Volume 2 - Issue 4, July - August 2026

This study proposes and empirically validates an advanced algorithmic trading framework for statistical arbitrage using a pairs-trading strategy enhanced with unsupervised machine learning. The framework is tested across four market cohorts: NIFTY 500 Long (2007–2025), NIFTY 500 Broad (2015–2025), S&P 500 Long (2007– 2025), and S&P 500 Broad (2015–2025). Principal Component Analysis (PCA) is applied to neutralize systematic risk factors and isolate idiosyncratic residual returns, after which the OPTICS (Ordering Points To Identify the Clustering Structure) algorithm performs density-based clustering to group assets with similar residual behaviour without prespecifying the number of clusters. The presence of a tradeable long-run relationship is confirmed through the EngleGranger two-step cointegration test. The backtesting engine incorporates institutional-level constraints, including a 15-basis-point round-trip transaction cost and a hard –5% stop-loss, and benchmarks performance against sovereign hurdle rates of 7.2% (India) and 4.2% (United States). The results show substantial alpha generation in the NIFTY 500 Long cohort, with a net ROI of 485.3% under the traditional benchmark, while the machinelearning-enhanced framework delivers a superior risk-adjusted outcome in the NIFTY 500 Broad cohort (Sharpe ratio of 1.090 versus 0.869) alongside a materially reduced maximum drawdown. In the S&P 500, the ML-enhanced framework improves the Sharpe ratio from 0.440 to 0.583 (Long) and 0.424 to 0.479 (Broad), though absolute magnitudes remain smaller, consistent with a more factor-cohesive and efficiently arbitraged developed market. Statistical significance tests, including a one-sided t-test, the Jobson-Korkie test, and bootstrap resampling, confirm that these performance gains are not attributable to random chance. The findings indicate that while market efficiency compresses arbitrage windows, the PCA-OPTICS framework is effective at isolating economically meaningful, cost-adjusted idiosyncratic opportunities in both emerging and developed equity markets.

Sankar Ganesh M, Dr.S.Johnsi,"Comparative Performance of ML-Enhanced Statistical Arbitrage Across the S&P 500 and NIFTY 500" International Journal of Advanced Multidisciplinary Research and Educational Development, V2(4): Page(278-289) July - August 2026. ISSN: 3107-6513. www.ijamred.com. Published by Scientific and Academic Research Publishing.