Desempeño de los mecanismos con CSI y con parámetros de QoS

In this section we question whether the forecasts of the currency excess returns can reveal the countercyclical nature of currency premia. According to theory (i.e. Ludvigson and Ng, 2009; Lustig et al., 2014) the forecasts of excess returns should be countercyclical reflecting the decrease in the global risk aversion during

good states of the world and vice versa. Lustig et al. (2014) show that currency premia exhibit countercyclical behavior that could be captured by forward dis- counts. In the term structure literature, Ludvigson and Ng (2009) find that the forecasts of bond risk premia demonstrate countercyclical movements only when they consider the macro factors. Thus, we attempt to see whether our domestic and global macro factors could help predict this behavior. To this end, figure A.1 (A.2) in the appendix show standardized 12-month moving average of carry and dollar carry trade excess returns - when considering the global (panel A) or domestic (panel B) factors - as well as the corresponding US (G7 countries) IP growth. We find that the inclusion of the global factors reveals the counter- cyclical nature of currency premia, while the domestic factors lead to acyclical or reverse results.25 In line with Lustig et al. (2014), the dollar carry trades exhibit a stronger countercyclical component (−0.82 correlation with the US IP growth) in comparison to the carry trade analogue (−0.22). This finding might be of interest to policy makers as it could help them adjust currency premia with the appropriate monetary policy or examine the interaction among risk premia, monetary policy and the economic environment (e.g. Ireland, 2014).26

25_{We come to a similar conclusion when we employ other predictors. The results are similar} for US and G7 IP growth because they are highly correlated. We also obtain similar results when, we exclude the US from the sample of the G7 countries.

1.6

Economic Evaluation of the Forecasts

In order to assess the economic value of the forecasts, we develop a strategy that resembles a decision rule. In particular, the investor is involved in one of the strategies at the end of month t if the forecast of the corresponding strategy is positive for the month t+ 1, otherwise she does not enter into a position. We use the forecasts of domestic and global factors as well as combination forecasts. Then, we examine the performance of the factors when investing in all strategies at the same time. In this case, identical weights are assigned to each strategy.

Table 8 displays Sharpe ratios (Panel A) and skewness (Panel B) of the condi- tional and unconditional payoffs. The unconditional payoff embodies the realized value of the payoff, while the conditional payoff is determined by a decision rule. As can be seen in the table, there is an overall significant increase in the Sharpe ratios and an improvement in the skewness profile of the payoffs both for whole sample and for the group of the Developed countries. In curly brackets we report p-values estimated based on 10,000 stationary bootstrap samples (Politis and Romano, 1994), for the null hypothesis that the Sharpe ratios of the conditional strategy do not exceed (statistically) the unconditional counterparts, which take a position in the FX strategy regardless of the sign of the prediction. With the exception of the momentum strategy, where there is no big improvement, the forecasts provide strong out-of-sample economic value for an investor who ap- plies the strategies of interest. In addition, the mixed strategy that combines all the three strategies provides exceptionally high annualized Sharpe ratios of around 1.06 with positive skewness.

Figure 4 illustrates rolling Sharpe ratios, using a 12-month window for carry, dollar carry and momentum strategies as well as the mixed strategy. The solid lines represent rolling Sharpe ratios of conditional payoffs obtained from the

forecasts of the optimal subset of factors (black) and the combination forecasts (blue). The dashed line displays the realized value of the payoffs. There is clearly an improvement in the rolling Sharpe ratios, especially during the crisis. Our decision rule shows that an investor could gain very high Sharpe ratios during the recent financial turmoil (2008-2009) if she had taken into account the domestic and global macroeconomic environment.

Overall, the out-of-sample study revealed a strong economic value in the pay- offs to carry, dollar carry and momentum strategy. In addition, the consideration of the factors improves not only the overall Sharpe ratio and the skewness profile of the payoffs, but also helps to mitigate the downside risk experienced during the recent global financial crisis.

Dynamic Asset Allocation. The decision rule does not amalgamate the investors risk preferences into the asset allocation decision. Thus, we ask whether our forecasts can benefit a risk-averse investor with mean-variance preferences who allocates her wealth on a monthly basis across risky assets (i.e. equities and currency strategies) and risk-free assets (i.e. U.S. Treasury bills). Particularly, we ask whether an investor could benefit from a currency investment strategy that it is appended by a traditional institutional investors 60/40 portfolio.27 To this end, we estimate the certainty equivalent return (CER), following Campbell and Thompson (2008) and Ferreira and Santa-Clara (2011).

The investor rebalances her portfolio at the end of month t, forming the weights of the currency strategies (wi

t) for investing at time t+ 1 as:

wi_t= 1 γ ˆ ψi t+1 ˆ σ2 i,t+1 ! for i=HM L, U SD, W M L (1.15)

where ˆψ_ti₊₁is the forecast of the payoff for thei-thstrategy, ˆσ2_i,t+1the correspond- ing forecast of the variance and γ denotes the investors absolute risk aversion. Therefore, the portfolio return at time t+ 1 is given by:

Ri_p,t+1 =w_tiψ_ti₊₁+Rp60/40t+1 for i=HM L, U SD, W M L (1.16)

where Rp60/40t+1 is the return of a traditional 60/40 portfolio that allocates 60%

on equities (i.e. S&P 500) and 40% on risk free bonds at time t+ 1 . As in Campbell and Thompson (2008) the variance of the payoffs is estimated on the basis of a five-year rolling window, the risk aversion coefficient equals five and the weights for the risky asset are confined in a particular interval (i.e. between 0 and 1). In this way, we do not allow for leverage. Thus, the average realized utility or CER is defined as:

CERi_p = ˆµi_p− γσˆ 2 i,p 2 for i=HM L, U SD, W M L (1.17) where ˆµi

p is the mean and ˆσ2i,p is the variance of the portfolio when investing in

each of the three strategies over the out-of-sample period.

The certainty equivalent return is the risk-free return that a mean-variance investor would consider sufficient in order to avoid investing in the strategy. The CER gain represents the difference between the average realized utility of the forecasts and the corresponding value of the historical average. It can be interpreted as the fee that an investor is willing to pay in order to utilize the forecasts rather than relying on the historical mean. Thus, a positive value of the CER means that the investor prefers the forecasts over the estimate of the historical mean when forming expectations with regard to the strategies of interest.

mentum strategy. In agreement with the R2_OOS and the one-sided p-values of the MSPE-adj, theCER gains are positive with the exception of the momentum strategy and the carry trade strategy only when we consider all the sample of all countries. Thus, there is a predictable component in the carry and dollar carry trade strategy that provides strong economic value to a risk-averse investor with mean-variance preferences.