The Macroeconomic Impact of Financial and Uncertainty Shocks

K.7 The Macroeconomic Impact of Financial and Uncertainty Shocks Caldara, Dario, Cristina Fuentes-Albero, Simon Gilchrist and Egon Zakrajsek Please cite paper as: Caldara, Dario, Cristina Fuentes-Albero, Simon Gilchrist and Egon Zakrajsek (2016). The Macroeconomic Impact of Financial and Uncertainty Shocks. International Finance Discussion Papers 1166. http://dx.doi.org/10.17016/IFDP.2016.1166 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1166 May 2016

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1166 May 2016 The Macroeconomic Impact of Financial and Uncertainty Shocks Dario Caldara, Cristina Fuentes-Albero, Simon Gilchrist and Egon Zakrajˇsek NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. 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.

The Macroeconomic Impact of Financial and Uncertainty Shocks Dario Caldara∗ Cristina Fuentes-Albero† Simon Gilchrist‡ Egon Zakrajˇsek§ February 2016 Forthcoming in the European Economic Review Abstract TheextraordinaryeventssurroundingtheGreatRecessionhavecastaconsiderabledoubton thetraditionalsourcesofmacroeconomicinstability. Intheirplace, economistshavesingledout financial and uncertainty shocks as potentially important drivers of economic fluctuations. Empiricallydistinguishingbetweenthesetwotypesofshocks,however,isdifficultbecauseincreases ineconomicuncertaintyarestronglyassociatedwithawideningofcreditspreads, anindication of a tightening in financial conditions. This paper uses the penalty function approach within theSVARframeworktoexaminetheinteraction betweenfinancial conditionsandeconomic uncertainty and to trace out the impact of these two types of shocks on the economy. The results indicate that (1) financial shocks have a significant adverse effect on economic outcomes and that such shocks were an important source of cyclical fluctuations since the mid-1980; (2) uncertainty shocks, especially those implied by uncertainty proxies that do not rely on financial asset prices, are also an important source of macroeconomic disturbances; and (3) uncertainty shocks have an especially negative economic impact in situations where they elicit a concomitant tightening of financial conditions. Evidence suggests that the Great Recession was likely an acute manifestation of the toxic interaction between uncertainty and financial shocks. JEL Classification: E32, E37, E44 Keywords: time-varying uncertainty; financial conditions; structural vector autoregression; optimization-based identification We are especially grateful to Johannes Pfeifer and Jim Stock for detailed comments and suggestions. We also thank Nicholas Bloom, Robert Kollman, Eric Leeper, Tommaso Monacelli, Werner Roeger, and participants at numerous seminars and conferences for helpful comments. George Gu, Edward Kim, Shaily Patel, and Rebecca Zhang provided outstanding research assistance at various stages of this project. All errors and omissions are our own responsibility. The views expressed in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or of anyone else associated with the Federal Reserve System. ∗Federal Reserve Board of Governors. Email: dario.caldara@frb.gov †Federal Reserve Board of Governors. Email: cristina.fuentes-albero@frb.gov ‡Boston University and NBER. Email: sgilchri@bu.edu §Federal Reserve Board of Governors. Email: egon.zakrajsek@frb.gov

1 Introduction The acute turmoil that swept through global financial markets during the 2008–09 financial crisis and the depth and duration of the associated economic downturn, both in the United States and abroad, have cast a considerable doubt on the traditional sources of business cycle fluctuations. In response, recent theoretical and empirical research aimed at understanding these extraordinary eventshaspointedtofinancialanduncertaintyshocks—ortheircombination—asalternativedrivers of economic fluctuations (Bloom, 2009; Bloom et al., 2012; Arellano et al., 2012; Christiano et al., 2014; Gilchrist et al., 2014). Empirically distinguishing between these two types of shocks, however, is difficult because increases in financial market volatility—a widely used proxy for macroeconomic uncertainty— are frequently associated with significant increases in credit spreads. A stark illustration of this empiricalchallengeisdepictedinFigure1,whichshowstherelationshipbetweenthedailychangein the option-implied volatility on the S&P 500 stock futures index (the VIX) and the daily change in the speculative-grade CDX index during the recent financial crisis.1 Clearly evident is the fact that episodesofacutefinancialdistressareassociatedwithspikesinassetpricevolatility. Indeed,intheir comprehensiveempiricalanatomyoftheGreatRecession,Stock and Watson(2012)explicitlysingle out the high (positive) correlation between credit spreads and proxies for economic uncertainty and conclude that “[T]hese two sets of instruments do not seem to be identifying distinct shocks.” Within a structural vector autoregressive (SVAR) framework—the workhorse of empirical macroeconomics—this high degree of comovement between indicators of financial distress such as credit spreads and uncertainty proxies significantly complicates the identification of financial and uncertainty shocks, as both types of variables are “fast moving.” As a result, it is difficult to impose plausible zero contemporaneous restrictions to identify these two types of disturbances. It it also difficult to impose sign restrictions on the impulse response functions in order to achieve an economically plausible identification because financial and uncertainty shocks have theoretically the same qualitative effects on both prices and quantities in most instances. In this paper, we use the penalty function approach developed initially by Faust (1998) and Uhlig (2005) to examine the interaction of economic uncertainty and financial conditions and to trace out the impact of the associated shocks on the macroeconomy. Within our SVAR framework, these two structural innovations are identified using a criterion that each shock should maximize the impulse response of its respective target variable over a pre-specified horizon. In economic terms, our identified uncertainty and financial shocks generate a prolonged period of heightened 1TheVIXindexisacommonlyusedproxyformacroeconomicuncertainty(Bloom,2009,2014). ThespeculativegradeCDXindexisatradablecreditderivativeindexusedwidelybyinvestorsforhedgingofandinvestingincorporate creditrisk. Buyingandsellingofthecreditderivativeindexiscomparabletobuyingandsellingportfoliosofcorporate bonds: By buying the index, the investor takes on the credit exposure—is exposed to defaults—a position similar to that of buying a portfolio of bonds; by selling the index, the credit exposure is passed on to another party. The speculative-gradeCDXindexreferences100(5-year)creditdefaultswap(CDS)contractsonfirmsthathavea“junk” rating from either Moody’s or Standard & Poor’s. The component firms must have highly liquid single-name CDS trading in their name, and the composition of both indexes, which is determined by a dealer poll, is representative of the U.S. corporate sector. 1

Figure 1: Financial Market Volatility and Credit Risk During the Great Recession 15 Correlation = 0.61 Acute phase of the crisis 10 5 0 -5 -10 -1.0 -0.5 0.0 0.5 1.0 Change in the Speculative-Grade CDX (pps.) ).spp( XIV eht ni egnahC Note: Sample period: daily data from 12/01/2007 to 06/30/2009. The scatter plot depicts the relationship betweenthedailychangeintheoption-impliedvolatilityontheS&P500stockfuturesindex(VIX)andthedaily changeinthe5-year(on-the-run)speculative-gradeCDXindex. Theperiod09/01/2008to08/05/2009marksthe acutephaseofthecrisis,whichreachedacriticalstageinearlySeptember2008,whenanevaporationofliquidity in the global credit markets threatened the solvency of several major financial institutions. The end date of the acute phase follows the release of the results from the Supervisory Capital Assessment Program (the so-called bank stress test) at 5 p.m. EST on May 7, 2009. economic uncertainty and a persistent tightening of financial conditions, respectively. Moreover, our identifying assumptions allow for financial conditions to react immediately to an uncertainty shock, while financial shocks can also have a contemporaneous effect on the level of economic uncertainty.2 Compared with identification schemes based on sign restrictions, this framework allows us to distinguish empirically between shocks that have otherwise very similar qualitative effects on the economy. Ourapproach, however, stillrequiresasequentialidentificationofthesetwoshocks. Asaresult, we implement the penalty function criterion in two steps. Under the baseline identification scheme, 2Tothebestofourknowledge,thereareonlytwostudiesthatanalyzetheeffectsofbothfinancialanduncertainty shocks in the VAR context. Focusing on the German economy, Popescu and Smets (2010) identify financial and uncertainty shocks using a recursive ordering, in which the uncertainty proxy is placed after the macro block but beforethefinancialmarketriskindex—thatis,theyallowuncertaintyshockstoelicitanimmediatechangeinfinancial conditions but not vice versa. Gilchrist et al. (2014) use U.S. data to explore the macroeconomic implications of uncertaintyandfinancialshocksusingalternativeorderingsfortheuncertaintyandfinancialstressproxies. Thekey finding that emerges from their analysis is that the economic significance of uncertainty shocks hinges crucially on whether they have been orthogonalized with respect to the contemporaneous information in credit spreads. This result highlights the need for an approach that allows for a contemporaneous feedback between financial conditions and economic uncertainty. 2

we first search for an innovation that maximizes the response of the uncertainty proxy over a given horizon—this optimization step identifies what we call an “uncertainty shock.” In the second step, we search for an innovation that maximizes the response of an indicator of financial conditions over the same horizon and that is orthogonal to the uncertainty shock identified in the first step—we call this shock a “financial shock.” To examine the robustness of the assumptions underlying our baseline identification scheme, we also consider an alternative strategy that reverses the ordering of the two penalty function steps used to identify these two disturbances. We implement the penalty function approach to multiple shock identification in the context of a standard monetary VAR, augmented with a measure of the tightness of financial conditions and an uncertainty proxy. To measure financial market strains, we use the excess bond premium, an indicator of the effective “risk-bearing capacity” of the financial intermediary sector, developed recently by Gilchrist and Zakrajˇsek (2012). Reflecting the amorphous concept of economic uncertainty,weexaminethemacroeconomicimplicationsofuncertaintyshocksimpliedbysixwidelyused uncertainty proxies. Three of these proxies are based on stock market volatility, one is the wellknown index of economic policy uncertainty developed by Baker et al. (2015), while the remaining two proxies are based on real economic activity; the latter two uncertainty measures correspond to the survey-based measure of forecast dispersion constructed by Bachmann et al. (2013) and a measure of dispersion in forecast errors constructed from a statistical model developed recently by Jurado et al. (2015). Webeginouranalysiswithasimpleforecastingexercise,inwhichwecomparetheabilityofthese sixuncertaintymeasurestoforecasteconomicactivityrelativetotheexcessbondpremium.3 While the excess bond premium provides economically important and statistically significant gains in predicting the course of future economic activity, we find that only the two measures of uncertainty based on economic activity provide marginal improvements in forecasting power. We then turn to an analysis of the role that uncertainty and financial shocks play in explaining macroeconomic outcomes in the VAR framework described above. Under the identification scheme that orders uncertainty first, we find that uncertainty shocks have an economically and statistically significant impact on both the stock market and real economic activity. This result holds true for all six uncertainty proxies, though the estimated declines in economic activity are appreciably larger in response to uncertainty shocks implied by the two uncertainty measures based on real economic data. Importantly, under this identification scheme, an increase in uncertainty also leads to a deterioration in financial conditions as measured by an increase in the excess bond premium, a result that highlights the close relationship between swings in economic uncertainty and changes in financial market conditions. Under the alternative identification scheme—in which uncertainty is ordered after the excess bond premium—the effect of an uncertainty shock varies significantly across the different uncertainty proxies. For the two proxies based on real economic data, we find that uncertainty shocks 3Gilchrist and Zakrajˇsek(2012)provideextensiveevidenceonthepredictivecontentofcreditspreadsforeconomic activity in general and the excess bond premium in particular. 3

again induce significant declines in real economic activity, although unsurprisingly, the effects are substantially attenuated relative to the identification scheme in which uncertainty is ordered first. Strikingly, under this second identification scheme, uncertainty shocks measured with either stock market data or economic policy news have no economically discernible effect on macroeconomic outcomes.4 Given the robustness of the results obtained using uncertainty proxies implied by real economic data, the remainder of our analysis explores the macroeconomic implications of uncertainty and financial shocks using the Jurado et al. (2015) measure of uncertainty, along with the excess bond premium to measure strains in financial markets. According to our results, both financial and uncertainty shocks have robust negative effects on economic activity that are quite similar in magnitude—they both imply a contraction in real industrial output between 0.6 percent and 1 percent, depending on the identification scheme. Forecast error variance decompositions imply that these two shocks account for 20 percent to 40 percent of the variability in industrial production at business cycle frequencies, again depending on the identification scheme. Historical variance decompositions further underscore the important role played by these two shocks in explaining both economic outcomes and fluctuations in the stock market. Notably and consistent with the findings of Stock and Watson (2012), the combination of financial and uncertainty shocks fully accounts for the contraction in economic activity and the collapse of the stock market during the Great Recession. The final step of our analysis considers an external validation exercise, in which we explore the extent to which shocks to economic uncertainty and financial conditions are, in fact, independent sources of macroeconomic instability, or whether they reflect endogenous responses to more traditional sources of business cycle fluctuations. Specifically, we study the extent to which both uncertainty and financial shocks are correlated with alternative measures of business cycle disturbances, such as unanticipated shifts in the stance of monetary policy and shocks to technology, oil prices, and government spending. Consistent with Stock and Watson (2012), we find that our shocks to both uncertainty and financial market conditions are completely uncorrelated with any such external instrument. The literature on the effects of uncertainty on macroeconomic outcomes has emphasized real options effects that lead to declines in business investment (Bloom, 2009; Bloom et al., 2012); financial mechanisms whereby the cost of external finance increases in response to a rise in uncertainty (Arellano et al., 2012; Christiano et al., 2014; Gilchrist et al., 2014); feedback mechanisms through which reduced economic activity leads to heightened macroeconomic uncertainty (Van Nieuwerburgh and Veldkamp, 2006; Fajgelbaum et al., 2014); and difficulties in forecasting economic variables during recessions (Orlik and Veldkamp, 2014). Our results imply that increases in uncertainty that are associated with a tightening of financial conditions have particularly pow- 4Recent work by Ludvigson et al. (2016) provides an alternative measure of financial uncertainty and a novel identification scheme, whichsuggeststhatitis financial, ratherthanreal, uncertaintythatdrivesthebusinesscycle. These disparate results may be due to the difference between financial uncertainty and stock market volatility and possibly account for the extent to which financial uncertainty comoves with the excess bond premium. 4

