Foreign economic policy uncertainty and U.S. equity returns

Board of Governors of the Federal Reserve System International Finance Discussion Papers ISSN 1073-2500 (Print) ISSN 2767-4509 (Online) Number 1401 December 2024 Foreign economic policy uncertainty and U.S. equity returns Mohammad. R. Jahan-Parvar, Yuriy Kitsul, Jamil Rahman, and Beth Anne Wilson Please cite this paper as: Jahan-Parvar,Mohammad. R.,YuriyKitsul,JamilRahman,andBethAnneWilson(2024). “Foreign economic policy uncertainty and U.S. equity returns,” International Finance Discussion Papers 1401. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/IFDP.2024.1401. NOTE: International Finance Discussion Papers (IFDPs) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the International Finance Discussion Papers Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Foreign economic policy uncertainty and U.S. equity returns Mohammad R. Jahan-Parvar Yuriy Kitsul Jamil A. Rahman Beth Anne Wilson∗ First draft: June 2024 This draft: September 2024 Abstract We document that foreign economic policy uncertainty (EPUF) has significant incremental predictive power for excess U.S. stock returns in the presence of domestic EPU, both in aggregate and for returns of portfolios constructed on firm characteristics, for 6 to 12-months-ahead horizons. We find that EPUF shocks primarily transmit to equity prices through cash flow news rather than the discount rate news channel. We examine whether responses of select macro-financial variables to an adverse EPUF shock are consistent with this transmission mechanism. Corporate investment outlays, payouts, and aggregate credit demand decline in response to such a shock. JEL Classification: G11, G12, C13, E20, E30. Keywords: Economicpolicyuncertainty, Cashflows, Discountrates, ICAPM,Returnpredictability, Transmission channels. ∗Jahan-Parvar, Kitsul, and Wilson are affiliated with the International Finance Division, Federal Reserve Board of Governors. Rahman is affiliated with the finance department, Yale School of Management. Please direct correspondence to Mohammad R. Jahan-Parvar (mohammad.jahan-parvar@frb.gov). We thank Sirio Aramonte, Andrew Detzel, Matteo Iacoviello, Oscar Jorda, Niklas Kroner, Sai Ma, Jackson Mills (FMA discussant), Juan-Miguel Londono, Dino Palazzo, Tatevik Sekhposyan, Emre Yoldas, and Stijn Van Nieuwerburgh for detailed discussions and seminarparticipantsatFederalReserveBoard,NorthAmericanSummerMeetingoftheEconometricSociety(2021), FMAannualmeeting(2021),CFE(2021),MidwestEconometricsMeeting(2022,MichiganStateUniversity),Math- Works Finance Conference (2023), and IRMC (2024, Milan). We thank Jake Harmon and Bill Lang for excellent research assistance. The analysis and conclusions are those of the authors, and do not reflect the views of other members of the research staff or the Board of Governors. 1

1 Introduction Seminal studies of Bloom (2009), Basu and Bundick (2017), and Baker, Bloom and Davis (2016) establish that economic uncertainty in general and economic policy uncertainty (EPU)–uncertainty about fiscal, monetary, regulatory and other economic policies–in particular, affect real economic decisions, including firms’ investment and hiring plans. Subsequently, P´astor and Veronesi (2012, 2013) and Brogaard and Detzel (2015) show that EPU predicts broad equity market index returns intheUnitedStates.1 Inaddition,thelatterstudyarguesthatEPUpredictabilityoperatesthrough the discount-rate channel. Atthesametime,thecross-countryinteractionsofeconomicpolicyuncertaintyandequityprices have received limited attention in academic literature, even as (1) the economies, financial markets, andbusinesseshavebeenbecomingincreasinglyinterconnected(Demirer,Diebold,LiuandYilmaz, 2018 and Candelon, Ferrara and Jo¨ets, 2021),2 (2) measures of economic activity, financial market volatility, and economic policy uncertainty across countries exhibit notable co-movement (Table 1), and (3) financial press, market analysts, and firm earnings call reports frequently cite uncertainty about economic conditions and policies abroad as affecting performance of the domestic equity returns and markets. For example, Hassan, Schreger, Schwedeler and Tahoun (2024) document thatU.S.-basedcompaniesfrequentlydiscusscountryrisksoriginatingfromBrazil, Canada, China, Japan,andMexico,basedontextualanalysisoftheirearningscalls. Thispaperclosestheimportant but so far neglected gap in research on the relationship between economic policy uncertainty and equity markets by investigating whether foreign EPU (EPUF) helps explain future excess equity returns in the U.S., as well as the channels through which EPUF shocks transmit to the U.S. stock returns. Inparticular,weask(1)whetherEPUF predictsvariousmeasuresofaggregateequityreturnsin the U.S.; (2) how the predictive ability of EPUF differs across returns of equity portfolios formed on firm characteristics that may affect returns sensitivity to foreign EPU (to better understand sources of aggregate predictability); (3) whether EPUF shocks transmit to equity prices through cash-flow or discount-rate channels; (4) what the responses of select financial and macro variables to EPUF shocks shed additional light on the transmission of EPUF shocks to U.S. stock prices. To construct a proxy for foreign economic policy uncertainty, we turn to the widely-used global EPUmeasureofBakeretal.(2016), whichisbasedonnewsarticlecountsin21countries, including the United States. Global EPU reflects perceived uncertainty about what economic policies will be implemented, who will implement the policy and when, and what impact the policy in question will have. To obtain the foreign EPU measure, we strip the U.S. component from the global EPU 1 BrogaardandDetzel(2015)alsodemonstratethatEPUcommandsariskpremiuminthecross-sectionofU.S.equity returns. 2 Note that trade in goods and services (exports plus imports) accounted for about 27% of the United States GDP in 2018, up from about 9.2% and 20% in 1960 and 1980, respectively. 1

by orthogonalizing global EPU (EPUG) with respect to its U.S. counterpart (EPUUS). We then investigate whether the constructed EPUF measure has incremental predictive power forexpectedU.S.equityreturnsinthepresenceofEPUUS andothercontrolvariables. Wefindthat EPUF predicts U.S. stock index returns at horizons between 9 to 12 months ahead. For returns of portfolios formed on firm characteristics, we find that these predictive effects are concentrated in companies that are typically larger, acquire more assets, have higher capital expenditures, and have higher foreign exposure than their peers. On balance, we observe that EPUF predicts longerhorizon U.S. returns compared to EPUUS, predictive power of which is concentrated for horizons less than six months. This is consistent with the delayed reaction mechanism (e.g. Hong and Stein, 1999), with information from abroad taking longer to diffuse to U.S. equity markets compared to domestic information. We next show that foreign EPU primarily affects the cash flow news component of the U.S. equityreturns,contributingtothedebateonwhetherdiscountrateorexpectedcashflownewsdrive equity prices (see Bianchi, Lettau and Ludvigson, 2022, Chen, Da and Zhao, 2013, and Cochrane, 2011). This finding is intuitive. It is less likely that changes in economic policy uncertainty abroad materially and consistently affect monetary policy, policy rates, and discount rates in the United States. On the other hand, domestic firms with material foreign exposure are likely to adjust their investment projects following the arrival of adverse foreign EPU news. These adjustments may lead to lower future cash flows from successful projects and, as a result, lower future payouts to shareholders. Weprovideadditionalevidenceconsistentwiththispotentialtransmissionmechanism by studying responses to such shocks of U.S. financial and macro variables that either affect or are affected by discount rates and future cash flows. We find that foreign EPU shocks appear to induce precautionary delays in demand for credit and capital expenditure. That is, after the arrival of an EPUF shock, in aggregate, firms reduce dividend distributions, as well as borrow and invest less. Our study contributes to the literature on effects of economic policy uncertainty on the macroeconomy and financial markets. Most of the literature focuses on domestic implications of changes in U.S. EPU. Examples include P´astor and Veronesi (2012, 2013) and Brogaard and Detzel (2015), which focus on the relationship between U.S. EPU and expected excess returns, as well as Kaviani et al. (2020) and Bonaime et al. (2018), which document effects of U.S. policy uncertainty on credit spreads and mergers and acquisition activity, respectively. Cross-country EPU spillovers have received limited attention, with the existing studies mostly focusing on cross-country spillovers of EPU measures. (Kl¨oßner and Sekkel, 2014, Shin et al., 2018). In contrast, ours is the first study on the cross-border effects of EPU in equity markets, an important question in a world of increasingly interconnected economies and financial markets. In addition, we validate our findings on crosscountry EPU spillovers in equity markets with results on spillovers to macro-financial variables. These results contribute to nascent literature on cross-border economic spillovers of economic un- 2

certaintymeasures, whetherataggregateuncertaintylevel(Londonoetal.,forthcoming, Greenland et al., 2019) or at firm level as in Hassan, Schreger, Schwedeler and Tahoun (2024). In particular, Hassan et al. (2024) document the relationship between perceived country risks transmitted from abroad and domestic firm-level corporate decisions. Moreover, our findings regarding the inverse relation between EPU levels and investment and capital expenditure are in line with Gulen and Ion (2015). By documenting that U.S. equities in aggregate are exposed to uncertainty spillovers from abroad, our findings also contribute to the growing literature on the cross-country interactions between political risk and financial markets. Examples include Boutchkova, Doshi, Durnev and Molchanov (2012) who study the relationship between national and foreign political risk and returns volatility, Kelly, P´astor and Veronesi (2016) who extract the political uncertainty protection embedded in options and find that the effects of political uncertainty spill over across countries, Kim (2019) studies the link between political uncertainty and financing costs using syndicated loan premiums. Brogaard, Dai, Ngo and Zhang (2020) find that political uncertainty measured by the United States election cycle spills over to equity prices abroad through the discount rate channel. It is intuitively plausible that political developments in the United States–as the largest and most significant global financial center–meaningfully transmit to other markets and affect investors’ risk toleranceanddiscountrates. Incontrast,ourstudyformallyteststhereverserelationship(spillovers from abroad to the U.S. equity market) and documents that the dominant transmission mechanism for spillovers from abroad is through cash flows. We also note that while political risk and EPU overlap, they measure different types of risk. Therestofthepaperproceedsasfollows. Wepresentourdata,constructionofforeigneconomic policy uncertainty measure, and the method of extraction of foreign EPU shocks in Section 2. We present aggregate index-level and portfolio returns predictability empirical findings in Section 3. Section 4 presents our findings regarding transmission channels of economic policy uncertainty shocks to financial and macroeconomic variables. Section 5 concludes. 2 Data We use the global and the Unites States EPU measures provided by Baker, Bloom, and Davis for January 1997 to May 2021.3 We use the 3-component index version of U.S. EPU (henceforth, EPUUS). This index is a weighted average of the news-based EPU (50%), tax-code expiration data, forecaster disagreement, and Federal/State/Local disagreement measures (each accounting for 1/6 of the remaining 50%). The global EPU index combines news article counts in 21 countries, including the United States, that account for about 75% of world output. EPUG reflects perceived 3 See https://www.policyuncertainty.com/index.html. While global EPU index is only available from January 1997, U.S. EPU is available for a longer period. 3

