Can reacting to short-term changes in asset return correlations improve efficient portfolio allocation? In their May 2012 paper entitled “The Role of Correlation Dynamics in Sector Allocation”, Elena Kalotychou, Sotiris Staikouras and Zhao Gang investigate the economic value of correlation timing in mean-variance optimal sector allocations. They test the usefulness of several correlation forecast methods by constructing dynamic, one-step-ahead (day, week or month) mean-variance optimal portfolios comprised of ten sectors in the Japanese, UK or U.S. equity markets (energy, basic materials, industrials, consumer goods, health care, consumer services, telecommunication, utilities, financials and technology). They use a static portfolio based on total-sample correlations as a benchmark. They use an initial subperiod (July 1996 through May 2005) to generate initial correlation forecasts and a later subperiod (Jun 2005 through May 2007) to implement recursive forecasts. They estimate sector index trading frictions for daily (monthly) portfolio rebalancing as approximately 0.07% (0.09%) for U.S., 0.30% (0.32%) for Japanese and 0.50% (0.52%) for UK sector indexes. Using daily prices for the ten sector indexes for each of the Nikkei 225, FTSE-All and S&P 500 during July 1996 through May 2007, along with corresponding interbank and U.S. Treasury bill yields as risk-free rates, they find that:
- Over the entire sample period, average pairwise sector correlations for Japan, UK and U.S. markets are 0.55, 0.44 and 0.52, respectively. Consumer services (0.62) and industrials (0.58) exhibit the highest average correlations with other sectors, while utilities (0.36) exhibit the lowest.
- Dynamic sector allocation portfolios generally beat the static portfolio based on gross Sharpe ratio with daily, weekly or monthly rebalancing to reflect shifting correlation outlooks.
- However, trading frictions for daily rebalancing offset advantages for dynamic portfolios in all three markets, and weekly rebalancing is advantageous only in the U.S. market.
- The gross benefits of correlation timing are more pronounced for monthly rebalancing (6.0% to 22.8% annualized) than daily rebalancing (5.5% to 18.9% annualized) and are robust to reasonable trading frictions. Depending on the correlation forecast model and level of investor risk aversion, breakeven trading frictions for optimal correlation-timed portfolios with monthly rebalancing range from 0.36% to 0.92% per trade.
In summary, evidence indicates that correlation timing implemented with monthly rebalancing may outperform static mean-variance optimal allocation on a net basis.
It would be interesting to use a portfolio with equal sector weightings as a benchmark.
Cautions regarding findings include:
- Testing of several different correlation forecasting methods and three different rebalancing intervals introduces data snooping bias, such that the best combination incorporates sample-specific luck.
- The out-of-sample test period is small for monthly measurements (24 months) and does not encompass a variety of equity market conditions (ending before the financial crisis).
- Data collection and processing efforts for tracking and implementing correlation changes appear substantial (or costly if delegated to an investment manager).
- Some investors may not be able to achieve even the largest break-even trading frictions estimated in the study.
