Errors in survey expectations display waves of pessimism and optimism and significant sluggishness. This paper develops a novel theoretical framework of time-varying beliefs capturing these empirical facts. In our model, the dynamic beliefs arise endogenously due to agents’ attitude toward alternative models. Decision-maker’s distorted beliefs generate countercyclical risk aversion, procyclical portfolio weights, countercyclical equilibrium asset returns, and excess volatility. A calibrated version of our model is shown to match salient features in equity markets.

This paper studies asymmetry in economic activity over the business cycle. It develops a tractable multisector model of the economy in which complementarity across inputs causes aggregate activity to be left skewed with countercyclical volatility. We then examine implications of the model regarding the time-series skewness of activity at the sectoral level, cyclicality of dispersion and skewness across sectors, and the conditional covariances of sector growth rates, finding support for each in the data. In the data, the skewness of employment growth, industrial production growth, and stock returns increases with the level of aggregation, which is consistent with the model’s implication that it is the nonlinearity in the production structure of the economy that generates the skewness. Other prominent models of asymmetry are not able to simultaneously match the range of empirical facts that the production network model can. *Dew-Becker: Northwestern University and NBER, Email: i-dewbecker@kellogg.northwestern.edu; Tahbaz-Salehi: Northwestern University and CEPR, Email: alirezat@kellogg.northwestern.edu; Vedolin: Boston University, NBER, and CEPR, Email: avedolin@bu.edu. We appreciate helpful comments from seminar participants at the NBER Monetary Program and Wharton.

We provide a model-free framework to study the global factor structure of exchange rates.To this end, we propose a new methodology to estimate international stochastic discount factors (SDFs) that jointly price cross-sections of international assets, such as stocks, bonds, and currencies, in the presence of frictions.We theoretically establish a two-factor representation for the crosssection of international SDFs, consisting of one global and one local factor, which is independent of the currency denomination.We show that our two-factor specification prices a large crosssection of international asset returns, not just in-but also out-of-sample with R2s of up to 80%.

This paper analyses how limits to the complexity of statistical models used by market participants can shape asset prices. We consider an economy in which the stochastic process that governs the evolution of economic variables may not have a simple representation, and yet, agents are only capable of entertaining statistical models with a certain level of complexity. As a result, they may end up with a lower-dimensional approximation that does not fully capture the intertemporal complexity of the true data-generating process. We first characterize the implications of the resulting departure from rational expectations and relate the extent of return and forecast-error predictability at various horizons to the complexity of agents' models and the statistical properties of the underlying process. We then apply our framework to study violations of uncovered interest rate parity in foreign exchange markets. We find that constraints on the complexity of agents' models can generate return predictability patterns that are simultaneously consistent with the well-known forward discount and predictability reversal puzzles.

We provide a theoretical characterization of international stochastic discount factors (SDFs) in incomplete markets under different degrees of market segmentation. Using 40 years of data on a cross-section of countries, we estimate model-free SDFs and factorize them into permanent and transitory components. We find that large permanent SDF components help to reconcile the low exchange rate volatility, the exchange rate cyclicality, and the forward premium anomaly. However, integrated markets entail highly volatile and almost perfectly comoving international SDFs. In contrast, segmented markets can generate less volatile and more dissimilar SDFs. In quest of relating the SDFs to economic fundamentals, we document strong links between proxies of financial intermediaries' risk-bearing capacity and model-free international SDFs. We interpret this evidence through the lens of an economy with two building blocks: limited participation by households and financiers who face an intermediation friction.

In this paper, we argue that monetary policy in the form of central bank communication can shape long-term interest rates by changing risk premia. Using high-frequency movements of default-free rates and equity, we show that monetary policy communications by the European Central Bank on regular announcement days led to a significant yield spread between peripheral and core countries during the European sovereign debt crisis by increasing credit risk premia. We also show that central bank communication has a powerful impact on the yield curve outside regular monetary policy days. We interpret these findings through the lens of a model linking information embedded in central bank communication to sovereign yields.

This paper develops a theory of dynamic pessimism and its impact on asset prices. Notions of time-varying pessimism arise endogenously in our setting as a consequence of agents’ concern for model misspecification. We generalize the robust control approach of Hansen and Sargent (2001) by replacing relative entropy as a measure of discrepancy between models by the more general family of Cressie-Read discrepancies. As a consequence, the decision-maker’s distorted beliefs appear as an endogenous state variable driving risk aversion, portfolio decisions, and equilibrium asset prices. Using survey data, we estimate time-varying pessimism and find that such a proxy features a strong business cycle component. We then show that using our measure of pessimism helps match salient features in equity markets such as excess volatility and high equity premium.

