Equity Strategy Backtest

Project Overview

This project focuses on designing and backtesting a long-short equity strategy using supervised learning models on factor-based features. The system selects assets based on predicted future returns, constructs dynamic portfolios using quantile-based thresholds, and incorporates realistic cost assumptions and signal holding periods.

Pipeline Highlights

• Alpha Modeling: Trained on predictive factors using random forest regression to estimate forward returns
• Signal Construction: Long/short signals derived from prediction quantiles with configurable thresholds
• Portfolio Management: Positions held over a fixed horizon with trade-aware transaction cost modeling
• Backtest Engine: Daily returns aggregated across tickers to simulate strategy equity curves

Performance Metrics

The table below summarizes out-of-sample performance across different quantile thresholds:

Upper Q	Lower Q	Sharpe	Sortino	Annual Return	Volatility	Calmar	Max DD	Win Rate	Avg Positions
0.90	0.10	2.14	3.86	20.5%	8.7%	5.69	-3.6%	48.9%	98.9
0.95	0.05	2.00	4.19	11.8%	5.6%	6.81	-1.7%	51.0%	49.5
0.80	0.20	3.60	12.36	35.4%	8.4%	26.25	-1.3%	57.4%	197.5

Key Features

• Robust Risk Control: Strategy delivers strong risk-adjusted returns with low drawdowns
• Flexible Quantile Thresholds: Multiple entry/exit schemes evaluated to assess signal strength
• Daily Turnover Monitoring: Turnover and trade frequency tracked to balance alpha vs cost
• Benchmark Comparison: Strategy tested against equal-weighted buy-and-hold baseline

Conclusion

The project demonstrates the viability of systematic return forecasting and long-short signal execution under real-world trading assumptions. Through careful design of alpha signals, cost-aware execution logic, and disciplined evaluation, the system achieves consistently strong performance.

Code

Full pipeline and backtest code are available by request at schilamkur@gmail.com.