uncertainty about which economic policies will be implemented, who will implement the policy and when, and what impact the policy in question will have. It reflects uncertainty about a broad range of policies, including those by fiscal and monetary authorities, but also the potential economic effects of policies that are not traditionally viewed as economic, such as military actions. It is available in two versions: current price GDP-weighted and PPP-adjusted GDP-weighted. We use the current price GDP-weighted series. Our results are robust to using either measure. The series are scaled by 100, and then demeaned. Well-known alternatives including economic uncertainty measures of Jurado et al. (2015) or Ludvigson et al. (2021), activity measures such as Aruoba et al. (2009), monetary policy uncertainty measure of Husted et al. (2020), or trade policy uncertainty of Caldara, Iacoviello, Molligo, Prestipino and Raffo (2020) are U.S.-specific and construction of a global version of these measures is beyond the scope of this study. Caldara and Iacoviello (2022) and Londono, Ma and Wilson (forthcoming) provide several country-specific indexes for their geopolitical risk and real economic uncertainty measures, respectively. However, while trade uncertainty, geopolitical risk, real economicuncertainty, andeconomicpolicyuncertaintyareundoubtedlyrelated, theycapturedifferent types of uncertainty.4 Thus, we remain focused on EPU in this study. We collect monthly data for equity returns, relevant financial and accounting quantities, pricing factors and firm characteristics in the universe of the U.S. publicly traded companies in the NYSE/AMEX/NASDAQ exchanges from the merged Center for Research in Security Prices (CRSP),Compustat,CapitalIQ,andotherresourcesfromWhartonresearchdataservices(WRDS) between January 1997 and May 2021. We access other macroeconomic and financial variables from FRED database maintained by Federal Reserve Bank of St. Louis, and authors’ websites (such as Kenneth French and Robert Shiller, among others).5 We investigate the relationship between foreign EPU and U.S. equity returns in two steps; first using predictive regressions and then studying the responses of financial and macroeconomic quantities to foreign EPU shocks. The first step requires the construction of a proxy for foreign 4 ThesamplecorrelationsbetweenCaldaraetal.(2020)TPUindexandEPUUS andEPUG are0.37and0.57,respectively, in our sample. These correlations for Caldara and Iacoviello (2022) GPR index and EPUUS and EPUG are 0.50and0.74,andthesequantitiesforLondonoetal.(forthcoming)REUindexandEPUUS andEPUG are0.31and 0.35. 5 The following data were retrieved from FRED, Federal Reserve Bank of St. Louis: 1) from U.S. Bureau of Economic Analysis: Real Gross Domestic Product [GDPC1], https://fred.stlouisfed.org/series/GDPC1; Real Gross Private Domestic Investment [GPDIC1], https://fred.stlouisfed.org/series/GPDIC1; Unemployment Rate [UNRATE], https://fred.stlouisfed.org/series/UNRATE; 2) from Chicago Board Options Exchange: CBOE Volatility Index: VIX [VIXCLS], https://fred.stlouisfed.org/series/VIXCLS; 3) from Board of Governors of the Federal Reserve System (US), Commercial and Industrial Loans, All Commercial Banks [BUSLOANS], https://fred.stlouisfed.org/series/BUSLOANS. See https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data library.html for data on equity portfolio returns, https://shillerdata.com/ for data on cyclically adjusted price-to-earnings and dividend-price ratios, and https://mysimon.rochester.edu/novy-marx/data lib/index.html for data on profitability factors. Realized volatility data used in Table 1 are from Heber et al. (2009). 4

economic policy uncertainty. The second step requires extraction of foreign EPU shocks. We describe the methods for construction of foreign economic policy uncertainty and extraction of shocks next. 2.1 Construction and economic significance of foreign EPU Given the size of the U.S. economy, the centrality of U.S. financial markets, and the position of the U.S. dollar as the global reserve currency, we must address the weight of U.S.-related news in the construction and dynamics of EPUG. Kl¨oßner and Sekkel (2014) and Shin, Zhang, Zhong and Lee (2018) show that there are significant economic policy spillovers from the United States to other countries. In particular, they show that U.S. economic policy uncertainty news contaminate EPU measurement abroad. In Table 1, we report that EPUUS and EPUG are highly correlated, with a coefficient of correlation about 0.80. Thus, using both EPUG and EPUUS in predictive regressions of cumulative stock returns leads to correlated regressor problems. We orthogonalize EPUG and EPUUS by fitting the following regression model to data to address these issues:6 EPUG = κ +κ EPUUS +ν . (1) t 0 1 t t We rename the OLS regression residuals, νˆ, from fitting equation (1) to data “foreign EPU” or t EPUF. This variable captures the variation in EPUG that is, by construction, uncorrelated with t EPUUS at time t, but have a coefficient of correlation equal to 0.59 with EPUG. Thus, orthogonalization is necessary to disentangle the effects of global and U.S. economic policy uncertainties. Table 2 reports the summary statistics of these measures. The constructed EPUF measure is intended to be similar (in statistical properties) but linearly uncorrelated to EPUUS, as shown in the table. We note that all three EPU measures are considerably less persistent than long-term asset pricing variables commonly used for explaining future excess stock returns in empirical studies. For example, the first, fifth, and tenth-order autocorrelation coefficients for S&P500 price-earnings (P/E) ratios are 0.98, 0.95, and 0.80, respectively. Throughout the paper, we use EPUF and demeaned values of EPUUS in the analysis. Figure1displaysthethreeEPUmeasures. Asmentionedearlier,U.S.andglobalEPUmeasures are highly correlated and track each other closely. As the residual from the statistical model in equation (1), EPUF is positive valued when EPUG is greater than its fitted value implied by EPUUS and equation 1, and negative-valued when the reverse holds. We observe a significant negative spike in EPUF corresponding to September of 2001. EPUF remained in negative territory for the better part of the Global Financial Crisis (GFC) period. Notable positive spikes in EPUF include November 2011 (negative developments related to the Euro Area Crisis) and November 6 One could also use ridge regressions or other solutions. 5

2016 (the United States presidential election). EPUF remained in positive territory from 2018 through late 2020 as, among other factors, trade tensions between the United States and its major trading partners rose. The bottom left panel of Figure 1 reports the histogram of EPUF. In Table 3, we present the estimates of standardized linear regressions of EPU measures on commonly used economic state variables capturing economic conditions for the United States and euro area. We use standardized variables, thus, the magnitudes of estimated coefficients translate into a beta-standard-deviation in EPU for every standard-deviation change in the explanatory variable, all else equal. The estimated model is I J EPU∗ = γ∗+ (cid:88)(cid:88) γi,∗Xi +e∗, (2) t c j t−j t i=1 j=0 where EPU∗ is one of the three EPU measures, Xi are contemporaneous and lagged standardized t t−j variables, and e∗ are error terms. We include VIX and option-implied volatility for Euro STOXX t 50 index (VSTOXX), the spread between 10-year and 1-year U.S. Treasury bond and German bund yields (Spread), the spread between U.S. BBB and AAA non-financial corporate bond yields (default spreador Def.), the smoothed log price-dividend ratios (log(P/D)) forS&P500 andEURO STOXX 600 indexes. We also include Federal Reserve Bank of Chicago’s national activity index (CFNAI) and log values of the Baltic Dry Index (BDI), a shipping freight-cost index that acts as a bellwether for international merchandise trade. Table 3 reports the estimated slope parameters, γˆ s, and their Newey and West (1987) standard errors in a multivariate regression of equation (2).7 j WereportU.S.-andeuro-areaspecificestimationresultsinPanelsAandBofTable3, respectively. Wefindthattheestimatedslopeparameters,γˆ forcontemporaneousandupto3-monthslagged j values of VIX and VSTOXX for EPUUS are positive-valued and significant. Other estimated γˆ for j EPUUS that are statistically different from zero include U.S. and euro-area P/D ratios, German spreads, CFNAI, the BDI (negative-valued), and U.S. Treasuries’ spread (positive-valued). For EPUG (the middle of Table 3), we find statistically significant γˆ for U.S. default spread, U.S. P/D j ratios, German term spreads, the BDI (negative-valued), and VIX (positive-valued). The bottom panel of Table 3 reports estimated γˆ for EPUF. These parameters are statistically significant for j U.S. default spread, U.S. P/D, VIX, VSTOXX, German spreads, the BDI (negative-valued), and CFNAI and euro-area P/D (positive-valued). ThenegativerelationshipofallEPUmeasureswiththeBDIindexisinterestingandinformative: when trade intensity rises (recedes), both domestic and foreign EPU measures decline (rise). This findingvalidatestheresultsreportedbyLondono, MaandWilson(forthcoming)whodocumentthe strength of trade ties as a cross-country transmission channel for economic uncertainty. Negative relationships with P/D ratios imply that EPUF is high (low) when times are “bad” (“good”) 7 Univariate results are available, but not reported. 6

either in the U.S. or abroad. This countercyclical dynamics of EPUF may be behind the measure’s predictive power for the time-series of U.S. equity returns, which we document later in the paper. 2.2 Construction of foreign EPU shocks In our analysis, we are interested in the response of macro-financial quantities to foreign economic policy uncertainty shocks. The goal is to compute impulse-response functions for variables of interest, conditional on the realization of a positive shock or a sequence of positive shocks to the uncertaintymeasure: whentheleveloftheuncertaintymeasurerisesunexpectedly. Theconstructed EPUF is quite persistent. As shown in Table 2, the values of its first and tenth autocorrelation coefficients are 0.89 and 0.71, respectively. Thus, the moves in EPUF level values, albeit sizeable, arenotshocks. Asaresult,wemustextractthedesiredshocksfromtheeconomicpolicyuncertainty indexes. We use the method described in Diercks, Hsu and Tamoni (2024) for this purpose. This shock extraction procedure echoes many elements of EPUF construction. To recover foreign EPU shocks, we fit the data using the following vector auto-regressive (VAR) specification: Y = A +A Y +εY, (3) t 0 1 t−1 t whereY isthevectoroftime-seriesdatausedforasequential(orCholesky)decomposition,withthe t following ordering of variables: EPUG index, EPUUS index, the log of the MSCI ACWI (excluding the United States) index, and the federal funds rate. A is a vectors of constants, A is a matrix of 0 1 coefficients, and εY is the shocks matrix. We choose the lag-lengths for all variables in Y based t t−1 on the Akaike information criteria (AIC), and treat the first column vector in εˆY(= Y −Yˆ) as t t t EPUF shocks.8 The international orientation of our exercise and data limitations force us to choose a different set of variables compared to Diercks et al. (2024). First, instead of S&P500 index, we use MSCI ACWI (excluding the United States) index to account for foreign equity price movements. Second, since we must account for correlations between global and U.S. policy uncertainty measures, in addition to EPUG, we also include EPUUS in the procedure. Third, since global measures of macroeconomic quantities such as unemployment or industrial production are not available, we can not include them in the procedure. Finally, we include the federal funds rate as a measure of monetary policy in the United States and an influential factor in setting the global cost of capital (see Miranda-Agrippino and Rey, 2020). The right-hand panels in the middle and the bottom rows of Figure 1 display these shocks and their histogram, respectively. 8 The order of Cholesky decomposition assumes no feedback from EPUG to EPUUS, which could not always be the case. Therefore, the results shown later could be viewed as lower bounds for the effect of EPUF shocks. 7

