Pooling Forecast Weights Across Instruments and Market Groups
Summary
The document examines whether forecast weights should be fitted separately for each instrument, pooled across all markets, or pooled within similar groups, and whether blending these estimates can balance robustness with market-specific performance. It motivates the question with a system combining multiple momentum speeds, where fitting allocates risk across signals while keeping the number of free parameters limited. Risk normalization is intended to make instruments and forecasts comparable without separate volatility calibration.
The examples compare momentum signal correlations and estimated Sharpe ratios for corn and an equity index. Correlations appear similar, while estimated signal performance and fitted weights differ; robust optimization moderates some of those differences. The author also shows weights under separate fitting and pooled gross returns, preserving instrument-specific trading costs so costs can influence trading speed. Later tables compare individual, group, whole-system, and blended weights across strategy types. The supplied excerpt introduces these alternatives but omits much of the results and final summary, so it does not establish which fitting approach performs best. Estimated differences may also be noisy, and pooling choices depend on the similarity of markets and strategies.
Key ideas
- Pooling market data can make forecast weight estimates more robust by increasing the information available for fitting.
- Instrument-specific fitting can reflect differences in signal performance but may rely on less data.
- Risk normalization helps make signals and instruments comparable during forecast weight estimation.
- The examples show similar momentum signal correlations across two markets alongside differing estimated Sharpe ratios and weights.
- A blended estimate can combine individual, group, and system-wide fitting, while instrument-specific costs remain relevant.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.