Predictability of Growth in Emerging Markets: Information in Financial Aggregates

K.7 Predictability of Growth in Emerging Markets: Information in Financial Aggregates Banegas, Ayelen Please cite paper as: Banegas, Ayelen (2016). Predictability of Growth in Emerging Markets: Information in Financial Aggregates. International Finance Discussion Papers 1174. http://dx.doi.org/10.17016/IFDP.2016.1174 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1174 July 2016

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1174 July 2016 Predictability of Growth in Emerging Markets: Information in Financial Aggregates Ayelen Banegas NOTE: International Finance and Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors.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 IDFPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from Social Science Research Network electronic library at www.ssrn.com. The views expressed in this paper are solely the responsibility of the author and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System.

Ayelen Banegas∗ Federal Reserve Board July 2016 Abstract This paper tests for predictability of output growth in a panel of 22 emerging market economies. We use pooled panel data methods that control for endogeneity and persistence in the predictor variables to test the predictive power of a large set of financial aggregates. Results show that stock returns, the term spread, default spreads and portfolio investment flows helppredictoutputgrowthinemergingmarkets. Wealsofindevidencethatsuggeststhatglobal aggregates such as the performance of commodity markets, a cross-sectional firm size factor, and returns on the market portfolio contain information about the future state of the economy. We benchmark our results against those from the U.S. and find that there are differences in the ability of financial markets in predicting economic growth. Our results generalize to emerging markets previous findings in the empirical macro-finance literature on the linkages between financial market performance and the real economy. JEL Classification: E44, G15 Keywords: Output growth predictability, emerging markets, leading indicators, financial variables ∗Federal Reserve Board, Division of Monetary Affairs. Washington, D.C. 20551. Tel.: +1 202 452 3767. Maria.A.Banegas@frb.gov

1 Introduction The recent financial crisis triggered a revival of the empirical and theoretical macro-finance literature on the links between financial markets and the real economy. Among the questions that have been at the center of the debate is whether we could have predicted the crisis. As a result, in recent years, the debate on leading indicators as predictors of future economic activity has reemerged.1 Although much of the earlier empirical macro-finance literature on output growth predictability focuses on developed countries (i.e.Harvey (1988); Estrella and Hardouvelis (1991); Weber (1998); Estrella and Mishkin (1998); Hamilton and Kim (2002); Stock and Watson (1989, 2003); and Gilchrist, Yankov, and Zakrajsek (2009)); not much work studies the predictive relationship between financial markets and output growth in emerging markets.2 This is an important under-explored area of research as emerging markets have become a central player in the global economy. 3 Inthispaper, wecontributetotheempiricalmacro-financeliteraturebystudyingthepredictive dynamicsofeconomicgrowthandevaluatingtheextenttowhichfinancialaggregatescanbeusedas leading indicators of economic growth in emerging markets, without taking a stand on whether our finding indicate a causal relation or whether they reflect information in forward-looking financial aggregates. Specifically, we test whether stock returns, capital flows, interest rates, valuation measures, commodity prices, and a set of well-established risk factors, among other variables, contain information about the future state of the economy. To address our question, we construct a dataset for 22 emerging market countries that were selected based on Standard and Poor’s classification system. Our sample includes a large set of country-specific and global macroeconomic and financial variables, such as real GDP growth, interest rates, capital flows, valuation measures, 1In particular, there is a large body of work on early warning indicators. See Frankel and Saravelos (2012) for a detailed literature review. 2Mauro(2003)focusesonthecorrelationoflaggedstockreturnsandoutputgrowthinemergingmarketsandfinds evidence of a positive relationship in several countries. 3Overthepastfewyearswehaveseenemergingmarketsoutperformingandrecoveringfasterfromtherecentcrisis than developed economies. Many of the more succesful countries have undertaken economic and political reforms in the pre-crisis period that resulted in more favorable business conditions and stronger government balance sheets. Thesereformsincluded,butwerenotlimitedto,fiscalprogramsaimedatcontrollingpublicdeficits,tradeagreements, financial system regulations and monetary policy that led to lower inflation rates.

andcommodityindices, coveringtheperiodfrom1992to2010.4 Totestforpredictabilityofoutput growth we use pooled panel data methods that control for endogenous and persistent predictor variables. An extensive amount of literature documents the effects of endogeneity and persistence ontime-seriespredictiveregressions. Stambaugh(1999)showsthatOLSestimatesarebiasedinthe presence of endogenous and persistent regressors, inhibiting normal inference. Hjalmarsson (2010) finds that these econometric issues also apply to panel data forecasting models when fixed effects are included and proposes a robust estimator that corrects for the Stambaugh-bias. Even though Stambaugh-bias has been discussed exclusively in the return predictability literature, we find that it can also arise in the context of output growth predictability. In particular, valuation ratios such as the dividend yield and the price-earnings ratio, as well as the market capitalization ratio, prove to be highly persistent and endogenous. For this reason, we follow Hjalmarsson’s methodology to correct for these features when testing for output growth predictability. Moreover, as part of our empirical analysis, we test for predictability at different forecasting horizons by applying reverse regression methods proposed by Hodrick (1992) and extended by Wei and Wright (2010). Our results show that stock returns, capital flows, default spreads, the term spread, and returns on commodity prices can help predict output growth in emerging market economies. In particular, large capital inflows predict subsequent high GDP growth as does a high term spread. Conversely, higher default spreads on emerging market government debt signals lower future GDP growth. Our findings also suggest that some variables such as the term spread, stock returns, and capital flows contained higher predictive power during the ’90s, while the short interest rate and portfolio flows provemoreusefulduringthepastdecade. Furthermore, forregional-levelpanels, wefindthatLatin America exhibits the strongest evidence of output growth predictability across regions. We also extend previous findings on the links between output growth and Fama and French (1993) risk factors (i.e. Liew and Vassalou (2000)). In particular, we test whether book-to-market, size, and the market portfolio contain information about the future state of the economy. We find strong evidence of the predictive power of the size factor and returns on the market portfolio to be consistent with small firms being more sensitive to the business cycle than large firms (see, for example, Perez-Quiros and Timmermann (2001)). We compare our results with empirical findings in the US. Given the characteristics of emerging market economic, financial, and institutional systems, we can expect the transmission channels 4We selected the period under analysis based on data quality restrictions.

between financial markets and the real economy to work differently because of liquidity or credit market frictions, among other factors, in emerging markets. Overall, we find that there are differences in the ability of financial markets to predict future output growth. Variables that prove to convey information about the future state of the economy in developing countries, such as the term spread, stock returns, and commodity prices, lack any predictive power in the context of the US. Further, while we find a positive relationship between capital and portfolio flow ratios and future output in these transitional economies, results for the US suggest that increases in these ratios are followed by subsequent declines in GDP growth. These findings can be interpreted as a flight to quality effect where in periods of global turmoil investors move their capital away from riskier investments to apparently safer allocations in developed markets. In summary, by identifying a set of financial aggregates that convey information about future output growth, our analysis sheds some light on the relationship between financial markets performance and the state of the economy in developing economies. These findings can be useful to policymakers and market participants who try to understand and predict the business cycles of emerging markets. The paper is organized as follows: section 2 details the predictive regression methodology, section3describesourdataset, section4presentsourempiricalfindings, section5providesrobustness analysis and section 6 presents conclusions. 2 Estimation An extensive amount of literature documents the effects of endogeneity and persistence on timeseriespredictiveregressions. Stambaugh(1999)showsthatOLSestimatesarebiasedinthepresence of endogenous and persistent regressors, inhibiting normal inference. Hjalmarsson (2010) finds that these econometric issues also apply to panel data forecasting models when fixed effects are included and proposes a robust estimator that corrects for the Stambaugh-bias. Even though Stambaughbias has been discussed exclusively in the return predictability literature, as shown in tables 2 and 3, we find that it can also arise in the context of output growth predictability. In particular, there is evidence of high persistence and endogeneity when using the capitalization ratio, the dividend yield and the price earning ratio as regressors. Also, the term spread and short interest rate are

highly persistent. However, cross-correlations for these predictors lack statistical significance in most of the countries. For this reason, we follow Hjalmarsson’s methodology to correct for these features when testing for output growth predictability. Our panel data framework is represented by the following system of equations: y = α +β(cid:48)x +u (1) i,t i i,t−1 i,t x = A x +v i,t i i,t−1 i,t A = I +C /T i i wherey istherealGDPgrowthforcountryiattimet,x isanmx1vectorofpredictorvariables, i,t i,t u arecountry-specificinnovationsassumedtobemartingaledifferencesequenceswithfinitefourth i,t moments, A is an mxm matrix representing the near unit root assumption and C is the local-toi i unity parameter. The model allows for endogeneity by letting u and v be contemporaneously i,t i,t correlated and introduces persistency by defining the dependent variable, x , as an autoregressive i,t process of order 1 with roots being local to unity. By letting C differ across countries, we allow i for the time series to have different persistence levels. We find this property to be of relevance in our analysis because we are considering a heterogenous panel of countries, and can therefore expect regressors to have different cross-sectional characteristics. 5 Using panel data methodology results in more precise estimates than the analogous individual time-series regressions. Furthermore, even in the case where the slope coefficients differ across countries, the pooled estimator will provide information about the average slope coefficient. Also, pooling the data allows us to evaluate whether ourgroup of developing countries possessescommon characteristics. This methodology is of particular relevance to investors who consider investing in emergingmarketsasanassetclass, ratherthaninindividualstocks. Forexample, investorsgaining exposure to global emerging markets through mutual fund investing. In our baseline setting we consider three different scenarios for pooled estimation presented in Hjalmarsson’s (2010) study. First, the naive case where we estimate a standard pooled estimator. 5The near unit root assumption is preferred to a unit root set up because it is a less restrictive condition on the regressors when modelling non-stationary series. Also, as Hjalmarsson (2010) points out, under the near unit root assumption the degree of persistence in the predictive regressors will be carried over to the asymptotic distribution. This allows the small sample distribution to correspond with that predicted by asymptotic distribution theory.

We then allow for the possibility of individual effects by letting α differ across countries. Finally, i we consider a recursively demeaned estimator that corrects for the bias found in the fixed effects procedure. Thestandardpooledestimatordoesnotallowforindividualeffectsandisgivenbythefollowing equation: (cid:32) n T (cid:33)−1(cid:32) n T (cid:33) (cid:88)(cid:88) (cid:88)(cid:88) βˆ = x˜ x˜(cid:48) y˜ x˜ (2) Pool i,t−1 i,t−1 i,t i,t−1 i=1 t=1 i=1 t=1 n T 1 (cid:88)(cid:88) y˜ = y − y i,t i,t i,t nT i=1 t=1 n T 1 (cid:88)(cid:88) x˜ = x − x i,t i,t i,t nT i=1 t=1 Despite the fact that from an economic perspective we might think that country-specific intercepts should be considered, it is worth evaluating this simple case because of the econometric properties of this estimator. In particular, this estimator is asymptotically normally distributed therefore allowing for normal inference. The fixed effects estimator is defined by the following equation: (cid:32) n T (cid:33)−1(cid:32) n T (cid:33) (cid:88)(cid:88) (cid:88)(cid:88) βˆ = x x(cid:48) y x (3) FE i,t−1 i,t−1 i,t i,t−1 i=1 t=1 i=1 t=1 T 1 (cid:88) y = y − y i,t i,t i,t T t=1 T 1 (cid:88) x = x − x i,t i,t T i,t−1 t=1 Although economically more meaningful in the context of emerging markets, this estimator suffers from a second-order bias in the presence of endogenous regressors. The time-series demeaned regressor, x , incorporates information corresponding to periods after t − 1. As a result the i,t−1 demeaned regressor is correlated with u , generating a small sample bias. 6 This bias, also known i,t as the Stambaugh bias (Stambaugh (1999)), can be defined in terms of the limiting bias in the 6In the standard pooled estimation, the small sample bias is not an issue since the effects of endogeneity are eliminated when pooling the data.

autoregressive root of the predictor variable. As shown in Hjalmarsson (2007), the analogue of the Stambaugh bias in the panel case can be written as w plim T(βˆ −β) = plim 12 T(ρˆ −ρ) (4) FE FE w (T,n→∞)seq (T,n→∞)seq 22 where w and w are the average covariance between u and v and variance of v respectively, 12 22 i,t i,t i,t and ρˆ is the fixed effects estimator of the autoregressive root of the predictor variable.7 FE Therecursivelydemeanedestimatorthatcorrectsforthisbiasisgivenbythefollowingequation: (cid:32) n T (cid:33)−1(cid:32) n T (cid:33) (cid:88)(cid:88) (cid:88)(cid:88) βˆ = xdd x(cid:48) yddx (5) FD i,t−1 i,t−1 i,t i,t−1 i=1 t=1 i=1 t=1 T 1 (cid:88) ydd = y − y i,t i,s i,t T −t+1 s=t T 1 (cid:88) xdd = x − x i,t i,t T −t+1 i,s s=t This estimator uses x as an instrument and udd includes only data corresponding to periods i,t−1 i,t after t. Therefore, by forward demeaning, udd is independent of x and no second-order bias i,t i,t−1 arises.8 Many studies have shown that the predictive power of a variable may depend on the forecasting horizon under analysis. Variables that proved to be helpful predictors for short horizons may lack any predictive power when tested over longer horizons, and vice-versa. Hence, as part of our empirical analysis, we test for predictability of our set of financial aggregates using different horizons. In doing so, we use reverse regressions as proposed by Hodrick (1992) who shows that, under the null of no predictability and covariance stationary, in the usual long horizon predictive regression ∆y = α +β x +u (6) t:t+k k,1 k,1 t t:t+k with k representing the number of periods ahead considered in the forecast, and ∆y = t:t+k ∆log(y /y ) = ∆y +∆y +...+∆y the k-period GDP growth, the numerator t+k t t:t+1 t+1:t+2 t+k−1:t+k 7Note that when the local-to-unity parameters are unknown, direct estimation of the bias is not feasible. 8See Hjalmarsson (2010) for theorems and proofs of the properties of the above estimators.