3 Aggregate and portfolio excess returns predictability Throughout this section, we discuss predictability of U.S. broad equity index returns and those of portfolios constructed based on certain financial or economic characteristics by foreign economic policy uncertainty measure discussed above. To this end, we fit the following statistical model to the data: J ri = αi+βiEPUF +βiEPUUS + (cid:88) φiZj +εi , (4) t,t+k 1 t 2 t j t t,t+k j=1 whereri arethecumulativeexcessreturnsofanequityindexor(valueweighted)stockportfolio t,t+k between times t and t+k (k assumes values between 1 and 12, implying one month to one year ahead predictions) over 1-month U.S. Treasury Bill rate (henceforth, referred to as “returns” for brevity)9, EPUF is constructed as in Section 2.1, and EPUUS is the demeaned U.S. EPU measure t t of Baker et al. (2016), Zjs are control variables (discussed below), and finally εi is an error t t,t+k term. Since economic policy uncertainty indexes are ultimately measures of risk, we expect that estimatedslopeparametersforEPUF andEPUUS inequation(4), βˆi andβˆi, tobepositivevalued, 1 2 as in the conventional Merton (1973)-style risk-return trade off relationship. To avoid look-ahead bias, we construct foreign EPU measure at time t with data available only up to time t for each step of the analysis. Similar to Brogaard and Detzel (2015), We include the following control variables: the NBER recession indicator, 10-year over 1-year U.S. Treasury bond spreads, BBB - AAA corporate bond spreads, Shiller’s (log) cyclically-adjusted aggregate P/E (CAPE) ratios, monthly changes in VIX option-implied volatility index (∆VIX), monthly growth rates of the industrial production index (∆IP ), andtheCFNAIindex.10 WealsoincludetwoofFamaandFrench(1992)threefactors: size t (SMB) and value premium (HML), since market factor (CRSP value-weighted returns) is one of our test variables. Finally, we include Jegadeesh and Titman (1993) “momentum” and Jegadeesh and Titman (1993) and Chan, Jegadeesh and Lakonishok (1996) “reversal” factors. If all φ = 0, then we only observe the predictive power and the ability of EPU measures in j explaining returns’ variation. With either β or β set to zero, the model gauges the predictive 1 2 power of either EPUUS or EPUF. Ideally, we expect εis to be i.i.d. standard normal. However, t this is almost never the case in empirical research. Regression residuals demonstrate notable serial correlation and heteroscedasticity. To address this problem and similar to Bali et al. (2017) and Golez and Koudijs (2018), among others, we compute Newey and West (1987) heteroscedasticity and serial correlation-consistent (HAC) standard errors for the estimated parameters.11 9 This specification implies a holding period of one month, rolled over to the next. 10The CFNAI series does not have a time-trend component. Other option-implied measures for headline equity price indexes in major economies exist. These indexes tend to closely co-move with VIX, see Table 1. As a result, and to save space, we only report results based on VIX as an alternative aggregate volatility proxy in this study. 11Since EPU measures are less persistent than typical long-term pricing variables such as P/E or P/D ratios (Section 8

3.1 Aggregate excess returns As mentioned earlier, we expect that the increased interconnection of U.S. economy with the rest of the world should expose U.S. equities to sources of uncertainty stemming from abroad. If this assertion is true and a sufficiently large number of publicly traded and U.S.-listed firms are indeed exposed to foreign sources of economic policy uncertainty, then we should be able to detect predictive power for EPUF–which measures policy uncertainty unrelated to U.S. news–for broad equitymarketindexreturns. Ourgoalinthissectionistestingthishypothesis. WeshowthatEPUF indeed has predictive power for the U.S. aggregate index excess returns, even after controlling for and in the presence of EPUUS. Thus, we complement and extend the findings of P´astor and Veronesi (2013) and Brogaard and Detzel (2015), who already establish EPUUS’s predictive ability. Both studies are based on aggregate index returns heavily populated by larger companies (S&P500 and CRSP value-weighted, respectively). We report our findings in Table 4. As in previous studies, we investigate the predictability of value-weighted CRSP portfolio, comprising all traded stocks in the NYSE/AMEX/NASDAQ exchanges, and the S&P500 index returns, representing the largest U.S. corporations. We also study whether technology-heavy Nasdaq Composite and small cap-dominated Russell 2000 index returns are also predictable by EPU measures.12 We only report slope parameters for the EPU measuresinTable4tosavespace. WereportestimatedslopeparametersforthetwoEPUmeasures in the presence of pricing factors discussed above (Zjs), and when φ are set to zero. The reason t j for the latter is two-fold: first, we want to observe any loss of statistical significance due to the presence of control variables, and second, we wish to gauge the ability of the two EPU measures in explaining the variations in cumulative index excess returns (the incremental adjusted R2 between the model with controls and those without). WefirstestablishthatEPUUS predictsS&P500returnsatk = 2to12andCRSPvalue-weighted returns at k = 2 to 6 months ahead, confirming the findings of P´astor and Veronesi (2012) and Brogaard and Detzel (2015), respectively (not shown). We next fit equation (4) to data. In the presence of EPUUS and control variables listed above, we find statistically significant evidence that Nasdaq returns are predictable by EPUF at k ≥ 6 and for CRSP value-weighted and S&P500 returns at k ≥ 9. Absent control variables but with EPUUS still present, these results generally hold, albeitatslightly lowersignificance levels. EPUF predictsRussell 2000returnsonlyatk = 12, while these returns are predictable by EPUUS and estimated slope parameters are significant for 2.1),webelievethatNeweyandWest(1987)correctionstostandarderrorsaresufficienttoaddresstheissuesthatarise from using overlapping returns, generated variables, and serially correlated or heteroscedastic residuals in predictive regressions. As noted in Section 3.2, using bootstrapped standard errors as an alternative remedy do not materially change our empirical findings. 12NasdaqCompositeisavalue-weightedindexthattracksmorethan3,000firmstradedontheNasdaqStockMarket. The index is dominated by technology sector with about 50% weight, followed by consumer services and health care sectors with about 20 and 10% weights, respectively. 9

k ≤ 9withcontrolsvariables. Ingeneral, weobservesomeweakeningofEPUUS predictivepoweras k increases. This pattern is not present for EPUF, which implies that the opposite might hold for foreign policy uncertainty: that EPUUS and some control variables account for potential EPUF’s predictability in shorter horizons, but their power dilutes notably as k increases. Another plausible explanation, which as we show later is also borne by our empirical findings, is that these somewhat long predictability horizons are indicative of EPUF operating through the slower moving cash-flow channel of firms with statistically significant exposure to such risks. Since these effects take longer to appear in firms’ financial or operational communications, there is a delay between their arrival and market’s reaction, leading to the predictability patterns documented above.13 The estimated Student-t statistics for these findings, especially for k = 12 exceed the threshold advocated by Harvey et al. (2015). Adjusted R2s increase in k uniformly. For predictive horizons where estimated βˆ statistically 1 differentfromzero, EPUF andEPUUS togetheraccountforbetween7and12.8percentofvariation in returns (Nasdaq at k = 6 and CRSP value-weighted at k = 12, respectively). These values are based on models with no control variables, and thus adjusted R2s only gauge the ability of EPU measures in accounting for variations in returns. On balance, EPUF contributes to about half of these adjusted R2 values. Estimated βˆs for both EPUUS and EPUF are generally positive-valued. These results are coni t t sistentwiththeintuitionfromMerton(1973)’sintertemporalcapitalassetpricingtheorysuggesting that investors demand positive expected return compensation for bearing risk and uncertainty. Taken together, we summarize our findings as follows: • Sizemaynotbethesoledeterminingfactor. EPUF predictsbothCRSPandS&P500returns. While S&P500 returns by construction represent the largest U.S.-listed companies, CRSP returns (while influenced by larger corporations) comprise the entire universe of publicly traded firms. Many large corporations are multinationals with significant international sales, links, andoperations. Butthesecompaniesalsohavetheresourcestohedgemostglobalrisks. Wenotethatwithcontrolvariablespresent, estimatedβˆ sforNasdaqreturnsarestatistically 2 not different from zero, while βˆ s are statistically different from zero for k ≥ 6 regardless of 1 control variable’ presence. This observation means that EPUF has notable predictive power for the broad but tech- and IPO-heavy Nasdaq Composite index. In addition, we have some evidence (albeit much weaker) that small-cap Russell 2000 returns might be predictable at longer horizons by EPUF. Thus, we believe that EPUF appears to predict returns for a non-negligible subset of medium-sized (or small) companies in the Unites States, implying notable exposure to global economic policy risks. We conduct a thorough search to identify 13Incontrast,asdocumentedbyCampbellandAmmer(1993)andPindyckandRotemberg(1993)amongmanyothers, the time required for discount rate shocks to affect prices are notably shorter. 10

the characteristics of companies that are affected by changes in foreign policy uncertainty in later sections. • There are similarities and also intriguing differences in patterns of predictability between EPUF and EPUUS. Based on similarities, they are both uncertainty factors that have predictive power for U.S. aggregate returns. However, based on their notable differences, especially with respect to prediction horizon k and their interactions with control variables mentioned above, they appear to operate in different horizons and affect different drivers of asset prices. These observations are in line with the findings of P´astor and Veronesi (2013) and Brogaard and Detzel (2015), and imply that EPUF predicts longer-duration elements of asset prices. We investigate the channels responsible for this observation in subsequent sections. • We find unambiguous, positive, and statistically significant risk-return trade off between current levels of economic policy uncertainty and future aggregate returns. Inclusion of commonpricingfactorsdoesnotweakenourresults. Thesign,size,andstatisticalsignificance of the estimated slope parameters are generally invariant to inclusion or exclusion of these factors. 3.2 Robustness checks for aggregate returns Wecarryoutteststoassesstherobustnessofpredictiveresultsdiscussedsofar. Ourresultssurvive these tests and we thus conclude that EPUF indeed predicts cumulative aggregate excess returns of headline U.S. equity market indexes. We discuss two primary robustness checks. We report these results in the Appendix. First, a concern is using generated variables in predictive regressions. As constructed, EPUF t is a generated variable and its inclusion may lead to under-rejection of the null hypothesis that βi = 0. A common remedy is to use bootstrapped standard errors in constructing Student-t 1 statistics, instead of Newey-West standard errors. We follow the methods discussed in Ruiz and Pascual (2002) and generate bootstrapped standard errors for predictive regressions. We find that the statistical significance of EPU measures’ slope parameters do not change much, and that the estimated slope parameters for EPUF, βˆ1, continue to remain statistically different from zero. 1 Another reasonable concern is whether domestic pricing factor (such as size, value, or momentum) are adequate control variables in estimated models and whether predictive results are robust to the inclusion of international pricing factors. Since international factors are available for developed economies, we replace domestic Fama-French size, value premium, and momentum factors their developed-economy counterparts. Our predictive results are robust to the inclusion of developed-economy factors, and similar to the findings discussed in Section 3.1. 11

