Statistical Arbitrage

Key Takeaways

• Overfitting led to inflated Sharpe ratios observed during backtesting.

• The time-based stop-loss approach proved to be extremely risky, as tail risk can lead to tremendous losses.

• The portfolio of pairs did not adequately account for correlations between pairs.

Introduction

Objective: 

• I began trading an account with approximately USD 50,000 using statistical arbitrage, implemented through pair trading.

Approach:

• Developed an initial prototype system to identify highly correlated stocks from a predefined basket.

• Filtered correlated pairs by testing for Engle-Granger cointegration.

• Conducted preliminary backtests using open-source platforms.

• Automated live execution through PairTradingLab's PTL Trader.

Period:

• July 16, 2023 - October 16, 2023 (55 trading days).

Markets Traded:

• Equities.

Trading Methodology

Core Strategy:

• Pair trading with a focus on mean reversion, identifying statistically significant divergences between historically related assets.

• Identified historically related assets whose price spread diverged significantly, creating opportunities for convergence.

Risk Management:

• Constructed a portfolio of pairs diversified across sectors and strategies.

• Applied a time-based stop-loss (maximum holding period of 20 days).

Execution Rules:

• For a given pair (e.g. ABC/XYZ):

  • Short the outperforming stock when the spread exceeds a positive z-score threshold.

  • Long the underperforming stock when the spread falls below a negative z-score threshold.

  • Exit positions once the spread mean-reverts toward zero or when the stop-loss condition is triggered.

Trade Report

• • Total Returns  (before fees) (USD): 5890.65 

  • Sharpe Ratio: 1.68

  • Sortino Ratio: 5.60

  • Max Drawdown (%): -5.42

  • Hit Rate (%): 65.68

  • Profit Factor: 1.66

  • Average Trade Return and Expectancy (USD): 20.22