Carry Trade Returns with Support Vector Machines

• Author: Gianfranco Forte,Emilio Colombo,Roberto Rossignoli
• DOI: http://doi.org/10.1111/irfi.12186
• Published date: 01 September 2019
• Date: 01 September 2019

Carry Trade Returns with Support

Vector Machines*

EMILIO COLOMBO

†

,GIANFRANCO FORTE

‡

AND ROBERTO ROSSIGNOLI

§

†

DiSEIS, Università Cattolica del Sacro Cuore, Milano, Italy

‡

DiSEADE, Università Milano-Bicocca, Milano, Italy and

§

Moneyfarm, London, UK

ABSTRACT

This paper proposes a novel approach to directional forecasts for carry trade

strategies based on support vector machines (SVMs), a learning algorithm

that delivers extremely promising results. Building on recent findings in the

literature on carry trade, we condition the SVM on indicators of uncertainty

and risk. We show that this provides a dramatic performance improvement

in strategy, particularly during periods of financial distress such as the recent

financial crises. Disentangling the measures of risk, we show that condition-

ing the SVM on measures of liquidity risk rather than on market volatility

yields the best performance.

Accepted: 17 February 2018

I. INTRODUCTION

Uncovered interest parity (UIP) is probably one of the simplest and most

intuitive equilibrium conditions in financial markets. For risk neutral investors

with rational expectations, the expected exchange rate change has to compensate

for the interest differential that may arise between two currencies. Such an equilib-

rium condition is most likely to hold in the FX market because it is the closest

approximation to the notion of market efficiency. Yet the empirical evidence fails to

support the UIP. On the contrary, the evidence shows the opposite: high-yielding

currencies tend to appreciate rather than depreciate, as theory predicts. This is

known as the “forward bias puzzle”

1

and has a natural implication: the possibility of

realizing excess returns from carry trade, that is, the practice of investing in high

yield currencies by going short on low yield ones.

Simple carry trade strategies deliver positive average excess returns for sub-

stantially long periods, coupled with Sharpe ratios significantly higher than

those measured in other financial markets (such as the US stock market), spur-

ring considerable attention from economists and practitioners alike. During

* We thank participants to Infiniti, Money, Macro and Finance, ICMAIF and IFABS conferences for

helpful comments.

1 The name derives from the fact that the failure of UIP results in the forward rate being a

biased predictor of future spot rates.

© 2018 International Review of Finance Ltd. 2018

International Review of Finance, 19:3, 2019: pp. 483–504

DOI: 10.1111/irfi.12186

the last decades, a large body of literature has investigated the reasons under-

lying the UIP puzzle and explanations for the excess of returns from carry

trade.

2

As an equilibrium condition, the carry trade does not require any model.

As such, it is a rather naïve strategy and we know that more sophisticated

investors would use any information that they may find useful, particularly

if it is from some established model. Unfortunately, the literature is of little

help in this respect. In fact, in a carry trade strategy, the only unknown is

the exchange rate and improving the strategy necessarily implies making a

correct guess about the future change in the exchange rate. However, since

the seminal work by Meese and Rogoff (1983), researchers struggled to find a

model capable of sufficient predictive power for the exchange rate. Recently,

there have been some improvements; Della Corte et al. (2009) show that for-

ward premia and stochastic volatility have predictive ability for exchange

rates; Li et al. (2014) reaffirm the out of sample predictive power of eco-

nomic fundamentals; and Molodtsova and Papell (2009) stress the role of

Taylor rule fundamentals. Nonetheless, as shown by the recent survey by

Rossi (2013), we are far from having a good predictive model for exchange

rates.

In explaining excess returns from carry trade, prior studies follow two main

approaches. On the one hand, the “traditional view”emphasizes the impor-

tance of fundamentals in providing useful information for exchange rate fore-

casting; in particular, Jordà and Taylor (2012) show that conditioning the carry

trade strategy on the predictions of a simple fundamental equilibrium exchange

rate model yields a better performance with a significant improvement in the

Sharpe ratio. On the other hand, building on the fact that carry trade returns

are negatively skewed, the “risk view”posits that they are essentially a compen-

sation for currency crashes, reflecting a sort of “Peso problem.”Indeed,

Brunnermeier et al. (2009), Burnside et al. (2011), and Farhi et al. (2015) show

that crash risk has accounted for a high proportion of the carry trade risk pre-

mium in advanced countries over the last 20 years.

3

Menkhoff et al. (2012) find

that excess returns to carry trades are a compensation for time-varying risk, and

in particular, for global foreign exchange volatility risk. More recently, Cened-

ese et al. (2014) provide some theoretical underpinnings to the relationship

between volatility and carry trade returns using an intertemporal capital asset

pricing model and show that conditioning the carry trade strategy on FX risk

measures results in a clear performance improvement, even accounting for

transaction costs.

Our paper contributes to this line of research providing innovative contribu-

tions in several domains: first, we propose a novel approach to directional

2 See Engel (2014) for a recent survey.

3 Using different techniques, Jurek (2014) downsizes the importance of crash risk. See also

Smales and Kininmonth (2016), who stress the role of investors’fear.

© 2018 International Review of Finance Ltd. 2018484

International Review of Finance