3.3 Portfolio excess returns So far, we have established that EPUF has predictive power for aggregate equity returns in the UnitedStatesandthisresultisrobusttothepresenceofvariousassetpricingfactors. Animportant question is which assets are sensitive to fluctuations in foreign economic policy uncertainty? For example, Gulen and Ion (2015) and Greenland et al. (2019) show that higher EPU is associated with declines in corporate investments, while, as Hou et al. (2014) discuss, high-investment firms have better investment opportunities and higher future cash flows. These firms may be better positioned to withstand policy uncertainty shocks by adjusting their asset holdings or investment projects. At the same time, some of their investment opportunities are likely to be abroad, making them more exposed to foreign EPU shocks. Thus, a relevant question is whether (foreign) EPU affects future returns for high-investment firms more than returns for low-investment or financially constrained firms. In addition, firms with high foreign sales are more likely to see their current equity prices decline (and expected returns increase) in the face of a foreign EPU surge. The question than follows if the predictive relationship between foreign EPU and future returns of firms with high foreign sales is stronger than that for firms with low foreign sales. Werespondtothesequestionsinthissection. Ourstrategyistouseexcessreturnsofportfolios, constructed along certain firm characteristics, over 1-month Treasury Bill rates to isolate features thatsignalstocks’sensitivitytoEPUF. Tothisend, wetestwhetherforeignordomesticEPUmeasures predict portfolio returns at horizons up to 12 months ahead. In particular, we are interested in EPU measures’ predictive power for particular segments of a portfolio–for example the top or the bottom terciles or quartiles. Thus, we investigate the properties of linear asset-pricing model presented in equation (4), fitting it to data where ri are returns on a particular portion–or t,t+k leg–of a portfolio, or they are the difference between two such legs. Given the findings discussed in Section 3.1, we sort all companies in CRSP/Compustat universe at the end of June of year t on the following firm characteristics or fundamental values: size, investment, capital expenditure on plant and equipment (CapEx) to price, cash-flow to price, and theratioofforeigntototalsales.14 Weconsiderbothsingle-sortedportfoliosconstructedbasedona single pricing factor (or fundamental) and double-sorted portfolios constructed on the intersection of two factors (primarily on size factor and on one of the factors listed above). We re-balance these portfolios at the end of June of year t+1, following the methodology of Fama and French (1992). The control variables are the same as those in Sections 3.1. We do not report estimated φi parameters to save space, but they are available upon request. Tables 6 to 9 summarize our j empirical findings. Based on our findings presented in Section 3.1, we have established that policy uncertainty 14We also report results for the following factors in the Appendix: book-to-market ratio, operating profitability, and idiosyncratic volatility. 12

measurespredictreturnsofequityindexesdominatedbylargecorporations. Arethesepredictability results concentrated in large firms only? We first present the estimation results for the familiar size factor single-sorted portfolio returns in Table 5, that point to predictability for the top 30 percent portfolio returns, comprised of the largest companies by market valuation, by EPUF for k ≥ 9. The bottom 30 percent portfolio returns do not display such predictability by EPUF, although EPUUS predicts both large and small company portfolio returns at k ≤ 9 horizons. The difference between the low-30% and the high-30% size portfolios, the SMB returns indicating the size premium, are negatively predictable by EPUF at all horizons (primarily a product of EPUF loadings for small-size portfolio returns that are not statistically different from zero, and loadings for large-size portfolio returns that are). EPUUS has statistically significant and positive-valued loadings for SMB portfolio returns for k = 6 or 9. We next turn to portfolios formed on investments introduced by Hou et al., 2014 and defined as the change in total assets from the fiscal year ending in year t−2 to the fiscal year ending in t−1, dividedbyt−2totalassets. Thisisabroaddefinitionofinvestments. Westudycapital-expenditurebased portfolios separately. High-investment firms have better investment opportunities, greater appetite to acquire assets, and higher expected future cash flows. They may be better positioned to withstand policy uncertainty shocks by adjusting their asset holdings or investment projects. At the same time, some of their investment opportunities are likely to be abroad, making the firms more exposed to changes in foreign EPU. In comparison, low-investment companies may have less room to adjust their assets and projects in response to domestic EPU shocks. On the flip side, low-investmentfirmsarealsolesslikelytohaveinvestmentprojectsabroadandthuslesssusceptible to foreign EPU shocks. We report the results for investment portfolios in Table 6 with the findings for single-sort investment portfolio returns available on the first three columns and the results for double-sorted investment and size (to isolate the effects of firm size) portfolio returns displayed in the last three columns. Estimated slope parameters for EPUF are positive-valued and statistically significant at horizons k ≥ 9 for the top 30% single-sorted investment portfolios and for portfolio returns of large and high-investment firms. EPUF does not predict bottom 30% single-sort and small/low investment portfolio returns. For six months and above horizons, the differences between EPUF loadings of low- and high-investment portfolio returns are negative and statistically significant, since the EPUF loading for low-investment portfolio returns is not statistically different from zero, while that of high-investment portfolio is positive-valued and significant. In contrast, for the same horizons, the differences between EPUUS loadings of low- and high-investment portfolio returns are positive and statistically significant, primarily driven by relatively high loading of low-investment portfolios. In the face of higher EPUF, high-investment firms may need to partially adjust investments 13

and thus give up a portion of their future cash flows. As a result, current equity prices (expected returns) would decrease (increase) more for high-investment portfolios. On the other hand, lowinvestment firms are not affected by higher EPUF, as a result, their current equity prices and expected returns remain little changed. In contrast, when faced with a domestic EPU shock, high-investment firms can adjust their assets and positions to partially smooth the shock, while low-investment firms tend to be constrained in that regard, and as a result, their current equity prices (expected returns) would decrease (increase) more than higher-investment portfolios. We now turn to capital expenditure-based portfolio returns. Greenland et al. (2019) document the negative impact of an increase in economic policy uncertainty on investments and exports for a panel of 14 countries (including the United States).15 In their influential study, Gulen and Ion (2015) convincingly establish a robust, negative relationship between changes in U.S. EPU level and firm-level capital investment, CapEx, with this relationship being stronger for firms with a larger share of irreversible investments and firms the revenues of which are more dependent on government spending. We next examine the relationship between measures of U.S. and foreign EPU and future returns of high- and low-CapEx firms. A priori, we expect to find patterns similar to those for portfolios formed on broader investment measure considered above. We investigate the role of corporate capital and plant investments in driving the relationship between EPUF and equity returns by forming single-sorted portfolios (on CapEx) and doublesorted portfolios (on size and CapEx). The construction of these portfolios closely follows the Fama-French approach. The value of investment on plant and equipment, CapEx, is available from Compustat annual schedules. We aggregate end of June of the year t CapEx values for firms reporting positive values over the period in the cross-section of the Compustat universe. We then normalize the aggregated values by dividing them by the aggregate market capitalization of reporting firms. We form two sets of portfolios. We start with three single-sorted portfolios: below 30th percentile, between 30th and 70th percentile, and above 70th percentile and follow with six double-sorted portfolios formed on size (below and above the median) and CapEx (top and bottom 30% and the middle 40%). We then fit equation (4) using CapEx portfolio returns and report the results in Table 7. EPUUS predicts CapEx portfolio returns for k ≤ 3, and EPUF does so for k > 6. In contrast to our null hypothesis, we find larger loadings for low-CapEx returns compared to high-CapEx. EPUF predictive results are sensitive to size sorting, and are not detected for small, low-CapEx portfolio returns. In contrast to investment portfolio returns discussed earlier, EPUUS only predicts CapEx portfolio returns, both high and low, for k ≤ 3 and does not predict the differences in returns. Statistically significant estimated slope parameters are positive valued. Predictability of portfolio returns sorted on investments and CapEx imply that higher EPU, 15Londono et al. (forthcoming) also document that rising real economic uncertainty negatively affects macroeconomic quantities,suchasindustrialproduction,bothdomesticallyandinternationally,corroboratingtheresultsofGreenland et al. (2019). 14

both foreign and domestic, results in declines (increases) in equity returns (premia) for a sizable– due to their link to physical private corporate investment–economically significant subset of U.S. companies. The predictability operates in different horizons for domestic and foreign EPU measures, but it is present for both EPU measures and across investment and CapEx characteristics. In Section 4 we show that through this EPUF transmits to the real economic quantities through cash flow-related channels, where investment is a prominent variable. But what can we say about financially constrained firms? Do firms’ cash flows indicate sensitivity of equity returns to changes in foreign EPU? When EPUF is high, do investors require higher expected excess returns to hold stocks with low cash flows relative to stocks with high cash flows? Weaddressthisissuebystudyingpredictabilityofportfolioreturnsconstructedoncash-flow factor (CFP), studied by many including Fama and French (1992) and Ball et al. (2016).16 Our empirical findings are available on Table 8. In this exercise, predictability is largely confined to single-sortportfolioreturnsonCFPfactor,andweobservethebynowfamiliarpatternofweakening EPUUS/strengthening EPUF predictive power as k increases, but primarily for the low-CFP portfolio returns. EPUF does not have predictive power for either segment of double-sorted CFP/size portfolio returns shown in the table, and EPUUS’s predictive power dissipates as k increases. In other words, low cash-flow to price (a sign of financial constraints) firms are sensitive and their returns are affected by changes in EPUF. Less constrained firm are not affected by EPUF shocks, and once we sort on size and CFP, this predictability vanishes. It is natural to test whether there is a predictive relationship between EPUF and stock returns for firms with significant foreign activity. Such firms earnings are likely to be more exposed to foreign EPU and, when EPUF is high, investors may require higher expected excess returns to hold stocks of such firms compared to stock returns of firms with low foreign exposure. That said, it is difficult to construct a satisfactory measure of foreign exposure that quantifies the myriad ways that a company might be exposed to risks stemming from abroad. In a first attempt, we use foreign sales as a starting point. CRSP/Compustat reports foreign sales for a notable portion of U.S.-based firms. Similar to CapEx, firms voluntarily disclose their foreign sales information on a quarterly basis, and thus reporting gaps exist. We use the ratio of foreign sales to total sales to normalize the data at firm level, then aggregate the data and build single-sorted and double-sorted portfolios as we did for CapEx earlier. WereportourempiricalfindingsforfittingreturnsonforeignsalesportfoliosinTable9. EPUUS predicts foreign sales portfolio returns for k ≤ 6, while EPUF predicts the returns for k ≥ 9. The pattern of decreasing predictive power for EPUUS and increasing power for EPUF as k increases is observable here too. The predictability of low foreign sales portfolio disappears once we we sort 16Additional analysis based on the related concept of operating profitability of the firm–studied by Ball et al. (2015) and Novy-Marx (2013)–is available in the Appendix. 15

stocks on both foreign sales and size. The following observations may explain this phenomena: A subset of firms may not have a large foreign sales share, but they may have other links (through dependence on foreign-sourced intermediate goods and services, funding, etc.) to the rest of the world that generates this predictability, or there might be a subset of large firms that may not have significant foreign sales, but due to their size, they may be susceptible to policy shifts that affect their suppliers, subsidiaries, etc. Thus once we control for size, this predictability vanishes. These two narratives are not mutually exclusive. Summarizing our findings in this section, we observe that: • Stock returns of firms that accumulate more assets (higher investment factor), have higher CapEx, and have relatively larger foreign sales shares with respect to their total sales are more sensitive to EPUF changes. Among such firms, all else equal, larger firms’ returns are more sensitive to changes in EPUF level. • Firms that have lower cash flow to price ratios are more sensitive to EPUF changes. • In many instances, we observe a distinct pattern of predictability, where EPUUS’s predictive power declines and that of EPUF rises as k increases, with EPUUS predictability generally concentrated in k ≤ 6 and that of EPUF in k ≥ 9. 4 Transmission channels for foreign EPU shocks Thus far, we have established that measures of foreign and U.S. economic policy uncertainty have predictive power for aggregate market index returns, as well as returns for various portfolio sorts and return horizons, in the United States. In the next step, we study the channels for transmission of foreign policy uncertainty shocks to aggregate equity returns, as well as whether responses of macro-financial variables to EPUF shocks confirm such channels. We use local projections (LP), pioneered by Jord`a (2005) and an increasingly popular method for the estimation of responses of macro-financial variables or their components to various shocks, including uncertainty shocks. Among others, Diercks et al. (2024) use LP to recover impulseresponses (IR) of macro-financial variables subject to uncertainty shocks–including EPU shocks–in a closed-economy setting. In Jord`a (2005), the LP model is presented as: P (cid:88) y = α(k)+β(k)ε + γ(k)w +u , (5) t+k x,t t−i (k)t+k i=1 where ε are shocks to variable x and extracted in an intermediate step, w are lagged control x,t t t−i variables, u are i.i.d. errors, and α(k),β(k) and γ(k) are parameters to be estimated. (k)t+k 16

