We Tested Whether More Diversification Helps a Mean-Reversion Strategy. It Doesn't.
We tested 5 position-sizing configs for our NSE SmallCap mean-reversion strategy — from 5 to 15 concurrent positions. More diversification hurt every metric. The 7-position limit acts as a quality filter, not a risk concentration.
We recently demoed our smallcap strategies to a seasoned institutional investor — someone who's spent over 15 years across public markets, venture capital, and private equity at some of Asia's most rigorous capital allocators.
He gave us sharp, specific feedback on both strategies.
This post is about one of those challenges.
After reviewing SmallCap Dislocation — our mean-reversion overlay that catches forced selling in NSE SmallCap 250 stocks — he looked at the portfolio construction and said:
"Seven positions at 3.5% each? For a statistical mean-reversion strategy, that's highly concentrated. You're relying on correlation, not causation — you need more positions to let the law of large numbers work."
The intuition is sound.
Classical mean-reversion theory says: each trade has a probabilistic edge, not a deterministic one.
The more positions you run, the more your realized returns converge to your expected edge.
Seven positions is a small sample on any given day.
We didn't have data to agree or disagree. So we ran the test.
The Strategy in Brief
SmallCap Dislocation looks for stocks that have been hit by forced selling — panic, margin calls, fund redemptions — and enters after the selling exhausts. The signal pipeline:
- Shock detection: Stock's 5-day return falls into the bottom 7% of the entire universe (or 10-day return in the bottom 10%). Percentile-based, not absolute — adapts automatically to market volatility.
- Failed follow-through: Wait at least 3 days. If the stock never closes below the shock-day low, sellers are exhausted. That's the entry signal.
- Exits are mechanical: +6% target, -15% stop loss, 30-day time stop. No discretion.
Position sizing: 3.5% per trade, maximum 7 concurrent positions (25% gross exposure), no more than 2 per sector, 30-day cooldown after exit.
The strategy runs as an overlay on Catalyst's idle cash — it doesn't need its own capital allocation.
The 7-position limit was never independently tested.
It was derived from arithmetic: 25% gross cap divided by 3.5% position size equals 7.
The 25% and 3.5% were reasoned about qualitatively — 25% prevents the overlay from taking over the portfolio during stress, 3.5% keeps worst-case single-trade impact under 1%.
But nobody had asked: what if we spread the same 25% across more positions?
What We Tested
Five configurations, all holding gross exposure near 25%, over January 2023 to October 2025 — roughly 713 trading days across 250 NSE SmallCap stocks:
| Configuration | Position Size | Max Positions | Max Per Sector | Gross Exposure |
|---|---|---|---|---|
| 5 @ 5.0% | 5.0% | 5 | 2 | 25% |
| 7 @ 3.5% (current) | 3.5% | 7 | 2 | 25% |
| 10 @ 2.5% | 2.5% | 10 | 3 | 25% |
| 12 @ 2.4% | 2.4% | 10 | 3 | 24% |
| 15 @ 1.7% | 1.7% | 15 | 3 | 26% |
Everything else stayed constant — same shock thresholds, same entry window, same exit rules, same cooldown, same liquidity filters.
The only variable was how we sliced the 25% gross exposure across positions.
If the fund manager was right, more positions should improve the Sharpe ratio and reduce max drawdown — even if average P&L per trade drops, the portfolio-level risk-adjusted return should improve through diversification.
The Results
| Config | Trades | Win Rate | Avg P&L | Max DD | Sharpe | Total Return |
|---|---|---|---|---|---|---|
| 5 @ 5.0% | 242 | 59.5% | +0.80% | -12.3% | 0.33 | +3.8% |
| 7 @ 3.5% | 340 | 60.6% | +1.43% | -7.4% | 1.14 | +11.3% |
| 10 @ 2.5% | 488 | 58.6% | +0.82% | -10.4% | 0.47 | +4.4% |
| 12 @ 2.4% | 485 | 56.7% | +0.67% | -8.6% | 0.33 | +2.7% |
| 15 @ 1.7% | 707 | 54.5% | +0.48% | -9.3% | 0.06 | +0.5% |
The current configuration — 7 positions at 3.5% — dominates every alternative on every metric that matters. Highest Sharpe (1.14 versus 0.47 for the next best).
