The information content of risk reversals in emerging market currencies ( O conteúdo informacional dos “ risk reversals ” em moedas de paı́ses emergentes )

This paper analyzes the information content of risk reversals for ten emerging market currencies. In contrast to the findings for major developed currencies, it is found that in some cases risk reversals (RR) are helpful in predicting currency returns, but in general RR are predicted by but do not predict carry trade returns. Evidence based on country vector autoregressions (VARs) and a panel VAR (PVAR) indicate that RR react in a procyclical way to carry returns, i.e., it is cheaper (more expensive) to buy protection against currency weakness after positive (negative) total returns. All in all, it is found that crash risk accounts for a small share of carry trade returns variance, which seems to be more related to global risk aversion shocks. A sentiment indicator of crash risk in emerging market currencies is highly correlated with the VIX.


Introduction
Is it possible to predict future returns of emerging market currencies based on information embedded in options prices?This paper investigates whether risk reversals (RR) -a measure of the skewness of exchange rate expected distribution -contain information about future returns of emerging market currencies.
Risk reversals consist in the difference between the implied volatilities of out-of-the-money call and put, of a given maturity and delta.With these parameters fixed, it implies a long position in a call and a short position in a put.Thus, by showing if the call or put has a higher cost in terms of implied volatility, it reveals the expectation of the market regarding the distribution of the exchange rate underlying the option.Risk reversals are often described as the slope of the volatility smile curve, which plots, for a given maturity, the implied volatility associated with each delta for calls and puts.
Information embedded in options prices receives considerable attention of both practitioners and authorities, since they convey forward looking information that may be useful for central banks and market participants in general.
The predictability of foreign currencies through risk reversals have been tested for major currencies.Kurbanov (2010) examined risk reversals for the EURUSD exchange rate, using weekly data.Using univariate meth-404 Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016 ods, he found a positive contemporaneous relationship between changes in risk reversals and EURUSD returns.In terms of predictability, he found a negative association between past RR changes and exchange rate returns, but with limited explanatory power, casting doubt on the usefulness of risk reversals for forecasting exchange rate returns.
Along the same lines, Dunis and Lequeux (2001) analyzed the information content of risk reversals for the pairs USDJPY, USDCHF, GBPUSD and AUDUSD, using daily data and multivariate time series methods, also finding little evidence for the usefulness of risk reversals in assessing the future evolution of exchange rates.Brunnermeier et al. (2009) investigated crash risk of currencies, i.e., sudden and abrupt depreciations of the target currency of the investment.They found that the interest rate differentials are negative associated with realized skewness, interpreting this result as a measure of crash risk.On the other hand, carry returns are positively associated with risk reversals, meaning that investors buy insurance against crash risk.Galati, Heath and McGuire (2007) argued that risk reversals capture directional uncertainty about the exchange rate, and associate this with carry trade risks.Gagnon and Chaboud (2007) documented that risk reversals increase after episodes of presumed unwinding of carry trades, in which the Yen is sharply appreciated, meaning that investors are frightened by losses, but they also found that there is no evidence of purchase of protection through risk reversal when carry trade positions are building up.Farhi et al. (2015) developed a model to estimate a time series of the compensation for global disaster risk exposure based on exchange rate data of major currencies.The disaster risk exposure measure obtained in the model showed a strong correlation with risk-reversals.As in Brunnermeier et al. (2009), they associated risk reversals with the level of interest rate differentials, meaning that expected carry trade returns compensate investors for the risk of a sharp depreciation of the target currency.Carr and Wu (2007) investigated exchange rate dynamics of the pairs USDJPY and GBPUSD, developing a model that delivers stochastic volatility and skewness, influenced by the time-varying feature of risk reversals found in the data.As in the afore-mentioned studies, they also documented a positive and strong correlation between currency returns and risk reversals, but the correlation between butterfly spreads and currency returns is much lower.Finally, De Bock and Carvalho Filho (2015) found that risk reversals are positively correlated with currency weakness during risk-off Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016 episodes, appearing significantly in regressions that explain the drivers of currency movements during the periods analyzed.
Despite the evidence for major currencies, at present no research has focused on the empirical content of risk reversals in emerging market currencies.The main contribution of this paper is to provide such evidence.From a theoretical point of view, this paper is related to the ongoing research on disasters and its implications for potentially solving asset pricing problems (Barro,2006;Barro and Ursúa, 2008;Tsai and Watcher,2015).Disasters are usually modeled as a shock that induces negative skewness in the distribution of output.As mentioned previously, as the price of skewness, risk reversals capture the expected skewness in the context of foreign exchange (FX) markets.