ComparedtotraditionalVAR,LPyieldsmoreflexibleimpulse-responsessinceitimposesweaker assumptionsonthedynamicsofthedata,andstrikesabalancebetweenefficiencyandrobustnessto model misspecification. However, the nonparametric nature of LP causes a notable efficiency cost, and in practice, the LP estimator may suffer from excessive variability. Barnichon and Brownlees (2019) address this issue in their smooth local projections (SLP) method. The SLP fitted model, as presented in Barnichon and Brownlees (2019), follows J J P J (cid:88) (cid:88) (cid:88)(cid:88) y ≈ a B (k)+ b B (k)ε + c B (k)w +u , (6) t+k j j j j x,t ij j t−i (k)t+k j=1 j=1 i=1 j=1 wherey areresponsesofthevariableofinteresttoashock,ε areshockstox asinequation(5), t+k x,t t w are lagged explanatory variables that could include lagged values of the responding variable t−1 (y ), u are i.i.d. shocks, B (k) is a set of B-spline basis functions, a and b are sets of scalar t−i (k)t+k j j j parameters. In the context of this study, ε = ε where ε are EPUF shocks extracted x,t EPUF,t EPUF,t usingthe VARspecificationdescribedin Section2.2and obtained fromfittingequation (3)todata. If the following relationships hold, J J J (cid:88) (cid:88) (cid:88) a B (k) ≈ α(k), b B (k) ≈ β(k), and c B (k) ≈ γ(k), j j j j j j j=1 j=1 j=1 then equation (6), introduced by Barnichon and Brownlees (2019), is approximately the same as equation (5), proposed by Jord`a (2005). We use the procedures shared by Barnichon and Brownlees (2019) to generate smooth impulseresponse functions for the desired variables and recover their respective confidence intervals. 4.1 Transmission of EPUF shocks to equity return components and other financial variables What are the channels of transmission for EPUF shocks to aggregate equity returns? Brogaard and Detzel (2015) show that EPUUS does not have predictive power for aggregate dividend growth rates at various predictive horizons. They further argue that, at least partially based on this result, EPUUS shocks affect equity prices through the discount factor channel and not through the cash-flow channel. Basu and Bundick (2017) introduce a New Keynesian general equilibrium model that produces a decline in policy rates in response to uncertainty shock, matching empirical patterns. Interest rates are among commonly-used predictors of stock returns. Early studies (e.g. Campbell, 1987) suggest that high interest rates generally predict low excess equity returns. Campbell and Ammer (1993) and Bernanke and Kuttner (2005) provide additional discussions of monetary policy, discount rates, and their transmission to equity prices. Thus, both empirical and 17

theoretical evidence support the hypothesis that domestic uncertainty shocks elicit a monetary policy response that affects discount rates, and thus transmits to equity prices. However, it is less likely that changes in economic policy uncertainty abroad consistently and materially affect U.S. monetary policy, policy rates, and discount rates. To the best of our knowledge, suchmechanismhasnotbeendocumentedintheliterature.17 Yet, wehaveshownthroughout thepaperthatEPUF haspredictivepowerforvariousU.S.broadequitymarketindexandportfolio returns. Thus, a plausible alternative channel of transmission could be through cash flows. We investigate this assertion in two steps. We first decompose monthly S&P500 returns into cash flow and discount rate news, following Campbell and Shiller (1988b,a), by fitting excess returns, long-run log price-dividend ratios, and CAPE in a standard VAR system, recovering the news using a Cholesky decomposition. Figure 2 displays the responses of these cash flow and discount rate news to a one-standard deviation shock in EPUF. It is immediately clear that the responses of these quantities to EPUF shocks are muted and statistically not different from zero for the first 5 months. Starting in the sixth month, they both demonstrate statistically significant responses that last until about the 10th month. However, their responses move in opposite directions with cash flow responses first dropping notably in the 6th month, and then rising; and the discount rate response–while statistically significant–is muted, rises a bit around the 6th month, and then declines. Taken together, this figure points to both the reaction of equity prices to changes in EPUF and the timing of predictability to be rooted in the more sizable changes in cash flow news. As a robustness check, we carry out the same exercise using Cenesizoglu and Ibrushi (2023) decomposition of S&P500 returns, based on a different set of variables in the fitted VAR.18 The results have the same general contours seen in Figure 2, but confidence intervals, especially for discount rate news, are a bit wider. That said, the same general patterns–including significant responses at the sixth month–are present. Since the Campbell and Shiller (1988b,a) decomposition of equity returns does not directly yield equity premia embedded in prices, we also study the response of Cieslak and Pang (2021) hedging and common premium news of S&P500 returns to EPUF shocks. We find that the responses of hedging and common premium news to EPUF shocks are generally muted and not statistically different from zero. This finding provides additional support for the claim that foreign economic policy uncertainty affects U.S. equity prices through cash flows, and not through discount rates or premia. This result shows that in addition to the discount rate transmission channel documented by Brogaard and Detzel (2015) for domestic EPU, there is a complementary channel for transmission 17Infact,studiessuchasMiranda-AgrippinoandRey(2020)forcefullyarguetheopposite,claimingthatU.S.monetary policy affects the global cost of capital, leverage levels of global financial intermediaries, the provision of domestic credit globally, international credit flows, and foreign financial conditions. 18We thank the authors for generously sharing their data. 18

of foreign policy uncertainty shocks to asset prices that operates through cash-flow news and over longer horizons. Thus, our findings in an open-economy setting with foreign uncertainty shocks, resemble those documented by Chen, Da and Zhao (2013) who emphasize a significant role for cash flow news (also over longer horizons). Next, we investigate whether responses of select financial and macro economic variables to EPUF shocks lend further support to the cash-flow transmission channel. Figure 3 reports the SLP responses of a select and important set of financial variables to EPUF shocks. The variables are: aggregate, year on year dividend growth rates for S&P500 index (from from Robert Shiller’s website), four-quarter average of stock repurchases to assets ratio (from Capital IQ), year-on-year changes in aggregate commercial and industrial (C&I) loans, 1-Year U.S. Treasury bond yields, federal funds rates, and the broad dollar index (all from FRED data bank at Federal Reserve Bank of St. Louis). The first three variables proxy stocks’ cash flows or reflect on firms’ investment opportunities, while the last three variables are related to discount rates as they represent various measures of interest rates or a variable directly affected by cross-country interest-rate differentials (the value of the U.S. dollar.) Starting on the top left, we note the statistically significant decline in dividend growth rates. Dividend growth rates decline for up to four months after the shock’s arrival. The size of this decline is small, about 0.10 percentage point–we do not expect a foreign uncertainty shock to have an outsize effect on U.S. dividend distributions–but statistically significant. This negative response of dividend growth is consistent with the cash-flow channel for the transmission of foreign EPU shocks to U.S. equities. Thus, as mentioned earlier, this finding complements and extends P´astor and Veronesi (2013) and Brogaard and Detzel (2015). The response of stock repurchases to assets ratio, albeit small at about 0.04 percent, is immediate after the arrival of the shock, persistent, and statistically significant for about 5 months. It implies that a non-negligible number of firms would reduce distributions to shareholders in the form of stock repurchases when faced with a positive EPUF shock. Similarly, the growth of C&I loans (middle row, left-hand side) slows by as much as 0.4 percentage point for up to about 4 months, indicating reluctance of firms to add leverage. Taken together, these three responses point to firms’ precautionary motives, with firms reducing investor payouts and borrowing for new projects in response to unexpected increases in EPUF, potentially depressing future cash flows. We revisit this issue and its implications in Section 4.2. The response of 1-Year Treasury bond yields to an EPUF shock is muted and statistically insignificant for the first 8 months after the shock. It then turns negative between months 6 and 11, by as much as 8 basis points, before turning insignificant again. An EPUF shock does not elicit statistically significant responses from Federal Funds rates or the broad dollar index (the bottom row on Figure 3), in line with the claim that foreign uncertainty shocks do not materially impact U.S. monetary policy. This result is in line with the conclusions of well-established studies such as 19

Miranda-Agrippino and Rey (2020). 4.2 Transmission of EPUF shocks to real variables Thus far, we have proposed a plausible transmission channel for foreign policy uncertainty shocks to U.S. equity returns though cash flows. We now turn to macroeconomic variables and examine whether their responses to EPUF shocks are consistent with those of stock prices and the cash-flow news transmission channel. We focus on the following variables: year-on-year changes in quarterlyadjusted real gross domestic private investment series (from U.S. Bureau of Economic Analysis), annualized changes in aggregate CapEx expenditure,19 the ratio of aggregate CapEx expenditures to total assets (both from Capital IQ), seasonally adjusted, monthly unemployment rate (from U.S. Bureau of Labor Statistics), and finally the log values of total number of employees reported by corporations (from Capital IQ). We consider the responses of variables from national accounts (investmentsandunemploymentrate)andfromaggregatedfirmleveldata(CapExandthenumber of people on payroll). We previously documented that EPUF predicts future stock returns through cash flow channel. Investment (capital) and employment (labor) are likely to have implications for firms’ future cash flows. Therefore, we ask the question of whether EPU affects these variables. The SLP responses of macroeconomic and aggregate corporate series discussed above are available on Figure 4. Using year-on-year real investment growth rates, top left panel, we find statistically significant declines, by as much as 1 percentage point, between 3 and 7 quarters after the EPUF shock. Investment growth rises after 9 quarters. This plot shares many features with investment response plot reported by Basu and Bundick (2017), confirming a similar empirical response of investments to uncertainty shocks and possibly similar general-equilibrium mechanisms. Given that gross private investment is measured at national level, we investigate the sources of these declines in investment growth by looking at corporate capital expenditures. The top right panel on Figure 4 shows the response of aggregate corporate CapEx growth to an EPUF shock. This response is statistically significant and negative valued, ranging between -1 to slightly below -2 percentage points for up to about 7 quarters after the shock. In addition, we find out that in response to an EPUF shock, the ratio of CapEx expenditure to total assets declines by as much as 0.04 percentage point (a notable number) between 2 and 10 quarters after the shock. The declines in this interval are statistically different from zero. AnEPUF shockdoesnotelicitstatisticallysignificantresponsesfromeithernationalunemployment rate (middle row, on the right) or (log) total number of employees reported by the corporate sector (the bottom figure). These observations indicate some reluctance by the corporate sector to adjust their workforce, in the United States, in response to uncertainty shocks abroad. Thus, the 19Similar to other voluntarily furnished corporate data, these series are lumpy, have notable gaps, and volatile. Thus we use aggregated and 4-quarter smoothed series for analysis. 20

transmission channel appear to be through adjustments in capital expenditure and investments, rather than labor. It also provide additional support for the presence of precautionary delays in investment in the presence if uncertainty shocks, documented in this study as well as in Gulen and Ion (2015). These delays, given the corporate sector’s reluctance to adjust labor, are likely to be driven by potentially higher degree of investment irreversibility by firms that delay investment. All in all, our findings in Section 4.1 and here regarding the statistically significant declines in variables associated with future cash flows (dividend growth, share buybacks, and demand for C&I loans, corporate CapEx expenditure and private investment) in response to an adverse EPUF shock are consistent with our finding that EPUF affects future equity returns through cash-flow components. 5 Concluding remarks Financialeconomicliteraturehasestablishedthatmeasuresofdomesticeconomicpolicyuncertainty transmit to financial asset prices and affect a variety of financial decisions in the United States. In particular, influential studies such as P´astor and Veronesi (2012, 2013) and Brogaard and Detzel (2015) have established that domestic EPU has both significant time-series predictive power for U.S. aggregate stock returns and is a priced factor in the cross-section of returns. With the rising interconnectedness of the United States economy and financial markets with the rest of the world, a significant number of American companies face risks that stem from economic policies abroad. Focusing on this salient feature of the U.S. economy, we show that foreign EPU has significant predictive power for market-wide equity index excess returns and for a notable number of factorbased portfolio returns. In particular, we show that firms with higher capital expenditure, foreign sales, investment, as well as low cash flow firms are sensitive to changes in foreign EPU. In addition, we investigate transmission channels of foreign EPU shocks to U.S. equity returns. Studies that focus on the impact of EPU on asset prices suggest that discount rates are the main transmission channel of EPU shocks. Given that notable domestic uncertainty shocks generally result in policy responses that affect investors’ discount rates, this is a plausible narrative. These studies generally do not find a significant role for the cash-flow news channel as a transmission conduit from policy uncertainty shocks to equity prices. In contrast, we find that foreign EPU shocks operate through cash-flow news channel, they do not affect discount rates or equity premia, and that aggregate credit demand and investment outlays respond significantly toa an adverse foreign EPU shock. Taken together, our results extend the existing literature and establish that foreign EPU is an economically significant uncertainty factor. 21

