10 Hypotheses, 0 Adopted — One Month of Validation

A system you can't trust is one that merely "reuses rules that happened to fit the past." So we test ours constantly. This is the first such record.

The question over the past month was simple — is there a way to push the Regime Strategy's returns higher? We tested more than ten improvement hypotheses and adopted none.

What we tested, and the results

New signals (8) — Eight indicators the system doesn't currently use (price action, volume, foreign-ownership ratio, and others), each tested individually. All eight rejected. In most cases the information was already reflected in existing signals, so no added explanatory power.

Dynamic position sizing (1) — Adjusting capital by market volatility. Rejected. It overlapped with the existing regime-detection function and added no independent effect.

Separating flow participants (1) — Measuring the predictive power of foreign vs. institutional net buying separately. Rejected. No statistically significant difference between the two (t = 1.05, p = 0.29). As an aside, the common belief in an "institutional information edge" did not reproduce in our data.

Expanding the universe (2) — KOSDAQ was deferred: delisted-stock data was too sparse to correct for survivorship bias. Mid-caps were tested fully but rejected — they underperformed the large-cap system.

Why "0 adopted" matters

Some hypotheses looked good over specific windows. But we judged that to be overfitting — tailoring a system to fit one slice of the past, which usually collapses in live trading — and did not adopt them.

That all ten converged on "no effect" suggests the current system has little room left to improve within available data. This is a conclusion from validation, not a guess.

Based on these results, the Regime Strategy's logic remains unchanged.

This post is for informational purposes and is not investment advice. All figures are backtest results on historical data.