erful adverse effects on real economic activity and the stock market, a finding that is consistent with models in which uncertainty shocks are amplified via a reduction in the supply of credit. It is also consistent with the recent theoretical work of Brunnermeier and Sannikov (2014), in which a deterioration in borrowers’ balance sheet conditions can induce greater financial market volatility. It is worth emphasizing, however, that all of the channels describe above are, to a large extent, complementary mechanism, which may simultaneously lead to heightened uncertainty, a tightening of financial conditions, and a collapse in economic activity. Nonetheless, the finding that both uncertainty and financial shocks are uncorrelated with other measurable external shocks suggests that there are limitations to the argument that increased uncertainty and the concomitant deterioration in financial conditions are purely endogenous responses to fluctuations in economic activity over the normal course of the business cycle. Rather, our results underscore that the interplay between heightened uncertainty and increased financial fragility have independent and deleterious consequences for both asset valuations and macroeconomic outcomes. 2 Uncertainty, Financial Conditions, and Economic Activity 2.1 Measuring Uncertainty and Financial Conditions Economic uncertainty is difficult to quantify. As a result, the empirical literature is awash with different uncertainty proxies and new measures crop up all the time (Bloom, 2014). Because there is little consensus among economists on what is the best measure of economic uncertainty, rather than taking a stand on any particular indicator, we consider six different proxies, which span the range of methodological approaches used to infer fluctuations in economic uncertainty at monthly frequency. Our first three uncertainty proxies are based on stock market data. Because financial asset prices, in principle, encompass all aspects of the firm’s environment that the investors view as important, much research in this area relies on stock prices to infer fluctuations in both the microeconomic and macroeconomic uncertainty. The first uncertainty proxy used in our analysis is the realized stock market volatility (RVOL), a model-free measure that is especially simple to construct. We also consider the option-implied volatility on the S&P 100 stock futures index constructed by the Chicago Board of Option Exchange (VXO).5 While intuitive and readily available, the RVOL and VXO uncertainty proxies are a combination of two factors: the actual realized or expected stock market volatility—that is, stock market uncertainty—and a factor containing information about risk and risk aversion. As emphasized by Bekaert et al. (2013), the risk component is highly countercyclical. In our context, therefore, using these two uncertainty proxies to jointly identify uncertainty and financial shocks is likely to confound their respective effect on the macroeconomy. Gilchrist et al. (2014) try to circumvent 5We use the VXO, as opposed to the VIX option-implied volatility, because the VXO is available starting in January 1986, compared with January 1990, the starting date for the VIX. The correlation between these two indicators, however, is almost 0.99 at a monthly frequency. 5

Figure 2: Financial Market Conditions Percentage points 3.5 2.5 1.5 0.5 -0.5 -1.5 1973 1977 1981 1985 1989 1993 1997 2001 2005 2009 2013 Note: Sample period: monthly data from 1973:M1 to 2015:M3. The solid line shows the excess bond premium, anindicatorofthetightnessofbroadfinancialmarketconditions;seeGilchrist and Zakrajˇsek(2012)andthetext for details. The shaded vertical bars denote the NBER-dated recessions. this problem by purging (excess) stock returns of their systematic variation using a set of standard empirical risk factors. Their uncertainty measure (IVOL), which captures common shocks in the idiosyncratic volatility of equity returns, is our third proxy and one that is less likely to reflect the countercyclical nature of contractual and informational frictions associated with financial shocks. Nevertheless, we also look beyond the stock market to infer fluctuations in macroeconomic uncertainty. Our first such measure is the economic uncertainty index proposed by Jurado et al. (2015), which is based on the implied forecast errors for real economic activity derived from a factor model that utilizes hundreds of economic and financial series (JLN).6 Next is the widely citedindexofeconomicpolicyuncertaintydevelopedbyBaker et al.(2015)(BBD),whichcaptures the frequency of words in major U.S. newspapers associated with uncertainty regarding economic policy. And our last proxy for macroeconomic uncertainty, put forth by Bachmann et al. (2013), is a measure of forecast dispersion constructed using the Philadelphia Fed’s Business Outlook Survey (BES). Details concerning the construction of the various uncertainty measures are contained in 6Throughout the paper, we use the JLN measure of uncertainty at the 3-month forecast horizon; our results, however, were robust to both shorter (1-month) and longer (12-month) horizons. 6

Table 1: Cross-Correlations Between the EBP and Different Uncertainty Proxies Lag/Lead (h) RVOL IVOL VXO JLN BBD BES −3 0.47∗∗∗ 0.25∗∗∗ 0.51∗∗∗ 0.46∗∗∗ 0.37∗∗∗ 0.02 −2 0.52∗∗∗ 0.28∗∗∗ 0.54∗∗∗ 0.49∗∗∗ 0.39∗∗∗ 0.08∗ −1 0.58∗∗∗ 0.33∗∗∗ 0.58∗∗∗ 0.51∗∗∗ 0.40∗∗∗ 0.09∗ 0 0.59∗∗∗ 0.34∗∗∗ 0.59∗∗∗ 0.52∗∗∗ 0.42∗∗∗ 0.12∗∗ 1 0.60∗∗∗ 0.39∗∗∗ 0.59∗∗∗ 0.52∗∗∗ 0.39∗∗∗ 0.12∗∗∗ 2 0.54∗∗∗ 0.33∗∗∗ 0.53∗∗∗ 0.51∗∗∗ 0.32∗∗∗ 0.14∗∗∗ 3 0.52∗∗∗ 0.31∗∗∗ 0.48∗∗∗ 0.50∗∗∗ 0.28∗∗∗ 0.14∗∗∗ Note: The entries in the table denote the cross-correlations between the EBP in month t and the specified uncertainty proxy in month t+h. RVOL = realized equity volatility (1973:M1–2015:M3, T =507); IVOL = idiosyncratic equity volatility based on Gilchrist et al. (2014) (1973:M1–2015:M3, T = 507); VXO = optionimplied volatility on the S&P 100 stock futures index (1986:M1–2015:M3, T = 351); JLN = uncertainty measure based on Jurado et al. (2015) (1973:M1–2015:M3, T = 507); BBD = uncertainty measure based on Baker et al. (2015) (1985:M1–2015:M3, T =363); and BES = uncertainty measure based on Bachmann et al. (2013) (1973:M1–2011:M12, T =468). ∗ p<.10, ∗∗ p<.05, and ∗∗∗ p<.01. Appendix A. To measure the tightness of financial market conditions, we rely on corporate bond credit spreads.7 Specifically, we use the excess bond premium (EBP) of Gilchrist and Zakrajˇsek (2012) (GZ hereafter), an estimate of the extra compensation demanded by bond investors for bearing exposure to U.S. nonfinancial corporate credit risk, above and beyond the compensation for expected losses.8 As emphasized by GZ, the corporate cash market is served by major financial institutions and fluctuations in the EBP thus capture shifts in the risk attitudes of these institutions and their willingness to bear credit risk and to intermediate credit more generally.9 The solid line in Figure 2 shows this indicator of broad financial conditions over the 1973:M1– 2015:M12 period, while Table 1 displays the pairwise cross-correlations between the EBP and the different uncertainty proxies at various leads and lags. Note that the cross-correlations between the tightness of financial conditions and economic uncertainty are all positive and tend to be the highestath = 0,thatis,contemporaneously. Moreover,thereisasubstantialdegreeofcomovement between the EBP and various uncertainty proxies at the near-term horizons, underscoring the close relationship between changes in financial conditions and swings in economic uncertainty over the course of a business cycle. 7Thefactthatcorporatebondcreditspreadsarehighlyinformativeaboutthetightnessoffinancialconditionsin the economy and thus the implied degree of departure from the Modigliani–Miller paradigm of frictionless financial markets is supported by large empirical literature, which shows that credit spreads form the most informative and reliable class of financial indicators for future economic activity and that unanticipated increases in credit spreads havelargeandpersistentadverseeffects onthemacroeconomy(Gilchrist et al.,2009;Gilchrist and Zakrajˇsek,2012; Boivin et al., 2013; Faust et al., 2013). 8The EBP is effectively a measure of credit spreads net of an estimate of default risk and thus has a natural interpretation of an expected credit return, In netting out default risk, we make a small modification to the original GZ methodology. Specifically, we allow the conditional variance of the (log) credit spread “pricing errors” to vary over time. However, all of the results reported in the paper are virtually the same if the conditional variance of the error term in the GZ credit spread pricing regression is assumed to be constant over time. 9See Adrian and Shin (2010), and L´opez-Salido et al. (2016) for related empirical evidence. 7

This holds true even for uncertainty proxies such as JLN and BBD, which are not explicitly basedonfinancialmarketdata. OneclearexceptiontothispatternistheBESuncertaintyproxy—a measure based on the survey respondents’ disagreement regarding future economic outcomes—the past values of which appear to be largely uncorrelated with future fluctuations in the EBP; on the other hand, a tightening of financial conditions is mildly indicative of a future increase in this uncertainty indicator. In sum, these results highlight the need for analysis that explicitly incorporates the interaction between economic uncertainty and financial conditions. 2.2 Uncertainty and Financial Conditions as Predictors of Economic Activity As a warm-up exercise, we first explore the relative roles of uncertainty and financial conditions as predictors of the near-term course in economic activity. Specifically, letting Y denote a monthly t measure of economic activity, we estimate the following forecasting regression: h ∆ Y = α+β σ +β EBP + γ ∆Y +ǫ , (1) h t+h 1 t 2 t i t−i t+h Xi=1 where ∆ Y ≡ 1200 ln Yt+h and h ≥ 0 is the forecast horizon (∆ ≡ ∆); σ denotes one of our h t+h h+1 (cid:16) Yt−1(cid:17) 0 t six uncertainty proxies in month t; EBP is the excess bond premium in the same month; and ǫ t t+h is the forecast error. For each of the six uncertainty proxies, we estimate the forecasting regression (1) separately by OLS. Measures of monthly economic activity considered include manufacturing industrial productionindexandprivate(nonfarm)payrollemployment. Tofacilitatethecomparisonofthepredictive power of economic uncertainty and financial conditions, we report the standardized estimates of the coefficients β and β . The results for the forecast horizon of 3 months are tabulated in Table 2, 1 2 while those for the 12-month horizon are shown in Table 3. AccordingtoTable2, thepredictivecontentofvariousuncertaintymeasuresforeconomicactivity is fairly uneven at the 3-month horizon. While increases in the realized stock market volatility are associated with statistically and economically significant slowdown in the growth of real industrial output and employment, the information content of the IVOL and VXO uncertainty proxies is quite limited. Similarly, the BBD economic policy uncertainty index appears to be uninformative about the near-term course of economic activity, conditional on the EBP. The JLN and BES uncertainty measures, by contrast, are strong predictors of the growth in both real industrial output and employment over the 3-month horizon. In comparison with these uncertainty measures, the EBP appears to be a considerably more reliablepredictorofnear-termeconomicdevelopments. Inallspecifications, theestimatedcoefficients imply an economically and statistically significant negative relationship between the tightness of financial conditions and subsequent economic activity. For example, using a coefficient estimate of 0.40, a value in the range of the central tendency of the point estimates reported in panel (a), an increaseof50basispointsintheEBPinmontht—amoveofaboutonestandarddeviation—implies 8