References Aruoba, S.B., Diebold, F.X., Scotti, C., 2009. Real-time measurement of business conditions. Journal of Business & Economic Statistics 27, 417–427. Baker, S.R., Bloom, N., Davis, S.J., 2016. Measuring Economic Policy Uncertainty. The Quarterly Journal of Economics 131, 1593–1636. Bali, T.G., Brown, S.J., Tang, Y., 2017. Iseconomicuncertaintypricedin thecross-sectionof stock returns? Journal of Financial Economics 126, 471 – 489. Ball, R., Gerakos, J., Linnainmaa, J.T., Nikolaev, V., 2016. Accruals, cash flows, and operating profitability in the cross section of stock returns. Journal of Financial Economics 121, 28–45. Ball, R., Gerakos, J., Linnainmaa, J.T., Nikolaev, V.V., 2015. Deflating profitability. Journal of Financial Economics 117, 225–248. Barnichon, R., Brownlees, C., 2019. Impulse Response Estimation by Smooth Local Projections. The Review of Economics and Statistics 101, 522–530. Basu, S., Bundick, B., 2017. Uncertainty shocks in a model of effective demand. Econometrica 85, 937–958. Bernanke, B.S., Kuttner, K.N., 2005. Whatexplainsthestockmarket’sreactiontoFederalReserve policy? The Journal of Finance 60, 1221–1257. Bianchi, F., Lettau, M., Ludvigson, S.C., 2022. Monetary policy and asset valuation. The Journal of Finance 77, 967–1017. Bloom, N., 2009. The impact of uncertainty shocks. Econometrica 77, 623–685. Bonaime, A., Gulen, H., Ion, M., 2018. Does policy uncertainty affect mergers and acquisitions? Journal of Financial Economics 129, 531 – 558. Boutchkova, M., Doshi, H., Durnev, A., Molchanov, A., 2012. Precarious Politics and Return Volatility. The Review of Financial Studies 25, 1111–1154. Brogaard, J., Dai, L., Ngo, P.T.H., Zhang, B., 2020. Global Political Uncertainty and Asset Prices. The Review of Financial Studies 33, 1737–1780. Brogaard, J., Detzel, A., 2015. The asset-pricing implications of government economic policy uncertainty. Management Science 61, 3–18. Caldara, D., Iacoviello, M., 2022. Measuring geopolitical risk. American Economic Review 112, 1194–1225. Caldara, D., Iacoviello, M., Molligo, P., Prestipino, A., Raffo, A., 2020. The economic effects of trade policy uncertainty. Journal of Monetary Economics 109, 38–59. Campbell, J.Y., 1987. Stock returns and the term structure. Journal of Financial Economics 18, 373–399. Campbell, J.Y., Ammer, J., 1993. What moves the stock and bond markets? a variance decomposition for long-term asset returns. The Journal of Finance 48, 3–37. 22

Campbell, J.Y., Shiller, R.J., 1988a. The dividend-price ratio and expectations of future dividends and discount factors. The Review of Financial Studies 1, 195–228. Campbell, J.Y., Shiller, R.J., 1988b. Stock prices, earnings, and expected dividends. The Journal of Finance 43, 661–676. Candelon, B., Ferrara, L., Jo¨ets, M., 2021. Global financial interconnectedness: A non-linear assessment of the uncertainty channel. Applied Economics 53, 2865–2887. Cenesizoglu, T., Ibrushi, D., 2023. Time Variation in Cash Flows and Discount Rates. Journal of Financial Econometrics 21, 1557–1589. Chan, L.K.C., Jegadeesh, N., Lakonishok, J., 1996. Momentum strategies. The Journal of Finance 51, 1681–1713. Chen, L., Da, Z., Zhao, X., 2013. What drives stock price movements? The Review of Financial Studies 26, 841–876. Cieslak, A., Pang, H., 2021. Common shocks in stocks and bonds. Journal of Financial Economics 142, 880–904. Cochrane, J.H., 2011. Presidential address: Discount rates. The Journal of Finance 66, 1047–1108. Demirer, M., Diebold, F.X., Liu, L., Yilmaz, K., 2018. Estimating global bank network connectedness. Journal of Applied Econometrics 33, 1–15. Diercks, A.M., Hsu, A., Tamoni, A., 2024. When it rains it pours: Cascading uncertainty shocks. Journal of Political Economy 132, 694–720. Fama,E.F.,French,K.R.,1992. Thecross-sectionofexpectedstockreturns. theJournalofFinance 47, 427–465. Golez, B., Koudijs, P., 2018. Fourcenturiesofreturnpredictability. JournalofFinancialEconomics 127, 248–263. Greenland, A., Ion, M., Lopresti, J., 2019. Exports, investment and policy uncertainty. Canadian Journal of Economics/Revue canadienne d’´economique 52, 1248–1288. Gulen, H., Ion, M., 2015. Policy Uncertainty and Corporate Investment. The Review of Financial Studies 29, 523–564. Harvey, C.R., Liu, Y., Zhu, H., 2015. ... and the Cross-Section of Expected Returns. The Review of Financial Studies 29, 5–68. Hassan,T.A.,Schreger,J.,Schwedeler,M.,Tahoun,A.,2024. SourcesandTransmissionofCountry Risk. The Review of Economic Studies 91, 2307–2346. Heber, G., Lunde, A., Shephard, N., Sheppard, K.K., 2009. Oxford-Man Institute’s realized library. Technical Report. Oxford-Man Institute, University of Oxford. URL: https://web.archive.org/web/20220903214048/https://realized.oxford-man.ox.ac.uk/. Hong,H.,Stein,J.C.,1999. Aunifiedtheoryofunderreaction,momentumtrading,andoverreaction in asset markets. The Journal of Finance 54, 2143–2184. 23

Hou, K., Xue, C., Zhang, L., 2014. Digesting Anomalies: An Investment Approach. The Review of Financial Studies 28, 650–705. Husted,L.,Rogers,J.,Sun,B.,2020. Monetarypolicyuncertainty. JournalofMonetaryEconomics 115, 20–36. Jegadeesh, N., Titman, S., 1993. Returns to buying winners and selling losers: Implications for stock market efficiency. The Journal of Finance 48, 65–91. Jord`a, O., 2005. Estimation and inference of impulse responses by local projections. American Economic Review 95, 161–182. Jurado, K., Ludvigson, S.C., Ng, S., 2015. Measuring uncertainty. American Economic Review 105, 1177–1216. Kaviani, M.S., Kryzanowski, L., Maleki, H., Savor, P., 2020. Policy uncertainty and corporate credit spreads. Journal of Financial Economics 138, 838–865. Kelly, B., P´astor, L., Veronesi, P., 2016. The price of political uncertainty: Theory and evidence from the option market. The Journal of Finance 71, 2417–2480. Kim, O.S., 2019. Does political uncertainty increase external financing costs? measuring the electoral premium in syndicated lending. Journal of Financial and Quantitative Analysis 54, 2141–2178. Kl¨oßner, S., Sekkel, R., 2014. International spillovers of policy uncertainty. Economics Letters 124, 508–512. Londono, J.M., Ma, S., Wilson, B.A., forthcoming. The global transmission of real economic uncertainty. Journal of Money, Credit & Banking n/a. Ludvigson, S.C., Ma, S., Ng, S., 2021. Uncertainty and Business Cycles: Exogenous Impulse or Endogenous Response? American Economic Journal: Macroeconomics 13, 369–410. Merton, R.C., 1973. An intertemporal capital asset pricing model. Econometrica 41, 867–887. Miranda-Agrippino, S., Rey, H., 2020. U.S. Monetary Policy and the Global Financial Cycle. The Review of Economic Studies 87, 2754–2776. Newey, W.K., West, K.D., 1987. A Simple, Positive Semi-definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix. Econometrica 55, 703–708. Novy-Marx,R.,2013. Theothersideofvalue: Thegrossprofitabilitypremium. JournalofFinancial Economics 108, 1 – 28. P´astor, L., Veronesi, P., 2012. Uncertainty about government policy and stock prices. the Journal of Finance 67, 1219–1264. P´astor,L.,Veronesi,P.,2013. Politicaluncertaintyandriskpremia. JournalofFinancialEconomics 110, 520 – 545. Pindyck, R.S., Rotemberg, J.J., 1993. The comovement of stock prices. The Quarterly Journal of Economics 108, 1073–1104. 24

Ruiz, E., Pascual, L., 2002. Bootstrapping Financial Time Series. Journal of Economic Surveys 16, 271–300. Shin, M., Zhang, B., Zhong, M., Lee, D.J., 2018. Measuring international uncertainty: The case of Korea. Economics Letters 162, 22–26. 25

Table 1: Growth, Volatility, and Economic Policy Uncertainty Measure Correlations GDP Growth (Quarterly, SA) U.S. Euro Area Germany U.K. Japan U.S. 1.0000 0.9029 0.8339 0.9302 0.7369 Euro Area 1.0000 0.9248 0.9505 0.7727 Germany 1.0000 0.8842 0.7787 U.K. 1.0000 0.7592 Japan 1.0000 Realized Volatility U.S. Euro Area Germany U.K. Japan U.S. 1.0000 0.7679 0.6942 0.7252 0.6305 Euro Area 1.0000 0.7789 0.8372 0.5968 Germany 1.0000 0.8238 0.5042 U.K. 1.0000 0.5035 Japan 1.0000 Option-Implied Volatility U.S. Euro Area Germany U.K. Japan U.S. 1.0000 0.8896 0.9277 0.9394 0.7116 Euro Area 1.0000 0.9681 0.9501 0.7403 Germany 1.0000 0.9306 0.7276 U.K. 1.0000 0.6466 Japan 1.0000 EPU U.S. Euro Area Germany U.K. Japan Global U.S. 1.0000 0.6715 0.7332 0.4332 0.4637 0.8050 Euro Area 1.0000 0.8764 0.8734 0.4058 0.8502 Germany 1.0000 0.6564 0.3969 0.8414 U.K. 1.0000 0.2753 0.7096 Japan 1.0000 0.4302 Global 1.0000 This table reports correlations between GDP growth rates for the United States, euro area, Germany, the United Kingdom,andJapan;correlationsbetweenrealizedandoption-impliedvolatilitymeasuresfortheUnitedStates,euro area, Germany, the United Kingdom, and Japan; and correlations between EPU indexes for the United States, euro area, Germany, the United Kingdom, Japan, and the global EPU. 26

