From GARCH models to reinforcement gamers: a quant notebook

What it does

QuantResearch is a curated collection of Jupyter notebooks and Python scripts exploring the mechanics of quantitative trading and asset management. It covers the full pipeline: portfolio optimization, risk estimation, statistical arbitrage, time-series forecasting, and derivatives pricing. Each example is explicitly paired with a long-form blog post—when available—so the code sits next to the narrative. The repo also includes a backtesting environment and references a separate live-trading engine.

The interesting bit

Most quant repos pick a lane—classical stats or deep learning—but this one treats them as a continuum, placing Kalman filters, MCMC regression, and ARIMA-GARCH models in the same folder as RNN predictors and RL-based option pricers. It is less a framework and more a practitioner’s scrapbook of techniques, with the “Reinforcement Gamer to Reinforcement Trader” entry capturing the spirit of the whole thing: borrow ideas from other domains and see if they survive contact with market data.

Key highlights

• Covers classical methods (linear regression in Bayesian, MCMC, Kalman-filter, and TensorFlow flavors) alongside modern ML.
• Includes specific trading strategies: mean reversion, cointegration pairs trading, and hidden Markov chain regimes.
• Applies reinforcement learning to backtests, American option pricing, and portfolio management (the last marked work-in-progress).
• Provides utility scripts for downloading free historical market data and computing market/volume profiles.
• Links nearly every notebook to a detailed blog post for context.

Caveats

• Several notebooks (e.g., Fama-French, Vector AutoRegression, Volume Factor Evaluation) have no linked blog post, so context is thinner.
• The reinforcement-learning portfolio manager is explicitly marked “wip.”
• It is a collection of standalone recipes, not a unified API or backtest framework; the quanttrader backtest engine lives in a separate repository.

Verdict

Worth bookmarking if you are a developer or student who learns by implementing—especially if you want to see how classical quant methods relate to modern ML. Skip it if you need a production-grade trading framework or a single, opinionated strategy to deploy out of the box.