of the slope coefficient estimated by cov[(∆y +...+∆y );x ] (7) t:t+1 t+k−1:t+k t is equivalent to cov[∆y ;(x +...+x )] (8) t:t+1 t t−k+1 where (8) is the numerator of the slope coefficient in the following predictive regression ∆y = α +γ x +u (9) t:t+1 k,1 k,1 t−k+1:t t:t+1 with x = x +...+x . The coefficient in the forward regression, β , can be expressed as t−k+1:t t t−k+1 k a linear function of the coefficient in the reverse regression, γ k V˜ xx β = γ k k V xx where V˜ and V are the covariance matrices of the regressors in the reverse and forward rexx xx gressions respectively. As a result, testing the null of no predictability β = 0 is equivalent to k,1 testingγ = 0. However, underthereverseregressionsettingofequation(9), inferencewillleadto k,1 less size distortions. Recent work by Wei and Wright (2010) extends these findings by considering persistency in the predictive regressors. They model the predictors as autoregressive processes of order 1 with near unit roots and find that reverse regression methods can still be performed in the context of nearly non-stationary processes. Furthermore, their results confirm previous findings that inference is more robust in small samples under the reverse regressions than with typical long horizon regression methods (i.e., equation (6)). We use this methodology to perform inference for long-term horizon growth rates by regressing the one-period ahead return on the sum of the regressors over the past k=4, and k=8 periods. 3 Data Our study focuses on twenty-two countries that were classified as emerging market economies by Standard and Poor’s as of December 2009. We develop a comprehensive dataset of country-specific and global macroeconomic and financial variables from December 1992 to March 2010. We use quarterly series that include real GDP growth rates, inflation, stock returns, valuation measures,

interest rates, capital flows, and commodity returns. GDP series are from the International MonetaryFund’sInternationalFinancialStatistics(IFS)databaseexceptinthecasesofMexico, Taiwan, and South Africa where the source is Global Financial database (GFD). We deflated these series using the corresponding GDP deflators from the IFS database. We tested and removed seasonality using the X-12-ARIMA Seasonal Adjustment Program from the U.S. Census Bureau. Table 1 presents descriptive statistics for the GDP growth series for the fourth quarter of 1992 to the first quarter of 2010. During the span of our sample all countries report positive average output growth rates. On average, quarterly GDP rates reach 1.13%, with China exhibiting the highest average growth rate (3.05%) and Mexico the lowest (0.61%) within emerging markets.9 In terms of volatility, Morocco presents the highest average volatility level (4.2%) while Hungary shows the most stable growth pattern (1.05%). Stock market index returns for our set of countries are from the Global Financial Database. They are inflation adjusted, include both changes in price and dividends, and are expressed in local currency terms. With respect to valuation measures, we consider country-specific dividend yields and price-earning ratios from GFD. These aggregates are based on large cap stocks which represent about 75% of the capitalization of the country. We also study the information content of stock market capitalization on subsequent output growth. In doing so, we employ capitalization growth rates and a ratio of capitalization to GDP from GFD. We evaluate different measures of financial flows from the balance of payments in our panel of developingcountries. Atthebroaderlevel,weconsidercapitalandportfolioflows. Capitalflowsare the net sum of direct investment, portfolio investment, financial derivatives, and other investments from the financial account. Portfolio investment flows are a subset of the capital flow variable and account for transactions with nonresidents in financial securities. We disaggregate this category into debt and equity flows. Debt flows are the net sum of both inflows and outflows of assets and liabilities covering fixed income instruments such as bonds and money market securities. Equity securitiesarethesumofassetsandliabilitiescoveringstocks,andsimilarinstrumentsthatrepresent ownershipofequity. Withincapitalflows,wealsoevaluatethepredictivepowerofdirectinvestment capital flows. This aggregate is the sum of direct investment abroad and direct investment in the reporting country. The source of our flow aggregates is the IFS database. Our interest rate variables are from GFD. We consider the term spread, the real short interest 9The number of observations for GDP series vary across countries.

rate, and returns on JPMorgan’s Emerging Markets Bond Index Plus country indices (EMBI+). ThisfamilyofindicesreportsthespreadbetweenlocalandUSgovernmentdebtinstruments. More specifically, they track the spread between the average yield on the securities of the local country against that of US treasuries. The local debt instruments include US-dollar denominated Brady bonds, eurobonds and traded loans issued by sovereign entities. We also test whether US interest rate aggregates lead the business cycle of emerging market countries. In doing so, we consider the US default spread, short interest rate and term spread. The default spread is measured as the difference between Moody’s Corporate BAA and AAA bond yields, the short interest rate is the yield on the 3-month Treasury Bill, and the term spread is the difference between the yield on the 10-year Government Bond and the 3-month Treasury Bill. We also consider the performance of a set of commodity indices. We use the Goldman Sachs commodity index (GSCI) as a proxy for the performance of commodity markets at the aggregate level. GSCI is a composite index of commodity sector returns weighted by world production. We also drill down to the sector level and include returns covering agricultural, energy, livestock and precious metal commodity markets from the S&P GSCI family indices. Finally, we build and test the predictive power of proxies for Fama and French risk factors. The size factor is the spread between returns on the MSCI Emerging Markets Small Cap and the MSCI Emerging Markets Large Cap total return indices. The book-to-market factor is represented by the return difference between the MSCI Emerging Markets Standard Growth and the MSCI Emerging Markets Standard Value total return indices. Lastly, we use the MSCI Emerging Markets total return index as a proxy for the market portfolio. 4 Empirical Results This section presents the empirical findings for our set of predictive regressions. Tables 4 through 8 introduce our baseline pooled and time series results for the set of return and valuation measures, interest rates and flows. The pooled regressions test the predictive power of country-specific financial aggregates to forecast subsequent, one-year-ahead, and two-year-ahead output growth.10 Following the empirical macro-finance literature (i.e.,Stock and Watson (2003)), we include lagged 10Note that, by construction, out-of-sample excercises are not feasable the forward demeaning estimator.

GDP growth and the variable of interest as regressors. However, for convenience, we report results only for our set of financial predictors.11 We also normalized our series to allow for comparison acrossvariables. Intables9through12,weshiftourfocusawayfromcountry-specificregressorsand assess the forecasting ability of a group of global aggregates. This set of variables includes returns on commodity markets, US interest rates and Fama and French (1993) risk factors. Finally, we also evaluate the differences and similarities on the role of capital markets to predict future economic growth in emerging markets and in the US. In doing so, we test for output growth predictability in the US over the same sample period and use analogous predictive regressors of our emerging market analysis. These results are described in table 13. 4.1 Returns and Valuation Measures Although the goal of this paper is not to establish causality in the traditional economic sense, our results can be related to financial theories that relate stock prices to the present value of future cash-flows. Inparticular, theoriessuchasthe“passiveinformant”hypothesissuggeststhatpositive news about future output growth will be translated into higher equity market prices, resulting in a positive relationship between stock returns and future GDP growth. Empirical evidence on the predictive power of stock returns is mixed. An early study by Harvey (1989) finds that about only 5% of variation in output growth is explained by stock market related variables in the US over the 1953-1989 period. On the other hand, Estrella and Mishkin (1998) find evidence that stock price indexes are good predictors of US recessions. Specific to emerging markets, Mauro (2003) focuses on the correlation of lagged stock returns and output growth and finds this relationship to be positive and significant in several countries. Moreover, his findings suggest that countries with a high market capitalization to GDP ratio exhibit stronger correlations.12 Although there is no consensus about the empirical success of equity prices and related variables to predict future economic activity, they are widely considered by policy makers and investors. In our baseline case, we find strong evidence of the predictive power of stock market returns in 11Details are available upon request. 12There is also a relatively new literature that studies stock return predictability in emerging markets. Harvey (1995, 1994) shows that equity returns are predictable in some developing countries when conditioning on country specificandglobalinformationvariablessuchasdividendyields,pastreturns,currencyindexandearning-priceratio. Bekaert, Harvey, and Lundblad (2007) find that local market liquidity is an important driver of returns in these economies.

emerging markets. As presented in table 4, the slope coefficients are statistically significant across models, suggesting that returns are positively related to future output growth. Furthermore, when we look at longer forecasting horizons our results are validated. Both panels B and C show that there is information content in equity market returns at the one- and two-year horizons. Strongest results are for the one year ahead forecasts.13 Time series regressions in table 5 point in the same direction. However, only ten of the countries present statistically significant coefficients. We also evaluate the predictive relationship between market capitalization, commonly used as an indicator of stock market development, and GDP growth. In doing so, we consider a ratio that relates capitalization to GDP. Results from our pooled estimations in table 4 lack statistical significance when forecasting one quarter and two-year-ahead economic activity. Only the slope coefficient for our capitalization ratio is significant under the fixed effects model at the one-yearahead horizon. This result suggests a negative predictive relationship. However, as shown in tables 2 and 3, this variable is both highly persistent and endogenous. As a result, our fixed effects estimator is biased.14 Therefore, we focus on the forward demeaned estimator that corrects for this small sample bias. Our country level results present no clear pattern about the relationship between capitalization and subsequent output growth. As table 5 shows, slope coefficients lack statistical significance in most of the countries. Estimates are significant only in Argentina, Egypt, and Malaysia, where we find a positive relationship between capitalization and subsequent output growth. Twowidelytestedvaluationmeasuresinthereturnpredictabilityliteraturearetheprice-earning ratio and the dividend yield. As shown in tables 2 and 3, we find that both aggregates suffer from a second order bias. We correct for this bias in our demeaned estimation. As presented in table 4, panelsAthroughC,wefindnoevidenceofthepredictivepowerintheprice-earningratio. Onlythe fixed effects estimator is statistically significant when forecasting two years ahead. However, when we correct for the Stambaugh bias, we fail to reject the null of no predictability.15 Time series regressions in table 5 provide no clear pattern of the predictive power of P/E ratios. Similarly, 13LaggedGDPgrowthincludedinthesebivariateregressionslackstatisticalsignificanceatanyforecastinghorizon and under all model specifications. 14Notethatsincethedirectionofthebiasdependsonthesignofthecross-correlationsoftheinnovationprocesses, and in this case cov(u ,v )>0, the bias is negative and therefore we see β >β . i i FD FE 15LaggedGDPcoefficientsarepositiveandstatisticallysignificantacrossmodelspecificationsandhorizons. These results are available upon request.

we find no predictive relationship between lagged dividend yields and future output growth in our robustestimations. Boththestandardpooledandthefixedeffectscoefficientssuggestevidenceofa negative predictive relationship between these aggregates when forecasting one quarter ahead ( see panel A of table 4). However, when we allow for individual effects and correct for endogeneity and persistence, signs of predictability disappear. This result is of particular relevance in the context of a highly persistent variable such as the dividend yield, since findings from our fixed effects estimation could result in misleading interpretations.16 Time series regressions in table 5 lack statistical significance in a large number of countries. However, when significant, results suggest a negative predictive relationship between dividend yields and subsequent output growth. When we compare the results of emerging markets and from the United States (table 13), we find similar patterns for capitalization and the dividend yield. In particular, there is no sign of predictability in these two measures; slope coefficients lack statistical significance when forecasting one-quarter-ahead GDP growth. On the other hand, we find that stock market returns, which lead economic growth in emerging markets, have no predictive power in the context of the US. Also, the price-earning ratio, which presents no evidence of predictive power in emerging markets, can be a helpful predictor of subsequent GDP in the US. Finally, although there are differences in the informationcontentofvaluationmeasuresacrossmarkets,wefindsomesimilaritiesinthecharacteristics of the data. Specifically, aggregates that show high levels of persistence and cross-correlations in emerging markets present the same patterns in the US. This is the case for capitalization ratio, the dividend yield, and the price-earning ratio. 4.2 Interest Rates Table 4 presents results for the real short interest rate, term spread and default spread. As panels A through C show, we find strong evidence of predictability in the spreads when predicting future GDP growth. In particular, higher default spreads on emerging market government debt signal lower output growth.17 This counter-cyclical predictive relationship is supported by literature on credit channels linking the financial and real side of the economy, where worsening in the quality 16In this bivariate setting, lagged GDP coefficients are both positive and statistically significant. 17ArellanoandMendoza(2002)documentthatepisodesofsuddenstopsinemergingmarketsarecharacterizedby sharp and unexpected declines in EMBI+ spread indices (our measure of default spreads) and plummeting equity markets.