Table 2: Summary statistics of demeaned EPU measures Panel A: EPU Summary Statistics EPUUS EPUG EPUF Std Dev (%) 15.66 23.34 13.85 Skewness 1.60 1.56 1.07 Kurtosis 6.67 5.57 4.26 AR(1) 0.83 0.91 0.89 AR(5) 0.64 0.80 0.82 AR(10) 0.48 0.68 0.71 Panel B: EPU Correlations EPUUS EPUG EPUF EPUUS 1.00 0.80 −1.41e−16 EPUG 1.00 0.59 EPUF 1.00 The top panel in this table reports summary statistics of demeaned U.S., global, and foreign EPU measures. We report sample standard deviations, skewness, and kurtosis values. AR(p) values reports pth autocorrelation values. These autocorrelation values are statistically different from zero at the 95% confidence level or better. The bottom panel reports sample correlations between demeaned U.S., global, and foreign EPU measures. 27

selbairaV laicnaniF-orcaM dna serusaeM UPE :3 elbaT selbairavngierofdnalabolG:BlenaP selbairavcfiiceps-.S.U:AlenaP )%(2R .jdA IDB )D/P(gol daerpS XXOTSV IANFC )D/P(gol daerpS XIV 77.93 ***42.0- **04.1- **94.0- ***74.0 31.0- ***50.2- **82.0 ***85.0 .pmetnoC SUUPE )55.3-( )91.2-( )44.2-( )28.3( )18.0-( )63.5-( )46.2( )81.4( 28.93 ***22.0- **05.1- **05.0- ***64.0 *13.0- ***20.2- **72.0 ***85.0 1=gal )32.3-( )95.2-( )75.2-( )75.3( )67.1-( )31.5-( )55.2( )39.3( 80.82 **81.0- ***85.1- **94.0- ***63.0 **72.0- ***46.1- **62.0 **24.0 3=gal )45.2-( )60.3-( )76.2-( )22.3( )72.2-( )16.3-( )93.2( )56.2( 35.81 *31.0- **55.1- **05.0- *02.0 **72.0- **63.1- **42.0 12.0 6=gal )56.1-( )82.3-( )89.2-( )17.1( )23.2-( )00.3-( )80.2( )65.1( 59.31 50.0- ***66.1- **93.0- 41.0 31.0- **25.1- 51.0 *42.0 9=gal )07.0-( )84.3-( )34.2-( )01.1( )30.1( )68.2-( )03.1( )07.1( 17.44 ***93.0- 84.0- ***37.0- 41.0 60.0- ***26.2- 21.0 **73.0 .pmetnoC GUPE )46.5-( )57.0-( )29.3-( )21.1( )04.0-( )94.6-( )60.1( )20.3( 16.64 ***83.0- 85.0- ***47.0- 51.0 31.0- ***17.2- 21.0 ***24.0 1=gal )64.5-( )79.0-( )80.4-( )90.1( )48.0( )46.6-( )30.1( )51.3( 20.93 ***63.0- 26.0- ***27.0- 40.0 80.0- ***54.2- 11.0 *52.0 3=gal )78.4-( )90.1-( )12.4-( )23.0( )57.0-( )87.5-( )19.0( )37.1( 51.43 ***33.0- 85.0- ***76.0- 80.0- 90.0- ***82.2- 80.0 70.0 6=gal )34.4-( )50.1-( )63.4-( )16.0-( )78.0-( )34.5-( )46.0( )25.0( 46.92 ***03.0- 86.0- ***65.0- 21.0- 40.0- ***53.2- 20.0 60.0 9=gal )31.4-( )52.1-( )58.3-( )09.0-( )74.0-( )77.4-( )61.0( )63.0( 68.54 ***31.0- **44.0 ***32.0- ***61.0- 30.0 ***66.0- 70.0- *60.0- .pmetnoC FUPE )06.5-( )47.2( )23.5-( )43.3-( )87.0( )05.3-( )95.1-( )37.1-( 50.64 ***31.0- **34.0 ***22.0- **51.0- **80.0 ***37.0- 70.0- 30.0- 1=gal )45.5-( )56.2( )91.5-( )11.3-( )44.2( )48.3-( )65.1-( )68.0-( 61.84 ***41.0- **44.0 ***12.0- ***61.0- **90.0 ***67.0- 70.0- 60.0- 3=gal )72.5-( )75.2( )90.5-( )63.3-( )37.2( )81.4-( )06.1-( )44.1-( 72.74 ***51.0- **44.0 ***81.0- ***61.0- **80.0 ***97.0- *70.0- *60.0- 6=gal )83.5-( )63.2( )81.4-( )54.3-( )12.2( )76.4-( )66.1-( )38.1-( 03.74 ***71.0- **44.0 ***71.0- ***61.0- 40.0 ***67.0- 70.0- **01.0- 9=gal )02.6-( )72.2( )46.3-( )64.3-( )87.0( )39.4-( )64.1-( )35.2-( UPEngierofehtdna)6102(.laterekaBfoserusaemUPElabolgdna.S.UdezidradnatsgnissergermorfdeniatbosetamitseretemarapepolsstroperelbatsihT noitpoeraselbairavehT .selbairavlaicnanfidnacimonoceorcamnaeporuEdna.S.Ufoseulavdeggaldnasuoenaropmetnocno,1.2noitceSnidebircsederusaem 005P&S rof oitar ecirp-dnedivid gol s’rellihS treboR ,)daerpS( sdaerps ngierevos namreG dna .S.U raey 1-revo-01 ,)XXOTSV dna XIV( seitilitalov deilpmi ytivitcalanoitandeFogacihC,)PI∆(sexedninoitcudorplairtsudnifosetarhtworgaera-oruedna.S.U,xedni006XXOTSORUEroftnelaviuqeehtdnaxedni ni srorre dradnats tnetsisnoc-CAH )7891( tseW dna yeweN no desab scitsitats t-tnedutS troper eW .)IDB( xednI yrD citlaB fo seulav gol dna ,)IANFC( xedni stropernmuloctsalehT .ylevitcepser,slevelecnedfinoctnecrep1dna,5,01ta0= βtahtsisehtopyhllunehtfonoitcejertneserper***dna,**,* .sisehtnerap i .s2R detsujda 28

Table 4: Index-level predictability Lags CRSP S&P500 NASDAQ Russell 2000 EPUF 0.46 0.54 4.28 -0.16 (0.27) (0.33) (1.65) (-0.06) k = 3 EPUUS 4.79*** 4.46*** 3.36 5.30** (2.83) (2.77) (1.65) (2.35) Adj. R2(%) 19.62 19.04 15.55 10.96 EPUF 2.73 2.43 11.01*** 1.26 (0.97) (0.91) (3.01) (0.33) k = 6 EPUUS 7.00** 6.60*** 3.59 8.83** (2.26) (2.65) (1.10) (2.24) Adj. R2(%) 25.87 26.61 26.93 21.01 EPUF 6.28** 5.72** 18.00*** 3.99 (2.19) (2.13) (3.52) (1.64) k = 9 EPUUS 8.83* 8.20** 3.25 10.72* (1.94) (1.90) (0.69) (1.81) Adj. R2(%) 32.85 32.76 29.73 25.23 EPUF 11.36*** 10.57*** 25.25*** 7.68* (3.43) (3.47) (3.92) (1.68) k = 12 EPUUS 7.27 6.80* 1.83 9.44 (1.55) (1.51) (0.38) (1.59) Adj. R2(%) 39.54 39.33 33.75 31.38 This table reports slope parameters for EPUUS and EPUFfrom fitting equation (4) to data. We use excess returns t t from CRSP value-weighted portfolio, S&P500, NASDAQ, and Russell 2000 indexes over the 1-month T-Bill rate. Our control variables are NBER recessions, 10-Year minus 1-year Treasury spread, BBB - AAA corporate spread, (the log value of) Shiller’s aggregate cyclically adjusted PE ratio (CAPE), changes in VIX, Chicago Fed’s CFNAI, growthrateofindustrialproductionindex,Fama-Frenchsize(SMB),market-to-bookratio(HML),momentum,and long-term reversal factors. Value of k ranges between 1 to 12 months ahead. We report Student-t statistics, based on Newey and West (1987) HAC-consistent standard errors in parentheses. *, **, and *** represent rejection of the null hypothesis that β =0 at 10, 5, and 1 percent confidence levels, respectively. i 29

Table 5: Predictability of portfolio returns formed on size Single-sort size High 30 Low 30 SMB EPUF 0.34 -2.47 -2.81** (0.20) (-1.02) (-2.11) k = 3 EPUUS 4.65*** 7.26*** 2.61 (2.66) (2.93) (1.58) Adj. R2 (%) 14.80 13.73 17.81 EPUF 2.73 -1.59 -4.32* (0.93) (-0.37) (-1.77) k = 6 EPUUS 6.49** 10.33** 3.84* (2.09) (2.30) (1.69) Adj. R2 (%) 21.17 11.30 15.35 EPUF 6.21* -0.26 -6.47* (1.91) (-0.06) (-1.94) k = 9 EPUUS 8.09* 14.71** 6.62** (1.74) (2.06) (2.00) Adj. R2 (%) 25.08 13.47 19.41 EPUF 10.87*** 3.47 -7.40* (2.95) (0.62) (-1.83) k = 12 EPUUS 6.40 12.37 5.98 (1.29) (1.55) (1.36) Adj. R2 (%) 30.85 17.99 22.07 ThistablereportsslopeparametersforU.S.andforeignEPUmeasuresinequation(4),wherethedependentvariables are returns for portfolios formed on Fama and French size (market value) factor. Our control variables are NBER recessions,10minus1-yearU.S.Treasuryspreads,BBB-AAAcorporatespread,(thelogvalueof)Shiller’saggregate cyclically adjusted PE ratio (CAPE), changes in VIX, Chicago Fed’s CFNAI, growth rates of the U.S. industrial productionindex,momentum,andlong-termreversalfactors. Student-tstatistics,basedonNeweyandWest(1987) HAC-consistent standard errors are in parentheses. *, **, and *** represent statistical significance at 10, 5, and 1% confidence levels, respectively. 30

