative but insignificant in the specification of R&D CAP EX, indicating that R&D investments of firms with more exposure to economic policy uncertainty are not significantly more sensitive to the uncertainty than capital expenditures are. Overall, our results highlight the importance of cost of capital in transmitting uncertainty about economic policy to the real economy. 6 Investment Irreversibility Channel The results thus far support the prediction that uncertainty about economic policy increases firms cost of capital, which reduces innovation activities. Another possible explanation is that managers, who face uncertainty, have an incentive to postpone investments until more information is revealed if investments are costly and (partial) irreversible. It is interesting to see whether cost of capital affects the economic policy uncertainty and R&D investment relationship after controlling for the investment irreversibility channel. To investigate the marginal effect of the cost-of-capital channel, we control for the degree of irreversibility in investment using several industry-level proxies as in Gulen and Ion (2016). The first proxy is the asset redeployability measure proposed by Hyunseob and Kung (2014). We use the 1997 Bureau of Economic Analysis capital flow table to construct the proxy. The table contains expenditures on 180 asset categories by firms in 123 industries. The first 24

26 irreversibility measure is constructed in three steps. First, a redeployability score for each asset category is computed as the number of industries using the asset scaled by 123. An asset is considered as being used by an industry if the industry s expenditure on the asset is more the 0.1% of the total expenditure on the asset by all industries. Second, an industry-level redeployability index is constructed as a weighted average of the redeployability scores of the invested asset categories. The weight for each asset category is an industry s expenditure on the asset divided by the industry s total expenditures. Firms in industries with a higher redeployability index have more redeployable assets and are more likely to recover a higher fraction of investment. Third, the irreversibility measure for each industry is calculated as the inverse of redeployability index. The second irreversibility measure is constructed based on cyclicality of firms sales following Sharpe (1994) and Almeida and Campello (2007). We first calculate the correlation between each firm s sales and gross national product over the sample period. The firm-level correlations are then averaged at the three-digit SIC level. The industry-level irreversibility measure is a dummy variable that is one for industries with correlations above the median, and zero otherwise. It is harder for firms in cyclical industries to recover their investments because potential buyers of their assets are likely to be also affected by negative shocks during economic downturns. The third proxy of irreversibility is an industry-level measure of sunk costs as in Farinas and Ruano (2005). Investments in industries with more sunk costs is harder to reverse. We first measure firm-level sunk costs using three proxies: rent expense, depreciation expense, and past sale of PPE, all normalized by the beginning-of-fiscal-year PPE. The industry-level means of these three proxies in each year are then obtained at the three-digit SIC level. We 25