of government balance sheets are translated into higher default risk, resulting in higher borrowing rates. These tighter financial conditions negatively impact future economic growth. Conversely, a high term spread predicts a future increase in GDP growth (i.e., short interest rate declines relative to the longer rate are followed by future rises in economic activity). Estimates are both positive and statistically significant across models when forecasting subsequent and oneyear-aheadoutputgrowth. Thispredictiverelationshipcanbeexplainedintermsofanexpectation and a term premium effect. In the first case, spreads are expected to convey information about future short rates. The positive predictive relationship is based on the expectations hypothesis of the term spread and the impact of monetary policy. For example, easing of monetary policy is expected to positively impact interest sensitive sectors, boosting economic activity. On the other hand, the term premium effect signals changes in the term premium related to the compensation demanded for holding long bonds, influenced by risk and liquidity premium among other factors.18 Overall, ourresultsareinlinewithalargebodyofresearchthatfindsthatthetermspreadcanbea usefulpredictoroffutureGDPgrowthindevelopedmarkets. Forexample,EstrellaandHardouvelis (1991) find evidence that the slope of the yield curve can predict changes in real output. Harvey (1988) documents that the expected real term structure forecasts consumption growth. 19 Regarding the real short interest rate, by adding lagged GDP into the regression, the information content of this aggregate vanishes. Our estimates lose predictive power across models and horizons.20 Timeseriesregressions, asshownintable6, lackstatisticalsignificanceformostofouremerging market countries. Nevertheless, when they are significant, the real short term rate and default spread signal a negative relationship with future GDP growth. These findings do not extend to the term spread where there is no clear pattern in the country level estimates. Our empirical results also suggest that there are some differences in the role of interest rates 18Hamilton and Kim (2002) decompose the contribution of the term spread into the expected changes in interest rates and the term premium and find that both factors are relevant for forecasting real GDP growth in the US. 19See Stock and Watson (2003) for a literature review. 20Astrandofthemacro-literaturelooksattherelationshipbetweentherealinterestratesemergingeconomiesface ininternationalmarketsandthebusinesscycle. Fernndez-Villaverde, Guerron-Quintana, Rubio-Ramirez, andUribe (2011)buildrealinterestratesonloansdenominatedinUSdollarsasthesumofaninternationalrateandacountry riskspreadandfindthat,increasesininterestratesvolatilityhaveanegativeeffectonoutputgrowth. Previouswork byNeumeyerandPerri(2005)findsthatrealratesarecountercyclicalandleadthebusinesscycleofagroupofsmall open developing economies.

in predicting output growth in the US and emerging markets. In particular, the term spread that proved to be a helpful predictor in emerging markets shows no predictive power in the US over our sample period.21 On the other hand, the default spread and the real short interest rate present similar patterns across markets.22 4.3 Flows Historically, we have seen capital flight from developed countries into emerging markets seeking investment opportunities with higher expected returns. These inflows boosted economic growth, relaxed liquidity constraints, raised asset prices and broadened and increased the depth of many of these financial markets. On the negative side, capital outflows have preceded financial and economic crisis in emerging markets.23 In this section we use country-level data on net capital flows and its components (i.e. direct and portfolio investment flows) to empirically evaluate the predictive power of capital flows in forecasting GDP growth. Results for our measures of flows are presented in tables 4, 7 and 8. Table 4 introduces pooled results at the aggregate level, table 7 looks at components of portfolio investment and capital flows and table 8 presents results for time series predictive regressions. As described in table 4, we find evidence of the predictive power of country flows in forecasting economic activity in emerging markets. Both capital and portfolio flow ratios present statistically significant slope coefficients when we forecast one-quarter-ahead output growth. Results suggest that periods of capital and portfolio inflows anticipate future output growth, and in turn, periods dominated by large outflows are followed by declines in economic activity. Furthermore, these resultsareextendedintable7. Wefindthatdebtandequitysecurityflowsaswellasdirectinvestment flows can be useful predictors of future economic growth. Slope coefficients are statistically significant signaling a positive relationship when predicting subsequent growth. Although the causality 21Work by Haubrich and Dombrosky (1996), Dotsey (1998), and Kucko and Chin (2009) examine the evidence of the yield curve as predictor of future economic activity in the US and find that the predictive power of the yield curve has deteriorated in recent years. 22Notethatinthecaseofthedefaultspreadwecannotperformaprecisecomparisonbecauseweconsidercorporate bondyieldspreadsintheUS,whileinemergingmarketswetrackspreadsbetweenlocalgovernmentdebtinstruments and US treasuries. 23There is a large body of research on sudden stops that models the effects of capital flows on output growth in emerging markets, e.g., Calvo (1998), Reinhart and Calvo (2000), Ferretti and Razin (2000), among others.

question is beyond the scope of this paper, the positive predictive relationship of portfolio flows suggests an information and portfolio optimization effect. As investors become optimistic about opportunities in emerging markets, we see financial inflows into local debt and equity securities. Overall, results suggest that investors are well informed and have the ability to react to news and expectations about future economic conditions in emerging markets. Net inflows to equity and debt markets predict subsequent output growth. Conversely, net portfolio outflows signal lower GDP growth. Also, direct investment capital flows present evidence of predictability over subsequent and one-year horizons, signaling a positive relationship between flows and future economic activity. These results can be related to the effect of demand shocks, where an exogenous increase in investments will be associated with a higher demand and a subsequent increase in the level of output. Our country level results, when statistically significant, support findings for our pooled models. Asdescribedintable8, bothcapitalandportfolioinvestmentflowspresenthighlyeconomicallysignificantslopecoefficientssignalingapositiverelationshipbetweeninflowsandsubsequenteconomic activity. When we extend our forecasting window to one and two years, we lose predictability in our aggregate measure of portfolio investment flows. Moreover, if we disaggregate further, we find a similar pattern in the case of debt security flows. On the other hand, fixed effects estimates for equity flows are statistically significant when predicting one-year-ahead GDP growth. We also find a reversal in the predictive relationship between capital flows and GDP when forecasting two years ahead. As seen in many south East Asian and Latin-American countries during the 90’s, strong capital inflows in weak financial systems that are unable to absorb flows productively can lead to financial bubbles, capital flow reversals, inflation and higher vulnerability to external shocks, among other unintended outcomes that negatively impact economic growth. As shown in table 13, results for the US are strikingly different when forecasting one-quarteraheadGDPgrowth. Inparticular,whilewefindapositiverelationshipbetweencapitalandportfolio flow ratios and subsequent output growth in emerging markets, our results suggest that in the US, increases in the ratios of flows to GDP are followed by declines in output growth. On the other hand, net capital and portfolio outflows signal subsequent increases in GDP growth. This could be interpreted as a flight to quality effect. In periods of global turmoil, investors will move their capital away from perceived as riskier investments in international markets, to safer allocations in

the US.24 4.4 Commodities It is well known that commodity markets are a vital engine for economic growth in many emerging market countries. Periods of higher prices in international commodity markets are expected to be related to higher output growth, while market declines will negatively impact the growth prospects ofmanyemergingeconomies. Intable9, panelA,weinvestigatethepredictivepowerofcommodity market returns at the global and sector level to forecast economic activity. The set of commodity sectors include agriculture, energy, livestock and precious metals. Overall, we find strong evidence of predictability both at the aggregate market level and across sectors. In particular, coefficients for commodity markets are economically and statistically significant across models and horizons at the aggregate level. Nevertheless, the predictive power of commodity market returns appears to be strongest when forecasting subsequent output growth. Given the inelastic nature of demand and supply for commodities, these findings could be interpreted in terms of the effect of demand shocks. As demand for commodities increases, commodity prices rise, having a positive impact on the terms of trade and output of export countries. These results are in line with Chen, Rogoff, and Rossi (2010) who find that commodity prices Granger-cause exchange rates in-sample when regressions are robust to parameter instability.25 At the sector level, agricultural and precious metal commodities appear as the most economicallysignificantpredictorswhenforecastingsubsequentGDPgrowth. Alsoresultssuggestapositive predictive relationship between energy commodities and future output, with economically and statistically significant coefficients across horizons. However, as we extend the forecasting horizon there is evidence of a decrease in the information content of sector commodity prices with regard to future economic growth. Furthermore, in the two-year forecasts, the livestock sector loses predictability across models. Findings from our time series regressions validate our pooled results. As described in table 10, slope coefficients of commodity markets are statistically significant in twelve of our countries, 24Results for the US can also be related to recent work by Bernanke, Bertaut, DeMarco, and Kamin (2011) who argue that capital inflows into the United States, through their effect on assets’ yield, likely played a role in the housing bubble and subsequent financial crisis. 25Theyalsoanalyzethereversepredictiverelationshipandfindin-andout-of-sampleevidencethatexchangerates are robust predictors of commodity prices.

suggesting a positive relationship between lagged returns on commodities and output growth. The only exception is Morocco, which presents a negative slope coefficient. This result does not contradict our findings given the specific characteristics of this country. More than half of their GDP comes from the service sector while the industry sector represents around 30% of their output. All sector level predictive regressions point in the same direction as the aggregate market. Higher commodity prices are followed by higher economic growth in emerging markets. For example, we find a positive and significant relationship between energy prices and output growth for Russia as well as for other countries. WecompareourresultsinemergingmarketsagainsttheUnitedStates. Asshownintable13,we findsignificantdifferencesintheinformationcontentofcommoditymarkets. Whiletheperformance of global and sector commodity markets convey information about future business conditions in our panel of developing countries, they show no forecasting ability when predicting output growth in the US. Slope coefficients are neither economically nor statistically significant. These results can be related to structural differences between the US and emerging market countries. While in many developing countries, economic growth is heavily dependent on commodities exports, growth in the United States is mostly related to services (around 77% of its GDP) and to the industry sector to a lesser extent (22%). Moreover, extensive literature finds evidence of predictability of energy commodities, such as oil prices in the US (Hamilton, 2003, 2010). However, their results suggest a nonlinear relationship between oil prices and output growth. 4.5 US Interest Rates A large literature finds that the US economy leads the cycle of many developed economies. Traditionally, trade has been considered the most common transmission channel. Periods of growth in the United States would be accompanied by increases in its imports, which in turn would boost economic growth of the trade partner. An extreme example of this direct relationship could be Mexico, whose GDP growth is known to be highly correlated with that of the United States. However, as financial markets become more integrated, they can be expected to be more relevant in the transmission of shocks from the US to the rest of the world. In this section we also test whether expectations about the US economy that are contained in interest rate aggregates can help predict output growth in emerging countries. As shown in panel B of table 9, we find evidence that the term and corporate default spreads contain information

about subsequent GDP growth. In particular, the US term spread has a positive relationship with future growth in emerging markets. Conversely, we find that increases in the US default spread are followed by subsequent declines in output growth. Furthermore, under the pooled and fixed effects settings, the predictive power of the term spread increases as we extend our forecasting horizon to two years. On the other hand, the default spread loses its predictive power as we move to longer term horizons. The short interest rate, represented by the 3-month US treasury bill, shows no signs of predictability in the short term. However, its predictive power emerges when we look at a two years ahead forecasting horizon, where we find economically and statistically significant negative slope coefficients. As described in table 11, evidence of predictability is limited when we look at individual time series regressions. Results on the term spread lack statistical significance across countries. The US short interest rate helps predict output growth only in the case of China. Finally, the default spread, has a economically and statistically significant negative relationship with growth in only 8 of our 22 countries. 4.6 Book-to-Market, Size, and The Market Portfolio This subsection discusses whether well-known factors such as book-to-market, size, and the market portfolioconveyinformationaboutthefuturestateoftherealeconomy. Aspresentedintable12,we findstrongevidenceofthepredictivepowerofthesizefactorandmarketportfolio. Slopecoefficients forthesizefactorareeconomicallyandstatisticallysignificantacrossmodelsovertheone-andtwoyear-ahead forecasting horizons. Positive slope coefficients suggest that outperformance of small over large capitalization stocks precede periods of output growth. Conversely, periods in which small capitalization stocks underperform are followed by declines in future economic activity. As described in table 12, we also find a positive predictive relationship between returns on the market portfolio and future GDP growth. Slope coefficients are statistically significant across horizons, with one-year-ahead forecasts presenting the strongest economic significance. On the other hand, the book-to-market factor shows no evidence of predictive power over one-quarter and one-year horizons. Only when forecasting two years ahead, and in the context of the forward demeaning model, we find a negative relation between book-to-market and future output growth. These results confirm and extend some previous findings of Liew and Vassalou (2000). Using

time-series regression analysis and Fama and French risk factors (Fama and French (1993)), Liew andVassalou(2000)testforoutputgrowthpredictabilityinagroupof10developedmarkets. They documentthatthebook-to-market, sizefactor, andmarketportfolioarepositivelyrelatedtofuture GDP growth. In the context of emerging markets, we find similarities in the cases of the size factor and market portfolio, where both variables contain information about the future state of the real economy. However, as shown in table 13, while in emerging markets size is positively related with future growth, results suggest a countercyclical relationship in the US during the sample period. 5 Robustness Analysis This section presents a series of additional exercises we perform to evaluate the robustness of our results to different scenarios. First, we allow for global innovations in our data generating process. Second, we split our sample into sub-periods, 1992-1999 and 2000-2010. Third, we test for output growth predictability at the regional level. Fourth, we group our set of countries based on GDP per capita. Fifth, we run our analysis in a multivariate setting. Finally, we perform a cross-validation exercise to check the stability of our results. As in our baseline scenario, we run bivariate regressions including lagged GDP growth and the variable of interest as regressors.26 We also normalized our series by their standard deviation to allow for comparison across variables. 5.1 Accounting for Common Factors We use Hjalmarsson’s (2010) framework that in addition to idiosyncratic factors allows for the effect of common shocks to returns and proposes a set of pooled estimators that are robust to cross-sectional dependence. This is of particular relevance when considering the phenomenon of emerging markets because these countries are known to be more vulnerable to external shocks and episodes of contagion than developed countries. As a result, there will be shocks to output growth that will impact emerging markets as a whole while others will be country-specific. The data generating process accounting for common factors can be described as follows: y = α +β(cid:48)x +u +δ(cid:48)f (10) i,t i i,t−1 i,t i t 26Forconveniencewereportresultsonlyforoursetoffinancialpredictors. ResultsforlaggedGDPcoefficientsare available upon request.