Table 2: Uncertainty, Financial Conditions, and Economic Activity (3-month Forecast Horizon) Predictor RVOL IVOL VXO JLN BBD BES (a) Industrial Production σ −0.15∗∗ −0.06 0.02 −0.34∗∗∗ −0.08 −0.26∗∗∗ t [2.51] [1.20] [0.03] [4.09] [1.27] [3.95] EBP −0.31∗∗∗ −0.38∗∗∗ −0.46∗∗∗ −0.30∗∗∗ −0.42∗∗∗ −0.40∗∗∗ t [3.65] [4.00] [3.33] [4.19] [3.37] [4.49] Adj. R2 0.36 0.35 0.45 0.41 0.45 0.41 (b) Payroll Employment σ −0.12∗∗∗ −0.09∗∗ −0.05 −0.22∗∗∗ −0.03 −0.19∗∗∗ t [4.01] [2.25] [1.32] [3.89] [0.74] [3.74] EBP −0.23∗∗∗ −0.27∗∗∗ −0.25∗∗∗ −0.23∗∗∗ −0.27∗∗∗ −0.27∗∗∗ t [5.84] [4.26] [3.86] [4.27] [3.61] [4.14] Adj. R2 0.67 0.67 0.80 0.69 0.81 0.70 Note: Thedependentvariableineachspecificationis∆3 Y t+3,theannualizedlog-differenceinthespecifiedindicator of economic activity from month t−1 to month t+3. The entries in the rows of the table corresponding to σ t denotethestandardizedestimatesoftheOLScoefficientsassociatedwiththespecifieduncertaintyproxyinmontht: RVOL = realized equity volatility (1973:M1–2015:M3, T = 507); IVOL = idiosyncratic equity volatility based on Gilchrist et al.(2014)(1973:M1–2015:M3,T =507);VXO=option-impliedvolatilityontheS&P100stockfutures index (1986:M1–2015:M3, T =351); JLN = uncertainty measure based on Jurado et al. (2015) (1973:M1–2015:M3, T = 507); BBD = uncertainty measure based on Baker et al. (2015) (1985:M1–2015:M3, T = 363); and BES = uncertainty measure based on Bachmann et al. (2013) (1973:M1–2011:M12, T = 468). The entries in the rows of thetablecorrespondingtoEBPt denotethestandardizedestimatesoftheOLScoefficientsassociatedwiththeEBP in month t. In addition to σ t and EBPt, each specification also includes a constant and lags 1, 2, and 3 of ∆Y t (not reported). Absolute t-statistics reported in brackets are based on the heteroskedasticity- and autocorrelationconsistent asymptotic covariance matrix computed according to Newey and West (1987) with the automatic lag selection method of Newey and West (1994): * p<.10; ** p<.05; and *** p<.01. a nearly 3 percentage points (annualized) drop in the growth rate of real industrial output over the following three months. In the labor market (panel (b)), such a tightening of financial conditions is estimated to reduce employment growth by more than one-half of a percentage point over the same horizon. Theunevenforecastingperformanceoftheuncertaintyindicatorsisevenmoreevidentatlonger forecast horizons (Table 3). Conditional on the EBP, uncertainty proxies based on stock market data (RVOL, IVOL, and VXO) have no predictive power for the growth of industrial production one-year ahead (panel (a)); and the same is true of the BBD economic policy uncertainty index. Once the state of current financial market conditions is taken into account, it is only the JLN and BES uncertainty measures that are informative about the trajectory of real industrial output over the next 12 months. A similar picture emerges when we turn to the labor market (panel (b)). The RVOL, IVOL, and VXO uncertainty measures have essentially no marginal predictive power for the year-ahead employmentgrowth,nordoestheBBDuncertaintymeasureappeartobeausefulindicatoroffuture labor market developments. Again, only the JLN and BES uncertainty measures help predict 9

Table 3: Uncertainty, Financial Conditions, and Economic Activity (12-month Forecast Horizon) Predictor RVOL IVOL VXO JLN BBD BES (a) Industrial Production σ −0.08 −0.01 0.05 −0.62∗∗∗ 0.01 −0.32∗∗∗ t [1.25] [0.06] [0.46] [5.51] [0.10] [3.16] EBP −0.37∗∗∗ −0.41∗∗∗ −0.52∗∗∗ −0.28∗∗∗ −0.49∗∗∗ −0.39∗∗∗ t [2.89] [3.34] [2.77] [2.78] [3.09] [3.84] Adj. R2 0.20 0.19 0.24 0.39 0.23 0.28 (b) Payroll Employment σ −0.10∗ −0.13 −0.00 −0.34∗∗∗ 0.07 −0.35∗∗∗ t [1.83] [1.39] [0.01] [4.03] [1.27] [3.59] EBP −0.36∗∗∗ −0.38∗∗∗ −0.36∗∗∗ −0.31∗∗∗ −0.39∗∗∗ −0.37∗∗∗ t [3.77] [4.13] [3.63] [3.79] [4.24] [4.45] Adj. R2 0.45 0.46 0.61 0.52 0.61 0.54 Note: The dependent variable in each specification is ∆12 Y t+12, the annualized log-difference in the specified indicatorofeconomicactivityfrommontht−1tomontht+12. Theentriesintherowsofthetablecorresponding to σ t denote the standardized estimates of the OLS coefficients associated with the specified uncertainty proxy in month t: RVOL = realized equity volatility (1973:M1–2015:M3, T = 507); IVOL = idiosyncratic equity volatility basedonGilchrist et al.(2014)(1973:M1–2015:M3,T =507);VXO=option-impliedvolatilityontheS&P100stock futures index (1986:M1–2015:M3, T = 351); JLN = uncertainty measure based on Jurado et al. (2015) (1973:M1– 2015:M3, T = 507); BBD = uncertainty measure based on Baker et al. (2015) (1985:M1–2015:M3, T = 363); and BES=uncertaintymeasurebasedonBachmann et al.(2013)(1973:M1–2011:M12,T =468). Theentriesintherows of the table corresponding to EBPt denote the standardized estimates of the OLS coefficients associated with the EBPinmontht. Inadditiontoσ t andEBPt,eachspecificationalsoincludesaconstantandlags1,2,...,12of∆Y t (not reported). Absolute t-statistics reported in brackets are based on the heteroskedasticity- and autocorrelationconsistent asymptotic covariance matrix computed according to Newey and West (1987) with the automatic lag selection method of Newey and West (1994): * p<.10; ** p<.05; and *** p<.01. employment growth over the 12-month horizon. In contrast, movements in the EBP continue to provide unambiguous and informative signals about the evolution of the year-ahead economic outlook,withatighteningoffinancialconditionsportendingamarkeddecelerationinrealindustrial output and a significant deterioration in labor market conditions.10 The results from the above forecasting exercises—which, of course, are completely silent on the causalrelationshipbetweenuncertainty,financialconditions,andeconomicactivity—areinstructive for two reasons. First, they underscore the fact that measures of financial conditions based on corporatebondcreditspreadsarehighlyinformativeabouttheeconomicoutlook. Second,although the information content of various uncertainty indicators is decidedly mixed, the available evidence neverthelesssuggeststhatsomeoftheseindicatorscontaineconomicallyandstatisticallysignificant marginal predictive power for economic activity. In combination with the fact that episodes of 10Asarobustnesscheck,wealsoestimatedforecastingregressionsthatinadditiontotheEBPandanuncertainty proxy also conditioned on the level of real risk-free interest rates and the slope of the yield curve (i.e., the term spread). The inclusion of these additional financial variables, however, yielded the same conclusions regarding the relative predictive power of the EBP and various uncertainty proxies. 10

financial distress are closely associated with periods of heightened economic uncertainty and that theoretical mechanisms based on frictions in financial markets imply an important interaction between changes in financial conditions and fluctuations in uncertainty, a natural question raised by this evidence concerns the relative importance of financial and uncertainty shocks in business cycle fluctuations. To answer this question empirically, however, one has to take a stand on the joint identification of these two types of shocks, the subject of the remainder of the paper. 3 Identifying Uncertainty and Financial Shocks To identify uncertainty and financial shocks, we employ the penalty function approach (PFA) proposed initially by Faust (1998) and Uhlig (2005) in the context of the VAR-based identification of monetary policy shocks. This approach was later extended by Mountford and Uhlig (2009) to jointly identify multiple structural disturbances. In brief, the PFA selects a structural VAR model bymaximizingacriterionfunctionsubjecttoinequalityconstraints. Thecriterionfunctionconsists of the sum of impulse response functions (IRFs) of selected variables from horizon 0 to horizon H, while the inequality constraints correspond to sign restrictions on these IRFs. In this section, we first provide a general formulation of the PFA. We then discuss the rationale underlying our two identification schemes—that is, the choice of the penalty functions—and the estimation details. 3.1 The SVAR and the Penalty Function Consider the following SVAR: p y′A = y′ A +c+ǫ′; t = 1,...,T, (2) t 0 t−i i t Xi=1 where y is an n×1 vector of endogenous variables, ǫ is an n×1 vector of structural shocks, A , t t i i = 1,...,p, is an n×n matrix of structural parameters with A invertible, c is a 1×n vector of 0 parameters, p is the lag length, and T is the sample size. Conditional on past information and the initial conditions y ,...,y , the vector of structural shocks ǫ is assumed to be Gaussian with 0 1−p t mean zero and covariance matrix I , the n×n identity matrix. n The SVAR model in equation (2) can be written more compactly as y′A = x′A +ǫ′, t 0 t + t where A′ = A′ ...A′ c′ and x′ = y′ ...y′ 1 . The dimension of A is m × n, where + 1 p t t−1 t−p + m = np+1, a(cid:2)nd the elemen(cid:3)ts of matric(cid:2)es A and A (cid:3)correspond to the structural parameters of 0 + the VAR system. The reduced-form representation of the VAR is given by y′ = x′B+u′, t t t 11

where B = A A−1, u′ = ǫ′A−1, and E u u′ = A A′ −1 ≡ Ω; the matrices B and Ω are the + 0 t t 0 t t 0 0 reduced-form parameters. (cid:2) (cid:3) (cid:0) (cid:1) The impulse response functions are defined as follows. Let {A ,A } represent arbitrary struc- 0 + tural parameters. Then, the IRF of the i-th variable to the j-th structural shock at a finite horizon h, denoted by L (A ,A ) , corresponds to the element in row i and column j of the h 0 + ij matrix A−1J′FhJ ′ , where 0 (cid:2) (cid:3) A A−1 I ··· 0 I 1 0 n n  . . . . . . ... . . .  0 F = and J = . . A A−1 0 ··· I   . .   p−1 0 n        A A−1 0 ··· 0 0  p 0    Note that the matrix of IRFs upon impact is given by L A ,A = A−1. As in Uhlig (2005), 0 0 + 0 we characterize the set of all possible IRFs using an n×n o(cid:0)rthonorm(cid:1)al matrix Q ∈ O(n), where O(n) denotes the set of all orthonormal n × n matrices. To see this, let T denote the lower triangular matrix from the Cholesky factorization of Ω. Then for any orthonormal matrix Q, the matrix A˜−1 = TQ is also a decomposition of Ω that satisfies A˜ A˜′ −1 = Ω. Identification of 0 0 0 the SVAR thus amounts to specifying a set of restrictions on t(cid:2)he mat(cid:3)rix Q. For instance, Q = I imposes identification based on the recursive ordering of the VAR—the widely used Cholesky n decomposition—whilesignrestrictionsontheIRFsinvolvespecifyingasetofadmissibleQmatrices. It is worth emphasizing that our identification strategy does not identify all n structural shocks—that is, the entire Q matrix. Rather, we identify a subset k ≤ n of shocks, represented by q = Qe , j = 1,...,k, where e denotes the j-th column of I . Specifically, let {A ,A } be any j j j n 0 + drawofthestructuralparametersandconsideracasewheretheidentificationofthej-thstructural shock restricts the IRF of a set of variables indexed by I+ ⊂ {0,1,...,n} to be positive and the j IRF of a set of variables indexed by I− ⊂ {0,1,...,n} to be negative. Furthermore, assume that j the restrictions on variable i are enforced for H ≥ 0 periods. The identification of q then amounts j to solving the following optimization problem: q∗ = argmin Ψ(q ) (3) j j qj subject to e′L (T−1,BT−1)q > 0, i ∈ I+ and h = 0,...,H; (4) i h j j e′L (T−1,BT−1)q < 0, i ∈ I− and h = 0,...,H; (5) i h j j Q∗′ q = 0, (6) j−1 j where H e′L (T−1,BT−1)q H e′L (T−1,BT−1)q Ψ(q ) = − i h j + i h j , (7) j (cid:18) ω (cid:19) (cid:18) ω (cid:19) i X ∈I+Xh=0 i i X ∈I−Xh=0 i j j 12

