Building RevDog: Systematic small-cap quant strategy
We extended our walk-forward factor validation from 6 to 15 months on NSE SmallCap 250. Signal got stronger — IC improved, stability increased, and two new persistent factors emerged that a shorter window couldn't detect.
Most mean-reversion strategies break when you touch the parameters. This one didn’t. After 18 sensitivity tests, removing logic, and tightening constraints, SmallCap Dislocation held up—because it’s driven by forced selling, not curve-fitting.
Indian small caps misprice during forced selling, not fundamentals. Using a day-by-day portfolio simulation, we tested a dislocation strategy that exploits liquidity shocks—64.8% win rate, −1.02 skew, and 3.1% max drawdown with strict risk controls.
Most NSE small-cap traders obsess over indicators. That’s a mistake. The real difference between mean reversion and trend following isn’t RSI or breakouts—it’s the return distribution you’re betting your capital and psychology on.
We tested multiple exit models in a small-cap strategy and found that binary exits destroy value. A score-driven decay framework with hysteresis improved returns by 7.4pp, raised Sharpe, and cut turnover by 46%.
Most quants chase high IC factors. That’s a mistake. This article breaks down, with real numbers, how low-IC quality signals can outperform through volatility suppression, regime behavior, and correct portfolio construction.
Should risk-on markets mean aggressive concentration? We tested 8–10 stock portfolios across regimes using live and backtested data. The result surprised us: concentration destroyed Sharpe. Here’s what the data actually says.
We tested a common belief: avoid stocks far above their 200DMA. The data showed the opposite—momentum persisted. More importantly, stocks 20%+ below their 200DMA delivered 2–4× higher monthly returns in extreme regimes. That non-linear insight reshaped our model.
Most investment systems record outputs, not reasoning. Decision traces capture why a system acted at the moment of commitment—inputs used, constraints applied, actions taken—turning opaque models into legible, self-auditing decision engines.
Backtests and walk-forward validation serve different purposes. One measures historical outcomes; the other tests whether signals actually persist out-of-sample. Robust systems require both—used correctly and in the right order.
Most factor models fail not because signals disappear, but because they are trusted for too long. This note explains how regime-aware factor gating, conditional growth, and governance caps were implemented in RevDog V3 to manage drawdowns in Indian equities.
RevDog tested sector-relative vs absolute scoring in its V2→V3 transition. With a 0.95+ rank correlation across 111 stocks, results showed normalization mattered far less than regime-aware factor gating—leading to a simpler, more robust V3 design.