x = A x +v +Γ(cid:48)z i,t i i,t−1 i,t i t z = A z +g t g i,t−1 t A = I +C /T g g wheref representsthecommonfactorintheerrorterm,andz isthecommonterminthepredictor t t x . z is an AR(1) process with near unit roots defined by A .27 i,t t g To perform estimation and inference in our panel setting, we use the following estimators that account for common factors: Robust Pooled Estimation: (cid:32) n (cid:33)−1(cid:32) n (cid:33) (cid:88) (cid:88) βˆ = X(cid:48) M X X(cid:48) M Y (11) RP i,−1 H¯ i,−1 i,−1 H¯ i i=1 i=1 M = I −H¯(H¯(cid:48)H¯)−1 H¯ where Y is T x 1, X is T x m and H¯ is also T x m with H¯ = 1 (cid:80)n x i i t n i=1 i,t−1 Robust Fixed Effects Estimation: (cid:32) n (cid:33)−1(cid:32) n (cid:33) βˆ = (cid:88) Xˆ(cid:48) Xˆ (cid:88) Xˆ Yˆ (12) RFE i,−1 i,−1 i,−1 i i=1 i=1 with Yˆ=M Y , Xˆ =M X , and Yˆ and Xˆ being their demeaned versions. i H¯ i i,−1 H¯ i,−1 i i,−1 Robust Recursive Demeaning Estimation: (cid:32) n (cid:33)−1(cid:32) n (cid:33) (cid:88) (cid:88) βˆ = Xˆ(cid:48)dd Xˆ Xˆ(cid:48) Yˆdd (13) RRD i,−1 i,−1 i,−1 i i=1 i=1 where Yˆdd and Xˆ(cid:48)dd are the recursively demeaned versions of Yˆ and Xˆ . i i,−1 i i,−1 Table 14 presents results for this setting. As in our baseline case, we run bivariate regressions including lagged GDP and the variable of interest to forecast subsequent GDP growth.28 Overall, our baseline results are validated when we allow for common factors. In particular, our set of valuation measures, interest rates and flow aggregates present similar patterns as in table 4. We 27All variables defined in the estimation section remained unchanged. 28For convenience we only present estimates and t-statistics for our set of country-specific variables. Results for lagged GDP are available upon request.

find a positive predictive relationship between stock returns and subsequent output growth. On the other hand, our measure of capitalization ratio shows no evidence of predictability. Similarly, the dividend yield and price-earnings ratio do not present any predictive power when we allow for individual effects and correct for endogeneity and persistence. We also find that the term spread and real short interest rate convey information about future economic activity. However, under this new setting, the government default spread loses its predictive power. Finally, results for capital and portfolio flows also suggest a strong positive relationship with future economic growth in emerging markets. 5.2 Sub-sample Analysis Our sample encompasses very different market conditions. We cover the financial liberalization process of many developing countries, the Tequila crisis (1995), the Asian crisis (1997), the Ruble crisis (1998), the Tech bubble (1999-2000), the expansion of the European Union (2004), the rise of China and the last global financial and economic crisis, among other key episodes. To gain further insight on the stability of our forecasts, we split our sample into sub-periods. Panels A and B of table 15 present results for the periods 1993-1999 and 2000-2010. Panel C, tests for output growth predictability excluding the period covering the last global crisis,1993-2007. As shown in panels A and B, we find that financial aggregates that lead economic growth in our baseline case are validated when we test for predictability in our subsample analysis. In particular, results for stock returns, the term spread, dividend yield and capital flows ratio, among other variables, present some similar patterns in both subsamples. However, there are differences. For example, the real short interest rate, portfolio investment flows and returns on global commodity markets show evidence of predictability only in the last decade, a period of higher economic growth andstrongerstockmarketperformancethanthe90’s. 29 Atthecommoditysectorlevel,resultsalso suggestthatenergyandlivestockcommoditiesconveyinformationaboutoutputgrowthonlyduring the 2000-2010 period. On the other hand, we also find that the price-earnings ratio, an aggregate that lacks predictive power in our baseline scenario, presents evidence of predictability during the 29During2000-2010,emergingmarkets’annualaverageGDPgrowthreached5.7%whilefor1993-1999GDPgrowth was 3.7%. Also, average annual returns on emerging equity markets represented by the MSCI Emerging Markets index was 17.6% for the last decade versus 14% during the 90’s. In terms of stock market volatility, we find similar volatility levels ( 39.5% and 40.3% ) across sub-samples.

90’s, signaling a positive relationship with GDP growth. However, predictability vanishes in the last decade when correcting for the small sample bias. The most striking differences between our subsample and baseline results are related to our set of risk factors. In particular, in our baseline setting estimates for the book-to-market factor lack statistical significance when forecasting subsequent GDP growth. However, when we split our sample into the 90’s and 00’s we do find evidence of predictability. Results suggest a negative predictive relationship during the 90’s while, in line with previous findings in developed countries (Liew and Vassalou (2000)), it becomes positive during the last decade. Finally, we test for output growth predictability excluding the period covering the last global crisis. As shown in panel C, we find similar patterns as in our baseline results. However, the dividend yield that lacks any predictive power in our baseline case, now shows evidence of predictability even when we correct for the small sample bias. Results suggest a negative predictive relationship with future GDP growth. Also, although the direction of the relationships remains unchanged, we find slight differences in the predictive power of the term spread and commodity returns. The information content of the term spread decreases when we include the period covering the crisis. On the other hand, the predictive power of returns on commodity markets increases when we consider the full sample in our analysis. 5.3 Regional Analysis To gain further insights on the relationship between financial aggregates and future economic activity we split our panel of countries according to their geographic location. Following standard classifications we define three different regions, Asia, Latin America and Europe, Middle East and Africa (EMEA).30 We find strong evidence that our set of financial aggregates can be used as leading indicators of economic activity across regions. However, Latin America presents the strongest evidence of output growth predictability. As table 16 shows, results suggest that return andvaluationmeasuressuchasstockreturnsandcapitalizationratiohavethestrongestforecasting ability in Latin America. In particular, slope coefficients are larger in this region than in Asia and EMEA.Also,portfolioinvestmentflowsmeasuredasaratioofflowstoGDP,althougheconomically 30Asia comprises China, Indonesia, India, South Korea, Malaysia, Philippines, Taiwan and Thailand. The EMEA region is composed of Czech Republic, Egypt, Hungary, Israel, Morocco, Poland, Russia, South Africa and Turkey. Latin America comprises Argentina, Brazil, Chile, Mexico and Peru.

and statistically significant across regions, present the strongest results in Latin America. As in our baseline case, we find a positive relationship between lagged portfolio flows and GDP growth. Furthermore, interest rate variables such as the short interest rate and the term spread also proved to have higher predictive power in this region than in the rest of the group. On the other hand, we find that the information content about future economic activity of the default spread and capital flows is the strongest in Asia, while returns on global commodity markets appear to play a bigger role in EMEA. For robustness, we also test for predictability in Asia excluding China, a key driver of regional and global growth in the past years. The goal is to analyze whether the predictive relationships in the region are mainly driven by this country. As panel D shows, results present similar predictive relationships in both scenarios. However, we do find a slight decrease in the information content of stock return when we exclude China. Conversely, the predictive power of the default spread and global commodity returns improves under this new scenario. 5.4 GDP per Capita As part of our robustness analysis, we split our panel based on GDP per capita. We grouped the countries in our sample into three categories: high, middle and low income.31 Table 17 presents results for this new setting. Overall, results for our set of financial aggregates are validated. However, we do find some differences across groups. In particular, stock and commodity returns present the strongest predictive power in our group of middle-income countries. Flows prove to be a helpful predictor across groups, with strongest economic significance in low-income countries. On the other hand, the set of interest rate variables show evidence of predictability in only the middleand low-income countries. 31High income is composed by countries with a PPP-adjusted GDP per capita higher than 13,000 USD per year. This group includes Taiwan, South Korea, Israel, Czech Republic, Hungary and Poland. GDP per capita for middle incomecountriesrangesfrom13,000USDto9,000USDperyear. Russia,Argentina,Chile,Malaysia,Mexico,Turkey, Brazil and South Africa fall in this category. GDP per capita for low income countries covers those with less than 9,000 USD per year. In this group we consider Peru, Thailand, China, Egypt, Morocco, Indonesia, Philippines and India.

5.5 Multivariate Regressions We also test the predictive power of the set of financial aggregates to forecast GDP growth in a multivariate setting. In doing so, we group our variables into four categories; namely, return and valuation measures, interest rates, flows and commodities. We use principal component analysis to reduce the number of predictors while still accounting for most of the variance in the observed variables. As shown in table 18, results suggest that our four sets of variables can be useful predictors of one-quarter-ahead GDP growth. In particular, coefficients related to commodities and return and valuation measures are positive and significant when forecasting one quarter ahead output growth. On the other hand, interest rate variables show a negative predictive relationship with subsequent GDP. Also, estimates for flows are positive and statistically significant under the pooled and fixed effects estimation. When we extend the forecasting horizon to one year, only commodities, andreturnandvaluationestimatescontaininformationaboutfuturegrowth. Finally, when forecasting two years ahead, only our measure of flows shows evidence of predictive power. In particular, results suggest a reversal in the predictive relationship of flows and economic activity. 5.6 Cross-Validation Analysis For robustness, we also perform a cross validation exercise to check the stability of our estimates. The goal is to study whether results are driven by a particular country. In a bivariate setting that includes lagged GDP and the variable of interest as the regressors, we randomly exclude three countries from our sample and estimate coefficients and t-statistics. We repeat this exercise twenty times for each variable. Table 19 presents results for this analysis.32 For each variable we report the average estimate, its standard deviation and the number of times (out of the twenty random draws)thatthecoefficientisstatisticallysignificant. Resultsprovideevidenceinfavorofthestability of our estimates across models and horizons. The predictive ability of the variables that proved helpful in our baseline case are validated by this analysis. 32For convenience, we present results only for financial aggregates. Results for lagged GDP are available upon request.

6 Conclusion This paper contributes to the empirical macro-finance literature by studying the predictive relationship between financial aggregates and economic growth in emerging markets. To address this question, we develop a comprehensive data set on output growth, country-specific and global financial aggregates for a large panel of developing countries, and test for output growth predictability using panel methods that are robust to endogenous and persistent predictors (Hjalmarsson, 2010). Our analysis provides evidence of predictability and shows that a set of financial variables can be used as leading indicators of economic activity in emerging markets. In particular, we find evidence that stock returns, portfolio flows, sovereign default spreads, and commodity prices, among other financial aggregates, can be useful predictors of emerging markets growth. 33 Our results are also robust to the inclusion of global shocks, which are of particular relevance in the context of emerging markets, as it is well known that these countries can be more vulnerable to external shocks and episodes of contagion. We also recognize the heterogeneity of the countries in the panel and the different economic conditions encompassed in our sample and perform a series of checks aimed at testing the robustness of our results under different escenarios. These checks include testing for predictability using subsamples based on income level and regional location. We also break the panel into subperiods and find that results are robust to the most recent financial crisis. Overall, byidentifyingasetofleadingindicatorsforemergingmarkets, ourresultscanbeuseful to policy makers and market participants who try to understand and predict the business cycles of these developing economies. 33Wetakenostandonwhetherthisfindingindicatesacausalrelationorwhetheritsimplyreflectsinformationin forward-looking financial aggregates.

References Arellano, C., and E. G. Mendoza (2002): “Credit Frictions and ’Sudden Stops’ in Small Open Economies: An Equilibrium Business Cycle Framework for Emerging Markets Crises,” NBER Working Papers 8880, National Bureau of Economic Research, Inc. Bekaert, G., C. R. Harvey, and C. Lundblad (2007): “Liquidity and Expected Returns: Lessons from Emerging Markets,” Review of Financial Studies, 20(6), 1783–1831. Bernanke, B. S., C. C. Bertaut, L. DeMarco, and S. B. Kamin (2011): “International Capital Flows and the Return to Safe Assets in the United States, 2003-2007,” SSRN eLibrary. Calvo, G. A. (1998): “Capital Flows and Capital-Market Crises: The Simple Economics of Sudden Stops,” Journal of Applied Economics, 0, 35–54. Chen, Y.-C., K. S. Rogoff, and B. Rossi (2010): “Can Exchange Rates Forecast Commodity Prices?,” The Quarterly Journal of Economics, 125(3), 1145–1194. Dotsey, M. (1998): “The predictive content of the interest rate term spread for future economic growth,” Economic Quarterly, (Sum), 31–51. Estrella, A., and G. A. Hardouvelis (1991): “The Term Structure as a Predictor of Real Economic Activity,” Journal of Finance, 46(2), 555–76. Estrella, A., and F. S. Mishkin (1998): “Predicting U.S. Recessions: Financial Variables As Leading Indicators,” The Review of Economics and Statistics, 80(1), 45–61. Fama, E. F., and K. R. French (1993): “Common risk factors in the returns on stocks and bonds,” Journal of Financial Economics, 33(1), 3–56. Fernndez-Villaverde, J., P. A. Guerron-Quintana, J. Rubio-Ramirez, and M. Uribe (2011): “Risk Matters: The Real Effects of Volatility Shocks,” American Economic Review, (01(6): 253061). Ferretti, G. M. M., and A. Razin (2000): “Current Account Reversals and Currency Crises, Empirical Regularities,” in Currency Crises, NBER Chapters, pp. 285–326. National Bureau of Economic Research, Inc.

