In the brief August 2011 paper entitled “Paired-switching for Tactical Portfolio Allocation”, flagged by a subscriber, Akhilesh Maewal and Joel Bock investigate the efficiency of a simple momentum strategy applied to pairs of exchange-traded funds (ETF) with negative return correlations. Every 13 weeks (four times per year), they rank the performances of the two funds over the prior thirteen weeks and buy the fund that has the higher return. They ignore trading frictions. Using weekly adjusted closing levels of SPDR S&P 500 (SPY), iShares Barclays 20+ Year Treas Bond (TLT), iShares MSCI EAFE Index (EFA) and Vanguard Total Stock Market ETF (VTI) over the period from about October 2002 through about June 2011, they find that:
- The simple momentum strategy applied to negatively correlated fund pairs generates high returns with low volatility:
- The SPY/TLT pair (correlation -0.84) generates an annualized gross return of 15.0%, with annual standard deviation 7.6% and worst year 6.7%, far outperforming both SPY and TLT.
- The EFA/TLT pair (correlation -0.79) generates an annualized gross return of 20.1%, with annual standard deviation 9.5% and worst year 6.2%, far outperforming both EFA and TLT.
- The VTI/TLT pair (correlation -0.84) generates an annualized gross return of 15.6%, with annual standard deviation 7.6% and worst year 6.9%, far outperforming both VTI and TLT.
- A longer-term test using the Vanguard 500 Index Investor (VFINX) and Vanguard Long-Term Treasury Investor (VUSTX) mutual funds starting in 1991 (correlation -0.21) generates an annualized gross return of 11.3%, with annual standard deviation 9.3% and worst year -11.9%, comfortably outperforming VFINX and VUSTX.
- The strategy substantially improves the performances of “lazy” ETF portfolios such as: 60/40 SPY/TLT since 2003; Andrew Tobias 3-Fund (SPY, EFA, TLT) since 2003; Swensen (0.3 SPY, 0.2 VNQ, 0.2 EFA, 0.3 TLT) since 2005; and Bernstein Basic (SPY, IWB, EFA, TLT) since 2003.
- Since switching frequency is no more than four times per year, trading frictions for a reasonably sized portfolio would be small.
- For some pairs, ranking and holding intervals other than 13 weeks substantially improve results (with attendant data snooping bias).
In summary, evidence from simple tests suggests that investors may be able to extract high performance from simple momentum strategies by restricting the universe of choices to two assets with very negatively correlated returns.
Cautions regarding findings include:
- The study apparently uses the same sample to select fund pairs (calculate return correlations) and to measure momentum strategy performance. An investor operating in real time would have only data older than the trading period to select negatively correlated pairs, and pair correlations (such as for stock market and bond market proxies) vary over time. Specifically, an investor may not pick very good pairs for a 2003-2011 momentum strategy test based on correlations from before 2003. This look-ahead bias undermines confidence in findings.
- The number of pairs tested is very small. Repeated tests for many pairs with different (past) return correlations would be needed to confirm a reliable relationship between lagged pair return correlations and future pair switching strategy returns.
- There may be data snooping bias (luck via optimization) in the selection of the 13-week ranking/holding interval.
- As noted in the paper, backtests ignore trading frictions.
- Given the variability of returns, the sample periods are short compared to the ranking interval (27-35 quarterly observations).
- Statistical reliability tests assume tame return distributions.
