Conditional currency hedging

• Date: 01 December 2020
• DOI: http://doi.org/10.1111/fima.12287
• Published date: 01 December 2020
• Author: Melk C. Bucher

DOI: 10.1111/fima.12287

ORIGINAL ARTICLE

Conditional currency hedging

Melk C. Bucher

PartnersGroup & University of St. Gallen (HSG)

Correspondence

MelkC. Bucher, Partners Group, Zugerstrasse 57,

6341Baar, Switzerland.

Email:Melk.bucher@unisg.ch

Thepaper has mostly been written during

PhDstudies at the University of St. Gallen and

ColumbiaUniversity.

Theopinions expressed in this publication are

thoseof the author alone.

Abstract

I propose a simple and robust approach to hedge currency risk

that can be directly applied by international investors in diverse

asset classes. Compared to current mean-variance approaches, it

is robust to overfitting and thus better anticipates risk-minimizing

currency positions for global equity, bond, and commodity investors

out of sample. Furthermore, correlations among currencies, equi-

ties, and commodities can be predicted by lagged implied foreign

exchange volatility. This allows investors to dynamically adjust

their hedges, resulting in significantly lower risk compared to other

hedging alternatives while maintaining or even improving Sharpe

ratio, particularly during crisis periods.

KEYWORDS

dynamic currency hedging, implied volatility, mean-variance analy-

sis, overfitting

1INTRODUCTION

Since the demise of the postwar Bretton Woodssystem in the 1970s, currency fluctuations have been a major financial

risk. International securities can provide better risk-reward trade-offs, but in a regime of floating foreign exchange

(FX) rates, any investordealing with a foreign currency is inherently exposed to adverse FX movements. It is therefore

natural to ask whether and how one can optimally hedge this currency risk.

I propose a simple and robust mean-variance approach to hedge FX risk for investors globally—dynamic condi-

tionalcurrency hedging (DCCH). This entails three contributions. First, I document that current mean-variance hedging

approaches break down out of sample due to their in-sample bias and tendency to overfit. Second, I propose DCCH as

an alternate hedging mechanism that is less prone to overfit and performs favorably both ina nd outof sample. Com-

pared to the alternatives discussed in the literatureto date, this approach leads to lower standard deviation and down-

side risk while maintaining or improving Sharpe ratio. Third, correlation patterns between major currency pairs and

other assets, such as equities and commodities, can be strongly predicted by implied FX volatility (IVOL). Using this as

conditional information for DCCH further reduces risk and improves the risk–return relationship of hedged returns.

This paper contributes to three related streams of literature on international diversification, global risk factors, and

systematic FX investment styles. The benefits of international diversification havelong been established in the 1960s

c

2019 Financial Management Association International

Financial Management. 2020;49:897–923. wileyonlinelibrary.com/journal/fima 897

898 BUCHER

and 1970s with work by Grubel (1968), Levy and Sarnat (1970), and Solnik (1974), for example. However,as noted

in Kroencke, Schindler, and Schrimpf (2014) and Christensen and Varneskov (2016), the inherent currency exposure

in international investments and its diversificationpotential has often been neglected. Only with Perold and Schulman

(1988), who find risk reductions for full relative to no hedging of international equity positions, has the diversification

literature started to explicitly analyze the question of currency exposure.Since then, Campbell, Serfaty-de Medeiros,

and Viceira (2010), de Roon, Nijman, and Werker(2003), Glen and Jorion (1993), and Kroencke et al. (2014) have ana-

lyzed the optimal FX positions of international investors in a mean-variance framework,with different results depend-

ing on the specific analysis and sample at hand.1Although all authors conclude that for an international equity investor

full hedging reduces risk relative to no hedging, the first two papers find no additional benefit of unconditional hedg-

ing while Campbell et al. (2010) find significant further risk reduction. Further,they disagree on whether a conditional

hedging strategy based on interest rate differentials—essentially a form of carry trade—furtherreduces risk.

The current mean-variance approaches suffer from in-sample bias. The risk-minimizing FX positions are estimated

with data of the entire sample that are only available ex post. This is unrealistic and thus makes it impossible to apply

the approach in reality. Investors in 1975 at the beginning of Campbell et al. (2010) sample could not possibly have

anticipated the equity market correlation of currencies decades into the future. Acknowledging that correlation pat-

terns between currencies and equities change over time, the in-sample results reported in previous literature are not

a realistic benchmark for achievable hedging results. Thus, the feasible risk reduction through mean-variance hedging

remains an open empirical question.

This paper developsa hedging framework that can be directly employedby real investors. Such an ex ante approach

necessarily can make use of only past or current data. In line with Christensen and Varneskov(2016), I find substan-

tial time variation in the optimal currency hedging in G7 currencies for an international investor. Giventhe changing

correlation properties of currencies, it is unsurprising that realistically feasible risk reductions are clearly lower than

in sample. Importantly, risk reduction through DCCH hedging for an equity and commodity investoris still significant

relativeto its alternatives, such as full hedging. The DCCH risk reduction is robust to changes in the estimation window

as demonstrated in Section 6.

The inter-asset covariance matrix of FX–equity–commodity returns is not just time varying but also stochastic—

the true underlying data generating process is unknown and is estimated with noise. An investor should be cautious

not to overfit his/her mean-variance hedging model to past data when choosing optimal FX positions. The investorhas

to separate the signal—the true underlying correlation between investment portfolio and currencies involved—from

the random noise. Tothis end, I introduce a simple yet effective regularization approach (the least absolute shrinkage

and selection operator: Lasso) to translate the backfitted optimal in-sample FX positions into conservative hedging

positions going forward. This simplifies the mean-variance hedging approach considerably as the investor now places

more weight on currencies with more stable correlation properties and takes zero positions (fully hedges) currencies

with a lot of noise. This leads to more stable and robust FX positions that are lower in absolute size (long or short),

which consequently lowers FX turnover and makes the hedging easier to implement in practice. Most importantly,it

also has a strong impact on the risk–return performance in out-of-sample testing, leading to significantly lower risk as

well as higher returns per unit of risk than unconstrained mean-variance models.

When the current mean-variance FX literature introduces conditioning information, it mostly uses backward-

looking and slowly moving variables, often related to FX investment styles. Most often interest differentials are being

used, essentially implementing a form of the carry trade (e.g., Campbell et al., 2010). More recently, Kroenckeet al.

(2014) have employed (backward looking) FX momentum and FX value as conditioning factors. These backward-

looking factors are not ideal conditioning information and often show nonexistentor marginal improvement to hedging

performance.2It seems a lot more intuitive to employ forward-looking risk factors for conditional risk management.

Christensen and Varneskov(2016) make a first step in the right direction by making use of intraperiod dynamics in the

1Basedon a methodology developed by Anderson and Danthine (1981).

2Seeresults of Campbell, Serfaty-de Medeiros, and Viceira (2010) and Glen and Jorion (1993).