Optimizing Trading Strategy Parameters and Testing Their Robustness
Summary
The article introduces parameter optimization for systematic strategies, using a moving average window as a simple example. It describes choosing a default value, search range, and step size, then comparing approaches such as sequential ascent, brute force, and genetic optimization. It also raises a key design question: whether selecting the single best objective score is a sound way to choose parameters, and points to splitting data into training and test periods or using walk-forward analysis to assess performance beyond the fitting sample.
A pairs trading example illustrates optimizing a hedge ratio, a z-score lookback, and entry spacing, with a genetic search and walk-forward settings. The article advises examining why an apparent effect might exist, whether it persists through time, whether simpler implementations capture it, and whether it appears across assets. It provides implementation examples but no reported comparative results. Optimization can overfit historical data, and the examples do not establish that any chosen parameters or strategy will work in live trading.
Key ideas
- Parameter searches require explicit defaults, ranges, and step sizes.
- Sequential ascent, brute force, and genetic methods offer different ways to search parameter combinations.
- A high in-sample objective score alone does not establish that parameters will generalize.
- Training and test splits and walk-forward analysis can help evaluate robustness.
- Investigate a strategy’s rationale, persistence, simplicity, and breadth across assets.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.