Classical oversold oscillators (RSI / Williams %R / MFI / Stochastic)

Thesis (rejected): extreme oversold readings on the standard oscillators (RSI < 30, Williams %R < -90, MFI < 20, Stochastic %K < 15) mark exhaustion of selling pressure. Buying on these readings — alone or as a 3-of-3 consensus — should produce a tradeable bottom.

Code: research/oversold_indicators.py Report: research/oversold_indicators_report.json (n=422,148 events across all four oscillators)

Why it failed: Even the strictest combination — all three of RSI / Williams / MFI in extreme oversold simultaneously, on a name already ≥ 30 % off its 252d high, OOS 2020+ — barely produces +0.8 % edge at 40 d. Short-horizon edges are negative: buying when an oscillator extreme prints reliably bleeds 1-2 % over the next week before any mean reversion shows up.

Oscillator extremes are noisy: a stock that's gone "very oversold" by the textbook oscillator definition is more often than not still in the middle of a slide, not at the bottom. The "buying-at-RSI-30" discipline is essentially the value-trap edge — flat-to-negative.

Verdict: Skip the oscillators as standalone bottom-timers. The direction of the oscillator reading isn't enough — what matters is the pattern over time (divergence; see Working signals below) or a consensus across distinct trigger mechanisms (consensus_capitulation).