This paper is organized as follows.Apart from this introduction, section 2 presents the data used and how some variables were constructed.Section 3 pursues with analysis of the relationship between currency returns and risk reversals, mainly on the forecast ability power from one variable to the other.Section 4 examines the relationship between global risk aversion, carry trade returns, risk reversals and interest rate differentials in VAR models estimated for each country.Section 5 constructs an indicator of crash risk sentiment in emerging market currencies and shows that it is highly correlated with the VIX.Section 6 continues with the analysis of the relationship between global risk aversion, carry trade returns, risk reversals and interest rate differentials, but with a PVAR model.Finally, section 7 concludes.Appendices present unit root tests, figures of the carry trade returns series used and robustness evidence for the findings.
I also collected some data on interest rate differentials for the currencies employed.I used 3-month interest rates.For Brazil, it is the implied NDF rate, for Chile is the CLF GVT Zero Yield, for India is the 3-month T-Bill yield, for Indonesia, Poland, Turkey and Israel I used the interbank 406 Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016 rate, for Mexico is the 90-day CETES rate, for South Korea is the 90-day NCD rate, and finally, for South Africa, it was the JIBAR 3-month rate.In order to construct the interest rate differential series, I subtracted each interest rate series from the 3-month US T-Bill rate.All data come from Datastream, and interest rate series used reflect availability in this program.Even though information about risk reversals for some of these currency pairs is available before the sample start date, I chose to begin the sample on the same date for all of them, for the sake of uniformity of the analysis.Brunnermeier et al. (2009) used 25 delta 1-month risk reversals.De Bock and Carvalho Filho (2015) used 25 delta at the 3-month maturity.Carr and Wu (2007) presented data for 10 and 25 delta risk reversals for many maturities.Kurbanov (2010) employed 1-month 10 and 25 delta risk reversals and butterflies.Lastly, Dunis and Luqueux (2001) analyzed 1-month 25 delta risk reversals.Aligned with the related literature, I have chosen to use 25 delta 3-month risk reversals.Table 1 reports a strong positive correlation across deltas and maturities, revealing a strong co-movement between risk reversals of a given currency pair.Figure 2 shows the evolution of risk reversals and exchange rates in the sample period.
I also constructed carry trade return series (variables z t ,as specified on equation 2 below) for the currency pairs used in the analysis, which encompasses both the currency returns and interest rate carry.Following Brunnermeier et al. (2008), the return series have been constructed using the following formulas: Where z t , the carry trade return, is the sum of FX and interest rate returns, and s t+1 is the currency return.A positive value for the currency returns means a depreciation of the domestic (emerging market) currencies relative to the dollar.For the interest rate differential, I used the series constructed and described above, considering a year with 52 weeks1 Therefore, I have daily data on total returns, exchange rate returns, 25-delta 3-month Figure 2 plots, for each currency pair, currency levels and risk reversals over the sample.Figure 3 presents interest differential data, while figures 9 and 10 in the Appendix show the carry trade return series used, respectively the variables z t constructed and the daily return series from Bloomberg.
Figure 1 shows that, with the exception of the INR, there is a positive correlation between the average level of risk reversals and average interest rate differentials over the sample period, implying that overall higher interest rate differentials lead to higher risks, or investors in high yielding currencies are exposed to crash risks -sudden or abrupt depreciations of the currencies in which investments were made-and to compensate for such risks they demand higher interest rates.For the INR, over the sample 408 Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016 With the exception of a small period for the PLN, risk reversals are always positive for all currencies analyzed in the sample, which means that the FX market always embedded some risk of abrupt depreciation of emerging market currencies.Since the convention of the currencies in this study is to quote them with the dollar (USD) as the base currency, upward movements in the pairs mean that emerging currencies are weakening against the dollar.Positive risk reversals mean that, for a given delta, calls on USD are more expensive than puts, in implied volatility units.The higher risk reversals are, higher it is the implied crash risk on the currency.
3. Do risk reversals help to forecast exchange rate returns in emerging markets ?
In this section, I examine the predictability of emerging market currencies based on risk reversals.First, I have examined the stationarity of risk  In a bivariate VAR as below: we say that variable y t does not Granger cause variable x t if d(L) = 0, in other terms, if we can remove the lags of y t of the equation of x t .
Table 9 in the Appendix presents the quantity of lags used in each bivariate VAR, and the corresponding Lagrange Multiplier statistic at that lag, to assess the autocorrelation.The lags were selected based on the information criteria.When the residuals still showed signs of autocorrelation at the indicated lag, more lags were added to the VAR, until the residuals showed Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016 no autocorrelation.Unit root tests, presented in the Appendix, indicated a unit root for the BRL risk reversal.For the other currency returns and risk reversals, tests indicated stationarity, and therefore entered the VAR in levels, while for the BRL the risk reversal entered the VAR in first differences.
The left side of table 2 shows the Granger causality tests when FX returns are removed from the risk reversals equations, while the right side of table 2 presents the Granger causality tests to investigate whether risk reversals help to predict currency returns.
The results indicate that for the BRL, INR and TRY, risk reversals are helpful in predicting the corresponding exchange rate returns.This finding contrasts with those for major developed currencies, since Dunis and Luqueux (2001) and Kurbanov (2010) did not find that risk reversals were helpful in predicting exchange rate returns.On the other hand, risk reversals for BRL, IDR, ILS, INR, KRW, MXN,TRY and ZAR are Granger-caused by their returns, suggesting that option market prices are influenced by the recent behaviour of exchange rate returns, but in many cases fail to anticipate their movements.