Frankel, J., and G. Saravelos (2012): “Can leading indicators assess country vulnerability? Evidence from the 200809 global financial crisis,” Journal of International Economics, 87(2), 216 – 231. Gilchrist, S., V. Yankov, and E. Zakrajsek (2009): “Credit market shocks and economic fluctuations: Evidencefromcorporatebondandstockmarkets,”JournalofMonetaryEconomics, 56(4), 471–493. Hamilton, J. D. (2003): “What is an oil shock?,” Journal of Econometrics, 113(2), 363 – 398. (2010): “Nonlinearities and the Macroeconomic Effects of Oil Prices,” SSRN eLibrary. Hamilton, J. D., and D. H. Kim (2002): “A Reexamination of the Predictability of Economic Activity Using the Yield Spread,” Journal of Money, Credit and Banking, 34(2), 340–60. Harvey, C. R. (1988): “The real term structure and consumption growth,” Journal of Financial Economics, 22(2), 305–333. (1989): “Forecasts of Economic Growth from the Bond and Stock Markets,” Financial Analysts Journal, 45(5), pp. 38–45. (1994): “Conditional Asset Allocation in Emerging Markets,” NBER Working Papers 4623, National Bureau of Economic Research, Inc. Harvey, C. R.(1995): “PredictableRiskandReturnsinEmergingMarkets,” Review of Financial Studies, 8(3), 773–816. Haubrich, J. G., and A. M. Dombrosky (1996): “Predictingrealgrowthusingtheyieldcurve,” Economic Review, (Q I), 26–35. Hjalmarsson, E.(2007): “TheStambaughBiasinPanelPredictiveRegressions,”SSRNeLibrary. (2010): “Predicting Global Stock Returns,” Journal of Financial and Quantitative Analysis, 45(01), 49–80. Hodrick, R. J. (1992): “Dividend Yields and Expected Stock Returns: Alternative Procedures for Inference and Measurement,” Review of Financial Studies, 5(3), 357–86.

Kucko, K., and M. Chin (2009): “The Predictive Power of the Yield Curve Across Countries and Time,” . Liew, J., and M. Vassalou (2000): “Can book-to-market, size and momentum be risk factors that predict economic growth?,” Journal of Financial Economics, 57(2), 221 – 245. Mauro, P. (2003): “Stock returns and output growth in emerging and advanced economies,” Journal of Development Economics, 71(1), 129–153. Neumeyer, P. A., and F. Perri (2005): “Business cycles in emerging economies: the role of interest rates,” Journal of Monetary Economics, 52(2), 345–380. Perez-Quiros, G., and A. Timmermann (2001): “Business cycle asymmetries in stock returns: Evidencefromhigherordermomentsandconditionaldensities,” Journal of Econometrics, 103(1- 2), 259 – 306. Reinhart, C., and G. Calvo (2000): “When Capital Inflows Come to a Sudden Stop: Consequences and Policy Options,” MPRA Paper 6982, University Library of Munich, Germany. Stambaugh, R. F. (1999): “Predictive regressions,” Journal of Financial Economics, 54(3), 375– 421. Stock, J. H., and M. W. Watson (1989): “New Indexes of Coincident and Leading Economic Indicators,” in NBER Macroeconomics Annual 1989, Volume 4, NBER Chapters, pp. 351–409. National Bureau of Economic Research, Inc. (2003): “Forecasting Output and Inflation: The Role of Asset Prices,” Journal of Economic Literature, 41(3), 788–829. Weber, C. E.(1998): “ConsumptionSpendingandthePaper-BillSpread: TheoryandEvidence,” Economic Inquiry, 36(4), 575–89. Wei, M., and J. H. Wright (2010): “Reverse Regressions and Long-Horizon Forecasting,” .

Table 1: GDP Growth Summary Statistics Country Dates Obs. Mean StdDev Skewness Kurtosis AR(1) Argentina Q1 1993-Q1 2010 68 0.896 2.940 -0.055 11.379 0.3546 Brazil Q1 1995-Q1 2010 60 0.859 2.435 0.758 16.564 -0.244 China Q4 1999-Q4 2009 40 3.047 2.326 4.222 24.328 0.0495 Chile Q1 1996-Q4 2009 55 0.887 1.226 -1.099 5.078 0.307 Czech Republic Q1 1994-Q1 2010 64 0.839 1.585 0.999 9.080 0.076 Egypt Q1 2002-Q4 2009 31 1.312 1.246 1.143 4.890 -0.2657 Hungary Q1 1995-Q1 2010 60 0.659 1.050 -1.409 6.071 0.5148 Indonesia Q1 1997-Q1 2010 52 0.862 2.256 -4.002 19.399 0.4188 India Q1 2004-Q4 2009 50 1.408 1.891 0.212 3.985 0.4869 Israel Q4 1992-Q1 2010 69 1.005 1.598 0.322 3.546 -0.1459 Korea Q4 1992-Q4 2009 68 1.220 1.876 -2.862 16.598 0.1797 Malaysia Q4 1992-Q4 2009 68 1.340 1.700 -1.477 5.785 0.5994 Mexico Q4 1992-Q1 2010 69 0.608 1.698 -2.338 9.797 0.3369 Morocco Q4 1992-Q4 2009 68 1.099 4.160 0.499 5.463 -0.4559 Peru Q4 1992-Q4 2009 68 1.283 1.526 -0.626 2.832 0.3127 Philippines Q4 1992-Q1 2010 69 1.284 1.271 3.023 20.185 0.1233 Poland Q1 1995-Q4 2009 59 1.141 1.525 0.039 6.921 -0.2384 Russia Q1 1995-Q3 2009 58 0.887 2.551 -2.316 11.466 0.097 South Africa Q4 1992-Q4 2009 68 1.131 1.689 0.319 6.888 -0.0804 Taiwan Q4 1992-Q1 2010 69 1.165 1.604 -1.621 8.458 0.3495 Thailand Q1 1993-Q1 2010 68 1.004 2.067 -0.855 5.238 0.393 Turkey Q4 1992-Q4 2009 68 0.976 2.870 -1.109 4.377 0.075 Table 1 shows descriptive statistics for real seasonally adjusted GDP growth of emerging market countries. AR(1) represents the autocorrelation coefficient. The sample spans from December 1992 to March 2010. GDP series are at the quarterly frequency.

setagerggA laicnaniF rof stneicffieoC noitalerrocotuA :2 elbaT swolF :C lenaP setaR tseretnI :B lenaP serusaeM noitaulaV dna snruteR :A lenaP swolF oiloftroP swolF latipaC tseretnI trohS mreT tluafeD dnediviD noitazilatipaC kcotS oitaR oitaR etaR daerpS daerpS oitaR EP dleiY oitaR snruteR 14.0 67.0 45.0 26.0 81.0 38.0 38.0 37.0 21.0anitnegrA 01.0 23.0 07.0 75.0 51.0 82.0 08.0 98.0 lizarB - - 49.0 - 40.0- 88.0 87.0 68.0 92.0 elihC - 60.0 26.0 37.0 60.0 65.0 87.0 68.0 50.0 anihC 82.0 22.0 18.0 26.0 - 95.0 18.0 78.0 cilbupeRhcezC - - 18.0 - 60.0 96.0 88.0 98.0 tpygE 60.0- 50.0 47.0 46.0 90.0 72.0 57.0 37.0 70.0 yragnuH 21.0 05.0 45.0 - 23.0 17.0 66.0 08.0 10.0aisenodnI 73.0 45.0 57.0 - - 19.0 97.0 47.0 12.0 aidnI 61.0 63.0 28.0 - - 31.0 87.0 19.0 80.0learsI 14.0 23.0 98.0 77.0 20.0 17.0 27.0 98.0 40.0 aeroK 72.0 02.0 86.0 86.0 91.0 38.0 28.0 78.0 21.0aysalaM 24.0 43.0 80.0- 46.0 71.0 76.0 07.0 88.0 50.0 ocixeM 60.0 52.0- 47.0 08.0 00.0 - - 29.0 01.0 occoroM 30.0 93.0 37.0 97.0 90.0 37.0 27.0 29.0 60.0 ureP 91.0 52.0 77.0 26.0 60.0 87.0 48.0 68.0 60.0 senippilihP 52.0 03.0 58.0 39.0 10.0- 85.0 48.0 39.0 43.0 dnaloP 43.0 73.0 23.0 95.0 31.0 95.0 66.0 59.0 40.0aissuR 12.0 74.0 68.0 28.0 21.0 38.0 97.0 09.0 90.0acirfAhtuoS - - 37.0 - - 18.0 88.0 - 60.0nawiaT 51.0 76.0 17.0 28.0 - 76.0 47.0 09.0 01.0dnaliahT 13.0 93.0 06.0 24.0 90.0- 52.0 36.0 58.0 30.0 yekruT 15.0 94.0 28.0 68.0 80.0 97.0 39.0 09.0 21.0 SU noitaulavdnasnruterkcotsrofstneicffieocstneserpAlenaP .setagerggalaicnanfifotesehtrofstneicffieocnoitalerrocotuacfiiceps-yrtnuocswohselbatsihT .soitarwoflrofClenaPdnaselbairavetartseretnirofBlenap,serusaem

setagerggA laicnaniF rof stneicffieoC noitalerroc-ssorC :3 elbaT swolF :C lenaP setaR tseretnI :B lenaP serusaeM noitaulaV dna snruteR :A lenaP swolF oiloftroP swolF latipaC tseretnI trohS mreT tluafeD dnediviD noitazilatipaC kcotS oitaR oitaR etaR daerpS daerpS oitaR EP dleiY oitaR snruteR 10.0 21.0 30.0 81.0- 10.0 61.0 60.0- 61.0 40.0 anitnegrA 10.0 70.0 12.0- 03.0 10.0 00.0 60.0- 03.0 . lizarB . . 10.0- . 00.0 60.0 30.0- 30.0- 11.0elihC . 00.0 90.0- 91.0- 00.0 10.0 91.0- 21.0 10.0anihC 00.0 30.0- 70.0- 80.0- . 10.0 02.0- 43.0 . .peRhcezC . . 41.0- . 51.0 50.0 82.0 40.0 . tpygE 10.0- 20.0- 40.0 61.0 00.0 10.0 71.0 61.0- 20.0yragnuH 40.0- 30.0- 61.0 . 70.0 60.0- 00.0 10.0 00.0 aidnI 60.0- 92.0 70.0 . . 90.0- 90.0- 70.0 10.0 aisenodnI 10.0 10.0 00.0 . . 40.0 10.0- 02.0 10.0 learsI 50.0 30.0 81.0 40.0- 10.0 62.0 52.0- 34.0 00.0 aeroK 01.0- 50.0- 10.0 21.0 40.0 21.0- 61.0- 41.0 50.0 aysalaM 80.0 70.0 30.0- 80.0- 50.0 60.0 31.0- 44.0 20.0ocixeM 00.0 80.0- 61.0 01.0- 20.0- . . 20.0- 00.0 occoroM 10.0- 50.0 31.0 20.0 10.0- 11.0- 11.0- 82.0 00.0 ureP 10.0 60.0 80.0 90.0- 40.0- 50.0 11.0 60.0 10.0 senippilihp 60.0 20.0 61.0 61.0- 20.0 10.0 61.0- 20.0 00.0 dnaloP 20.0 60.0 80.0- 01.0 10.0 30.0 32.0- 43.0 20.0 aissuR 10.0 21.0 72.0- 71.0 30.0 20.0 31.0- 30.0- 10.0 acirfAhtuoS . . 51.0- . . 72.0 93.0- . 40.0 nawiaT 10.0 30.0- 61.0- 00.0 . 40.0 10.0 43.0 20.0 dnaliahT 40.0- 00.0 80.0- 31.0- 10.0- 10.0- 92.0- 42.0 20.0yekruT 41.0 60.0 61.0- 21.0- 10.0 12.0- 73.0- 22.0 50.0- SU stneicffieoC .ssecorpevissergerotuaredro-tsrfis’rotciderpehtdnanoissergerevitciderpehtnismretrorreehtfostneicffieocnoitalerrocswohselbatsihT .rettebrolevel%5ehttaecnacfiingislacitsitatsetacidnidlobni

