Introductory Context
"Industry-normalized statistical arbitrage partitioning the Nifty 500 into 12 sector clusters to trade relative mean reversion without unwanted macro sector bias."
Industry-Normalized Relative Value: The Multi-Cluster Advantage
Simple whole-market statistical arbitrage strategies (such as ranking all 500 stocks in the Nifty 500 by raw 5-day RSI or Z-score) suffer from a fatal flaw: they inadvertently take massive directional sector bets. When the entire IT sector sells off due to US tech weakness, a whole-market mean reversion system will load 100% of its capital into IT stocks, mistaking a structural macroeconomic sector downgrade for a temporary statistical bounce.
The Multi-Cluster Industry Mean Reversion Strategy solves this by partitioning the equity universe into distinct, highly correlated industry clusters (e.g. Private Banks, PSU Banks, IT Services, Pharmaceuticals, Auto OEMs, Metal Producers, Real Estate). Statistical residual Z-scores are computed strictly within each cluster, and long/short or market-neutral portfolios are constructed cluster-by-cluster.
• The Multi-Cluster Mean Reversion Rule in One Sentence
Segment the Nifty 500 into 12 distinct industry clusters; within each cluster, calculate the 10-day rolling price residual relative to the cluster's equal-weight mean; buy the oversold tail (Z < -1.8) and sell the overbought tail (Z > +1.8), rebalancing weekly with strict sector-neutral budget constraints.
Mathematical Formulation of Cluster-Relative Residuals
Let Ck be an industry cluster containing Nk liquid stocks. The cluster benchmark return on day t is the equal-weight mean:
For each stock i ∈ Ck, the relative cumulative spread over lookback window L (default 10 trading days) is:
Standardize into a rolling 60-day Cluster Z-Score (Zi):
- Oversold Entry (Zi ≤ -1.8): Stock i has significantly underperformed its direct peer group. The desk initiates a long position.
- Overbought Entry (Zi ≥ +1.8): Stock i has decoupled upward from its cluster. The desk takes a short position (via stock futures on NSE) or underweights the stock in a long-only basket.
- Mean Reversion Exit (|Zi| ≤ 0.2): Position is closed as the stock's performance converges back to cluster parity.
Worked Indian Example: Banking Cluster Decoupling on NSE
Consider the Nifty Private Bank cluster during a 10-day earnings divergence on the National Stock Exchange:
| Stock Name | 10-Day Return | Cluster Mean | Relative Spread | Cluster Z-Score | Quant Action |
|---|---|---|---|---|---|
| ICICI Bank | +3.80% | +1.20% | +2.60% | +1.15 | NEUTRAL |
| Axis Bank | +5.40% | +1.20% | +4.20% | +2.10 (Overbought) | SHORT / UNDERWEIGHT |
| Kotak Mahindra Bank | -3.20% | +1.20% | -4.40% | -2.25 (Oversold) | LONG / OVERWEIGHT |
| HDFC Bank | +1.10% | +1.20% | -0.10% | -0.05 | NEUTRAL / BENCHMARK |
| IndusInd Bank | -1.10% | +1.20% | -2.30% | -1.05 | NEUTRAL |