Blog - Investing Notes

April 11, 2008 - Using Commitments of Traders Reports to Time Asset Allocations

Is the aggregate sentiment of commodity traders predictive for associated asset returns? In their June 2006 paper entitled "How to Time the Commodity Market", Devraj Basu, Roel Oomen and Alexander Stremme investigate whether information in the weekly Commodity Futures Trading Commission's Commitments of Traders (COT) reports enable successful market timing. These reports tabulate the size and direction of the positions taken by different categories of futures traders in different assets. "Commercial" traders use futures contracts for hedging, "non-commercial" traders use them for other types of speculation and "non-reportable" traders operate below the individual reporting threshold. The study utilizes "hedging pressure" (the fraction of positions that are long) for the S&P 500 index (focusing on commercial and non-reportable hedging pressure) and for copper and oil (focusing on non-commercial hedging pressure) to adjust portfolio allocations among the S&P 500 index, copper and oil futures and a one-month certificate of deposit. Using COT reports and associated asset price data for the period January 1993 to May 2006, they conclude that:

• The data in COT reports offer reliable signals for switching between equities and commodities.
• Dynamic asset allocation strategies (weekly rebalancing) based on this data exhibit strong out-of-sample performance, beating the S&P 500 index by about 15% per year with Sharpe ratios over 1.0, as follows:
  • A long-only dynamic allocation strategy based on the predictive power of linear regressions of returns with respect to COT data from January 1993 through April 2000 generates an out-of-sample annual average portfolio return of 16.8% with Sharpe ratio 1.23 during May 2000 to May 2006 (see the chart below). A long-only static portfolio based only on historical means, volatilities and correlations of asset returns generates an average annual return of just 1.4% with Sharpe ratio -0.14.
  • A long-only dynamic allocation strategy based on the predictive power of linear regressions of returns with respect to COT data from January 1993 through April 2002 generates an out-of-sample annual average portfolio return of 18.5% with Sharpe ratio 1.64 during May 2002 to May 2006. A long-only static portfolio based only on historical means, volatilities and correlations of portfolio assets generates an average annual return of just 7.3% with Sharpe ratio 0.45.
• The interaction among different categories of futures traders, rather than the behavior of any one category, conveys the information supportive of successful market timing.

The following chart, taken from the paper, describes the out-of-sample performance of a dynamic strategy (weekly rebalancing) that allocates funds among the S&P 500 index, copper and oil futures and a one-month certificate of deposit during May 2000 to May 2006. The strategy is based on the linear relationships between asset returns and associated COT report data during January 1993 through April 2000.

The top graph shows cumulative portfolio return (Portfolio) compared to the cumulative return for the S&P 500 index (Index). This dynamic portfolio has an annualized Sharpe ratio of 1.23 and an "alpha" relative to the S&P 500 index of 13.9% per year.

The second and third graphs show the associated dynamic portfolio weights for the S&P 500 Index, a one-month certificate of deposit (CD or Risk-Free Asset) and copper and oil futures.

This study apparently excludes trading costs from return calculations.

In summary, the information in Commitments of Traders reports regarding aggregate positions of different categories of futures traders may support successful timing of associated markets.

For related research, see Blog Synthesis: Sentimental Journey and Blog Synthesis: Investing/Trading in Commodities and Commodity Futures.