It is common practice to divide currency carry strategies into two separate baskets: one for developed and one for emerging market currencies. It is often assumed that this helps manage risk in an investment strategy that is known to be vulnerable to negative tail events.
But is this just a myth? When we tested this assumption we found quite the opposite.
Across markets carry is thought of as the expected return of an asset assuming that its price does not change. As an investment strategy, currency carry usually involves selling low-interest-rate currencies and investing in high-interest-rate currencies. Over the past 20 years, a diversified portfolio of these positions would have delivered a risk-reward ratio of around 0.5-0.6. The source of risk in this strategy is, of course, that the price of currencies can change—sometimes markedly and rapidly. If the long currencies crash or the short currencies spike, considerable losses may be incurred.
Superficially, then, it may seem intuitive to separate currency carry strategies into developed and emerging market baskets. Because EM currencies tend to be higher-yielding than DM currencies, if you blend the two universes you will tend to sell DM currencies to buy EM currencies. When you separate the two universes, you will always sell EM currencies (say, TWD or CZK) to buy other EM currencies (perhaps BRL or TRY). EM currencies are generally more volatile, and therefore it makes sense for both sides of the trade be an EM currency—that way, if the currency you are buying crashes heavily, the currency you are selling is more likely to be crashing heavily, too.
In testing this, however, we found that the split portfolio underperformed a blended portfolio even on a risk-adjusted basis. In fact, the evidence is mixed as to whether separating developed and emerging currencies reduces risk at all.
Constructing the currency carry trade with separated emerging and developed constituents makes for a less attractive strategy because the reduction in carry is certain and often substantial, but any risk reduction is uncertain and often modest. Having initially observed this in mean variance optimization examples, we ran an analysis using all freely floating currencies with home countries that were classified as “Emerging Market” or “Developed” by the MSCI classification system, between 2000 and 2018. We excluded the Pakistan rupee and the Venezuelan bolivar, for which forward rate data is not available. We excluded pegged currencies because the risks stemming from pegs are structurally idiosyncratic and can lead to biases in in-sample results, a phenomenon named the “peso problem.”
This universe was ranked by three month interest rates and rebalanced weekly. Then we ran two distinct strategies: in the full universe strategy we were long the top-four highest-carry currencies and short the four lowest-carry currencies; in the separate-baskets strategy we were long and short the top- and bottom-two currencies in the emerging and developed baskets, respectively. All currencies were assumed to be equally weighted.
The results are shown below.
Hypothetical Backtest Performance Results
Source: Bloomberg, Neuberger Berman analysis.
Hypothetical Backtested Methodology: The simulated portfolio seeks to achieve its objective by investing in global currencies. The investment process represented in the model portfolio has multiple steps: The exposure to the carry risk premium is determined for all currencies in the investable universe. The relative attractiveness of each currency is then expressed as a ranking based on the currency’s exposure to the carry risk premium. On a weekly basis, the position sizes are systematically increased or decreased based on each currency’s ranking. Key assumptions used in the simulation include: no transaction costs; currency returns derived from spot rate return and carry return derived from available three month interest rate as per the covered interest rate parity; a portfolio starting value of 100 million USD; and max target trade size under 6.25% of exposure per currency. The model assumes weekly rebalancing. The period of the test was January 1, 2000 to September 30, 2018.
See important hypothetical backtested performance disclosures at the end of this material.
We observed that not only was the expected return of the full universe strategy higher, but the reduction in volatility achieved by separating the baskets was modest—and this outperformance persisted even when it included extremes such as the Argentine peso devaluation that occurred when the currency was free-floating and classified as EM. We found that separating emerging and developed markets increased tail risk: for example, maximum drawdown, which occurred in 2008, increased significantly from 21% to 31%, suggesting that systematic risk was not reduced. The changes in skewness and kurtosis were also unfavourable.
Risk management is an important consideration in currency carry strategies, which are vulnerable to both systematic tail events and idiosyncratic currency crashes. This is often taken as the basis for separating emerging and developed currencies in carry portfolios, but we do not believe that the backtest evidence supports this approach. Instead, we maintain that a thoughtfully designed approach to currency carry, which includes trading emerging and developed market currencies together, can capitalize on the upside of the carry risk premium while actively helping to manage the drawdowns to which it is prone.