Same Condition, Different Universe

The Myth of the Context-Free Screen

When researchers draft a screening formula in TC2000 or custom backtesting tools, they usually treat conditions as mathematically neutral filters. Testing that same condition across two distinct market universes immediately invalidates that premise. An identical technical condition—such as a 20-day high breakout coupled with a volume surge—behaves like two completely unrelated algorithms when shifted from large-cap index constituents to an expansive small-cap universe.

The mathematical expression remains completely static, yet the ambient distribution of the underlying universe dictates the hit rate and character of the result set. Large-cap stocks possess tighter spreads, dominant institutional sponsorship, and lower baseline idiosyncratic volatility. Small-cap issues introduce episodic illiquidity, wide bid-ask gaps, and frequent micro-structure false signals. The screening logic did not alter its mechanics, yet the resulting dataset tells a contradictory analytical narrative.

Empirical Drift Between Baskets

During benchmark audits across trailing market regimes, applying a simple volatility expansion trigger produced an average of 4.2 passing candidates per trading session within the liquid large-cap pool. Transferring the exact code without any threshold adjustment into an uncurated small-cap basket expanded average daily candidate volume to 38.6 candidates. The volume distribution alone creates an insurmountable qualitative difference in how an analyst inspects individual charts.

This dramatic dispersion demonstrates how baseline asset characteristics reshape binary criteria. Passing candidates in the large-cap group consistently demonstrated orderly multi-day continuation along primary trends. Conversely, candidates passing within the broader universe frequently displayed sharp exhaustion moves, erratic spread widening, and sudden mean reversion on following bars. Without specifying the denominator universe, citing a screener’s pass rate provides zero meaningful insight.

The Trap of Compensatory Filtering

Analysts regularly make the mistake of modifying their conditions when moving across watchlists. Seeing noisy or overwhelming candidate counts in broader baskets, they reflexively add secondary filters, such as stricter moving average slopes or proprietary volume multipliers. This response misdiagnoses the problem: the syntax of the condition was never defective, but the underlying population simply contains distinct distributional characteristics. Tweaking parameters to compensate for uncurated input sets introduces curve-fitting and disguises fundamental selection errors.