Hedge Research Perspective

Tactical Allocation in Commodity Futures Markets:
Combining Momentum and Term Structure Signals
Ana-Maria Fuertes, Joëlle Miffre, Georgios Rallis
May 2008

Link to SSRN

Data

The authors use standard futures data covering 35 commodities during the period 1979-2007.. The sample include some unusual contracts such as Milk and Lumber. Results do not take slippage cost into account.
The authors follow the a sorting approach of Fama-French type. At the end of each month, commodities are divided into quintiles with high and low momentum (backwardation) over the previous R months. The top portfolio of 5-7 contracts is then held long against the bottom portfolio.

Momentum results

Most profits are made on the long side, but short positions are also profitable on average. The reported Sharpe Ratio is 0.5 or lower. The maximum drawdown is large (-60%) roughly three times the annualised volatility (20%).
Returns are positively correlated with the GSCI (corresponding to a 60% weight to a long only portfolio). Trends thus seem to happen more in commodity bull markets. Returns are uncorrelated to equity and bond returns
The ‘favourite’ strategy holds contracts 1 month based on 1 month momentum.

Term Structure results.

The risk/return profile is similar to momentum based trading. The reported Sharpe Ratio is 0.5 or lower, the maximum drawdown is large (-60%). It appaers that the front contract has a stronger carry signal than back contracts.
Performance is positively correlated with GSCI returns (corresponding to a 50% weight to a long only portfolio.) Term structure results have a 0.30 - 0.40 correlation to momentum results. Presumably this is due to energy having high volatility and often being long.
Returns are uncorrelated to equity and bond returns
The ‘favourite’ strategy sorts contracts on the front month carry.

Double Sort Results

Here the available contracts are sorted on both Momentum and backwardation.  The reported Sharpe Ratio is around 0.7 is in line with expections from two uncorrelated strategies with a Sharpe ratio of 0.5. The maximum drawdown did not improve compare to the two substrategies  and remained around 60%. Returns are still correlated with the GSCI.

Other results.

Using trading volume as signal or conditioner, does nothing for performance.

Comments

The strategy lacks risk control and could therefore not be traded in real life. The results are, however, interesting as a backdrop for the analysis of commodity trading models. The reader should keep in mind, that only the strongest results are presented so there is a clear risk of data mining.

Since the strategy is long/short, it is surprising that its correlation with a long only benchmark is so high.

What we learn

• Momentum is a weak predictor.
• Commodity carry helps to predict commodity prices.
• The strategy works for both liquid and less liquid markets. Possibly a bit better for more liquid markets.
• Energy is likely the driver behind the results and it would be interesting to seem how this model would have performed as Energy prices fell in 2008. (i.e. out of sample).