This project develops and evaluates daily options-volatility strategies on index and single-stock underlyings. Research spans the full pipeline: data engineering and quality checks, implied-volatility surface modelling, volatility forecasting, and strategy backtesting. Backtests use realistic execution assumptions (bid/ask, slippage, commissions, position sizing, and risk limits) and are documented with reproducible notebooks and published reports.
Notebook reports (GitHub Pages): https://anthonymakarewicz.github.io/volatility-trading/
- Clone the repository:
git clone https://github.com/anthonymakarewicz/volatility-trading.git
cd volatility_trading- Create a virtual environment and install runtime dependencies (Python 3.12+):
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt- Optional (development setup: runtime + dev tools + tests):
pip install -r requirements-dev.txtrequirements-dev.txt already includes all runtime dependencies from
requirements.txt.
- Set credentials (ORATS):
cp .env.example .envThen set ORATS_API_KEY, ORATS_FTP_USER, and ORATS_FTP_PASS in .env.
Pipeline steps:
- API download
- API extract
- FTP download
- FTP extract
- Build options chain
- Build daily features
- QC options chain
- QC daily features
Use --dry-run to validate config, paths, and credentials before running writes/network.
orats-api-download --config config/orats_api_download.yml --dry-runFor the full command sequence, see Data pipeline.
Run unit tests (default):
pytest -qRun integration tests:
pytest -q -m integrationSee Testing guide for layout and conventions.
GitHub Actions runs:
- Ruff lint + format checks
- Pyright type checks
- Unit tests by default
- Integration tests on pushes to
main(and manual runs)
See CI workflow.
Common commands are available via Makefile:
make lint
make format
make check
make typecheck
make test
make test-unit
make test-integration
make sync-nb
make sync-nb-all
make ciFor full setup and tooling details, see the Documentation index. Notebook HTML reports are built in GitHub Actions and published to GitHub Pages.
See Documentation index for the full docs map. Most-used pages:
We build a 21-day realized variance forecasting model on ES futures (2010–2025) and compare:
- Naive RV benchmark – carry current 21D RV forward
- HAR-RV – classic daily / weekly / monthly RV lags
- HAR-RV-VIX – HAR-RV + VIX as a forward-looking volatility proxy
- Random Forest (RF) – non-linear benchmark on the same feature set
Before model selection, we run a feature-importance & stability analysis (SFI, Lasso, RF, permutation importance) to keep only parsimonious, economically sensible predictors for the linear and RF models.
All metrics are computed on log 21D RV, using an expanding walk-forward with 3-year rolling re-fit and a 21-day purge.
| model | R² | MSE | QLIKE | Var_res | R²_oos |
|---|---|---|---|---|---|
| Naive_RV | 0.0943 | 0.5078 | 0.2791 | 0.5080 | 0.0000 |
| HAR-RV | 0.2920 | 0.3970 | 0.2086 | 0.3920 | 0.2182 |
| HAR-RV-VIX | 0.3676 | 0.3546 | 0.1788 | 0.3549 | 0.3017 |
-
HAR-RV-VIX is the final candidate model: it clearly beats both Naive RV and HAR-RV in OOS
$R²$ , MSE and QLIKE, and delivers a ~30%$R²_{oos}$ vs the naive benchmark. - The Random Forest does not improve on HAR-RV-VIX in the validation period and is therefore not carried forward to the final walk-forward evaluation.
- All modelling choices (features, models, hyper-parameters) were fixed on 2010–2020; the 2021–2025 walk-forward backtest is run once to avoid backtest-tuning bias.
Trade the 30-day to expiry, 25 Delta SPX put–call skew via a delta-hedged risk reversal:
-
Synthetic Skew – Interpolate across expiries to build a continuous “30 DTE / 25 Δ” skew series.
-
Entry / Exit – Short RR when skew z-score ≥ 1.5 (too steep) – Long RR when skew z-score ≤ –1.5 (too flat) – Exit when |z-score| ≤ 0.5
- VIX Filter: Block entries if VIX > 30
- IV Percentile: Trade only when ATM IV is within its 20–80 historical percentile
- Skew Percentile: Trade only when skew is below its 30th (for longs) or above its 70th (for shorts) percentile
We run a walk-forward backtest on daily SPX options (2016 – 2023), starting with $100 000 of capital.
-
Entry Rule 50-day z-score mean-reversion on normalized skew (|z| ≥ 1.5 triggers a risk reversal; exit when |z| falls below 0.5)
-
Signal Filters
- VIX Filter: Skip trades when VIX > 30
- Skew Percentile: Only go long when normalized skew falls below its 30th percentile, and short when it rises above its 70th percentile
-
Execution Costs
- Bid/Ask & Slippage: Fill sells at the bid, buys at the ask + $0.01 slippage per leg
- Commissions: $1 per option leg
-
Risk Controls
- Delta Hedging: E-mini S&P 500 futures (ES=F) used to neutralize net Δ (lot size = 50)
- Position Sizing: Dynamically scale trade size by signal strength (base 1% of equity at entry-threshold, +0.5% per additional 0.5σ) and cap it at 2%
- Risk Floor: Enforce a minimum $750 worst-case risk per contract to prevent oversized position sizing when Greek-based risk is very low
- Stop-Loss & Take-Profit: SL at 100% of notional, TP at 70% of notional
- Holding Period Cap: 3 business days (skip negative theta trades on Fridays if 2-day θ decay > 200)
| Metric | Value |
|---|---|
| Sharpe Ratio | 0.61 |
| CAGR | 10.73% |
| Average Drawdown | –1.62% |
| Max Drawdown | –14.61% |
| Max Drawdown Duration | 287 days |
| Historical VaR (99%) | –0.87% |
| Historical CVaR (99%) | –2.64% |
| Total P&L | $68,754.99 |
| Profit Factor | 2.66 |
| Trade Frequency | 15.2 trades/yr |
| Total Trades | 78 |
| Win Rate | 62.82% |
| Avg Win P&L | $2,271.04 |
| Avg Loss P&L | –$1,444.52 |