ω is the standard deviation of variable i, and Q∗ = q∗...q∗ , for j = 1,...,n.11 i j−1 1 j−1 In line with the existing literature, our characteriza(cid:2)tion of the(cid:3)penalty function given by equation (7) assumes that the vector q rotates the IRFs associated with the Cholesky factor matrix j T. As in Uhlig (2005) and Mountford and Uhlig (2009), the constraints (4) and (5) correspond to sign restrictions on the IRFs that enter the penalty function Ψ(q ). Note that the PFA selects a j single rotation matrix Q∗ for each draw of the structural parameters {A ,A }, the same as in the 0 + standardpointidentification. However, thePFAdiffersfromthestandardsignrestrictionapproach because the IRFs corresponding to each draw of the structural parameters {A ,A } are computed 0 + using the rotation matrix Q∗ that minimizes the penalty function (7). The constraints (4) and (5) do not identify a set of structural models—that is, a set of Q matrices—rather, they define a set of admissible rotation matrices from which the matrix Q∗ is selected.12 In our implementation of the PFA, however, the sign restrictions are never violated, and the rotation matrix Q∗ always lies in the interior of the admissible set. Following Mountford and Uhlig (2009), we jointly identify multiple structural shocks sequentially by specifying a penalty function (7) for each shock and imposing—via the constraint (6) in the optimization problem (3)—that shock j is orthogonal to shocks 1,...,j −1. This sequential approach is reminiscent of a recursive ordering implicit in the Cholesky decomposition—in fact, this approach returns a Cholesky factorization of Ω if the penalty function that identifies shock j contains only the impact response (H = 0) of variable j, for j = 1,...,n. In all other cases, the sequential identification of the shocks using the PFA does not impose any zero restrictions on the structural parameters {A ,A } or on the IRFs at any horizon. 0 + This identification strategy, however, is not invariant to the ordering of the shocks. Next we describe the two identification schemes used in our analysis, which share the same penalty function anddifferonlyintheorderingoftheuncertaintyandfinancialshocks. Forthepurposeofdescribing thetwoidentificationschemes,itsufficestosaythatwithoutlossofgenerality,weordertheEBPand an uncertainty measure first and second, respectively, in the vector of the endogenous variables y . t 3.2 Identification and Estimation Both of our identification schemes involve two steps. In the baseline identification scheme, the first step identifies the uncertainty shock as an innovation that generates the largest increase in a measure of uncertainty for the first six months. The penalty function associated with this shock is given by 5 e′L (T−1,BT−1)q Ψ(q ) = − 1 h 1 , 1 (cid:18) ω (cid:19) Xh=0 1 11We follow the convention by letting Q∗ 0 equal the n×n null matrix; to obtain ω i, we compute the standard deviation of the OLS residuals associated with the i-th variable. 12This identification strategy differs from the pure sign restriction identification approach, in that both of our identificationschemesidentifyasinglestructuralVARspecification,ratherthanasetofmodels(Caldara and Kamps, 2012; Arias et al., 2013). 13

with e′L (T−1,BT−1)q > 0, h = 0,...,5, 1 h 1 where j = 1 because we identify the first shock in the system; i = 1 because the uncertainty measure is the first variable in the system; I+ = {1} and I− = ∅ because we only impose positive 1 1 restrictions on the uncertainty measure; and H = 5 because we restrict the IRFs of the uncertainty measure for the first six months (in our notation the impact response occurs in period 0).13 The second step in the scheme identifies financial shocks. Specifically, a financial shock is an innovation that generates the largest increase in the EBP for the first six months and is orthogonal to the uncertainty disturbance identified by the first step. The penalty function associated with this shock is given by 5 e′L (T−1,BT−1)q Ψ(q ) = − 2 h 2 , 2 (cid:18) ω (cid:19) Xh=0 2 with e′L (T−1,BT−1)q > 0, h = 0,...,5; 2 h 2 Q∗′q = 0, 1 2 where j = 2 because we are identifying the second shock and i = 2 because the EBP is the second variable in the VAR. Implicit in this identification scheme—which we call the σ–EBP identification—is the argument that fluctuations in economic uncertainty are driven mainly by uncertainty shocks, while unanticipated worsening in financial conditions is due primarily to an adverse financial shock. Moreover, such shocks elicit a persistent increase in their corresponding endogenous variable over the nearterm horizon, rather than just a maximal impact response. These identifying assumptions do not rule out the possibility that financial conditions might react contemporaneously to a change in economicuncertaintyinducedbyanuncertaintyshock; bythesametoken, macroeconomicuncertainty is allowed to change immediately in response to a financial shock. Although more general than the identification strategy based on the recursive ordering of the VAR system, the σ–EBP identification scheme still imposes a timing restriction—namely, the optimizationproblemthatidentifiesthefinancialshockissolvedconditionalonsolvingtheoptimization that identifies the uncertainty shock. Given the close relationship between changes in financial conditions and swings in economic uncertainty, this sequential ordering may lead to a concern that the identified uncertainty shocks are to some extent contaminated by financial shocks. This concern may be especially relevant in the case of uncertainty proxies based solely on financial asset prices (i.e., RVOL, IVOL, and VXO), movements in which reflect the confluence of fluctuations in the underlying uncertainty and changes in risk aversion. To take into account this possibility, our second identification scheme, which we refer to as 13Alltheresultsreportedinthepaperwerequalitativelyandquantitativelyverysimilarifuncertaintyandfinancial shocks were identified using a longer response horizon. 14

the EBP-σ identification, reverses the ordering of the two optimization problems. In other words, we first identify the financial shock as an innovation that generates the largest increase in the EBP—a persistent tightening of financial conditions—for the first six months. In the second step, an uncertainty shock, orthogonal to the financial disturbance implied by the first step, is identified as a shock that induces the largest increase in an uncertainty measure over the same horizon. We acknowledge that neither scheme fully resolves the difficult problem of how to identify these two types of shocks in a VAR context. However, in the absence of valid external instruments, we view the two approaches as providing useful bounds on the role of uncertainty and financial shocks in business cycle fluctuations. In spirit, our implementation of the PFA is similar to the identification strategy used by Uhlig (2003), Barsky and Sims (2011), and Kurmann and Otrok (2013), all of whom identify structural shocks by maximizing—over a pre-specified forecast horizon—the shock’s contribution to the forecast error variance of a given variable. Our identification schemes, in contrast, identify shocks by maximizing the impulse response of a given variable over a pre-specified horizon. We chose this approach because it implies that the identified financial shocks generate a persistent increase in the EBP—a prolonged period of financial distress—while uncertainty shocks lead to a persistent increase in an uncertainty proxy; that is, a period of heightened economic uncertainty, rather than just a one-off spike in volatility. Selecting shocks that maximize the forecast error variance of these twovariables, bycontrast, doesnotguaranteethattheirIRFswillnotswitchsignsovertheforecast horizon. However, the results reported below indicate that the two shocks identified using the PFA also explain a vast majority of the forecast error variance of their respective variables at business cycle frequencies, which suggests that these two approaches are unlikely to yield very different conclusions. To implement the two identification schemes, we employ Bayesian estimation techniques. Our monthly VAR specification consists of 10 endogenous variables: (1) an uncertainty proxy; (2) the EBP; (3) the log-difference of manufacturing industrial production index; (4) the log-difference of private (nonfarm) payroll employment; (5) the log-difference of real personal consumption expenditures(PCE);(6)thelog-differenceofthePCEpricedeflator; (7)thenominal2-yearTreasuryyield; (8) the nominal 10-year Treasury yield; (9) the value-weighted total stock market (log) return; and (10)thelog-differenceoftheS&PGoldmanSachsCommodityIndex. WeimposeaMinnesotaprior on the reduced-form VAR parameters by using dummy observations (Del Negro and Schorfheide, 2011)andselectthehyper-parametersthatgoverntheirpriordistributionsandtheVARlaglengthp by maximizing the marginal data density; to perform this optimization, we use the CMA-ES evolutionary algorithm proposed by Hansen et al. (2003). The resulting specification, which includes a constant, is estimated over the 1975:M1–2015:M3 period using six lags of the endogenous variables.14 14Details about the prior specification are provided in Appendix B. We use the first two years of the sample (1973:M1–1974:M12) as a training sample for the Minnesota prior. All the results reported in the paper are based on50,000drawsfromtheposteriordistributionofthestructuralparameters,wherethefirst10,000drawswereused as a burn-in period. 15

4 Economic Implications of Financial and Uncertainty Shocks This section presents our main results. In presenting the results, we adopt the following exposition scheme, which is best viewed in color. Specifically, the results based on the σ–EBP identification use a green-based color motif, while the results based on the EBP–σ identification of these two shocks use an orange-based color motif. The solid lines in panel (a) of Figure 3 show the median IRFs of the six uncertainty proxies to their own shocks under the σ–EBP identification scheme, while the shaded bands represent the corresponding90-percentpointwisecrediblebands. Consistentwithouridentifyingassumptionthat uncertainty shocks should elicit more than a one-off jump in uncertainty, the identified shocks in all cases induce a fairly persistent rise in their corresponding uncertainty proxy. With the exception of the JLN measure, which exhibits a hump-shaped response with a notably greater persistence, the responses of other proxies peak upon the impact of the shock, which then dies out within about 24 months. The impact of these shocks on the EBP—our measure of broad financial conditions—is shown in panel (b) of Figure 3. In response to each uncertainty shock, the EBP jumps upon impact, indicatinganimmediatetighteningoffinancialconditions. Thedeteriorationinfinancialconditions is somewhat more severe in response to uncertainty shocks based on equity valuations (RVOL, IVOL, and VXO), though shocks to other uncertainty proxies also lead to a sizable increase in the EBP. The macroeconomic effects of these uncertainty shocks are summarized in Figure 4.15 As shown in panel (a), industrial production, an especially cyclically sensitive measure of economic activity, respondsveryquicklytoanincreaseineconomicuncertainty: realindustrialoutputstartsdeclining almost immediately and, depending on the uncertainty proxy used to identify the corresponding shock, bottoms out between 0.5 percent and 1 percent below trend about 14 months after the impact of the uncertainty shock. According to panel (b), the increase in economic uncertainty is alsoverybadnewsforthestockmarket: inalmosteverycase, thepersistentincreaseinuncertainty and the associated tightening of financial market conditions lead to a sharp and immediate drop in the broad stock market; the one exception to this pattern is the JLN uncertainty measure—the response of equity prices to the JLN uncertainty shock is not nearly as sharp and immediate as that implied by uncertainty shocks based on other proxies. Figures5and6presentthesameinformationundertheEBP–σ identificationscheme. Asshown in panel (a) of Figure 5, the IRFs of the different uncertainty proxies to their own shocks under this alternative identification scheme are very similar to those reported in the corresponding panel of Figure 3. However, as shown in panel (b), the effect of uncertainty shocks on financial conditions differssignificantlyundertheEBP–σ identificationscheme. Inthiscase,adverseuncertaintyshocks do not lead to a systematic deterioration in financial conditions. In fact, only uncertainty shocks 15To conserve space, wereport only the IRFs of industrial production and the broad stock market. The responses ofotherendogenousvariablestobothuncertaintyandfinancialshocksallhavesignsandpatternsthatareconsistent with economic intuition and other related research. 16

Figure 3: Uncertainty Shocks Based on Different Uncertainty Proxies (σ–EBP Identification) Uncertainty measure: RVOL Uncertainty measure: IVOL Uncertainty measure: VXO Percent Percent Percent 6 12 4 4 8 3 2 2 4 1 0 0 0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Uncertainty measure: JLN Uncertainty measure: BBD Uncertainty measure: BES Percent Percent Percent 3 28 6 21 2 4 14 2 1 7 0 0 0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (a) Response of different uncertainty proxies to their own shocks Uncertainty measure: RVOL Uncertainty measure: IVOL Uncertainty measure: VXO Basis points Basis points Basis points 15 15 15 10 10 10 5 5 5 0 0 0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Uncertainty measure: JLN Uncertainty measure: BBD Uncertainty measure: BES Basis points Basis points Basis points 15 15 15 10 10 10 5 5 5 0 0 0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (b) Response of the EBP to uncertainty shocks based on different uncertainty proxies Note: The solid lines in panel (a) depict median responses of the specified uncertainty proxy to its own shock of 1 standard deviation, while those in panel (b) depict median responses of the EBP to the specified uncertainty shockof1standarddeviation;theshadedbandsrepresentthe90-percentpointwisecrediblesets. Seethetextand notes to Table 1 for details. 17

Figure 4: Economic Effects of Uncertainty Shocks Based on Different Uncertainty Proxies (σ–EBP Identification) Uncertainty measure: RVOL Uncertainty measure: IVOL Uncertainty measure: VXO Percent Percent Percent 0.0 0.0 0.0 -0.5 -0.5 -0.5 -1.0 -1.0 -1.0 -1.5 -1.5 -1.5 -2.0 -2.0 -2.0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Uncertainty measure: JLN Uncertainty measure: BBD Uncertainty measure: BES Percent Percent Percent 0.0 0.0 0.0 -0.5 -0.5 -0.5 -1.0 -1.0 -1.0 -1.5 -1.5 -1.5 -2.0 -2.0 -2.0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (a) Response of industrial production to uncertainty shocks based on different uncertainty proxies Uncertainty measure: RVOL Uncertainty measure: IVOL Uncertainty measure: VXO Percent Percent Percent 2 2 2 0 0 0 -2 -2 -2 -4 -4 -4 -6 -6 -6 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Uncertainty measure: JLN Uncertainty measure: BBD Uncertainty measure: BES Percent Percent Percent 2 2 2 0 0 0 -2 -2 -2 -4 -4 -4 -6 -6 -6 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (b) Response of the stock market to uncertainty shocks based on different uncertainty proxies Note: The solid lines in panel (a) depict median responses of industrial production to the specified uncertainty shock of 1 standard deviation, while those in panel (b) depict median responses of aggregate stock prices to the specified uncertainty shock of 1 standard deviation; the shaded bands represent the 90-percent pointwise credible sets. See the text and notes to Table 1 for details. 18