Global Risk Aversion, Carry Trade Returns and the Price of Skewness
In this section there is an investigation of the role of risk reversals in a more dynamic setting.Along the same lines of Brunnermeier et al. (2009), 412 Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016 for each currency I estimate VAR´s with the following variables: the VIX index, interest rate differential, carry trade returns (variables z t ) and risk reversal (the price of skewness).
Attempting to investigate the carry trade -the strategy of borrowing in low yield currencies and investing in high yield currencies -and major failures of the uncovered interest parity (UIP) -firstly documented by Fama (1984) -Brunnermeier et al. (2009) estimated a standard recursive system with the following order, using quarterly data: interest rate differential, total return z t ,-as in expression 2-, realized skewness and futures positioning, as a proxy for carry trade activity for five currencies: Canadian Dollar (CAD), Japanese Yen (JPY), Swiss Franc (CHF), Great Britain Pound (GBP) and Euro (EUR).
Since there are no positioning data for the currencies in our analysis, and data are on a daily frequency, a smaller specification has been opted, without positioning data and replacing the realized skewness for the risk reversals.Brunnermeier et al. (2009) constructed a quarterly measure of realized skewness from daily exchange rate returns.As our analysis is based on daily data, it was not possible to include this variable in the estimated models.Feroli et al. (2014) defend the use of daily data to assess a positive feedback loop between returns and bond flows2 .
Regarding positioning, their analysis is based on futures positioning from the Commodity Futures Trading Comission (CFTC), but they acknowledge the limitations of using this data3 , and also argue that returns tend to be highly associated with an increse in positioning 45 Given these arguments, we assume that the lack of positioning data is harmless in our analysis.This idea is also consistent with models and empirical evidence in mutual funds whereby a feedback loop emerges between flows and returns, in which high returns attract flows, which further enhance returns, until a crash happens (Feroli et al. 2014).
The VIX index, based on the implied volatility of the S&P 500, is widely recognized as a proxy of global risk aversion, risk on/risk off sentiment (Rey, 2013), and uncertainty (Bloom, 2009) and was included in the estimated VAR´s in order to control for these factors.Brunnermeier et al. (2009) argues that bouts of global risk aversion could trigger funding constraints through higher margins and capital requirements, associated with unwinding of carry trades.Consistent with the literature on uncertainty shocks (Bloom, 2009), it was ordered first in the estimated VAR´s under the Cholesky identification scheme.Thus, the order of the models implies that the VIX affects contemporaneously interest rate differentials, carry trade returns (variables z t ) and risk reversals, while shocks to risk reversals only impact the other variables with lags.This specification follows Brunnermeier et al. (2009), with the inclusion of the VIX ordered first in the VAR´s.
The Autocorrelation tests and lags that were used are shown on table 10 in the Appendix.All VAR´s passed the stability conditions, i.e., all eigenvalues lied inside the unit circle, and stability tests are reported on figures 14 and 15 in the Appendix.Finally, the normality of the residuals is frequently rejected, as pointed out by the Jarque-Bera test.
In the appendix is presented evidence of a robustness exercise, in which the constructed measure of carry trade returns (z t ) is replaced by the carry trade returns measure obtained on Bloomberg 6 .Since there was no daily carry return for the ILS and KRW on Bloomberg, I ended up with carry returns for 8 currencies, instead of the 10 with the previous measure.I have also estimated the VAR´s with a different ordering, in which risk reversals exerts a contemporaneously effect on z t , that is, this variable is ordered third in the VAR´s.The results, although not shown, do not reveal any substantial difference between those reported on Figures 4 , 5 and 6 below.
Table 3 presents Granger causality tests between carry trade returns and risk reversals.We found that in the case of the BRL, CLP, IDR and INR risk reversals help to forecast carry trade returns in the 4-variable VAR estimated for each country.On the other hand, with the exception of the PLN, in all other cases we found that carry trade returns Granger cause risk 6 Obtained on the screen WCRS 414 Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016 reversals.Overall, this evidence indicate that in general risk reversals are Granger-caused by the returns, implying that option prices, or the prices of skewness, tend to follow the behaviour of carry trade movements.On the other direction, for the BRL, CLP, INR, and IDR, it was found that risk reversals convey useful information in predicting carry trade returns, meaning that option prices contain useful information for the returns.Since many of these currencies are used as target in the carry trade, I found that risk reversals provide useful information for these kinds of transactions.Although Granger causality tests indicate that risk reversals are helpful in predicting both the exchange and carry trade returns of some currencies, they do not provide information about the shape of the relationship between these variables.For this reason, it is presented below the impulse response of risk reversals to a shock in carry trade returns (variable z t ), based on the estimated VAR´s.The order of the models implies that the VIX contemporaneously affect interest rate differentials, carry trade returns and risk reversals, while shocks to risk reversals only impact the other variables with lags.
