Bootstrap Inference for Comparing Sharpe Ratios

Summary

The document asks how to test whether a strategy’s Sharpe ratio differs significantly from another strategy and how to obtain a p-value. Its response emphasizes that strategy returns are often serially correlated, so ordinary resampling of individual observations may not preserve the dependence structure needed for inference.

It points to time-series bootstrap methods, including block and wild bootstrap approaches, as possible ways to estimate the sampling distribution for a comparison. It also mentions comparing performance with random portfolios or an approximation to them, using a naive benchmark as a reference. The document does not specify a null hypothesis, a resampling design, a test statistic, or implementation details, and it gives no empirical results. Those choices depend on the return process and comparison being studied, so the suggestions are starting points rather than a complete testing recipe.

Key ideas

• Serial correlation in strategy returns complicates inference about Sharpe ratios.
• Block bootstrap methods can preserve some time dependence by resampling return blocks.
• Wild bootstrap is another candidate approach for time-series inference.
• A random-portfolio comparison can benchmark a strategy against naive alternatives.
• The document does not provide a full null hypothesis or a procedure for calculating a p-value.

Tags

# How do i test the significance of Sharpe ratio of a strategy using bootstrap


# How do i test the significance of Sharpe ratio of a strategy using bootstrap












How do i test the significance of Sharpe ratio of a strategy whether it is any different from another strategy ?? How do i get a p-value out of it ?

What should be the H0 in the hypothesis testing ?

I was hinted by my professor that i would require bootstrap

## Answer by Dirk Eddelbuettel (score 5, accepted)

https://quant.stackexchange.com/a/11164

"It's compliated" because the trading strategy performance will depend on the data which is most likely serially correlated.

So you want to look into bootstrap approaches for time series such as the block bootstrap, or the wild bootstrap.

Another approach would be to look into 'random portfolios' or an approximation thereof. The basic idea is to test how much better your portfolios performs relative to a naive benchmark.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.