Skip to content
All library documents

Selecting Trading Rules Under a Target False Discovery Rate

Article Quant Q&A · Author: John
Read the original

Summary

The document asks how to apply the FDR+ method to select trading rules that appear to outperform. It says the author has computed p-values using a stationary bootstrap, following a cited study, but provides no worked calculation, code, or detailed definition of the FDR+ statistic.

The procedure quoted from the study begins with the positively performing rule with the smallest p-value. It adds rules in order of increasing p-value, recalculating portfolio FDR+ after each addition, and stops when the chosen threshold would be exceeded. The selected set is the largest sequence that remains at or below the predetermined FDR+ level. This is a brief description of a selection algorithm rather than a complete implementation guide. The document does not specify how to calculate FDR+ at each step, handle ties, or validate assumptions, so readers would need the cited paper for those details.

Key ideas

  • The author applies stationary-bootstrap p-values to trading rules.
  • Selection starts with the positive-performing rule that has the smallest p-value.
  • Rules are added in p-value order while recalculating portfolio FDR+.
  • The final set must remain within a predetermined FDR+ threshold.
  • The document does not explain the FDR+ calculation or provide implementation details.

Tags

StatisticsBacktestingMachine learning
Full text
# Trading rules: Controlling the portfolio FDR+ level


# Trading rules: Controlling the portfolio FDR+ level












I'm trying to apply the FDR+ (False Discovery Rate +) methodology from Bajgrowicz (2011) link another_link. I have computed the p-values with the stationary bootstrap as they did, however I am not sure how to use the FDR+ methodology in order to pick the trading rules that are 'truly' outperforming.

This is what they argue in their paper:

"We derive an algorithm that allows the construction of a portfolio of trading rules with a FDR+ level fixed at at predetermined rate. The algorithm starts with the rule having the smallest p-value (and a positive performance). Then, the rule corresponding to the next p-value is added and the FDR+ recomputed. This process is repeated until we reach the desired FDR+ rate, and we select the rules resulting in a FDR+ not greater than the predetermined level."

I'm trying to make the code in R, but some pseudocode would already be of great help.

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.