The goal was to investigate the interaction between global risk aversion, carry trade returns and the price of skewness, more specifically to see show risk reversals in emerging market currencies react to a selloff or a rally in some currency.In case the currency strengthens, there is a questioning if the price of insurance becomes more expensive, reflecting the risk of a Overall, with the exception of the PLN, I found that risk reversals react negatively to a shock in z t .The response of BRL risk reversal is slightly different, in the sense that risk reversals immediately fall after the shock, and then quickly recovers the original level, while for the other currencies the risk reversal response is more persistent after the shock.This difference might be due to the way the BRL risk reversals enter in the VAR -in first differences, while the others enter on levels.In any case, as in the others, the response after the shock is negative and statistically significant, with 95% confidence bands in gray.After the shock, risk reversals decline and 416 Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016 The information content of risk reversals in emerging market currencies slowly returns to the previous value, showing a lot of persistence.
This result implies that the market often follows the behaviour in carry trade returns, that is, when there is a positive shock to the returns in the carry trade, the price of insurance against a crash in currency markets actually falls.Therefore, it becomes cheaper to buy protection after a rally in the market, while whenever the market sells off, the price of skewness increases, suggesting that market gets spooked and purchase insurance against additional falls in the returns.
Figure 5 shows how risk reversals react to a positive shock in the VIX, taken as a measure of global risk aversion.Risk reversals respond in a positive way, meaning that episodes of risk off in the market lead to an increase in the price of insurance of a devaluation in emerging market currencies against the dollar.This evidence is consistent with the presented in Brunnermeier et al. (2009) for risk reversals of developed currencies, Finally, figure 6 reports the response of carry trade returns to a shock in the VIX.For all currencies carry trade returns fall right after the shock, and bounces back in a few days.This evidence is aligned with the interpretation of Brunnermeier et al. (2009) andFeroli et al. (2014), in which global risk off episodes leads to tighter funding liquidity, unwinding of carry trades and a procyclical spiral of flows and returns.
Based on the estimated VAR´s, it is also presented the variance decomposition of the variables in the estimated VAR´s.
According to results shown on table 4, risk reversals in emerging market currencies are mainly explained by their on shocks, carry trade returns and global risk aversion.I found that risk reversals in MXN, ZAR, PLN, CLP and KRW are very linked to global risk aversion movements, represented by shocks on the VIX.On the other hand, risk reversals for the BRL, INR and ILS are relatively insulated from global risk aversion, with less than 10% of their variance explained by shocks on the VIX.Carry trade returns are the main drivers of the TRY risk reversals, explaining 52% of its variability.Carry trade returns are also important determinants of the risk reversals of the KRW, MXN, CLP, ILS and BRL, explaining more than 10% of their variability in the estimated VAR´s.
Table 4 also indicate how much variation in the carry trade returns in due to currency crash risk embedded on risk reversals, or, in other terms, how much compensation for currency risk is included in carry returns for the period under study.As shown in the second panel, I found that crash risk shocks, or shocks to risk reversals, explain a small share of carry trade  returns.Rather than crash risk shocks, global risk aversion, represented by shocks on the VIX, are much more important for explaining carry trade returns.The MXN, ZAR and CLP exhibit the larger exposure to global risk aversion shocks, while the IDR and ILS are more insulated from global risk aversion.In a panel VAR setting, the share of risk reversals shocks on the variability of carry returns remain small.Results a described in more detail in section 6.Finally, interest rate differentials are mainly driven by their own shocks, In the case of ILS and CLP, rate differentials are more exposed to shocks on global risk aversion.On the former case, shocks on risk reversals explain around 7% of the variability of interest rate differentials.