)stluseR noitamitsE delooP( ytilibatciderP htworG tuptuO :4 elbaT C lenaP B lenaP A lenaP daehA sraeY owT daehA raeY enO daehA retrauQ enO DF EF delooP DF EF delooP DF EF delooP dna snruteR serusaeM noitaulaV 04.0 44.0 63.0 26.0 76.0 26.0 94.0 15.0 15.0 )%( snruteR kcotS 99.3 93.4 27.3 91.5 32.6 93.5 78.6 80.8 35.7 11.0 31.0- 70.0- 11.0 21.0- 80.0- 22.0 10.0 40.0oitaR noitazilatipaC 68.0 05.1- 65.1- 95.0 66.2- 94.1- 97.0 42.0 76.0- 04.1 20.0 20.0- 85.0 60.0- 70.0- 03.0- 03.0- 72.0dleiY dnediviD 95.0 03.0 83.0- 34.0 77.0- 42.1- 95.0- 95.3- 15.4- 10.0 70.0 50.0 20.0- 10.0 10.0- 60.0 60.0 50.0 oitaR EP 12.0 06.2 65.1 06.0- 83.0 72.0- 26.1 98.1 85.1 setaR tseretnI 21.0- 10.0 10.0 40.0- 40.0 10.0- 21.0- 90.0- 01.0etaR tseretnI trohS 38.0- 91.0 12.0 63.0- 93.0 60.0- 42.1- 21.1- 03.1- 40.0 32.0 01.0 42.0 74.0 22.0 33.0 74.0 82.0 daerpS mreT 22.0 02.1 60.1 49.1 53.2 44.2 14.4 49.4 07.3 53.0 23.0- 61.0- 92.0- 35.0- 24.0- 63.0- 54.0- 34.0daerpS tluafeD 41.2 45.4- 92.2- 95.1- 60.5- 08.3- 90.3- 51.4- 89.3swolF 41.0- 92.0- 91.0- 01.0 40.0- 50.0- 34.0 43.0 62.0 oitaR swolF latipaC 97.1- 65.3- 84.4- 93.1 55.0- 61.1- 74.3 01.3 67.2 10.0- 50.0- 30.0- 11.0 01.0 01.0 03.0 03.0 92.0 oitaR swolF oiloftroP 11.0- 47.0- 86.0- 83.1 03.1 54.1 41.3 12.3 21.3 gninaemeddrawrofdna,stceffedexfi,deloopdradnatsruorofscitsitats-tdnastneicffieocdetamitseswohselbatsihT tahtgnittesetairavibanielbairav-yb-elbairavdetamitseerasnoissergerevitciderpehT .noitamitsehtworgtuptuo CdnaBlenaPelihw,stsacerofdaeharetrauqenorofstlusersebircsedAlenaP .htworgPDGdeggalsedulcniosla %5ehttaecnacfiingislacitsitatsetacidnidlobnistneicffieoC .snozirohgnitsacerofdaehasraeyowtdnaenorof .rettebrolevel

Table 5: Country Level Estimation: Returns and Valuation Measures Stock Returns Cap. Ratio PE Ratio Dividend Yield Beta T-stat Beta T-stat Beta T-stat Beta T-stat Argentina 0.05 2.08 18.87 2.30 0.10 1.87 -0.47 -2.02 Brazil - - 0.45 1.57 -0.03 -0.59 -0.44 -2.16 China 0.03 1.72 0.03 0.17 -0.05 -2.45 0.84 1.64 Chile 0.01 0.70 0.45 1.28 0.08 0.86 -0.43 -1.44 CzechRep. - - -0.90 -1.10 0.00 0.73 -0.04 -0.55 Egypt - - 0.45 3.86 0.08 2.26 -0.08 -1.72 Hungary 0.01 0.74 0.00 -1.87 0.00 -3.01 -0.56 -4.00 Indonesia 0.01 1.20 1.07 1.71 0.03 0.66 -0.06 -0.24 India 0.00 0.12 -1.09 -1.57 -0.03 -0.44 0.75 1.22 Israel 0.04 3.82 0.09 0.75 0.00 0.23 -0.12 -1.17 SouthKorea 0.04 2.37 0.00 0.89 0.11 2.89 -1.86 -2.96 Malaysia 0.03 2.10 0.24 3.43 0.09 2.50 -0.48 -2.68 Mexico 0.07 2.96 0.00 0.44 0.01 0.04 -1.91 -2.68 Morocco -0.01 -0.46 -7.47 -0.49 - - - - Peru 0.03 2.80 0.58 0.68 0.04 2.10 -0.22 -1.74 Philippines 0.00 0.20 -0.01 -0.41 0.00 0.19 0.44 1.91 Poland 0.01 0.98 -0.33 -0.15 -0.01 -0.18 -0.09 -0.71 Russia 0.02 1.38 0.01 1.58 0.06 3.43 -1.15 -1.77 SouthAfrica 0.02 1.04 0.00 -0.47 -0.03 -0.51 0.38 1.05 Taiwan 0.03 2.48 - - 0.06 2.22 -0.17 -1.03 Thailand 0.04 4.68 3.79 1.78 0.04 1.49 -0.50 -2.40 Turkey 0.04 2.49 32.38 1.10 0.00 1.40 -0.25 -0.58 Thistablepresentsresultsforourtimeseriespredictiveregressionsovertheentiresampleperiod. Werun bivariateregressionsforthesetofreturnandvaluationmeasuresanduseNewey-Weststandarderrors. Coefficientsinboldindicatestatisticalsignificanceatthe5%levelorbetter.

Table 6: Country Level Estimation: Interest Rates Default Spread Term Spread Short Interest Rate Beta T-stat Beta T-stat Beta T-stat Argentina -0.03 -4.91 0.06 3.32 -0.08 -2.68 Brazil -0.01 -1.30 0.00 -0.13 -0.06 -2.67 China 0.00 0.27 - - -0.18 -1.14 Chile -0.01 -0.58 0.34 2.04 -0.03 -1.31 CzechRep. - - 0.07 0.70 -0.22 -1.59 Egypt 0.00 0.84 - - -0.06 -1.35 Hungary -0.01 -0.95 -0.23 -2.11 -0.15 -1.40 Indonesia -0.01 -2.69 - - 0.02 0.84 India - - - - 0.04 0.41 Israel - - - - 0.02 0.34 SouthKorea -0.02 -6.94 0.50 1.30 0.08 0.80 Malaysia -0.02 -3.79 0.25 1.09 0.09 0.38 Mexico -0.03 -2.01 0.17 1.13 -0.03 -1.02 Morocco -0.01 -0.68 0.26 1.02 0.03 0.24 Peru -0.02 -2.52 0.05 2.81 -0.05 -2.28 Philippines 0.00 -0.61 0.01 0.15 -0.07 -1.75 Poland 0.00 -2.67 0.09 1.68 -0.07 -1.32 Russia -0.01 -0.80 0.04 0.57 0.00 -0.35 SouthAfrica 0.00 -0.30 0.29 2.73 -0.02 -0.16 Taiwan - - - - 0.09 0.60 Thailand - - 0.54 3.26 -0.07 -0.64 Turkey -0.01 -0.36 0.37 0.91 0.00 0.30 Thistablepresentsresultsforourtimeseriespredictiveregressionsoverthe entiresampleperiod. Werunbivariateregressionsforthesetofreturnand valuationmeasuresanduseNewey-Weststandarderrors. Coefficientsin boldindicatestatisticalsignificanceatthe5%levelorbetter.

soitaR wolF latipaC tnemtsevnI tceriD dna oiloftroP :7 elbaT daehA sraeY owT daehA raeY enO daehA retrauQ enO DF EF delooP DF EF delooP DF EF delooP 30.0 51.0- 90.0- 70.0 50.0- 20.0- 91.0 41.0 31.0 seitiruceS tbeD 03.0 59.1- 76.1- 76.0 95.0- 83.0- 97.3 16.2 34.2 20.0 90.0 40.0- 11.0 21.0 20.0 41.0 80.0 40.0 tnemtsevnI tceriD 31.0 88.0 12.1- 08.1 21.2 15.0 33.2 92.1 18.0 13.0- 21.0 90.0 70.0- 51.0 21.0 60.0 61.0 51.0 seitiruceS ytiuqE 94.1- 41.1 32.1 37.0- 70.2 18.1 18.0 92.2 03.2 gninaemeddrawrofdna,stceffedexfi,deloopdradnatsruorofscitsitats-tdnastneicffieocepolsstneserpelbatsihT ytiuqedna,tbed,latipactnemtsevnitceridotdetalerswoflnitenforewopevitciderpehttseteW .sledom tceridfomusehtsilatipactnemtsevnitceriD .sdoirepdaehasraey-owtdna,raey-eno,tneuqesbusrevo .seitiruces dnaswoflnihtobfomustenehtsiseitiruceStbeD .ymonocegnitroperehtnitnemtsevnitceriddnadaorbatnemtsevni dnastessafomusehtsiseitiruceSytiuqE .stnemurtsniemocnidexfignirevocseitilibaildnastessafoswofltuo snoissergeretairavibnureW .ytiuqefopihsrenwotneserpertahtstnemurtsniralimisdna,skcotsgnirevocseitilibail ecnacfiingislacitsitatsetacidnidlobnistneicffieoC .htworgPDGdeggaldnatseretnifoelbairavehtedulcnitaht .rettebrolevel%5ehtta

Table 8: Country Level Estimation: Flows Ratios Capital Flows Portfolio Flows Beta T-stat Beta T-stat Argentina 8.74 0.44 27.70 1.43 Brazil 17.60 2.20 25.31 2.73 Chile -2.09 -0.62 - - CzechRep. 2.99 0.82 0.47 0.10 Hungary 2.66 1.85 3.75 1.24 Indonesia 31.93 3.47 3.75 2.16 India 17.38 0.80 42.57 1.12 Israel -2.48 -1.08 4.51 1.05 SouthKorea 21.41 2.43 6.82 1.11 Malaysia 5.07 1.67 7.14 1.78 Mexico 58.01 1.66 94.48 2.64 Morocco 5.33 1.13 38.12 1.01 Peru 8.66 2.35 18.49 2.55 Philippines -0.97 -0.93 7.69 2.84 Poland -0.97 -0.95 7.69 4.67 Russia 4.80 0.57 2.90 0.22 SouthAfrica 29.15 1.42 7.76 0.48 Thailand 8.83 2.09 -0.89 -0.08 Turkey -0.25 -0.04 11.20 0.51 Thistablepresentsresultsforourtimeseriespredictive regressionsovertheentiresampleperiod. Werun bivariateregressionsforthemeasuresofflowsanduse Newey-Weststandarderrors. Coefficientsinboldindicate statisticalsignificanceatthe5%levelorbetter.

)stluseR noitamitsE delooP( setagerggA labolG :9 elbaT daehA sraeY owT daehA raeY enO daehA retrauQ enO DF EF delooP DF EF delooP DF EF delooP stekraM ytidommoC :A lenaP 73.0 12.0 51.0 43.0 82.0 22.0 14.0 04.0 93.0 )%( xednI ytidommoC 12.5 73.3 63.2 00.5 32.4 00.3 17.6 03.7 00.7 83.0 30.0 10.0 53.0 12.0 81.0 93.0 53.0 53.0 )%( xednI larutlucirgA P&S 58.21 26.0 12.0 83.6 48.2 63.2 61.01 70.9 00.9 32.0 51.0 11.0 52.0 22.0 71.0 13.0 13.0 03.0 )%( xednI ygrenE P&S 83.3 48.2 59.1 93.4 70.4 27.2 19.4 04.5 11.5 10.0 01.0 60.0 91.0 42.0 12.0 41.0 61.0 61.0 )%( xednI kcotseviL P&S 81.0 55.1 19.0 08.2 94.3 80.3 15.2 28.2 17.2 21.0 02.0 02.0 91.0 03.0 03.0 74.0 63.0 73.0 )%( xednI lateM P&S 13.0 91.3 55.3 56.0 65.4 76.4 66.11 75.6 05.6 setaR tseretnI SU :B lenaP 04.0- 40.0 01.0 32.0- 70.0- 00.0 90.0 72.0- 32.0daerpS tluafeD s’ydooM 68.1- 17.0 86.1 55.1- 68.0- 10.0- 33.0 31.4- 63.3- 70.0 51.0- 41.0- 20.0 60.0- 50.0- 20.0- 40.0- 30.0llib-T SU 96.1 49.2- 08.2- 73.0 90.1- 19.0- 52.0- 86.0- 06.0- 01.0 02.0 12.0 80.0 80.0 01.0 71.0 90.0 90.0 daerpS mreT SU 62.1 98.3 21.4 58.0 40.2 93.2 80.2 40.2 12.2 htworgtuptuogninaemeddrawrofdna,stceffedexfi,deloopdradnatsruorofscitsitats-tdnastneicffieocepolsstneserpelbatsihT ecnamrofrepehtforewopevitciderpehtstsetAlenaP .setagerggalaicnanfilabolgforewopevitciderpehttsetewnehwnoitamitse ,raeyeno,retrauqenoredisnoceW .selbairavsetartseretniSUfotesehtrofstlusersebircsedBlenaPelihwstekramytidommocfo sedulcnioslatahtgnittesetairavibanielbairav-yb-elbairavdetamitseerasnoissergeR .snozirohgnitsacerofdaehasraeyowtdna .rettebrolevel%5ehttaecnacfiingislacitsitatsetacidnidlobnistneicffieoC .htworgPDGdeggal

Table 10: Country Level Estimation: Commodity Markets Commodities Energy Agriculture Livestock Precious Metals Beta T-stat Beta T-stat Beta T-stat Beta T-stat Beta T-stat Argentina 0.03 1.23 0.01 0.36 0.05 1.36 0.03 0.97 0.15 2.29 Brazil 0.05 2.71 0.02 1.26 0.08 3.54 0.04 1.38 0.08 3.02 China 0.00 0.05 -0.01 -0.48 0.04 1.53 0.05 1.43 0.12 1.97 Chile 0.03 3.28 0.02 2.22 0.04 1.77 0.01 0.63 0.04 1.09 CzechRep. 0.06 3.92 0.04 2.98 0.05 2.34 0.02 0.91 0.01 0.21 Egypt 0.00 0.10 0.00 -0.13 0.03 2.45 0.02 0.80 0.04 1.49 Hungary 0.04 2.27 0.02 2.01 0.02 1.01 0.02 1.50 0.00 0.11 Indonesia 0.04 1.38 0.03 1.40 0.01 0.65 0.00 0.13 0.06 1.29 India 0.02 1.42 0.01 0.91 0.02 0.79 0.08 2.27 0.00 0.03 Israel 0.03 1.44 0.02 1.36 0.03 1.56 -0.01 -0.27 0.03 1.05 SouthKorea 0.04 2.17 0.03 1.84 0.04 2.12 0.05 2.01 0.09 1.67 Malaysia 0.06 3.61 0.04 3.21 0.05 2.33 0.02 0.68 0.07 2.40 Mexico 0.06 2.12 0.04 1.92 0.03 1.00 0.02 0.69 0.02 1.02 Morocco -0.05 -2.03 -0.04 -1.88 0.00 -0.10 0.06 1.03 0.14 1.66 Peru 0.02 1.14 0.01 0.35 0.05 2.68 -0.03 -1.53 0.08 3.43 Philippines 0.02 3.21 0.01 2.45 0.01 0.94 0.06 1.84 0.03 1.46 Poland 0.02 2.60 0.01 2.16 0.04 3.90 -0.03 -1.11 0.03 1.28 Russia 0.10 2.72 0.07 2.37 0.04 1.40 0.08 2.22 0.07 1.30 SouthAfrica 0.02 1.33 0.00 0.29 0.05 2.75 0.05 1.46 0.03 1.52 Taiwan 0.04 1.88 0.02 1.31 0.05 1.59 0.01 0.78 0.07 1.95 Thailand 0.04 1.89 0.02 1.12 0.06 2.04 -0.02 -0.70 0.10 2.32 Turkey 0.07 2.83 0.05 2.28 0.07 2.42 0.00 0.06 0.05 0.68 Thistablepresentsresultsforourtimeseriespredictiveregressionsovertheentiresampleperiod. Werunbivariate regressionsforthesetofreturnsoncommoditymarketsanduseNewey-Weststandarderrors. Coefficientsinbold indicatestatisticalsignificanceatthe5%levelorbetter.