Lowest max drawdown (-7.4%).
Highest total return (+11.3%).
Highest win rate (60.6%).
More diversification didn't just fail to help.
It made everything worse. Monotonically.
Why More Positions Hurts
The fund manager's intuition was correct in theory.
In a universe with unlimited high-quality signals, more positions would diversify idiosyncratic risk and converge toward the expected edge.
But NSE SmallCap 250 mean-reversion doesn't produce unlimited high-quality signals.
It produces a limited number of genuine dislocations — and a long tail of marginal ones.
Here's what the data shows happening as you increase position count:
Win rate drops monotonically: 60.6% → 58.6% → 56.7% → 54.5%. With 7 slots, only the strongest shocks get through. When a shock fires but all 7 positions are full, it's blocked by the exposure limit. That blocked signal was likely a weaker one — less severe shock, less compelling failed follow-through. The 7-position limit acts as a natural quality filter. With 15 slots, every marginal signal enters the portfolio. The average quality of the trade book drops.
Average P&L per trade drops from +1.43% to +0.48%. This is the signal dilution in dollar terms. The strongest dislocations — the ones with the sharpest shocks and cleanest seller exhaustion — generate +1.4% average returns. Adding the 8th through 15th positions brings in trades that generate barely positive returns, dragging the portfolio average down.
Max drawdown actually increases with more positions. This is the counterintuitive finding. At 5 positions, max DD is -12.3% (too concentrated, a couple of bad trades compound). At 7, it drops to -7.4% (sweet spot). At 10-15, it climbs back to -8.6% to -10.4%. Why? Because more concurrent positions means more exposure to market-wide drawdowns. During a broad small-cap selloff, having 15 mean-reversion positions open — even at 1.7% each — creates more correlation exposure than having 7 at 3.5%. The diversification benefit of more names is overwhelmed by the correlation spike during stress.
The 5-position configuration confirms the lower bound. With only 5 slots at 5%, you take fewer trades (242 versus 340), miss genuine dislocations because the portfolio is full, and each bad trade hurts more. The win rate drops to 59.5% — not because signal quality is worse, but because 5 positions isn't enough to capture the full opportunity set.
The Natural Capacity of Dislocations in This Universe
The data suggests something specific about the NSE SmallCap 250: the universe produces roughly enough high-quality dislocation signals to sustain 5-7 concurrent positions at any given time during normal markets, and more during stress — but the additional signals during stress are lower quality because they're driven by correlation, not forced selling.
Seven positions appears to be the natural capacity of genuine dislocations in this universe.
Below 7, you're leaving real edge on the table.
Above 7, you're picking up noise and calling it diversification.
This has an important implication for scaling.
You cannot make this strategy bigger by adding positions.
The 25% gross exposure cap and 3.5% position size are not arbitrary constraints — they reflect the actual supply of high-quality mean-reversion opportunities in Indian small-caps.
Increasing AUM beyond the strategy's capacity doesn't diversify risk. It dilutes alpha.
What We're Keeping and Why
The current configuration stays: 7 positions at 3.5%, 25% gross exposure, 2 per sector, 30-day cooldown.
The sensitivity test validated what the arithmetic suggested but couldn't prove — that these parameters aren't just reasonable defaults, they're close to optimal for this universe.
We are, however, adding this test to our quarterly validation cycle.
If the SmallCap 250 universe changes meaningfully — new constituents, different sector composition, altered liquidity profile — the optimal position count could shift.
We want to detect that, not assume today's answer holds forever.
The Takeaway
The instinct to diversify a statistical strategy is correct in general.
But it assumes that additional trades carry the same expected edge as the first ones.
In a finite, liquidity-constrained universe like NSE SmallCap 250, they don't.
The 7-position limit isn't a risk concentration.
It's a quality filter.
The portfolio's exposure cap blocks weaker signals from entering — and that blocking is what protects the strategy's edge.
Not every constraint is a limitation.
Some constraints are the strategy.