Crash Risk Sentiment in Emerging Market Currencies and Global Risk Aversion
This section investigates how coordinated are risk reversal movements across different currencies.To assess the co-movements of risk reversals, a factor analysis was done, shown on table 5.The first factor explain more than half of the variability of risk reversals over the period.Observing uniqueness of the factors on table 5, we find that the MXN, PLN, IDR and BRL are more exposed to the common factor, while the CLP, INR and ILS are less sensitive to the common cycle, with around 30-40% of their variability driven by idiosyncratic factors.
Feroli et al. ( 2014) extract a principal component of bond flows and identify the first principal component as an indicator of bond market sentiment.They show that the derived bond sentiment indicator shows little correlation with the VIX, which they take as positive, implying that the indicator capture more than the volatility of equity markets, and is unrelated with the global financial cycle explored by Rey (2013) and uncertainty (Bloom, 2009).
I proceed in a similar way here, extending the analysis in Feroli et al. (2014)7 .Figure 7 compare the first factor extracted from the risk reversals, which I label FX crash risk sentiment indicator in emerging markets, and  2014) for bond flows, the crash risk sentiment indicator in emerging market currencies shows a strong correlation with the VIX.The correlation coefficient is 0.81, statistically significant at the 1% level.Thus, crash risk sentiment is highly associated with global risk aversion.This finding is consistent with risk off episodes in which uncertainty leads to higher volatility, tighter funding liquidity, unwinding of carry trades, and risk reversals reacting in a procyclical way, with the price of insurance increasing after the market sells off.