Table 6: Predictability of portfolio returns formed on investment and size/investment Single-sort investment Double-sort size and investment High 30 Low 30 LMH Big/Hi Inv. Small/Lo Inv. SL-BH Inv EPUF 0.49 -0.45 -0.95 0.61 -2.91 -3.52** (0.25) (-0.28) (-0.83) (0.31) (-1.20) (-2.55) k = 3 EPUUS 4.38** 5.02*** 0.64 4.26** 6.28*** 2.02 (2.28) (2.92) (0.94) (2.22) (2.67) (1.42) Adj. R2 (%) 19.53 17.88 22.79 19.14 26.45 37.48 EPUF 4.47 -0.06 -4.53** 4.70 -3.27 -7.97*** (1.42) (-0.02) (-2.10) (1.50) (-0.70) (-2.87) k = 6 EPUUS 5.27 7.76** 2.49* 5.01 9.42** 4.41* (1.50) (2.59) (1.88) (1.44) (2.07) (1.96) Adj. R2 (%) 25.66 20.06 22.14 26.28 18.60 28.81 EPUF 9.48** 2.28 -7.20** 9.82*** -2.73 -12.55*** (2.69) (0.83) (-2.42) (2.76) (-0.58) (-3.39) k = 9 EPUUS 6.21 10.15** 3.94** 5.74 13.67* 7.92** (1.21) (2.35) (2.13) (1.13) (1.94) (2.40) Adj. R2 (%) 30.65 26.60 20.96 30.93 20.88 31.21 EPUF 16.16*** 6.24** -9.92*** 16.62*** 0.63 -15.99*** (3.76) (2.01) (-2.76) (3.82) (0.12) (-3.77) k = 12 EPUUS 3.28 9.08** 5.80** 2.73 10.85 8.11* (0.62) (2.07) (2.45) (0.52) (1.44) (1.82) Adj. R2 (%) 35.72 31.03 24.01 35.49 23.94 33.98 ThistablereportsslopeparametersforU.S.andforeignEPUmeasuresinequation(4),wherethedependentvariables arereturnsforportfoliosformedontheinvestmentfactorofHouetal.(2014),andsizeandinvestmentfactors. Our control variables are NBER recessions, 10 minus 1-year U.S. Treasury spreads, BBB - AAA corporate spread, (the logvalueof)Shiller’saggregatecyclicallyadjustedPEratio(CAPE),changesinVIX,ChicagoFed’sCFNAI,growth rates of the U.S. industrial production index, Fama and French HML, excess market returns, momentum, and longterm reversal factors. Student-t statistics, based on Newey and West (1987) HAC-consistent standard errors are in parentheses. *, **, and *** represent statistical significance at 10, 5, and 1% confidence levels, respectively. 31

Table 7: Predictability of portfolio returns formed on Size/Capital Expenditures Single-sort CapEx Double-sort size and CapEx High 30 Low 30 LMH Big/Hi CapEx Small/Lo CapEx SL-BH EPUF -0.35 -0.70 -0.36 -0.31 -1.65 -1.34 (-0.22) (-0.27) (-0.27) (-0.19) (-0.72) (1.22) k = 3 EPUUS 4.14** 5.44** 1.29 4.14** 5.37*** 1.23 (2.31) (2.45) (1.43) (2.31) (2.61) (1.15) Adj. R2 (%) 19.44 20.00 18.13 19.56 22.91 18.16 EPUF 1.20 2.68 1.48 1.29 -0.41 -1.70 (0.50) (0.66) (0.69) (0.97) (-0.11) (0.90) k = 6 EPUUS 4.44 4.86 0.42 4.43 5.37 0.94 (1.42) (1.17) (0.26) (1.43) (1.37) (0.57) Adj. R2 (%) 27.36 17.74 13.10 27.56 14.79 14.35 EPUF 4.41* 7.94* 3.53 4.53* 2.64 -1.88 (1.88) (1.75) (1.30) (1.93) (1.09) (0.76) k = 9 EPUUS -0.12 -0.92 -0.80 -0.11 1.04 1.14 (-0.04) (-0.19) (-0.30) (-0.03) (0.21) (0.37) Adj. R2 (%) 35.21 25.52 18.90 35.33 21.67 17.64 EPUF 8.19*** 13.02** 4.83 8.33*** 5.74 -2.58 (2.96) (2.38) (1.42) (3.02) (1.25) (0.83) k = 12 EPUUS -1.68 -2.94 -1.26 -1.65 0.13 1.78 (-0.48) (-0.53) (-0.35) (-0.47) (0.02) (0.42) Adj. R2 (%) 38.46 28.18 16.09 38.51 24.46 15.94 ThistablereportsslopeparametersforforeignandU.S.EPUmeasuresinequation(4)wherethedependentvariables arereturnsforportfoliosformedoncapitalexpendituretomarketcapitalizationratio(CapEx)andsizeandCapEx. Control variables are SMB, HML, momentum, long-term reversal, NBER recessions, 10-Year over 2-year Treasury spreads, BBB - AAA corporate spreads, (the log value of) Shiller’s aggregate cyclically adjusted PE ratio (CAPE), changesinVIX,ChicagoFed’sCFNAI,andchangesintheIPindex. Student-tstatistics,basedonNeweyandWest (1987) HAC-consistent standard errors are in parentheses. *, **, and *** represent statistical significance at 10, 5, and 1% confidence levels, respectively. 32

Table 8: Predictability of portfolio returns formed on cash flow to price and size/cash flow to price Single-sort CFP Double-sort size and CFP High 30 Low 30 HML Big/Hi CFP Small/Lo CFP BH-SL EPUF -1.95 0.32 -2.28* -1.78 -1.64 -0.14 (-1.07) (0.18) (-1.78) (-0.98) (-0.77) (-0.12) k = 3 EPUUS 3.74** 4.71*** -0.98 3.63** 5.74*** -2.11** (2.24) (2.62) (-1.16) (2.18) (2.81) (-2.41) Adj. R2 (%) 19.23 19.06 29.31 18.86 24.01 31.74 EPUF -4.24 3.48 -7.73*** -4.02 -0.52 -3.50* (-1.18) (1.26) (-2.77) (-1.12) (-0.13) (-1.86) k = 6 EPUUS 6.40* 6.08** 0.32 6.26* 8.64** -2.38* (1.84) (2.07) (0.21) (1.81) (2.16) (-1.68) Adj. R2 (%) 18.21 27.78 29.99 19.64 17.34 26.27 EPUF -5.45 7.83*** -13.28*** -5.16 1.21 -6.37** (-1.32) (2.75) (-3.28) (-1.25) (0.29) (-2.52) k = 9 EPUUS 8.60 7.47* 1.12 8.29 12.33** -4.04** (1.54) (1.88) (0.44) (1.48) (2.00) (-2.29) Adj. R2 (%) 19.30 33.94 26.76 21.31 21.14 28.28 EPUF -4.52 13.57*** -18.09*** -4.21 5.18 -9.38*** (-0.99) (4.06) (-3.83) (-0.91) (1.11) (-2.74) k = 12 EPUUS 6.44 6.14 0.31 5.99 9.69 -3.70 (1.02) (1.48) (0.08) (0.94) (1.47) (-1.36) Adj. R2 (%) 22.82 36.76 28.68 25.21 24.98 24.37 ThistablereportsslopeparametersforU.S.andforeignEPUmeasuresinequation(4),wherethedependentvariables arereturnsforportfoliosformedoncashflowtopriceratioandsizeandcashflowtopriceratio. Ourcontrolvariables are NBER recessions, 10 minus 1-year U.S. Treasury spreads, BBB - AAA corporate spread, (the log value of) Shiller’saggregatecyclicallyadjustedPEratio(CAPE),changesinVIX,ChicagoFed’sCFNAI,growthratesofthe U.S.industrialproductionindex,FamaandFrenchHML,excessmarketreturns,momentum,andlong-termreversal factors. Student-tstatistics,basedonNeweyandWest(1987)HAC-consistentstandarderrorsareinparentheses. *, **, and *** represent statistical significance at 10, 5, and 1% confidence levels, respectively. 33

Table 9: Predictability of portfolio returns formed on Size/Foreign Sales Single-sort foreign sales Double-sort size and foreign sales High 30 Low 30 LMH Big/Hi FS Small/Lo FS SL-BH EPUF -0.13 -0.78 -0.65 -0.03 -2.15 -2.12 (-0.08) (-0.45) (-0.69) (-0.02) (-0.98) (-2.05) k = 3 EPUUS 5.04** 6.08*** 1.04 5.03** 6.04** 1.01 (2.55) (3.36) (1.46) (2.56) (2.54) (0.98) Adj. R2 (%) 18.72 21.48 5.46 18.77 20.92 27.22 EPUF 2.20 0.57 -1.63 2.39 -1.59 -3.97 (0.90) (0.21) (-1.01) (0.98) (-0.47) (-2.25) k = 6 EPUUS 5.53* 7.81** 2.29 5.54* 6.91 1.37 (1.40) (2.58) (1.54) (1.70) (1.60) (0.77) Adj. R2 (%) 26.13 26.14 11.13 26.85 15.52 21.32 EPUF 6.02*** 4.06 -1.77 6.27*** 1.14 -5.13 (2.67) (1.51) (-1.07) (2.80) (0.36) (-2.25) k = 9 EPUUS 0.88 5.86 4.98** 0.95 2.15 1.20 (0.30) (1.53) (2.37) (0.33) (0.49) (0.39) Adj. R2 (%) 36.61 31.32 17.17 37.23 26.16 26.92 EPUF 10.20*** 7.78** -2.20 10.53*** 4.49 -6.04 (3.83) (2.52) (-1.19) (3.97) (1.18) (-2.20) k = 12 EPUUS -0.88 6.27 7.14*** -0.78 0.58 1.37 (-0.28) (1.52) (2.78) (-0.25) (0.12) (0.36) Adj. R2 (%) 39.40 32.06 25.26 39.84 28.01 30.20 This table reports the estimated slope parameters for foreign and U.S. EPU in equation (4) where the dependent variablesarereturnsforportfoliosformedonforeignsalestototalsalesratioandforeignsalestototalsalesratio/size. Control variables are SMB, HML, momentum, long-term reversal, NBER recessions, 10-Year over 2-year Treasury spreads, BBB - AAA corporate spreads, (the log value of) Shiller’s aggregate cyclically adjusted PE ratio (CAPE), changesinVIX,ChicagoFed’sCFNAI,andchangesintheIPindex. Student-tstatistics,basedonNeweyandWest (1987) HAC-consistent standard errors are in parentheses. *, **, and *** represent statistical significance at 10, 5, and 1% confidence levels, respectively. 34

nosirapmoC UPE ngieroF dna ,labolG ,.S.U :1 erugiF ot 7991 yraunaJ morf skcohs UPE ngierof detcartxe dna ,erusaem UPE ngierof detcurtsnoc eht ,sexedni .S.U dna labolg )6102( .la te rekaB stolp erugfi sihT .skcohs FUPE detcartxe dna FUPE fo margotsih eht syalpsid lenap mottob ehT .snoissecer REBN era saera dedahS .4202 hcraM 35

Figure 2: Responses of cash flow and discount rate news to EPUF shocks Thefigureplotstheresponsesofcashflow(blue)anddiscountrate(red)news,extractedfromS&P500indexreturns using Campbell and Shiller (1988a,b) method, to EPUF shocks, based on Barnichon and Brownlees (2019) smooth local projection (SLP) method. Shaded areas represent their respective 95% confidence bands. 36

Figure 3: EPUF shocks and financial variable responses This figure plots responses of cash flow (blue) and discount rate (red) news, extracted from S&P500 index returns usingCampbellandShiller(1988b,a)method,toEPUF shocks. TheresponsesarebasedonBarnichonandBrownlees (2019) smooth local projection (SLP) method and shaded areas represent their respective 95% confidence bands. 37

Figure 4: EPUF shocks and macroeconomic variable responses This figure plots responses of cash flow (blue) and discount rate (red) news, extracted from S&P500 index returns usingCampbellandShiller(1988b,a)method,toEPUF shocks. TheresponsesarebasedonBarnichonandBrownlees (2019) smooth local projection (SLP) method and shaded areas represent their respective 95% confidence bands. 38