Figure 5: Uncertainty Shocks Based on Different Uncertainty Proxies (EBP-σ Identification) Uncertainty measure: RVOL Uncertainty measure: IVOL Uncertainty measure: VXO Percent Percent Percent 6 12 4 4 8 3 2 2 4 1 0 0 0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Uncertainty measure: JLN Uncertainty measure: BBD Uncertainty measure: BES Percent Percent Percent 3 28 6 21 2 4 14 2 1 7 0 0 0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (a) Response of different uncertainty proxies to their own shocks Uncertainty measure: RVOL Uncertainty measure: IVOL Uncertainty measure: VXO Basis points Basis points Basis points 5 5 5 0 0 0 -5 -5 -5 -10 -10 -10 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Uncertainty measure: JLN Uncertainty measure: BBD Uncertainty measure: BES Basis points Basis points Basis points 5 5 5 0 0 0 -5 -5 -5 -10 -10 -10 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (b) Response of the EBP to uncertainty shocks based on different uncertainty proxies Note: Thesolidlinesinpanel(a)depictmedianresponsesofthespecifieduncertaintyproxytoitsownshockof 1 standard deviation, while those in panel (b) depict median responses of the EBP to the specified uncertainty shock of 1 standard deviation; the shaded bands represent the 90-percent pointwise credible sets. See the text and notes to Table 1 for details. 19

Figure 6: Economic Effects of Uncertainty Shocks Based on Different Uncertainty Proxies (EBP-σ Identification) Uncertainty measure: RVOL Uncertainty measure: IVOL Uncertainty measure: VXO Percent Percent Percent 1.0 1.0 1.0 0.5 0.5 0.5 0.0 0.0 0.0 -0.5 -0.5 -0.5 -1.0 -1.0 -1.0 -1.5 -1.5 -1.5 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Uncertainty measure: JLN Uncertainty measure: BBD Uncertainty measure: BES Percent Percent Percent 1.0 1.0 1.0 0.5 0.5 0.5 0.0 0.0 0.0 -0.5 -0.5 -0.5 -1.0 -1.0 -1.0 -1.5 -1.5 -1.5 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (a) Response of industrial production to uncertainty shocks based on different uncertainty proxies Uncertainty measure: RVOL Uncertainty measure: IVOL Uncertainty measure: VXO Percent Percent Percent 3.0 3.0 3.0 1.5 1.5 1.5 0.0 0.0 0.0 -1.5 -1.5 -1.5 -3.0 -3.0 -3.0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Uncertainty measure: JLN Uncertainty measure: BBD Uncertainty measure: BES Percent Percent Percent 3.0 3.0 3.0 1.5 1.5 1.5 0.0 0.0 0.0 -1.5 -1.5 -1.5 -3.0 -3.0 -3.0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (b) Response of the stock market to uncertainty shocks based on different uncertainty proxies Note: The solid lines in panel (a) depict median responses of industrial production to the specified uncertainty shock of 1 standard deviation, while those in panel (b) depict median responses of aggregate stock prices to the specified uncertainty shock of 1 standard deviation; the shaded bands represent the 90-percent pointwise credible sets. See the text and notes to Table 1 for details. 20

based on the JLN measure induce a statistically and economically significant rise in the EBP under this identification. For all other uncertainty proxies, increases in the EBP in response to different uncertainty shocks are quite small in economic terms and are also estimated imprecisely.16 This difference in the response of the EBP to uncertainty shocks has important macroeconomic implications. As shown in panel (a) of Figure 6, uncertainty shocks implied by the RVOL, IVOL, VXO, and BBD measures have no significant real effects. Only shocks based on the JLN and BES uncertainty measures have an economically and statistically meaningful impact on industrial production under the EBP–σ identification scheme. And even in those two cases, the effect of uncertainty shocks on real industrial output is appreciably smaller compared with that implied by the corresponding shocks under the σ–EBP identification (see panel (a) of Figure 4). These results suggest that financial market conditions are an important part of the mechanism through whichuncertaintyshocksaffectthemacroeconomy. Andeventhoughourtwoidentificationschemes do not impose any zero restrictions on the contemporaneous responses of financial conditions to economic uncertainty and vice versa, just reversing the order of the two optimization problems has a first-order effect on the macroeconomic relevance of uncertainty shocks. This finding is particularly relevant for uncertainty proxies based on stock market data, where the sole effect of an uncertainty shock under the EBP–σ identification appears to be a short-lived decline in the stock market that has no consequences for real economic outcomes (see panel (b) of Figure 6). And while the JLN and BES uncertainty shocks both have real effects, the impact of the former shock on the stock market is more robust across the two identification schemes. This result may reflect the fact that the BES uncertainty proxy is based on the cross-sectional dispersion of indexes of expectations among survey participants and thus is more indicative of the degree of disagreement among the participants, rather than the actual forecast uncertainty, a point emphasized by D’Amico and Orphanides (2008). As a result, the remainder of the paper relies solely on the JLN measure of economic uncertainty. Under both of our identification schemes and consistent with the forecasting results reported in Section 2, this choice of the uncertainty proxy, in effect, gives economic uncertainty the maximum role in explaining business cycle fluctuations. Ausefulwaytojudgetheplausibilityandeconomicimportanceofuncertaintyshocksunderour twoidentificationschemesisdepictedinFigure7. Thesolidlinesshowtheamountofvariationinthe forecasterrorvarianceintheJLNuncertaintyproxy, theEBP,industrialproduction, andthebroad stock market that is attributable to the JLN uncertainty shocks under the σ–EBP identification scheme; the dotted lines, in contrast, show the same information for the JLN uncertainty shocks implied by the EBP–σ identification scheme. Recall that in both schemes, the identification of uncertainty shocks involved selecting orthogonalized innovations in the JLN uncertainty proxy that maximized the response of this proxy over the first six months after the impact of the shock. 16ThepuzzlingdropintheEBPthatoccursuponimpactofthedifferentuncertaintyshocks—especiallythosebased onstockmarketdata—isdueinlargeparttotheone-offspikesinfinancialmarketvolatility,suchasthestockmarket crash in October of 1987 and the “flash crash” in May of 2010. In those instances, stock market volatility increased sharply, but there was very little movement in credit spreads. In fact, when those observations are “dummied” out, the drop in the EBP upon the impact of uncertainty shocks is substantially attenuated. 21

Figure 7: Forecast Error Variance Decomposition of an Uncertainty Shock (σ–EBP vs. EBP–σ Identification) Uncertainty measure: JLN EBP Percent Percent 100 100 80 80 60 60 σ - EBP identification 40 40 EBP - σ identification 20 20 0 0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Industrial production Stock prices Percent Percent 100 100 80 80 60 60 40 40 20 20 0 0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Note: Thesolid(dotted)lineineachpaneldepictsthemedianestimateoftheportionoftheforecasterrorvariance of a specified variable at different horizon that is attributable to a 1 standard deviation JLN uncertainty shock undertheσ–EBP(EBP–σ)identificationscheme; theshadedbandsdenotethe90-percentpointwisecrediblesets corresponding to the σ–EBP identification scheme. And under both approaches, the identified uncertainty shocks account for the vast majority of the variation in the JLN measure at business cycle frequencies. This result is reassuring because it is consistent with our maintained assumption that fluctuations in economic uncertainty are due primarily to uncertainty shocks. According to this metric, uncertainty shocks are a significant source of economic fluctuations— they are estimated to explain between 20 percent to 40 percent of the variation in industrial productionatbusinesscyclefrequencies. Undertheσ–EBPidentificationscheme, theJLNuncertainty shocks also explain an economically meaningful amount of the variation in the stock market and about one-fifth of the forecast error variance in the EBP. Under the EBP–σ identification scheme, by contrast, the JLN uncertainty shocks are completely uninformative about future swings in the EBP;similarly, theircontributiontotheforecasterrorvarianceofthestockmarketisindistinguishable from zero. Both of these results are consistent with the view that changes in financial market conditions are an important conduit through which uncertainty shocks affect the real economy and that uncertainty shocks are an important independent source of cyclical disturbances. The macroeconomic implications of financial disturbances under both of our identification schemes are shown in Figure 8. According to the solid lines, an adverse financial shock under 22

Figure 8: Economic Effects of a Financial Shock (EBP–σ vs. σ–EBP Identification) EBP Uncertainty measure: JLN Basis points Percent 30 3 σ σ EBP - identification 2 - EBP identification 20 1 10 0 -1 0 -2 -10 -3 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Industrial production Stock prices Percent Percent 0.5 0 0.0 -2 -0.5 -4 -1.0 -1.5 -6 -2.0 -8 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Note: The solid (dotted) line in each panel depicts the median response of a specified variable to a 1 standard deviation EBP shock under the EBP–σ (σ–EBP) identification scheme; the shaded bands denote the 90-percent pointwise credible sets corresponding to the EBP–σ identification scheme. the EBP–σ identification scheme induces an economically significant and persistent increase in the EBP, a result consistent with our underlying identification strategy, which seeks to maximize the response of the EBP to its own shock. This significant tightening of financial conditions causes a mild increase in economic uncertainty and leads to a large and protracted decline in industrial production, as well as a sharp and immediate drop in the broad stock market. Under the σ–EBP identification scheme, the dotted lines, the response of the EBP to the financial shock is virtually identical to that implied by the EBP–σ identification. However, the worsening of financial conditions has no effect on economic uncertainty under the σ–EBP identification scheme. Nevertheless, an adverse financial shock still results in a significant and persistent decline in real industrial output and a considerable drop in the stock market. In Figure 9, we show the portion of the forecast error variance attributable to the financial shocks under our two identification schemes. In both cases, financial shocks account for the bulk of the variation in the EBP; at the same time, these shocks have only a limited impact on the JLN measure of economic uncertainty. Both of these findings are consistent with our assumptions that financial and uncertainty shocks—along with their potential interaction—are important independent sources of cyclical fluctuations. Indeed, according to our estimates, financial shocks explain 20 percent to 40 percent of the variation in real industrial output, the same amount as the JLN 23

Figure 9: Forecast Error Variance Decomposition of a Financial Shock (EBP–σ vs. σ–EBP Identification) EBP Uncertainty measure: JLN Percent Percent 100 100 80 80 60 60 40 EBP - σ identification 40 σ - EBP identification 20 20 0 0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Industrial production Stock prices Percent Percent 100 100 80 80 60 60 40 40 20 20 0 0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 Note: Thesolid(dotted)lineineachpaneldepictsthemedianestimateoftheportionoftheforecasterrorvariance ofaspecifiedvariableattributabletoa1standarddeviationEBPshockundertheEBP–σ (σ–EBP)identification scheme;theshadedbandsdenotethe90-percentpointwisecrediblesetscorrespondingtotheEBP–σidentification scheme. uncertainty shocks. Moreover, financial disruptions have an appreciably larger effect on the broad stockmarketcomparedwiththeJLNuncertaintyshocks—theEBPshocksareestimatedtoexplain about 40 percent of the forecast error variance in stock prices, on balance. Incombination,theaboveresultsareconsistentwiththeoreticalmechanismsthatemphasizethe presence of frictions in financial markets—and their effect on the effective supply on credit—as an important conduit through which fluctuations in uncertainty are propagated to the real economy (Arellano et al., 2012; Christiano et al., 2014; Gilchrist et al., 2014). In response to an unanticipated increase in uncertainty, distortions in financial markets induce a tightening of financial conditions—effectively reducing the supply of credit available to businesses and households—which leads to a decline in spending and production and a drop in the stock market. In addition, if we restrict our attention to uncertainty proxies that are not based solely on stock market data, our results also support the view that financial and uncertainty shocks are both important drivers of business cycles. 24