Panel VAR Evidence
This section presents further evidence on the relationship between risk reversals and carry returns.Unlike in section 4, in which it was showed results based on VAR´s estimated for each country, here I present evidence based on a panel VAR (PVAR), from Abrigo and Love (2015).Overall PVAR´s follow the same structure of VAR´s, treating all variables in the system as endogenous, but with the addition of a cross-section dimension, Crash Risk Sentiment in Emerging Currencies and the VIX which brings three features to the model: i) lags of all endogenous variables of all units enter the model for one specific unit, ii) the error terms are generally correlated across different units and iii) the intercept, the slope and the variance of the shocks may be unit specific (Canova and Ciccarelli , 2013).Thus, the model is: I have an unbalanced panel.With the exception of Chilean data, I have 1296 daily observations for a total of 8 currencies.For the CLP, our data began in the middle of 2011, more specifically on June 28, whereas for the other countries the dataset began on November 11, 2010.Along the same lines of section 4, the panel VAR was estimated using series of interest rate differentials, total returns (both constructed -variables z t -and from Bloomberg) and risk reversals, in this order.So, Yit = [interest rate dif f erentials, carry trade returns, risk reversals] for country i and day t.
The fixed effects variable, f i , which conveys the information about the individual heterogeneity, was removed by forward mean-differencing (fod), which removes the mean of forward future observations available for each country and day (Arellano and Bover, 1995), and is the default option in the package of Abrigo and Love (2015).I also present evidence considering the case when the fixed effect was removed by first differences (fd).
The information criteria8 (not shown) pointed to 1 lag for the panel VAR, with both the MBIC and MQIC achieving their lowest levels, while the MAIC achieved its lowest level with 3 lags.This happens both when the model is estimated with the variables z t and with the measure from Bloomberg.The PVAR (1) was estimated using the lags from 2 to 10 of interest rate differentials, total returns and risk reversals as instruments.The large number of observations in our dataset allowed us to be generous in this respect, allowing a large number of instruments.Using these instruments, I failed to reject the null hypothesis of the Hansen (1982) J-test of overidentifying restrictions9 .Therefore, I failed to reject the hypothesis that the instruments were valid instruments, uncorrelated with the error term in the generalized method of moments (GMM) estimation.All estimated PVAR models satisfied the stability condition, i.e., all engeinvalues from the estimated models lie inside the unit circle.
Concerning Granger causality tests in the context of the estimated PVAR models, broadly speaking it failed to point that some variable in the system Granger-cause the other.The exception was interest rate differentials in the equation of risk reversals, which means that interest rate differentials are helpful in predicting risk reversals at the 10% level (p-value = 0.078).In terms of variance decompositions, results show that carry trade returns (measured by variables z t ) account for up to 5% of risk reversals variance.
Since the Cholesky decomposition is used in the PVAR, implying that the first variables exerts a contemporaneous effect on the other variables in the system, I also estimated the system with a different order, with risk reversals before carry trade returns, in an attempt to provide robustness to the results.Ordering risk reversals before carry trade returns, I did not found a statistically significant negative response of risk reversals to carry trade returns shocks.
Table 6 presents the variance decomposition of the estimated PVAR´s.Crash risk, or risk reversal shocks, account for only a small portion of the variability of carry trade returns, with a maximum share of 3.7% when the PVAR( 1) is estimated with first-differences and with carry trade returns obtained from Bloomberg, suggesting that the role of crash risk shocks in explaining carry trade returns in emerging markets is probably small.On the other hand, carry trade returns shock account for up to 6% of the variability of risk reversals in the 10-day horizon, implying that carry trade return shocks explain a larger share of the variability of risk reversals than the other way around.
Following Brunnermeier et al. (2008), the analysis of the relationship between carry trade returns and risk reversals proceeded in a dynamic context with VAR models estimated for each country and PVAR´s.On the former, I found overwhelming evidence that risk reversals react in a negative way to carry trade return shocks: the price of skewness becomes cheaper (more expensive) after positive (negative) returns, which I take as indication that the market gets frightened after negative returns and purchase more protection against further losses.In turn, sentiment improves after positive returns, and the market fails to anticipate that a crash may come.On the latter, I also found evidence of the same kind of response, but not robust to a different ordering.
Finally, I have found that crash risk shocks, or shocks on risk reversals, account for a small share of the variance of carry trade returns in emerging market currencies in the context of the estimated models.The forecast error variance decompositions indicate that carry trade returns in emerging market currencies are more related to global risk aversion shocks, proxied the the VIX.Although in a different, model-free framework, this finding somehow contrasts those from Farhi et al. (2015), who have found a sizeable (around 30%) role for disaster risk in explaining carry trade risk premium for advanced economies in their model.Lastly, along the same lines but in contrast to the findings in Feroli et al. (2014), it was found that an indicator of crash risk in emerging markets strongly correlate with the VIX, suggesting that global risk aversion plays a major role on carry trades in emerging markets.
This paper thus shed some light on the role of risk reversals in emerging market currencies.As a byproduct, it also contributes, in an empirical way, to the ongoing research agenda of the role of crash risk shocks and their asset pricing implications (Barro,2006;Farhi et al, 2015;Tsai and Watcher,2015).

