Bootstrap Confidence Intervals for Trading Strategy Returns
Summary
The article explains how bootstrap resampling can estimate confidence intervals for trading system performance from out-of-sample results. It describes three approaches: pivot, percentile, and bias-corrected and accelerated intervals. A statistic such as mean returns or Sharpe ratio is recalculated across resampled datasets, and the resulting distribution is used to estimate a range for future performance. The accompanying MQL class lets users select an interval method, number of resamples, and statistic function.
The article motivates intervals as a way to judge whether observed profitability may be sufficient relative to risk and alternatives, rather than relying on a single test metric. It also recommends inspecting resampled statistic distributions for heavy tails. These estimates depend on the original sample representing future conditions; inadequate or unrepresentative data can make the intervals misleading. The methods offer probabilistic guidance, not precise forecasts or guarantees of live profitability.
Key ideas
- Bootstrap resampling creates replacement datasets by repeatedly drawing observations from the original sample.
- Pivot, percentile, and bias-corrected and accelerated methods provide different ways to construct confidence intervals.
- A user-defined statistic can be applied to each resample to estimate quantities such as mean returns or Sharpe ratio.
- Heavy-tailed bootstrap statistic distributions warrant caution when interpreting projected performance.
- Confidence intervals are only informative when the underlying sample represents conditions likely to occur in the future.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.