Table 11: Country Level Estimation: US Interest Rates Term Spread T-bill Default Spread Beta T-stat Beta T-stat Beta T-stat Argentina 0.19 0.57 -0.36 -1.60 -0.32 -0.67 Brazil 0.09 0.54 0.09 0.62 -0.38 -0.95 China 0.30 0.96 -0.41 -2.22 -0.24 -0.66 Chile 0.07 0.38 0.05 0.37 -0.72 -3.20 CzechRep. 0.01 0.07 0.08 0.64 -1.15 -3.29 Egypt -0.18 -1.89 0.07 0.60 -0.36 -2.24 Hungary -0.08 -0.51 0.11 0.93 -1.20 -4.25 Indonesia 0.28 1.01 -0.32 -1.19 0.56 0.90 India -0.12 -0.40 -0.24 -1.49 -1.47 -3.15 Israel -0.11 -0.95 0.13 1.75 -0.65 -3.18 SouthKorea 0.13 0.68 0.17 0.73 -0.28 -0.61 Malaysia 0.34 1.63 0.04 0.17 -0.66 -1.20 Mexico -0.14 -1.24 0.15 1.07 -1.25 -2.44 Morocco -0.11 -0.54 -0.10 -0.56 0.51 1.53 Peru 0.29 1.30 -0.13 -1.01 -0.57 -1.80 Philippines 0.08 0.56 -0.19 -1.94 0.26 0.71 Poland -0.18 -1.39 0.12 1.77 -0.48 -2.66 Russia 0.04 0.15 -0.12 -0.59 -1.61 -1.55 SouthAfrica -0.05 -0.28 -0.11 -0.81 -0.76 -2.87 Taiwan 0.19 0.97 0.21 0.94 -0.45 -0.85 Thailand 0.40 1.77 -0.08 -0.28 -0.25 -0.38 Turkey 0.25 0.76 0.17 0.62 -0.76 -1.12 Thistablepresentsresultsforourtimeseriespredictiveregressionsoverthe entiresampleperiod. WerunbivariateregressionsforthesetofUSinterest ratesanduseNewey-Weststandarderrors. Coefficientsinboldindicate statisticalsignificanceatthe5%levelorbetter.

srotcaF ksiR tekraM dna elytS ,eziS :21 elbaT daehA sraeY owT daehA raeY enO daehA retrauQ enO DF EF delooP DF EF delooP DF EF delooP 25.0 54.0 74.0 73.0 63.0 93.0 80.0- 30.0- 10.0eziS 19.7 26.6 11.7 46.3 00.4 55.4 32.1- 14.0- 12.0- 73.0- 30.0- 40.0- 50.0- 80.0 50.0 50.0 01.0 01.0 tekraM-ot-kooB 27.5- 95.0- 98.0- 88.0- 98.1 71.1 97.0 08.1 37.1 74.0 73.0 82.0 76.0 56.0 95.0 55.0 65.0 55.0 tekraM 56.6 63.5 30.4 91.9 28.9 94.8 28.8 26.8 83.8 drawrofdna,stceffedexfi,deloopdradnatsruorofscitsitats-tdnastneicffieocepolsstneserpelbatsihT eW .stekraMgnigremErofsrotcafhcnerFdnaamaFforewopevitciderpehttsetewnehwsledomgninaemed detamitseerasnoissergeR .ytilibatciderphtworgtuptuodaehasraey-owtdna,raey-eno,tneuqesbusroftset dlobnistneicffieoC .htworgPDGdeggalsedulcnioslatahtgnittesetairavibanielbairav-yb-elbairav .rettebrolevel%5ehttaecnacfiingislacitsitatsetacidni

Table 13: Output Growth Predictability in the US Beta T-stat Beta GDP T-stat Panel A: Return and Valuation Measures Stock Returns (%) 0.14 1.80 0.25 2.09 Capitalization Ratio 0.00 -0.01 0.31 2.35 Dividend Yield -0.04 -0.63 0.30 2.39 PE Ratio 0.03 0.38 0.33 2.04 Panel B: Interest Rates Short Interest Rate 0.18 1.90 0.26 2.39 Term Spread 0.00 -0.08 0.31 2.47 Default Spread -0.27 -3.17 0.12 1.14 Panel C: Flows Capital Flows Ratio -0.16 -2.04 0.30 2.23 Portfolio Flows Ratio -0.13 -2.12 0.29 2.35 Panel D: Commodity Markets Commodity Index (%) -0.04 -0.46 0.33 2.59 S&P Agricultural Index (%) -0.02 -0.27 0.32 2.51 S&P Energy Index (%) -0.05 -0.65 0.33 2.55 S&P Livestock Index (%) 0.05 0.98 0.31 2.48 S&P Metal Index (%) -0.02 -0.34 0.31 2.48 Panel E: Risk Factors Book-to-market 0.06 0.73 0.31 2.24 Size -0.24 -3.00 0.22 1.77 This table presents estimation results for the US over the period covering December 1992 to March 2010. FOr each variable we run bivariate regressions that include lagged GDP growth. We report estimates and t-statistics of the variable of interest and lagged GDP growth. Panel A shows results for returns and valuation measures, Panel B for our set of interest rates variables, Panel C for flows, Panel D for performance of global and sector commodity markets and Panel E for Fama and French risk factors.

Table 14: Common Factors Panel A: Returns and Valuation Measures Beta RP Beta RFE Beta RFD Stock Returns 0.33 0.26 0.24 3.03 2.51 2.63 Capitalization Ratio 0.22 0.18 0.42 1.56 1.21 0.10 Dividend Yield 0.01 -0.30 -0.30 0.07 -2.70 -1.05 PE Ratio 0.25 0.09 0.10 2.03 1.00 0.93 Panel B: Interest Rates Beta RP Beta RFE Beta RFD Short Interest Rate -0.11 -0.16 -0.22 -1.41 -2.05 -2.17 Term Spread 0.39 0.63 0.55 2.04 4.05 2.32 Default Spread -0.15 -0.24 -0.24 -0.66 -1.13 -1.04 Panel C: Flows Beta RP Beta RFE Beta RFD Capital Flows Ratio 0.30 0.31 0.39 3.31 2.89 2.78 Portfolio Flows Ratio 0.29 0.26 0.29 2.52 2.21 2.32 This table shows estimated coefficients and t-statistics for our standard pooled, fixed effects, and forward demeaning output growth estimation when we account for common factors in the data generating process. Panel A describes results for returns and valuation measures, Panel B covers interest rates variables, and Panel C portfolio and capital flows. The coefficients in bold indicate statistical significance at the 5 % level or better.

sisylanA elpmaS-buS :51 elbaT 7002-3991 :C lenaP 0102-0002 :B lenaP 9991-3991 :A lenaP DF ateB EF ateB delooP ateB DF ateB EF ateB delooP ateB DF ateB EF ateB delooP ateB 34.0 64.0 64.0 44.0 74.0 74.0 05.0 95.0 55.0 snruteR kcotS 12.5 81.6 47.5 82.4 79.4 68.4 15.3 85.4 08.3 30.0- 00.0 20.0- 57.31- 30.0 90.0- 10.0 51.0 60.0 oitaR noitazilatipaC 92.0- 40.0 03.0- 61.0- 69.0 97.1- 50.0 00.1 86.0 21.1- 52.0- 42.0- 20.0- 42.0- 02.0- 81.1- 08.0- 14.0dleiY dnediviD 44.2- 79.1- 36.2- 60.0- 77.2- 75.3- 89.0- 96.5- 97.3- 60.0- 50.0 40.0 20.0 30.0 20.0 06.0 14.0 43.0 oitaR EP 06.1- 46.1 63.1 21.1 87.2 94.1 54.2 49.1 68.1 81.0- 51.0- 51.0- 92.0- 31.0- 31.0- 01.0- 70.0- 90.0etaR tseretnI trohS 96.1- 93.1- 34.1- 12.5- 01.2- 96.2- 06.0- 94.0- 86.0- 54.0 06.0 33.0 71.0 43.0 81.0 07.2 04.1 65.0 daerpS mreT 43.5 11.6 28.3 88.3 30.7 08.3 80.4 26.2 71.4 93.0- 83.0- 63.0- 62.0- 34.0- 24.0- 19.0- 06.0- 65.0daerpS tluafeD 76.2- 64.2- 82.2- 10.2- 68.2- 58.2- 75.2- 35.1- 25.1- 24.0 23.0 52.0 72.0 61.0 90.0 21.1 07.0 46.0 oitaR swolF latipaC 25.2 50.2 19.1 06.2 03.1 49.0 06.3 44.2 67.2 03.0 32.0 22.0 72.0 72.0 62.0 05.0 93.0 93.0 oitaR swolF oiloftroP 71.2 67.1 57.1 26.4 36.4 63.4 27.1 93.1 43.1 41.0 02.0 02.0 14.0 34.0 93.0 40.0 91.0 81.0 seitidommoC 66.2 36.4 07.4 19.5 43.6 24.5 52.0 27.1 86.1 90.0 40.0 30.0 31.0 91.0 91.0 74.0- 24.0- 14.0tekram-ot-kooB 14.1 07.0 95.0 81.2 61.3 21.3 04.2- 14.2- 73.2- 21.0- 60.0- 50.0- 11.0- 20.0- 20.0- 21.0- 30.0- 10.0eziS 84.1- 29.0- 96.0- 35.2- 73.0- 53.0- 57.0- 61.0- 60.0- 83.0 44.0 44.0 94.0 15.0 94.0 75.0 95.0 95.0 tekraM 55.6 86.7 85.7 77.5 72.6 69.5 32.4 97.4 08.4 lacitsitatsetacidnidlobnistneicffieocehT .7002-3991dna0102-0002,9991-3991gnirevocsdoirepehtrofstluserelpmas-busswohselbatsihT .rettebrolevel%5ehttaecnacfiingis

stekraM gnigremE lanoigeR :61 elbaT MATAL :C lenaP AEME :B lenaP aisA :A lenaP DF ateB EF ateB delooP ateB DF ateB EF ateB delooP ateB DF ateB EF ateB delooP ateB 26.0 66.0 66.0 25.0 45.0 25.0 14.0 14.0 14.0 snruteR kcotS 94.6 07.7 69.7 04.5 54.6 10.6 90.4 52.4 32.4 90.1 50.0 21.0- 40.0 30.0- 10.0 11.41- 41.0 40.0oitaR noitazilatipaC 76.5 88.0 64.3- 94.0 80.1- 13.0 26.0- 18.3 06.0- 35.2- 85.0- 04.0- 53.0 92.0- 52.0- 11.0- 90.0- 01.0dleiYdnediviDf ¯ 55.1- 19.4- 12.3- 84.1 33.2- 48.2- 70.0- 57.0- 99.0- 51.0 42.0 72.0 90.0 80.0 60.0 40.0 41.0 02.0 oitaR EP 05.1 99.1 55.2 86.1 80.2 57.1 94.0 11.1 25.2 15.0- 44.0- 04.0- 30.0- 40.0- 40.0- 32.0- 10.0- 40.0etaR tseretnI trohS 02.5- 32.5- 93.5- 64.0- 68.0- 39.0- 70.2- 82.0- 02.1- 32.0 65.0 42.0 72.0 72.0 22.0 30.0- 44.0 52.0 daerpS mreT 85.2 74.6 84.3 08.4 46.2 04.2 11.0- 57.1 14.1 26.0- 46.0- 46.0- 11.0- 52.0- 52.0- 56.0- 17.0- 96.0daerpS tluafeD 83.5- 92.8- 94.8- 80.2- 36.3- 98.3- 91.4- 32.4- 68.3- 82.0 22.0 52.0 53.0 31.0 50.0 55.0 06.0 65.0 oitaR swolF latipaC 94.1 32.1 54.1 06.3 07.1 77.0 89.1 15.2 74.2 55.0 94.0 74.0 41.0 31.0 11.0 04.0 44.0 54.0 oitaR swolF oiloftroP 65.4 91.4 92.4 06.3 83.4 49.3 97.1 01.2 01.2 53.0 93.0 83.0 74.0 44.0 44.0 13.0 72.0 52.0 seitidommoC 01.3 97.3 57.3 16.3 65.3 65.3 87.5 46.6 69.4 80.0- 50.0- 50.0- 31.0 02.0 02.0 30.0 01.0 90.0 tekram-ot-kooB 58.0- 94.0- 05.0- 94.1 74.2 04.2 53.0 80.1 40.1 50.0- 40.0 40.0 32.0- 81.0- 81.0- 41.0 61.0 71.0 eziS 15.0- 35.0 45.0 00.3- 98.2- 08.2- 79.1 89.1 34.2 95.0 56.0 56.0 45.0 25.0 25.0 84.0 74.0 54.0 tekraM 60.6 34.6 83.6 78.4 43.4 83.4 80.5 91.5 10.5 ClenaP,acirfAdnatsaEelddiM,eporuErofBlenaP,stekramgnigremenaisArofstluserstneserpAlenaP .levellanoigerehttastluserswohselbatsihT .rettebrolevel%5ehttaecnacfiingislacitsitatsetacidnidlobnistneicffieocehT .anihC-xeaisArofDlenaPdnaaciremAnitaLrof

