This is the lesson that ties the whole Financial Machine Learning module together, and it is the one a senior interviewer is most likely to press on. Every tool in this module exists to stop you from fooling yourself, and the most common way quants fool themselves is backtest overfitting: trying many strategies, picking the one with the best Sharpe ratio, and forgetting that the best of many random numbers is large even when none of the strategies has any edge. This lesson frames that as a multiple-testing problem, derives how large a Sharpe ratio pure luck produces when you run many trials, then builds the two corrections that fix it: the deflated Sharpe ratio, which discounts an observed Sharpe for the number of trials and for non-normal returns, and the probability of backtest overfitting, which estimates how likely your selected strategy is to disappoint out of sample. It draws on the bias-variance idea, the purged cross-validation paths from the first lesson of the module, and the hypothesis-testing logic from the Probability course.
Table of Contents
Already have an account? Log in!