Figure 10: Historical Variance Decomposition of Selected Variables (σ–EBP Identification) EBP Uncertainty: JLN Basis points Index 300 40 Annual Annual Financial shock Uncertainty shock 200 Data 20 100 0 0 -100 -20 1975 1983 1991 1999 2007 2015 1975 1983 1991 1999 2007 2015 Industrial production growth Stock market return Percent Percent Annual Annual 10 35 0 0 -10 -35 -20 -70 1975 1983 1991 1999 2007 2015 1975 1983 1991 1999 2007 2015 Note: Sample period: annual data from 1976 to 2014. The shaded regions in each panel depict the historical contributions of shocks to the EBP (red) and the JLN uncertainty measure (green) to the specified variable; the twoshocksareorthogonalizedusingtheσ–EBPidentificationscheme. Theactualseries(solidlines)areexpressed in deviations from their respective estimated means. 4.1 Historical Significance of Financial and Uncertainty Shocks To put the above results into a historical perspective, this section examines the role of financial and uncertaintyshocksineconomicfluctuationsoverthepast40years. Thatis,fortheJLNuncertainty proxy, theEBP,industrialproduction, andstockprices, wecalculatetheportionoftheactualseries that is attributable to these two types of shocks over the 1976–2014 period.17 To better delineate the relative contributions of these two shocks to economic fluctuations, we present the data at an annual frequency.18 17We compute the historical variance decomposition at the OLS estimates of the reduced-form parameters. The sample used for the estimation includes actual data and the dummy observations used to implement the Minnesota prior (see Appendix B for details on the prior specification). 18TheEBPandtheJLNuncertaintyproxyaremeasuredasofDecemberofeachyear,whilethegrowthofindustrial production and the stock market return are expressed in year-over-year changes. 25

Theresultsofthisexercisefortheσ–EBPidentificationschemeareshowninFigure10. According to the historical decomposition implied by this identification scheme, our identified financial shocks account for the majority of the movements in the EBP—outside the Great Recession, the corresponding uncertainty shocks do not appear to have had much of an effect on financial market conditions. In contrast, the EBP shocks account roughly for about as much of the variability in the JLN measure of uncertainty as do the uncertainty shocks. Shocks emanating from the financial sector also shaped importantly the ups and downs in economic activity over our sample period. However, the economic significance of credit supply shockshasvariedconsiderablyoverthepastfourdecades. Suchshocksplayedadistinctlysecondary role in economic fluctuations during the first half of our sample, a result consistent with the heavy regulationoffinancialinstitutionsandmarketsduringthisperiod. Inaddition,thereisconsiderable empirical evidence showing that the economic downturns of the 1970s were influenced significantly bytheOPEC-inducedincreasesinoilprices(Hamilton,1983),whiletherecessionsoftheearly1980s owed importantly to the tightening of monetary policy under the then-Fed Chairman Volcker, who wasdeterminedtofightinflationandreversetheriseininflationexpectations(Lindsey et al.,2005). Financial shocks assumed new importance in the wake of financial deregulation and the associated financial deepening that took place during the second half of the 1980s and the early 1990s. Accordingtoourestimates,apronouncedtighteninginfinancialmarketconditionsoccurredinperiodssurroundingthe2001and2007–09cyclicaldownturns,andadversecreditsupplyshocksaccount for significant portions of the decline in real industrial output and equity valuations during these two recessions. On the other hand, easy financial conditions—at least in retrospect—characterized much of the mid-1990s and mid-2000s, and economic activity and the stock market were buoyed substantially by expansionary credit supply shocks. The Great Recession and its immediate aftermath is the one period in our sample during which economic uncertainty had an out-sized effect on macroeconomic outcomes. Adverse uncertainty shocks during this period contributed significantly to a tightening of financial conditions, a drop in industrial production, and the collapse in stock prices. In fact, it may be that the combination of these two types of shocks has an especially pernicious effect on the macroeconomy, which explains the prolonged slump. A qualitatively similar historical narrative emerges under the EBP–σ identification scheme. As shown in Figure 11, changes in credit market conditions over the past four decades are again driven primarily by financial shocks, although adverse uncertainty shocks contributed noticeably to the massive tightening of financial conditions experienced at the nadir of the 2008–09 crisis. Under this identification scheme, financial shocks continue to shape importantly the cyclical swings in economic uncertainty, especially during the latter part of our sample. And while these identifying assumptions assign a somewhat greater role in economic fluctuations to disruptions in the credit intermediationprocess,uncertaintyshocksremainasignificantsourceofmacroeconomicinstability, especially during the Great Recession. 26

Figure 11: Historical Variance Decomposition of Selected Variables (EBP–σ Identification) EBP Uncertainty: JLN Basis points Index 300 40 Annual Annual Financial shock Uncertainty shock 200 Data 20 100 0 0 -100 -20 1975 1983 1991 1999 2007 2015 1975 1983 1991 1999 2007 2015 Industrial production growth Stock market return Percent Percent Annual Annual 10 35 0 0 -10 -35 -20 -70 1975 1983 1991 1999 2007 2015 1975 1983 1991 1999 2007 2015 Note: Sample period: annual data from 1976 to 2014. The shaded regions in each panel depict the historical contributions of shocks to the EBP (red) and the JLN uncertainty measure (green) to the specified variable; the twoshocksareorthogonalizedusingtheEBP–σ identificationscheme. Theactualseries(solidlines)areexpressed in deviations from their respective estimated means. 4.2 The Role of the Great Recession The Great Recession is arguably the defining moment of U.S. post-war economic history and one in which financial and uncertainty shocks appeared to have played an especially prominent role, accordingtoourresults. Hence,itisnaturaltoasktowhatextentaretheaboveresultsinfluencedby the events surrounding this extraordinary period of economic and financial turmoil. To answer this question, we re-estimate our SVAR on the 1973:M1–2007:M12 subsample. Figure 12 summarizes the results of this exercise for the σ–EBP identification scheme, while Figure 13 does the same for the EBP–σ identification scheme. For comparison purposes, the solid lines and the shaded bands in both figures show the results based on the full (1973:M1–2015:M3) sample period. As shown by the dashed lines in panel (a) of Figure 12, the macroeconomic impact of the JLN 27

Figure 12: Economic Effects of Uncertainty and Financial Shocks (Subsample Analysis under the σ–EBP Identification) EBP Industrial production Stock prices Basis points Percent Percent 30 0.4 2 1973-2015 1973-2007 20 0 -0.4 10 -2 -1.2 -4 0 -6 -10 -2.0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (a) Responses of selected variables to a JLN uncertainty shock EBP Industrial production Stock prices Basis points Percent Percent 30 0.4 2 1973-2015 1973-2007 20 0 -0.4 10 -2 -1.2 -4 0 -6 -10 -2.0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (b) Responses to selected variables to an EBP shock Note: Thesolidanddottedlinesinpanel(a)depictmedianresponsesofselectedvariablestotheJLNuncertainty shock of 1 standard deviation (based on the 1973:M1–2015:M3 sample period), while those in panel (b) depict median responses of the same variables to the EBP shock of 1 standard deviation (again based on the 1973:M1– 2015:M3 sample period); both shocks are orthogonalized using the σ–EBP identification scheme. Responses are basedonthesameVARspecificationestimatedovertwosampleperiods: solid=1973:M1–2015:M3;anddotted= 1973:M1–2007:M12. The shaded bands represent the 90-percent pointwise credible sets based on the 1973:M1– 2015:M3 sample period. uncertaintyshocksisinfluencednotablybytheGreatRecession. Excludingthepost-2007datafrom our sample implies a more modest effect of uncertainty shocks on both the industrial production and the stock market. This difference appears to reflect the fact that during this period, the JLN uncertaintyshocksdonotelicitaneconomicallymeaningfultighteningoffinancialconditionsunder the σ–EBP identification. According to panel (b) of Figure 12, the macroeconomic impact of financial shocks is also influenced by the Great Recession, primarily reflecting the fact that the response of the EBP to its own shock is somewhat less persistent over the 1973–2007 period. The reduction in persistence implies a less severe decline in industrial production in response to a financial shock, especially over the first 12 months or so. The economic effect of financial shocks on stock prices is also somewhat attenuated in the immediate aftermath of such a shock. Figure 13 shows the same set of the IRFs under the EBP–σ identification scheme. As shown in 28

Figure 13: Economic Effects of Uncertainty and Financial Shocks (Subsample Analysis under the EBP–σ Identification) EBP Industrial production Stock prices Basis points Percent Percent 30 0.4 2 1973-2015 1973-2007 20 0 -0.4 10 -2 -1.2 -4 0 -6 -10 -2.0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (a) Responses of selected variables to a JLN uncertainty shock EBP Industrial production Stock prices Basis points Percent Percent 30 0.4 2 1973-2015 1973-2007 20 0 -0.4 10 -2 -1.2 -4 0 -6 -10 -2.0 0 6 12 18 24 30 36 0 6 12 18 24 30 36 0 6 12 18 24 30 36 (b) Responses of selected variables to an EBP shock Note: Thesolidanddottedlinesinpanel(a)depictmedianresponsesofselectedvariablestotheJLNuncertainty shock of 1 standard deviation (based on the 1973:M1–2015:M3 sample period), while those in panel (b) depict median responses of the same variables to the EBP shock of 1 standard deviation (again based on the 1973:M1– 2015:M3 sample period); both shocks are orthogonalized using the EBP–σ identification scheme. Responses are basedonthesameVARspecificationestimatedovertwosampleperiods: solid=1973:M1–2015:M3;anddotted= 1973:M1–2007:M12. The shaded bands represent the 90-percent pointwise credible sets based on the 1973:M1– 2015:M3 sample period. panels (a) and (b), when we estimate the VAR using only the data over the 1973–2007 period, the macroeconomic impact of uncertainty and financial shocks on both the real industrial output and the stock market—though economically and statistically still significant—is not as pronounced as that implied by the full sample period. As was the case under the σ–EBP identification scheme, uncertainty shocks under the EBP–σ identification also fail to elicit an economically significant tightening of financial conditions during the 1973–2007 period (panel (a)). By the same token, financial shocks induce, on balance, a less severe and protracted tightening of financial conditions, which accounts for their more limited macroeconomic effects (panel (b)). 29

4.3 Validation of Financial and Uncertainty Shocks The historical variance decomposition presented above clearly indicates that the identified financial and uncertainty shocks were important drivers of fluctuations in economic activity and swings in broad equity prices over the past four decades. At the same time however, a natural question that emerges from this analysis is whether these two types of shocks in fact represent distinct sources of cyclicalfluctuations orwhethertheyaresimplyemblematicoftraditional originsofmacroeconomic instability. To examine this hypothesis more formally, we look at the correlations between the identified financial and uncertainty shocks and other widely cited economic disturbances, all of which are external to our VAR system. At monthly frequency, we consider two types of popular shocks: monetary policy and oil price shocks. To measure unanticipated changes in the stance of monetary policy,werelyonhigh-frequencyfinancialmarketdata. Ourfirstmonetarypolicyshockcorresponds to the “target surprise” proposed by Kuttner (2001), which measures the unexpected change in the target federal funds rate associated with an FOMC announcement.19 In addition, we use changes in the (on-the-run) 2-year Treasury yield over a narrow window bracketing FOMC announcements as monetary policy surprises, which provide a more complete characterization of the unanticipated changes in the stance of policy, according to Gilchrist et al. (2015).20 Wealsoconsidertwomonthlymeasuresofoilpriceshocks. Thefirstsetofoilshockscorresponds to residuals from an AR(1) model of the log-difference of the real price of the WTI crude. The second measure of oil supply shocks is from Killian (2009), who employs a SVAR-based approach to identify the underlying demand and supply shocks in the global crude oil market. At quarterly frequency, we concern ourselves with technology and fiscal shocks. The first set of technology shocks corresponds to shocks to labor productivity identified by Mertens and Ravn (2011a) using a SVAR-based approach, which are orthogonal to tax shocks derived from the narrative approach of Romer and Romer (2010). As an alternative proxy, we also use residuals from an AR(1) model of the log-difference in the utilization-adjusted total factor productivity (TFP), which attempts to adjust measured TFP for a range of non-technological factors that can drive a wedge between TFP and technology (Basu et al., 2006). Onthefiscalfront, weconsiderthesurprisetaxpolicychangesfromMertens and Ravn(2011b), which are based on the narrative approach of Romer and Romer (2010); the anticipated tax policy changes of Leeper et al. (2013); and the unanticipated changes in the expected present value of government spending in response to military events from Owyang et al. (2013). All told, our list 19Specifically,theunanticipatedchangeinthefundsrateiscalculatedasthechange—withminoradjustments—in thecurrent-monthfederalfundsfuturescontractrateina30-minutewindow(10minutesbeforeto20minutesafter) around the FOMC announcement; see Kuttner (2001) for details. 20As emphasized by Gu¨rkaynak et al. (2005), using solely the target surprises to characterize the unanticipated changes in the stance of monetary policy is incomplete because such an approach omits the effect of changes in the futurepolicyratesthatareindependentoftheshocktothecurrenttargetrateandwhicharecloselyassociatedwith the FOMC statements that accompany changes in the target rate. As shownbyGilchrist et al. (2015), however, the first-order effects of monetary policy actions can be summarized adequately by the intraday changes in the 2-year nominal Treasury yield bracketing FOMC announcements. 30

