Effective Number of Trials

The effective number of trials is an estimate of how many independent strategy tests would represent the search breadth of a larger set of correlated variants.

If 500 strategies differ only slightly, treating all 500 as independent can overstate the amount of distinct research search. If the variants are strongly dependent, the effective count can be much lower than the raw count.

Raw trial count and independent search breadth are different

The distinction matters because search-aware statistics raise the hurdle for a reported winner as the research process explores more alternatives.

A raw count answers, "How many variants were tested?" An effective count asks, "How much independent search opportunity did those correlated variants create?"

That makes effective trial count closely related to family-wise error rate and to the distinction between nominal observations and effective sample size.

For the actual search-aware Sharpe calculation, use the Deflated Sharpe Ratio calculator. For an empirical benchmark showing how correlation among strategy variants changes the selected winner distribution, see Correlated Strategy Search.

Effective trial count is a model input, not an automatic truth

There is no universal conversion from an arbitrary strategy library to one exact independent-trial count. Dependence can be nonlinear, hierarchical, regime-dependent, and created by shared data or shared parameters.

A researcher should therefore document how the count was estimated and avoid silently substituting the raw number of tested variants.

Sources: Bailey and López de Prado, Deflated Sharpe Ratio and Grizzly Bulls correlated-strategy-search benchmark.

Explore more topics in the Financial Research Encyclopedia.