This paper analyzes how limits to the complexity of statistical models used by market participants can shape asset prices. We consider an economy in which agents can only entertain models with at most k factors, where k may be distinct from the true number of factors that drive the economy’s fundamentals. We first characterize the implications of the resulting departure from rational expectations for return predictability at various horizons. We then apply our framework to two applications in asset pricing: (i) violation of uncovered interest rate parity at different horizons and (ii) momentum and reversal in equity returns.

We establish four facts about skewness and conditional volatility in the economy: (1) aggregate activity is negatively skewed; (2) sector activity is negatively skewed, but less than aggregate; (3) the cross-sectional variance of output growth is countercyclical; (4) when a sector shrinks, it subsequently covaries more with other sectors. Those facts can all be generated qualitatively and quantitatively by a multisector equilibrium model with the key feature that production inputs are gross complements. Three alternative models that have been proposed to generate skewness and stochastic volatility are unable to simultaneously match all four facts even qualitatively.

In the short-run, bond risk premia exhibit pronounced spikes around major economic and financial crises. In contrast, long-term bond risk premia feature cyclical swings. We empirically examine the predictability of the market variance risk premium-a proxy of economic uncertainty-for bond risk premia and we show the strong predictive power for the one-month horizon that quickly recedes for longer horizons. The variance risk premium is largely orthogonal to well-established bond return pre-dictors-forward rates, jumps, and macro variables. We rationalize our empirical findings in an equilibrium model of uncertainty about consumption and inflation which is coupled with recursive preferences. We show that the model can quantitatively explain the levels of bond and variance risk premia as well as the predictive power of the variance risk premium, while jointly matching salient features of other asset prices.

We test the role of funding-constrained investors across developed financial markets. We compile direct measures of the severity of funding frictions, or illiquidity, from deviations of government bond yields from a fitted yield curve. Using these illiquidity measures, we first show that higher illiquidity is associated with a flatter security market line across markets. Exploiting the cross-section, we find that cross-country variation in illiquidity is associated with cross-country variation in alpha, in line with our theoretical predictions. Finally, we estimate a significant negative illiquidity risk premium that reveals a strong willingness of investors to hedge against the deterioration of funding conditions.

We revisit evidence of real effects of uncertainty shocks in the context of interest rate uncertainty. We document that adverse movements in interest rate uncertainty predict significant slowdowns in real activity, both at the aggregate and at the firm levels. To understand how firms cope with interest rate uncertainty, we develop a dynamic model of corporate investment, financing, and risk management and test it using a rich data set on corporate swap usage. We find that interest rate uncertainty depresses financially constrained firms’ investments in spite of hedging opportunities, because risk management by means of swaps is effectively risky. Received December 11, 2016; editorial decision January 26, 2018 by Editor Itay Goldstein. Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web Site next to the link to the final published paper online.

Preand Post-Announcement Returns: Table IA.I reports results of regressing individual currency returns on the announcement dummy over three different time windows: the entire day (4pm to 4pm), the pre-announcement window (4pm to 215pm), and the post-announcement window (215pm to 4pm). The results for the entire day, reported in Panel A, are in line with those presented in Table I in the main article: the difference between announcementand nonannouncement-day returns is statistically different from zero for all currencies except for the Japanese yen and the Norwegian krona. Panels B and C report the estimated coefficients for the announcement dummy for returns over the preand post-announcement windows, respectively. As the table shows (and consistent with our results for interest rate-sorted portfolios in Table IV in the main article), the difference between announcementand nonannouncement-day returns is positive and significant for a majority of the individual currencies over both time windows.

We propose a direct and robust method for quantifying the variance risk premium on financial assets. We theoretically and numerically show that the risk-neutral expected value of the return variance, also known as the variance swap rate, is well approximated by the value of a particular portfolio of options. Ignoring the small approximation error, the difference between the realized variance and this synthetic variance swap rate quantifies the variance risk premium. Using a large options data set, we synthesize variance swap rates and investigate the historical behavior of variance risk premia on five stock indexes and 35 individual stocks.

