liethon / statistical-arbitrage

High-frequency statistical arbitrage

Statistical-Arbitrage

In this project we provide a backtesting pipeline for intraday statistical arbitrage. Both traditional spread models (i.e. pairs trading with cointegration tests, time series analysis) and continuous time trading models (i.e. Ornstein-Uhlenbeck process) are used to model the spread portfolios.

Scripts

• data: intraday data files, including stocks, options, and dual listing stocks
• utils: arbitrage tool functions including cointegration tests and regression analysis
• models: simulations and parameter estimations for stochastic models and option greeks
  • BM.py: brownian motion related functions
  • Vasicek.py: OU-process related functions
  • BS_model.py: Black-Scholes model and option greeks
• statistical_arbitrage: notebook for realizing pair trading based on limit orderbook stock data
• res: results for positions, thresholds, and PnLs

Backtesting

Spread Portfolios

Threshold Analysis

Position Analysis

PnL Visualization

Note

• Higher the transaction costs, larger the optimal entry points for arbitrage, lower the trading frequency.
• Sensitivity analysis needs to be conducted with respect to the level of transaction costs (price impact, bid-ask spread, and commission fees).