Regional Emerging Markets (Continued) Panel D: Asia ex-China Beta Pooled Beta FE Beta FD Stock Returns 0.38 0.38 0.35 3.86 3.91 3.54 Capitalization Ratio 0.02 0.12 -14.34 0.88 3.38 -0.56 Dividend Yield -0.14 -0.13 0.20 -1.46 -1.12 0.25 PE Ratio 0.24 0.26 0.14 4.72 4.86 1.42 Short Interest Rate -0.01 -0.01 -0.24 -0.52 -0.43 -2.10 Term Spread 0.25 0.44 -0.03 1.41 1.75 -0.11 Default Spread -0.79 -0.79 -0.74 -4.17 -4.12 -4.65 Capital Flows Ratio 0.56 0.60 0.55 2.47 2.51 1.98 Portfolio Flows Ratio 0.45 0.44 0.40 2.10 2.10 1.79 Commodities 0.29 0.30 0.35 7.83 7.87 7.33 Book-to-market 0.05 0.05 -0.02 0.57 0.56 -0.23 Size 0.22 0.22 0.19 3.66 3.60 3.55 Market 0.47 0.48 0.48 4.85 4.88 4.68

atipaC rep PDG :71 elbaT emocnI woL :C lenaP emocnI elddiM :B lenaP emocnI hgiH :A lenaP DF ateB EF ateB delooP ateB DF ateB EF ateB delooP ateB DF ateB EF ateB delooP ateB 04.0 63.0 93.0 65.0 56.0 36.0 34.0 64.0 54.0 snruteR kcotS 29.2 72.3 32.3 90.5 20.6 86.5 08.3 75.4 12.4 64.0- 01.0- 10.0- 70.0 30.0- 60.0- 79.31- 62.0 41.0 oitaR noitazilatipaC 82.2- 06.2- 31.0- 55.0 87.0- 04.1- 36.0- 62.2 11.9 13.0- 01.0- 31.0- 75.0- 26.0- 14.0- 83.2- 42.0- 62.0dleiY dnediviD 75.0- 58.0- 34.1- 78.0- 30.6- 57.5- 85.0- 91.2- 08.2- 10.0- 01.0 91.0 90.0 90.0 90.0 70.0 40.0 20.0 oitaR EP 01.0- 76.0 55.2 03.1 65.1 65.1 50.1 91.1 29.0 92.0- 12.0- 32.0- 31.0- 70.0- 70.0- 01.0 50.0- 20.0etaR tseretnI trohS 27.2- 29.1- 10.2- 60.1- 28.0- 78.0- 80.1 84.0- 62.0- 43.0 15.0 64.0 43.0 84.0 52.0 02.0 60.0 30.0 daerpS mreT 28.1 50.2 43.2 06.3 08.6 26.3 53.1 25.0 84.0 80.0- 11.0- 21.0- 14.0- 05.0- 94.0- 15.0- 46.0- 46.0daerpS tluafeD 28.0- 80.1- 51.1- 17.2- 52.4- 21.4- 84.1- 78.1- 68.1- 16.0 25.0 15.0 02.0 22.0 61.0 15.0 72.0 41.0 oitaR swolF latipaC 04.2 02.2 44.2 98.1 20.2 49.1 54.2 13.1 00.1 35.0 15.0 15.0 72.0 13.0 03.0 51.0 31.0 21.0 oitaR swolF oiloftroP 20.2 19.1 29.1 75.2 73.3 93.3 93.4 64.4 49.3 71.0 02.0 91.0 65.0 65.0 65.0 84.0 14.0 14.0 seitidommoC 09.2 75.3 53.3 18.4 51.5 21.5 11.01 57.9 24.9 40.0- 10.0 00.0 80.0 41.0 41.0 90.0 71.0 61.0 tekram-ot-kooB 24.0- 11.0 20.0 87.0 33.1 23.1 50.1 62.2 42.2 30.0- 10.0 30.0 71.0- 80.0- 80.0- 10.0- 30.0 30.0 eziS 62.0- 80.0 72.0 91.2- 31.1- 11.1- 01.0- 33.0 43.0 93.0 24.0 14.0 27.0 57.0 57.0 74.0 44.0 44.0 tekraM 77.4 27.4 36.4 76.8 45.8 84.8 16.4 32.4 12.4 tekramgnigremerofstluserstneserpAlenaP.)DSU5002tnatsnocta(atipacrepPDGdetsujda-PPPybdepuorgslenaprofstluserswohselbatsihT PDGahtiwesohtrofClenaPdna,000,9$dna000,31$neewtebgnignarseirtnuocrofBlenaP,rehgihro000,31$foatipacrepPDGahtiwseirtnuoc .rettebrolevel%5ehttaecnacfiingislacitsitatsetacidnidlobnistneicffieocehT .000,9$nahtrewolatipacrep

sisylanA etairavitluM :81 elbaT daehA sraeY owT daehA raeY enO daehA retrauQ enO DF ateB EF ateB delooP ateB DF ateB EF ateB delooP ateB DF ateB EF ateB delooP ateB 77.1 51.0- 72.0- 03.0 43.0 23.0 04.0 34.0 04.0 dna nruteR 26.0 78.0- 16.2- 93.2 34.3 63.3 18.2 04.4 16.3 serusaeM noitaulaV 53.0- 80.0- 71.0- 61.0- 61.0- 61.0- 23.0- 03.0- 72.0setaR tseretnI 07.0- 92.0- 41.1- 62.1- 81.1- 73.1- 79.4- 06.3- 82.4- 01.0- 73.0- 82.0- 60.0 40.0 10.0- 01.0 90.0 60.0 swolF 51.2 44.2- 72.2- 94.0 72.0 21.0- 72.1 02.2 70.2 21.0- 92.0 92.0 22.0 70.0 70.0 13.0 92.0 92.0 seitidommoC 41.0- 34.1 76.1 40.3 29.0 40.1 62.5 84.5 42.5 tseretni,serusaemnoitaulavdnanruter,seirogetacruofotniselbairavruopuorgeW .sisylanaetairavitlumruorofstluserstneserpelbatsihT .rettebrolevel%5ehttaecnacfiingislacitsitatsetacidnidlobnistneicffieocehT .seitidommocdnaswofl,setar

Table 19: Cross-Validation One Quarter Ahead One Year Ahead Two Years Ahead Pooled FE FD Pooled FE FD Pooled FE FD Stock Returns 0.51 0.52 0.49 0.63 0.67 0.63 0.37 0.44 0.40 Stdv. 0.03 0.03 0.03 0.04 0.04 0.05 0.03 0.03 0.05 No. Sig. 20 20 20 20 20 20 20 20 20 Capitalization Ratio -0.04 0.02 0.48 -0.08 -0.13 0.30 -0.07 -0.15 3.12 Stdv. 0.03 0.06 0.94 0.03 0.05 0.71 0.03 0.10 10.80 No. Sig. 0 2 0 4 20 0 1 2 0 Dividend Yield -0.27 -0.30 -0.34 -0.07 -0.06 0.27 -0.02 0.02 1.35 Stdv. 0.03 0.04 0.20 0.02 0.03 2.13 0.02 0.04 1.08 No. Sig. 20 20 0 0 0 0 0 0 0 PE Ratio 0.05 0.06 0.06 -0.01 0.01 -0.03 0.05 0.07 0.01 Stdv. 0.01 0.01 0.01 0.01 0.01 0.01 0.02 0.02 0.02 No. Sig. 0 4 0 0 0 0 1 17 0 Short Interest Rate -0.11 -0.09 -0.12 -0.01 0.03 -0.05 0.01 0.01 -0.12 Stdv. 0.03 0.03 0.03 0.05 0.05 0.05 0.03 0.04 0.05 No. Sig. 0 0 1 0 0 0 0 0 2 Term Spread 0.29 0.48 0.34 0.23 0.50 0.26 0.11 0.27 0.05 Stdv. 0.03 0.06 0.04 0.04 0.12 0.05 0.05 0.10 0.05 No. Sig. 20 20 20 19 19 6 4 4 0 Default Spread -0.42 -0.44 -0.36 -0.42 -0.53 -0.30 -0.17 -0.33 0.36 Stdv. 0.05 0.05 0.05 0.04 0.03 0.08 0.03 0.03 0.11 No. Sig. 20 20 20 20 20 4 18 20 12 Capital Flows 0.26 0.33 0.44 -0.05 -0.05 0.11 -0.19 -0.29 -0.13 Stdv. 0.03 0.04 0.05 0.02 0.03 0.05 0.02 0.03 0.05 No. Sig. 20 20 20 2 0 2 20 20 5 Portfolio Flows 0.30 0.30 0.31 0.10 0.10 0.12 -0.04 -0.05 -0.01 Stdv. 0.03 0.04 0.03 0.03 0.03 0.04 0.02 0.03 0.04 No. Sig. 20 20 20 0 0 3 0 0 0 Thistablepresentsresultsforourcross-validationanalysis. Foreachvariablewereporttheaverageestimate,its standarddeviationandthenumberoftimes,outofthetwentyrandomdraws,thatthecoefficientis statisticallysignificant.

Cross-Validation (Continued) One Quarter Ahead One Year Ahead Two Years Ahead Pooled FE FD Pooled FE FD Pooled FE FD Debt securities 0.13 0.14 0.19 -0.02 -0.04 0.07 -0.09 -0.15 0.03 Stdv. 0.03 0.03 0.03 0.04 0.04 0.05 0.03 0.03 0.07 No. Sig. 16 18 20 0 0 1 6 7 0 Direct Inv. flows 0.04 0.08 0.13 0.02 0.11 0.10 -0.04 0.07 0.01 Stdv. 0.02 0.03 0.03 0.01 0.02 0.03 0.02 0.04 0.05 No. Sig. 0 2 11 0 7 4 2 0 0 Equity securities 0.16 0.16 0.07 0.12 0.15 -0.06 0.09 0.11 -0.29 Stdv. 0.03 0.03 0.03 0.03 0.04 0.03 0.04 0.06 0.08 No. Sig. 15 15 0 4 11 0 0 1 0 Size -0.01 -0.03 -0.08 0.39 0.35 0.37 0.47 0.45 0.51 Stdv. 0.02 0.02 0.03 0.04 0.04 0.05 0.03 0.03 0.03 No. Sig. 0 0 0 20 20 20 20 20 20 HML 0.09 0.10 0.05 0.05 0.08 -0.05 -0.05 -0.04 -0.37 Stdv. 0.02 0.02 0.02 0.01 0.01 0.03 0.02 0.02 0.05 No. Sig. 3 3 0 1 3 0 0 0 20 Market 0.11 0.10 -0.99 -0.08 -0.09 -1.44 -0.16 -0.17 -1.54 Stdv. 0.02 0.02 0.18 0.02 0.03 0.64 0.02 0.03 0.74 No. Sig. 12 9 20 2 3 3 20 20 0 Commodities 0.38 0.39 0.40 0.22 0.27 0.33 0.15 0.21 0.36 Stdv. 0.02 0.02 0.02 0.03 0.03 0.03 0.02 0.02 0.03 No. Sig. 20 20 20 20 20 20 14 20 20 S&P Agriculture 0.35 0.35 0.39 0.18 0.21 0.35 0.01 0.04 0.38 Stdv. 0.02 0.02 0.02 0.03 0.03 0.02 0.02 0.02 0.05 No. Sig. 20 20 20 17 20 20 0 0 20 S&P Energy 0.29 0.30 0.30 0.17 0.22 0.25 0.11 0.15 0.22 Stdv. 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.03 No. Sig. 20 20 20 19 20 20 8 20 20 S&P Livestock 0.15 0.16 0.14 0.21 0.24 0.19 0.06 0.10 0.01 Stdv. 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 No. Sig. 18 20 12 20 20 19 0 0 0 S&P MetalS 0.38 0.37 0.47 0.30 0.30 0.59 0.20 0.21 0.83 Stdv. 0.02 0.02 0.02 0.02 0.02 0.05 0.02 0.02 0.11 No. Sig. 20 20 20 20 20 20 20 20 20