Table 4: Correlations Between Uncertainty, Financial, and Other External Shocks Identification Scheme σ–EBP EBP–σ Uncertainty Financial Uncertainty Financial (a) Monthly External Shocks FFR target surprisesa 0.03 0.03 0.02 0.04 2-year Treasury yield surprisesb 0.08 0.08 0.05 0.09 Real price of oil shocksc −0.07 0.05 0.07 −0.05 Oil supply shocksd 0.05 0.02 −0.00 0.06 (b) Quarterly External Shocks Technology shockse −0.08 −0.08 −0.05 −0.11 TFP growth shocksf −0.16∗∗ 0.09 0.13 −0.12 Unanticipated tax shocksg −0.03 −0.04 −0.03 −0.07 Anticipated tax shocksh −0.06 −0.08 0.05 −0.08 Defense spending shocksi −0.05 0.08 −0.10 −0.02 Note: The entries in the table denote the pairwise correlations between the specified external shock and the financial (EBP) and uncertainty (JLN) shocks identified under the σ–EBP and EBP–σ identification schemes. Sampleperiodformonthlyfinancialanduncertaintyshocksis1975:M1to2015:M3;therespectivequarterlyshock series are the quarterly averages of the corresponding monthly values. * p<.10; ** p<.05; and *** p<.01. aUnanticipated changes in the target federal funds rate in the 30-minute window bracketing FOMC announcements; see Kuttner (2001) (1992:M2–2015:M3, T =278). bChangesinthe(on-the-run)2-yearTreasuryyieldinthe30-minutewindowbracketingFOMCannouncements; see Gilchrist et al. (2015) (1994:M2–2015:M3, T =254). cResiduals from a first-order autoregressive model of the log-difference in the real price of the WTI crude (1975:M1–2015:M3, T =483). dCrude oil supply shock from Killian (2009) (1975:M1–2007:M12 T =396). eTechnology shocks from Mertens and Ravn (2011a) (1975:Q1–2006:Q4, T =128). fResiduals from a first-order autoregressive model of the log-difference in the utilization-adjusted total factor productivity; see Basu et al. (2006) (1975:Q1–2015:Q1, T =161). gUnanticipated tax shocks from Mertens and Ravn (2011b) (1975:Q1–2006:Q4, T =128). hAnticipated tax changes from Leeper et al. (2013) (1975:Q1–2006:Q4, T =128). iDefense spending shocks from Owyang et al. (2013) (1975:Q1–2013:Q4, T =156). of external shocks spans the space of disturbances commonly considered to be the most important drivers of aggregate fluctuations. The pairwise correlations between these external shocks and our identified financial and uncertainty shocks are reported in Table 4. As evidenced by the entries in the table, there appears to be no systematic contemporaneous association between financial and uncertainty shocks and other typical macroeconomic disturbances under either identification scheme. Virtually all pairwise correlations are statistically indistinguishable from zero and all of them are very small in economic terms. These results provide strong corroborative evidence for the hypothesis that the identified financial and uncertainty shocks represent distinct sources of economic disturbances, independent of traditional business cycle shocks. 31

5 Conclusion This paper employs the penalty function approach to jointly identify shocks to financial conditions and economic uncertainty and to trace out the impact of these two types of disturbances on the economy. The two structural innovations are identified using the criterion that each shock should maximize the impulse response of its respective target variable over a pre-specified horizon, an approach that allows us to orthogonalize shocks that have otherwise very similar qualitative effects on the economy. Intuitively, we assume that a persistent tightening of financial conditions is due to an adverse financial shock, whereas prolonged periods of elevated economic uncertainty are driven by uncertainty shocks. Our identification strategy also does not rule out a contemporaneous response of financial conditions to uncertainty shocks or vice versa. We implement this approach in the context of a standard monetary VAR, augmented with the excessbondpremium—anindicatorofthetightnessoffinancialconditions—andvariousproxiesfor economic uncertainty. Our results indicate that financial shocks have a significant adverse effect on economic outcomes and that such shocks were an important source of cyclical fluctuations since the mid-1980. Inaddition, uncertainty shocks, especiallythose impliedby uncertainty proxies based on real economic data, are also an important source of macroeconomic disturbances; such uncertainty shocks appear to have a significant more pronounced economic impact in situations where they elicit a concomitant tightening of financial conditions. All told, the evidence presented in this paper provides a considerable support for the hypothesis that financial and uncertainty shocks have both played a significant role in business cycle fluctuations over the past four decades. This finding is buttressed importantly by the evidence that our identified financial and uncertainty shocks are uncorrelated with external instruments that serve as proxies for a range of traditional cyclical disturbances. In fact, according to our results, the combination of financial and uncertainty shocks fully accounts for the severe contraction in real industrial output and the collapse of the stock market during the Great Recession. References Adrian, T. and H. S. Shin (2010): “Liquidity and Leverage,” Journal of Financial Intermediation, 19, 418–437. Arellano, C., Y. Bai, and P. J. Kehoe (2012): “Financial Markets and Fluctuations in Volatility,” Staff Report No. 466, Federal Reserve Bank of Minneapolis. Arias, J., J. Rubio-Ramirez, and D. Waggoner(2013): “InferenceBasedonSVARsIdentified with Sign and Zero Restrictions,” Working Paper, Dept. of Economics, Duke University. Bachmann, R., S. Elstner, and E. Sims (2013): “Uncertainty and Economic Activity: Evidence From Business Survey Data,” American Economic Journal: Macroeconomics, 5, 217–249. Baker, S. R., N. Bloom, and S. J. Davis (2015): “Measuring Economic Policy Uncertainty,” NBER Working Paper No. 21633. 32

Barsky, R. B. and E. R. Sims(2011): “NewsShocksandBusinessCycles,” Journal of Monetary Economics, 58, 273–289. Bartels, R. (1982): “The Rank Version of von Neumann’s Ratio Test for Randomness,” Journal of the American Statistical Association, 77, 40–46. Basu, S., J. G. Fernald, and M. S. Kimball (2006): “Are Technology Improvements Contractionary?” American Economic Review, 96, 1418–1448. Bekaert, G., M. Hoerova, and M. Lo Duca (2013): “Risk, Uncertainty, and Monetary Policy,” Journal of Monetary Economics, 60, 771–788. Bloom, N. (2009): “The Impact of Uncertainty Shocks,” Econometrica, 77, 623–685. ——— (2014): “Fluctuations in Uncertainty,” Journal of Economic Perspectives, 28, 153–176. Bloom, N., M. Floetotto, N. Jaimovich, I. Saporta-Eksten, and S. Terry (2012): “Really Uncertain Business Cycles,” NBER Working Paper No. 18245. Boivin, J., M. Giannoni, and D. Stevanovic´ (2013): “Dynamic Effects of Credit Shocks in a Data-Rich Environment,” Staff Report No. 2013, Federal Reserve Bank of New York. Brunnermeier, M. K. and Y. Sannikov (2014): “A Macroeconomic Model with a Financial Sector,” American Economic Review, 104, 379–421. Caldara, D. and C. Kamps(2012): “TheAnalyticsofSVARs: AUnifiedFrameworktoMeasure FiscalMultipliers,”FinanceandEconomicsDiscussionSeriesPaperNo.2012-20,FederalReserve Board. Campbell, J. Y., M. Lettau, B. G. Malkiel, and Y. Xu (2001): “Have Individual Stocks Become More Volatile? An Empirical Exploration of Idiosyncratic Risk,” Journal of Finance, 56, 1–44. Carhart, M. M. (1997): “On Persistence in Mutual Fund Performance,” Journal of Finance, 52, 57–82. Christiano, L. J., R. Motto, and M. Rostagno (2014): “Risk Shocks,” American Economic Review, 104, 27–65. D’Amico, S. and A. Orphanides (2008): “Uncertainty and Disagreement in Economic Forecasting,” Finance and Economics Discussion Series Paper No. 2008-56, Federal Reserve Board. Del Negro, M. and F. Schorfheide (2011): “Bayesian Macroeconometrics,” in Handbook of Bayesian Econometrics, ed. by J. Geweke, G. Koop, and H. van Dijk, Oxford University Press, Chapter 7. Elliott, G. and U. K. Mu¨ller (2006): “Efficient Tests for General Persistent Time Variation in Regression Coefficients,” Review of Economic Studies, 73, 907–940. Fajgelbaum, P., E. Schaal, and M. Taschereau-Dumouchel (2014): “Uncertainty Traps,” NBER Working Paper No. 19973. Fama, E. F. and K. R. French(1992): “TheCross-SectionofExpectedStockReturns,”Journal of Finance, 47, 427–465. 33

Faust, J. (1998): “The Robustness of Identified VAR Conclusions About Money,” Carnegie- Rochester Conference Series on Public Policy, 49, 207–244. Faust, J., S. Gilchrist, J. H. Wright, and E. Zakrajˇsek (2013): “Credit Spreads as Predictors of Real-Time Economic Activity: A Bayesian Model-Averaging Approach,” Review of Economics and Statistics, 95, 1501–1519. Gilchrist, S., D. Lo´pez-Salido, and E. Zakrajˇsek (2015): “Monetary Policy and Real Borrowing Costs at the Zero Lower Bound,” American Economic Journal: Macroeconomics, 7, 77–109. Gilchrist, S., J. W. Sim, and E. Zakrajˇsek (2014): “Uncertainty, Financial Frictions, and Investment Dynamics,” NBER Working Paper No. 20038. Gilchrist, S., V. Yankov, and E. Zakrajˇsek (2009): “Credit Market Shocks and Economic Fluctuations: Evidence From Corporate Bond and Stock Markets,” Journal of Monetary Economics, 56, 471–493. Gilchrist, S. and E. Zakrajˇsek (2012): “Credit Spreads and Business Cycle Fluctuations,” American Economic Review, 102, 1692–1720. Gu¨rkaynak, R. S., B. Sack, and E. Swanson(2005): “DoActionsSpeakLouderThanWords? TheResponseofAssetPricestoMonetaryPolicyActionsandStatements,”InternationalJournal of Central Banking, 1, 55–93. Hamilton, J. D. (1983): “Oil and the Macroeconomy Since World War II,” Journal of Political Economy, 91, 228–248. Hansen, N., S. D. Mu¨ller, and P. Koumoutsakos (2003): “Reducing the Time Complexity of the Derandomized Evolution Strategy With Covariance Matrix Adaption (CMA-ES),” Evolutionary Computation, 11, 1–18. Jurado, K., S.C.Ludvigson, andS.Ng(2015): “MeasuringUncertainty,”AmericanEconomic Review, 105, 1177–1216. Killian, L. (2009): “Not All Oil Price Shocks are Alike: Disentangling Demand and Supply Shocks in the Crude Oil Market,” American Economic Review, 99, 1053–1069. Kurmann, A. and C. Otrok (2013): “News Shocks and the Slope of the Term Structure of Interest Rates,” American Economic Review, 103, 2612–2632. Kuttner, K. N. (2001): “Monetary Policy Surprises and Interest Rates: Evidence From the Fed Funds Futures Market,” Journal of Monetary Economics, 47, 523–544. Leeper, E. M., T. B. Walker, and S. S. Yang (2013): “Fiscal Foresight and Information Flows,” Econometrica, 81, 1115–1145. Lindsey, D. E., A. Orphanides, and R. H. Rasche (2005): “The Reform of October 1979: How It Happened and Why,” Review, Federal Reserve Bank of St. Louis, 87, 187–235. Lo´pez-Salido, D., J. C. Stein, and E. Zakrajˇsek (2016): “Credit-Market Sentiment and the Business Cycle,” NBER Working Paper No. 21879. 34

Ludvigson, S. C., S. Ma, and S. Ng (2016): “Uncertainty and Business Cycles: Exogenous Impulse or Endogenous Response?” Working Paper, Dept. of Economics, New York University. Mertens, K. and M. O. Ravn (2011a): “Technology-Hours Redux: Tax Changes and the Measurement of Technology Shocks,” in NBER International Seminar on Macroeconomics, Chicago: The University of Chicago Press, vol. 7, 41–76. ——— (2011b): “Understanding the Aggregate Effects on Anticipated and Unanticipated Tax Policy Shocks,” Review of Economic Dynamics, 14, 27–54. Mountford, A. and H. Uhlig (2009): “What Are the Effects of Fiscal Policy Shocks?” Journal of Applied Econometrics, 24, 960–992. Newey, W. K. and K. D. West (1987): “A Simple, Positive Semi-Definite, Heteroskedasticityand Autocorrelation-Consistent Covariance Matrix,” Econometrica, 55, 703–708. ——— (1994): “Automatic Lag Selection in Covariance Matrix Estimation,” Review of Economic Studies, 61, 631–653. Orlik, A. and L. L. Veldkamp (2014): “Understanding Uncertainty Shocks and the Role of Black Swans,” NBER Working Paper No. 20445. Owyang, M. T., V. A. Ramey, and S. Zubairy (2013): “AreGovernmentSpendingMultipliers Greater During Periods of Slack: Evidence from Twentieth-Century Historical Data,” American Economic Review, 103, 129–134. Popescu, A. and F. Smets (2010): “Uncertainty, Risk-Taking, and the Business Cycle in Germany,” CESifo Economic Studies, 56, 596–626. Romer, C.D.andD.H.Romer(2010): “TheMacroeconomicEffectsofTaxChanges: Estimates Based on a New Measure of Fiscal Shocks,” American Economic Review, 100, 763–801. Rousseeuw, P. J. and C. Croux (1993): “Alternatives to Median Absolute Deviation,” Journal of the American Statistical Association, 88, 1273–1283. Stock, J. H. and M. W. Watson(2012): “DisentaglingtheChannelsofthe2007–09Recession,” Brookings Papers on Economic Activity, 44, 81–135. Uhlig, H. (2003): “What Moves Real GNP?” Working Paper, Humboldt University Berlin. ——— (2005): “What Are the Effects of Monetary Policy on Output? Results From an Agnostic Identification Procedure,” Journal of Monetary Economics, 52, 381–419. Van Nieuwerburgh, S. and L. L. Veldkamp (2006): “Learning Asymmetries in Real Business Cycles,” Journal of Monetary Economics, 53, 753–772. Zellner, A. (1971): An Introduction to Bayesian Inference in Econometrics, New York: John Wiley & Sons Inc. 35

