Two Conditions Were Measuring Almost the Same Thing

The False Comfort of Multiple Filters

Screening workflows often suffer from the psychological urge to stack conditions. Researchers believe that requiring an asset to pass three separate technical gates guarantees higher quality results. In reality, combining momentum metrics derived from related lookback periods introduces extreme collinearity. A stock showing an elevated Relative Strength Index almost invariably presents an elevated Stochastic oscillator reading. The two metrics evaluate nearly identical mathematical dimensions within the raw time series.

When we audited a universe of 3,000 equities using both conditions simultaneously, the combined screen reduced the candidate list from 214 tickers down to 142. Superficially, this reduction felt like rigorous refinement. Yet examining the excluded 72 tickers revealed no meaningful structural weakness. Instead, minor tick-level differences in closing prices pushed those symbols just below an arbitrary decimal cutoff on one metric while comfortably clearing the other.

The Mathematics of Redundant Screening

To map criteria overlap quantitatively, we calculated the correlation between a 14-period RSI and a 20-period Fast Stochastic %K across 500 daily trading sessions. The Pearson correlation between their normalized outputs exceeded 0.91. That degree of collinearity means that nearly ninety percent of their variance stems from shared underlying price momentum. Adding the second indicator acts as a redundant confirmation loop rather than an orthogonal filter.

In an ideal research environment, each successive condition introduces an independent dimension of information, such as pairing price distribution with balance sheet debt or volume concentration. Stacking two price-ratio formulas merely creates double jeopardy for borderline assets. Assets meeting the core thesis get discarded simply because mathematical rounding artifacts penalize one calculation slightly more than the other.

Boundary Clipping and Unseen Survivor Bias

The real damage of criteria redundancy occurs at the distribution edges. When two conditions track the same force, their joint distribution forms a narrow diagonal ellipse. Applying rectangular cutoff boundaries to this diagonal cloud clips off dozens of valid data points sitting just outside the corner. Researchers mistake this artificial attrition for quality control, assuming the remaining subset represents superior candidates. In truth, the final list simply reflects arbitrary survival against identical mathematical hurdles.