A filtered currency carry trade

Highlights

• We propose a new index for measuring the predictive capacity of forward discounts on future currency returns.

• We find that the predictive capacity not only differs across currencies but also persists.

• We propose a novel filtering-signal-weighted carry trading strategy.

• We find that the new carry trade outperforms the conventional carry trade.

• The relative outperformance is significant for emerging market currencies and during tranquil periods.

Abstract

In this paper, we document that the predictive capacity of forward discounts on future currency returns not only differs across currencies but also persists. We then propose a new currency carry trade strategy that relies on the differential predictive capacity of forward discounts. We find that the new strategy offers a significant amount of profit improvement over the conventional currency carry trade strategy. We also find that emerging market currencies provide relatively large profit opportunities. While both strategies show decreasing carry trade profits as FX markets get volatile, the relative outperformance of the new carry trade strategy tends to be found in stable periods but disappears in volatile periods. The superiority of the new carry trade relative to the conventional carry trade is robust to various specification changes.

Introduction

In foreign exchange (FX) markets, empirical evidence has documented that future currency returns can be predicted by some currency characteristics. As a prominent example, the currency carry trading strategy utilizes the predictive power of interest rate differentials to form currency portfolios. Lustig, Roussanov, and Verdelhan (2011) and Lustig and Verdelhan (2007) sort currencies by interest rate differentials (or forward discounts), build a long–short currency portfolio by going long currencies with high interest rates and short currencies with low interest rates, and show the profitability of the currency carry trading strategy.

We can view an interest rate differential as a predictor variable for future currency returns in the conventional currency carry trading strategy and recognize that the profitability of the carry trade relies on the predictive power of interest rate differentials. In this paper, we investigate whether the predictive power significantly differs across currencies or not. Further, if the predictive power significantly differs across currencies, we will examine how to utilize the fact to improve the carry trade. To investigate this issue, we construct a new index that measures the predictive power of interest rate differentials. We then use the new information variable to form a new carry trade and show that the new method can improve the conventional carry trade.¹

Our method can be broadly regarded as a kind of empirical Bayesian methodology that learns about the signal validity according to past successes. For example, Stock and Watson, 1996, Stock and Watson, 1999, Stock and Watson, 2003, Stock and Watson, 2004 attribute out-of-sample forecasting failures for macroeconomics time series to structural instability and suggest to adapt past forecasting errors and also to combine forecasts to enhance forecasting performance. In finance, Rapach, Strauss, and Zhou (2010) employ the forecast combination method for out-of-sample equity premium prediction and report significant prediction gains. Recently, Barroso and Saxena (2018) explicitly utilize past out-of-sample errors to correct the portfolio weight estimates in portfolio optimization. In this paper, we propose a filter to the standard carry trade, that exploits past success in predicting returns. The filter is used to form a new filtered carry trade that essentially over-weighs currencies for which carry has shown more predictive power for returns in the past.

This paper is closely related to that by Suh (2019), who also employs a filtering method to exploit differential predictive power of interest rate differentials in the carry trade. While both Suh (2019) and this study attempt to improve the carry trade in a similar way, the two methods differ in measuring the predictive power and forming currency portfolios. In addition, this study also tests the relevance of the new trading signal in an adequate portfolio setting with parametric portfolio policies (PPPs). That is an important difference as few papers in the currency market literature use PPPs and this is a particularly robust optimization method. Furthermore, in that setting, this study tests statistical significance of the new filtered carry controlling for the strategy proposed in Suh (2019) and find that the improvement is robust.

There exists a strand of literature about improving carry trade profits. Examples include Bakshi and Panayotov (2013), Barroso and Santa-Clara (2015), Bekaert and Panayotov (2020), Burnside et al., 2006, Burnside et al., 2008, Burnside et al., 2011, Daniel, Hodrick, and Lu (2017) and Hassan and Mano (2019). This paper contributes to this literature by proposing a new method to improve carry trade profits. Further, the sorting-based currency trading strategy has become a standard method and has been employed in currency characteristics other than interest rate differentials (see, Della Corte, Ramadorai et al., 2016, Lustig et al., 2011, Menkhoff et al., 2012a). As the new method is also applicable to currency characteristics other than interest rate differentials, it potentially helps to improve other currency trading strategies.

We argue that the predictive power possessed by interest rate differentials may differ greatly across currencies. Unlike the conventional carry trade that forms equal-weighted currency portfolios, if the predictive power sufficiently differs across currencies, it would be better to put more (less) weight on currencies with high (low) predictive power for improving carry trade profits. We expect to find that this new predictive-power-weighted trading strategy could improve the profitability of the carry trade. To measure the predictive power of interest rate differentials for each currency, we first construct a dummy variable that indicates whether the sort on interest rate differentials correctly assigns the currency or not. The carry trade strategy specifies that correctly-assigned currencies with high (low) interest rates should exhibit high (low) realized future currency returns. We use historical information about portfolio-assignment errors to measure the predictive power of interest rate differentials. We then use the information variable to form predictive-power weighted currency portfolios instead of equal-weighted portfolios. We find that the new strategy offers significant additional returns for the carry trade.

We provide empirical evidence that our new carry trade outperforms the conventional carry trade. We statistically confirm that the new carry trade offers higher returns than the conventional carry trade in various settings by adjusting risks, controlling for the effect of transaction costs, and varying investment horizons. In order to check the robustness of our results, we consider an alternative method for forming the new carry trade, perform subperiod analysis, and analyze FX style-based investments for investors who want to use currency characteristics to form an optimal currency portfolio. Our results are robust to the various specification changes, confirming that our new carry trade offers investors greater profit opportunities than the conventional carry trade.

The outperformance of the new carry trade rests on some premises. First, the predictive power of interest rate differentials should sufficiently differ across currencies. We document evidence for the differentiated predictive powers across the cross section of currencies. Importantly, the predictive power is more differentiated across emerging market currencies than advanced currencies, and thus the new carry trade shows a better improvement for emerging market currencies than for advanced currencies. Second, the predictive power of interest rate differentials should persist. We find evidence for the persistence of portfolio-assignment accuracy. Consistent with intuition, the predictive power is negatively related with global FX market volatility, and we find that the relative improvement of the new carry trade is greater during tranquil periods than during volatile periods.

We show that equity or bond market risk factors cannot sufficiently explain the additional excess returns from the new carry trade. We also document that the outperformance of the new carry trade is unrelated with the carry trade crash risks. Our new carry trade not only deepens anomalous carry trade profits but also poses a further challenge for its explanation by adding more dimensions.

We develop an aggregate index by averaging the cross section of individual indexes at each time and find that it is positively correlated with future conventional carry trade returns. This aggregate index can serve as a new information variable for the return predictability of the conventional carry trade.

The remainder of this paper is organized as follows. Section 2 introduces the new carry trade and compares it with the conventional carry trade. Section 3 provides several empirical analyses. It first explains the data to be used, and presents performance results of both the new carry trade and the conventional one. In addition, it also explains the mechanism through which the new carry trade outperforms the conventional one, by presenting the cross sectional and time series properties of the predictive power index. It also tries to explain the outperformance of the new carry trade by analyzing whether equity or bond market risk factors can explain it or not and analyzing whether carry trade crash risks can explain it. The results for currency subgroup analysis and the relation to FX market volatility are also provided. In addition, it examines the validity of the new carry trade in an optimal currency portfolio choice problem. To provide the results for robustness checks and additional analyses, we also consider the effect of investment horizon, subperiod analysis, the information content of an aggregate predictive power index, and an alternative specification of the predictive power index. Section 4 concludes the paper.