We document that a trading strategy that is short the U.S. dollar and long other currencies exhibits significantly larger excess returns on days with scheduled Federal Open Market Committee (FOMC) announcements. We also show that these excess returns (i) are higher for currencies with higher interest rate differentials vis-a-vis the U.S.; (ii) increase with uncertainty about monetary policy; and (iii) intensify when the Federal Reserve adopts a policy of monetary easing. We interpret these excess returns as a compensation for monetary policy uncertainty within a parsimonious model of constrained financiers who intermediate global demand for currencies.

We characterize model-free international stochastic discount factors (SDFs) under various degrees of market segmentation in incomplete markets. Our SDFs can be factorized into permanent and transitory components and they minimize the SDF dispersion subject to international pricing constraints. We find that large permanent SDF components are essential to jointly reconcile the low exchange rate volatility, the exchange rate cyclicality and the forward premium anomaly, however, at the cost of highly volatile SDFs. Market segmentation in stock and bond markets induces a deviation from the asset market view which helps avoid implausibly large SDF dispersions. Hence, economies featuring some form of mild market segmentation and large martingale components can match salient features of exchange rates, bond, equity, and FX option markets.

We show that the cross-sectional dispersion of conditional foreign exchange (FX) correlation is countercyclical and that currencies that perform badly (well) during periods of high dispersion yield high (low) average excess returns. We also find a negative cross-sectional association between average FX correlations and average option-implied FX correlation risk premiums. Our findings show that while investors in spot currency markets require a positive risk premium for exposure to high dispersion states, FX option prices are consistent with investors being compensated for the risk of low dispersion states. To address our empirical findings, we propose a no-arbitrage model that features unspanned FX correlation risk.

We build a parsimonious international asset pricing model in which deviations of government bond yields from a fitted yield curve of a country measure the tightness of investors' capital constraints. We compute these measures at daily frequency for six major markets and use them to test the model-predicted effect of funding conditions on asset prices internationally. Global illiquidity lowers the slope and increases the intercept of the international security market line. Local illiquidity helps explain the variation in alphas, Sharpe ratios, and the performance of betting-against-beta (BAB) strategies across countries.

We study feedback from the risk of outstanding mortgage-backed securities (MBS) on the level and volatility of interest rates. We incorporate supply shocks resulting from changes in MBS duration into a parsimonious equilibrium dynamic term structure model and derive three predictions that are strongly supported in the data: (1) MBS duration positively predicts nominal and real excess bond returns, especially for longer maturities; (2) the predictive power of MBS duration is transitory in nature; and (3) MBS convexity increases interest rate volatility, and this effect has a hump-shaped term structure.

