Defining Redundancy in Screening Logic
Screening engines execute boolean evaluations across thousands of candidate instruments in sequential passes. A filter is formally redundant when its activation removes no candidates beyond those already excluded by previously evaluated conditions in the stack. In practice, researchers often layer multiple momentum or trend criteria believing each adds a layer of confirmation. When those criteria share mathematical roots, the secondary test simply confirms the first test under another name.
For instance, pairing a moving average slope condition with a short-period rate-of-change filter sounds like comprehensive trend confirmation. However, empirical testing reveals that 97% of instruments meeting the slope requirement already clear the rate-of-change hurdle. The candidate universe retains identical constituent membership, while the researcher assumes an enhanced level of selectivity that does not exist in the data.
A condition that filters out nothing new does not protect capital; it only complicates maintenance and blinds the researcher to genuine structural vulnerabilities.
Three Diagnostic Tests for Candidate Criteria
Isolating redundant filters requires stepping through each rule independently rather than running an entire screening package as a monolith. Screening teams can verify incremental contribution using three practical diagnostic steps:
- Marginal Exclusion Ratio: Run the full screen, toggle the candidate filter off, and record the exact change in result size. An exclusion shift below 3% flags near-total collinearity with remaining rules.
- Pairwise Cross-Tabulation: Compare candidate passes across individual criteria in isolation. If condition A yields 820 symbols and condition B yields 835 symbols with an 810-symbol overlap, one rule is mathematically superfluous.
- Temporal Stability Check: Test the rule across multiple market regimes. A filter that acts redundantly in trending markets might suddenly over-constrain the set during high volatility, causing erratic basket turnover.
Removing redundant criteria simplifies screener maintenance and reduces execution latency across large instrument tables. More importantly, it prevents the false confidence that arises from thinking five independent indicators validate a single setup when all five merely track short-term price momentum.