Skip to content

README.md

Latest commit

 

History

History
72 lines (58 loc) · 3.02 KB

README.md

File metadata and controls

72 lines (58 loc) · 3.02 KB
Raw

Equity Walk-Forward Backtest (US Equities)

A research-style, walk-forward backtesting notebook for simple equity signals using daily US equity/ETF data. The project focuses on leakage-safe feature construction, realistic next-day execution, transaction costs, and out-of-sample evaluation.

What this project does

  • Downloads daily OHLCV data from Stooq for a small universe of liquid US tickers
  • Builds rolling features (returns, rolling mean/std, z-scores)
  • Generates trading signals (mean reversion) and simulates next-day execution
  • Runs a rolling walk-forward evaluation (train window → test window)
  • Applies per-side transaction costs based on turnover
  • Reports out-of-sample performance metrics and plots equity/drawdown

Data

  • Source: Stooq daily CSV
  • Universe (20 tickers): SPY, AAPL, MSFT, AMZN, NVDA, GOOGL, META, JPM, XOM, JNJ, PG, KO, PEP, WMT, COST, HD, V, MA, UNH, LLY
  • Date range: 1970-01-02 to 2026-01-15

Strategy (Mean Reversion)

Features

  • Daily returns: r_t = P_t / P_{t-1} - 1
  • Rolling z-score over a window w:
    • z_t = (r_t - mean_w(r)) / std_w(r)

Entry-only variant

  • Enter long when z_t < entry_z
  • Position is applied next day to avoid lookahead: pos_{t+1} = signal_t

Entry + Exit variant

A stateful rule to reduce tail risk / overholding:

  • Enter long when z_t < entry_z
  • Exit when z_t > exit_z (fixed exit_z = -1 in current comparison)
  • Next-day execution: positions are shifted by 1 day

Backtest assumptions

  • Execution: next trading day (signals computed on day t, positions applied on day t+1)
  • Portfolio: equal-weight across tickers each day
  • Costs: 5 bps per side (0.0005) applied when a sleeve changes position (turnover-based)

Walk-forward validation

Rolling walk-forward backtest:

  • Train window: 504 trading days (~2 years)
  • Test window: 63 trading days (~3 months)
  • Step size: 63 trading days
  • Parameter selection on each train window using Sharpe (net of costs)
  • Test features computed using train + test history (realistic availability), but performance is measured on test dates only

Results (Out-of-sample, with costs)

Comparison from the current run:

Variant Sharpe (OOS) Max Drawdown (OOS) Avg Daily Turnover (OOS)
Entry-only 0.223 -0.255 0.044
Entry + Exit (exit_z = -1) 0.353 -0.122 0.039

Interpretation:

  • Adding an exit rule improved risk-adjusted performance and roughly halved max drawdown while slightly reducing turnover.

How to run

  • Open notebooks/01_stooq_walkforward.ipynb and run top-to-bottom:
    1. download prices
    2. compute features/signals
    3. run walk-forward backtests
    4. produce plots and metrics

Dependencies are listed in requirements.txt.

Next improvements

  • Add benchmark comparisons (SPY buy & hold, equal-weight buy & hold)
  • Add parameter sweeps for exit thresholds and windows
  • Add volatility targeting / position sizing
  • Move notebook logic into src/ and create a single scripts/run_backtest.py runner
  • Add unit tests for turnover/cost logic and leakage checks