A. Unit Root Tests
Tables 7 an 8 presents Dickey Fuller -GLS and Phillips Perron unit root tests.

B. Autocorrelation Tests
This section presents the LM autocorrelation tests on the estimated VAR´s.Table 9 shows the autocorrelation tests on the bivariate VAR´s, used to assess Granger causality tests of Section 3. Note: The null is that there is no serial correlation at the respective lag.Note: The null is that there is no serial correlation at the respective lag.
differentials, carry trade returns (either variable z t or the measure obtained from Bloomberg) and risk reversals.It is also shown the lag order selected for each model.

C. Carry Trade Returns
Figures 9 and 10 show the data for the carry trade return series used in the paper.

D. Some Robustness Tests
In this section we make some robustness analysis of the results obtained in section 4. We seek to investigate whether the results depend on the mea-  Variable Z, constructed using formula 2 for each currency sure of carry trade returs.I replaced variables z t in the estimated models by the daily carry trade return obtained on Bloomberg.Table 11 presents Granger causality tests between carry trade returns and risk reversals.Riks reversals help to predict carry trade returns for the BRL, CLP, IDR, INR and TRY.On the other hand, carry trade returns predict risk reversals in all currencies but the PLN.Again, this is consistent with a delayed response of risk reversals to carry trade returns.
Figure 11 shows that risk reversals fall after a carry trade return shock, and that the responses in general are very persistent, similar to those on figure 4. Figure 12 shows that risk reversals increase after a positive shock on the VIX, while figure 13 reveal that carry trade returns decline after a shock on global risk aversion.Overall, the responses are very consistent and qualitative similar to those in section 4.
Finally, forecast error variance decompositions of the estimated VAR´s again show that risk reversals are in good part driven by carry trade returns and global risk aversion.The latter is particularly important for the MXN, ZAR, PLN and CLP, while carry trade returns are relevant for explaining risk reversals of the TRY , MXN and CLP.For the variability of carry trade returns, it was found that in general global risk aversion shocks are more Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016  The information content of risk reversals in emerging market currencies

Figure 1
Figure 1 Risk Reversals and Interest Rate Differentials

Figure 2
Figure 2 Exchange Rates and Risk Reversals

Figure 3
Figure 3 Interest Rate Differentials

Figure 4
Figure 4 Response of Risk Reversals to a Carry Trade Return Shock (Variable zt)

Figure 5 Figure 6
Figure 5 Response of Risk Reversals to a Shock on the VIX

Figure 8
Figure 8Response of risk reversals to a carry trade return shock

Figure 9
Figure 9Variable Z, constructed using formula 2 for each currency

Figure 10
Figure 10 Daily Carry Trade Returns (Bloomberg)

434
Figure 11Response of Risk Reversals to a Carry Trade Return Shock (Bloomberg)

Figure 13
Figure 13Response of Carry Trade Returns (Bloomberg) to a Shock on the VIX

Table 1
Correlations between different risk reversals series

Table 2
Granger Causality Tests Between Currency Returns and Risk Reversals : RR stands for risk reversal and RET for the currency return s t .The null is that the excluded variable does not Granger cause the equation variable.***,** and * denotes, respectively, statistical significance at 1%, 5% and 10%. Note

Table 3
Granger Causality Tests Between Carry Returns (Variable zt) and Risk Reversals

Table 4
Variance Decomposition of the VAR´s

Table 5
Factor Analysis of Risk Reversals In contrast to the findings in Feroli et al (

Table 7
Unit Root Tests Note: ***,** and * denotes, respectively, statistical significance at 1%, 5% and 10%.The null of both tests is that the series is non stationary.

Table 9
Table 10 presents the autocorrelation tests of the country VAR´s between the VIX, interest rate 430 Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016 **,** and * denotes, respectively, statistical significance at 1%, 5% and 10%.The null of both tests is that the series is non stationary.Bivariate VAR´s between FX returns and risk reversals Bivariate VAR (returns and risk reversals) Lags in the VAR Rev. Bras.Financ ¸as (Online), Rio de Janeiro, V14, No. 3, July 2016