The Online Appendix contains additional results not includ ed in the main paper. In Section OA-1, we show that our FX correlation dispersion measure FXC is very robust to di fferent choices regarding its construction. Section OA-2 shows that our cross-sectional asset pricing tests are robu st to sing the non-traded ∆FXC factor instead of the traded HMLC factor and to using di fferent sample periods and sample currencies. In Section OA-3 , we confirm that our cross-sectional results with respect to the FX correlation risk premiums are robust to alternative construction metho ds. In Section OA-4, we discuss the e ff ct of jumps on implied FX correlation measures. Section OA5 presents summary statistics for FX variances and FX variance risk premiums. S ection OA-6 shows that sorting on exposure to the FX correlation risk factor is not subsumed by exposure to an FX v ariance risk factor. Finally, Section OA-7 explores the spanning properties of FX correlation risk. OA-1. Alternative construction of FXC In the paper, we document a strong cross-sectional associat ion between the average level and the cyclicality of conditional FX correlation: high average correlation FX pa irs become even more correlated during bad times, whereas the correlation of low correlation FX pairs falls. This empi rical result motivates the use of our FX correlation dispers ion measureFXC, which is defined as the di fference in average conditional FX correlation between the to p and the bottom deciles of FX pairs sorted on their conditional correlation . [Insert Figure OA-1 here.] Instead of using deciles, we can alternatively construct th e measure as the di fference in correlations between the top and bottom quintiles or quartiles ( FXCQuintile andFXCQuartile, respectively). Increasing the size of the top and bottom groups reduces the correlation spread and, thus, the averag e level of the dispersion measure. In Panel A of Figure OA-1, we plot the original measure, along with the two altern ative measures. Both alternative measures are very highly correlated with the original measure, both in levels (0.99 a nd 0.98, respectively, for FXCQuintile andFXCQuartile) and in first differences (0.95 and 0.94, respectively). Not surprisingly, t he portfolio results are very robust to using the alternative measures: Panels B and C present three G10 curre n y portfolios sorted on the alternative ∆FXC betas for various sample periods. The results for both alternative me asures are qualitatively the same as those for the original measure. In particular, for all subsamples, excess currenc y returns are decreasing in the ∆FXC betas and the spread between the high and the low ∆FXC beta portfolio is substantial and statistically significan t. OA-2. Cross-sectional asset pricing In Table 10 of the main paper, we present estimates of the mark et p ice of FX correlation risk using various test assets. Table OA-1 contains the first-stage regression esti mates for test asset sets (3) and (4) that are omitted in Table 10. Email addresses: p.mueller@lse.ac.uk (Philippe Mueller),astath@uw.edu (Andreas Stathopoulos), a.vedolin@lse.ac.uk (Andrea Vedolin) Preprint submitted to Journal of Financial Economics October 26, 2016 [Insert Table OA-1 here.] Instead of usingHMLC returns, a traded factor, we can also perform our asset prici ng tests using the non-traded factor∆FXC. Figure OA-2 illustrates the performance of the ∆FXC factor by plotting the predicted annualized excess returns for various test assets (G10 currencies in Panel A, c urrency portfolios using all currencies and developed country currencies, in Panels B and C, respectively) agains t the actual annualized mean excess returns. Compared with the results when using the traded factor HMLC , the second-stage regression R2s are lower, but still high: 0.79, 0.78 and 0.58, respectively. Table OA-2 compares the estimates f or the market prices of risk of two factors, HMLC returns andFXC innovations, for the samples of all currencies and develope d country currencies, using eight test assets (four currency portfolios sorted on FX correlation betas and four cur ency portfolios sorted on nominal interest rates) in each case. We consider three sample periods for each specific ation: January 1996 to December 2013, January 1984 to December 2013, and January 1996 to July 2007. Our results a re fairly robust across alternative sample periods and factor specifications. In particular, the market price o f risk for the non-traded correlation factor ∆FXC is always negative and often significant. Compared to our benchmark sa mple period (January 1996 to December 2013), the results for the January 1984 to December 2013 sample are slig htly weaker, while the price of risk estimates when we exclude the financial crisis (January 1996 to July 2007) are s trongly significant. Overall,FXC innovations perform reasonably well in pricing the cross section of currency ret urns, albeit somewhat worse than our traded factor, HMLC returns. [Insert Figure OA-2 and Table OA-2 here.] OA-3. Alternative definitions for FX correlation risk premi ums In the paper, we show that average FX correlation risk premiu ms and average FX correlations are negatively related in the cross section of FX pairs. We measure FX correlation ri sk premiums using information available at time t: we calculate the conditional risk-neutral correlations usin g currency option prices observed at time t, while our proxy for conditional correlations under the physical measure is the average realized correlation over the three-month period ending att, sampled daily. Instead of using a three-month window, we ca n instead proxy for the conditional FX correlations under the physical measure using an one-month wi dow of past daily exchange rates (i.e., daily data from t − 1 to t), or using an one-month window of future daily exchange rate s (i. ., daily data fromt to t + 1). Figure OA-3 provides scatter plots of average FX correlation risk premi ums against the average FX correlations constructed using each of those two alternative measures of physical measure c onditional FX correlations: Panel A refers to measures constructed using data over the month ending at t, whereas Panel B refers measures constructed using data bet weent andt + 1. As we can see, the negative cross-sectional association b etween average FX correlation risk premiums and average FX correlations is robust to alternative measures o f conditional FX correlation under the physical measure. [Insert Figure OA-3 here.] OA-4. The effect of jumps on implied FX correlation To construct the measures of model-free implied exchange ra te moments, we follow the methodology of Britten-Jones and Neuberger (2000). They impose no-arbitr ge conditions and show that the risk-neutral expected return variance of an underlying asset is fully specified by a continuum of call and put options on that asset, provided that the price of the underlying asset is a di ffusion process. Given recent empirical evidence of priced ju mp risk in exchange rates (see, e.g., Chernov, Graveline, and Zviadad ze (2016)), it is natural to ask whether that methodology remains valid when the underlying price process includes ju mps. In the following, we address this question in two ways. First, we show that the Britten-Jones and Neuberger (2 000) methodology is valid even in the presence of jumps, provided that the higher order moments of the jump distribut ion are not very large. Second, we also consider the approach of Martin (2016), who derives a measure of risk-neu tral expected variance that is robust to jumps, and we