Appendices – For Online Publication A Data Appendix This appendix provides a brief description of the six uncertainty proxies used in the analysis. RVOL measure of uncertainty. The realized equity volatility is calculated as the (annualized) standard deviation of the daily value-weighted total market (log) return from the Center for Research in Security Prices (CRSP) database. To mitigate the effects of large daily swings in equity prices—many of which occurred during the 2007–08 financial crisis—we use the robust estimator of scale proposed by Rousseeuw and Croux (1993) to calculate the monthly standard deviation of daily stock returns. IVOL measure of uncertainty. This uncertainty proxy is a monthly version—with minor modifications—of the quarterly measure proposed by Gilchrist et al. (2014). First, we extracted dailystockreturnsforallU.S.nonfinancialcorporationswithatleast500tradingdaysofdata. This selectioncriterionyieldedapanelof14,856firmsovertheperiodfromOctober1, 1972toMarch31, 2015. To ensure that our results were not driven by a small number of extreme observations, we eliminated all firm/day observations with a daily absolute return in excess of 50 percent. The estimate of uncertainty is based on the following three-step procedure. First, we remove the forecastable variation in daily excess returns using the standard (linear) factor model: (R −r f ) = α +β′f +u , (A-1) itd td i i td itd whereiindexesfirmsandt ,d = 1,...,D ,indexestradingdaysinmontht. Inequation(A-1),R d t itd denotesthe(total)dailyreturnoffirmi,r f istherisk-freerate,andf isavectorofobservablerisk td td factors. In implementing the first step, we employ a 4-factor model—namely, the Fama and French (1992) 3-factor model, augmented with the momentum risk factor proposed by Carhart (1997). Inthesecondstep,weusetherobustscaleestimatorofRousseeuw and Croux(1993)tocalculate the monthly firm-specific standard deviation of the daily idiosyncratic returns—that is, the OLS residualsuˆ fromequation(A-1). Denotedbyσ ,thisprovidesuswithanestimateoftime-varying itd it equityvolatilityforfirmi,ameasurethatispurgedoftheforecastablevariationinexpectedreturns. In the third step, we assume that the firm-specific measure of uncertainty σ follows an autoreit gressive process of the form: 3 logσ = γ +δ t+ ρ logσ +v +ǫ , it i i k i,t−k t it Xk=1 where γ denotes a firm fixed effect intended to control for the cross-sectional heterogeneity in σ , i it while the firm-specific term δ t captures secular trends in the idiosyncratic risk of publicly traded i U.S. nonfinancial firms documented by Campbell et al. (2001). The IVOL uncertainty proxy corresponds to the sequence of estimated time fixed effects vˆ, t t = 1,...,T, which captures shocks to idiosyncratic volatility that are common to all firms. As emphasized by Gilchrist et al. (2014), the presence of the common variation in the volatility of idiosyncratic equity returns is essential because if fluctuations in idiosyncratic volatility were themselves entirely idiosyncratic, the macroeconomic impact of such uncertainty shocks should wash out in the aggregate. 36

VXO measure of uncertainty. The VXO uncertainty proxy corresponds to the option-implied volatility calculated from a hypothetical at the money S&P 100 option 30 days to expiration. JLN measure of uncertainty. To measure economic uncertainty, Jurado et al. (2015) fit a factor model to a large cross section of macroeconomic and financial time series and use the estimated model to generate forecasts of all the series. Next, they assume that forecast errors of each individual series follow a univariate stochastic volatility process, and the cross-sectional average of these processes for a subset of variables pertaining to real economic activity becomes an estimate of macroeconomic uncertainty. BBDmeasureofuncertainty. Thisindexofeconomicpolicyuncertainty(EPU)isbasedonthe frequency of newspaper references to policy-related economic uncertainty; see Baker et al. (2015) for details. BES measure of uncertainty. To measure economic uncertainty, Bachmann et al. (2013) construct a measure of forecast dispersion using the Philadelphia Fed’s Business Outlook Survey. This monthly survey of manufacturing firms contains qualitative information on the current state of firms’ business conditions and their expectations of future business conditions. Bachmann et al. (2013) focus on two questions: 1. General Business Conditions: Whatisyourevaluationofthelevelofgeneralbusinessactivity six months from now versus [current month]? Answers: decrease; no change; increase. 2. Company Business Indicators: Shipments six months from now versus [current month]? Answers: decrease; no change; increase. The qualitative survey responses to both questions are coded into three discrete numerical categories: −1 = decrease; 0 = no change; and 1 = increase. For each question, Bachmann et al. (2013) define Frac+ as the (unweighted) proportion of firms t that responded with “increase” at time t and Frac− as the (unweighted) proportion of firms that t responded with “decrease” at time t. The cross-sectional forecast dispersion for any of the two questions is then computed according to D = Frac++Frac−− Frac+−Frac− 2 . t t t t t q (cid:0) (cid:1) The measure of time-varying business-level uncertainty used in this paper corresponds to the crosssectional forecast dispersion for the question pertaining to general business conditions. Figure A-1 shows the the time paths of the three uncertainty proxies that rely entirely on financial market data (RVOL, IVOL, and VXO), while Figure A-2 shows the time-series evolution of uncertainty measures not based exclusively on financial market data (JLN, BBD, and BES). 37

Figure A-1: Economic Uncertainty (Proxies Based on Financial Market Data) Percent 100 80 60 40 20 0 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 (a) Realized stock market volatility (RVOL) Percent 140 120 100 80 60 40 20 0 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 (b) Idiosyncratic stock market volatility (IVOL) Percent 80 60 40 20 0 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 (c) Option-implied volatility on the S&P 100 stock futures index (VXO) Note: Thepanelsofthefigureshowthethreedifferentmeasuresofeconomicuncertaintybasedonexclusivelyon stock market data. The shaded vertical bars denote the NBER-dated recessions. 38

Figure A-2: Economic Uncertainty (Proxies Not Based Exclusively on Stock Market Data) Index 1.3 1.2 1.1 1.0 0.9 0.8 0.7 0.6 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 (a) Uncertainty index implied by forecast errors (JLN) Index 300 250 200 150 100 50 0 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 (b) Economic policy uncertainty index (BBD) Standard deviations 1.0 0.8 0.6 0.4 0.2 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 (c) Uncertainty based on forecast dispersion (BES) Note: The panels of the figure show the three different measures of economic uncertainty that are not based on exclusively on stock market data. The shaded vertical bars denote the NBER-dated recessions. 39

Table A-1: Selected Characteristics of Different Economic Uncertainty Proxies Summary Statistic Uncertainty Proxy CV α α α q RVN 1 2 3 LL RVOLa 0.61 0.70 0.13 0.03 −11.48 4.18∗∗∗ IVOLb 0.38 0.63 0.18 0.05 −16.00 0.08 VXOc 0.40 0.83 0.02 0.06 −12.59 2.58∗∗ JLNd 0.12 0.99 −0.75 0.17 −11.83 0.19 BBDe 0.36 0.71 0.04 0.06 −8.98 1.24 BESf 0.14 0.80 0.26 0.05 −13.69 −0.12 Pairwise Correlations Uncertainty Proxy RVOL IVOL VXO JLN BBD BES RVOL 1.00 0.62∗∗∗ 0.82∗∗∗ 0.47∗∗∗ 0.50∗∗∗ 0.13∗∗∗ IVOL 1.00 0.57∗∗∗ 0.07 0.39∗∗∗ 0.23∗∗∗ VXO 1.00 0.55∗∗∗ 0.49∗∗∗ 0.19∗∗∗ JLN 1.00 0.31∗∗∗ 0.08∗ BBD 1.00 0.02 BES 1.00 Note: The entries in the table denote the specified summary statistic: CV = coefficient of variation; α = k partial autocorrelation at lag k; q LL = the Elliott and Mu¨ller (2006) test statistic of the null hypothesis that the autoregressive coefficients from an AR(3) model are constant over time; and RVN = the Bartels (1982) test statisticofthenullhypothesisthattheOLSresidualsfromanAR(3)modelaredistributedrandomly. ∗ p<.10, ∗∗ p<.05, and ∗∗∗ p<.01. aRealized equity volatility (1973:M1–2015:M3, T =507). bIdiosyncratic equity volatility based on Gilchrist et al. (2014) (1973:M1–2015:M3, T =507). cOption-implied volatility on the S&P 100 stock futures index (1986:M1–2015:M3, T =351). dUncertainty measure based on Jurado et al. (2015) (1973:M1–2015:M3, T =507). eUncertainty measure based on Baker et al. (2015) (1985:M1–2015:M3, T =363). fUncertainty measure based on Bachmann et al. (2013) (1973:M1–2011:M12, T =468). The top panel of Table A-1 provides some summary statistics for the six uncertainty measures used in our analysis. Not too surprising, the coefficients of variation for the uncertainty proxies derivedfromequityvaluationstendtobelargercomparedwiththoseofuncertaintyproxiesthatare not based on financial market data. As evidenced by the partial correlations, all measures exhibit significant positive first-order autocorrelation; the degree of serial dependence, however, dies off very quickly in every case. Letting each series follow an AR(3) process, there is no evidence of parameter instability in the autoregressive coefficients, according to the Elliott and Mu¨ller (2006) Quasi-Local Level test. Moreover, in most cases, the resulting residuals appear to be distributed randomly. Asshownbytheentriesinthebottompanel,thesixseries,ingeneral,exhibitsignificantpositive contemporaneous correlation. As expected, the highest degree of comovement is between the three uncertainty proxies based on the stock market data (RVOL, IVOL, and VXO). The other three measures(JLN,BBD,andBES)arealsopositivelycorrelatedwiththeirequity-basedcounterparts, though to a noticeably lesser extent. The pairwise correlations between JLN, BBD, and BES, in contrast, are appreciably lower. 40

B Estimation Appendix As discussed in the main text, we follow Del Negro and Schorfheide (2011) to implement the Minnesota prior on the reduced-form coefficients {B,Ω} through dummy observations. The Minnesota priorisspecifiedconditionalonfivehyper-parameters, denotedbyλ ,...,λ . Thehyper-parameter 1 5 λ controls the overall tightness of the prior; λ scales the prior standard deviation of the coeffi- 1 2 cients associated with the lags of the endogenous variables; λ controls the prior on Ω; λ controls 3 4 the prior for the intercept; and λ controls the prior correlation between the coefficients. 5 Table B-1: Values for the Hyper-Parameters of the Minnesota Prior Uncertainty Proxy (Sample Period) λ λ λ λ 1 2 4 5 JLN (1973:M1–2015:M3) 1.07 1.79 2.41 4.29 JLN (1973:M1–2007:M12) 1.17 1.78 2.36 4.16 RVOL (1973:M1–2015:M3) 1.64 1.79 2.46 2.52 IVOL (1973:M1–2015:M3) 1.70 1.67 2.48 2.20 VXO (1986:M1–2015:M3) 1.34 2.39 9.14 2.60 BBD (1985:M1–2015:M3) 1.83 2.43 5.37 2.18 BES (1973:M1–2011:M12) 1.76 1.65 2.54 2.00 Note: The entries in the table denote values for the specified hyper-parameter. We select the hyper-parameters and the VAR lag length p by maximizing the marginal data density. To perform this optimization, we use the version of the CMA-ES evolutionary algorithm proposed by Hansen et al. (2003). The only hyper-parameter we do not select optimally is λ , 3 which we set equal to 1. Table B-1 tabulates the model-specific hyper-parameters that maximize the marginal data density. The optimal VAR length for all models is p = 6. We use the Direct Monte Carlo Sampling algorithm of Zellner (1971) to obtain draws of the reduced-form parameters from the posterior distribution. 41