Machine Learning for Trading — Aamir Jahan
← All work
Coursework — Georgia Tech M.S. CS

Machine Learning for Trading

A market simulator and a set of trading strategies built from technical indicators, supervised learning, and reinforcement learning, then evaluated against historical equities data.

ML4T Reinforcement Learning Python
Overview

Coursework covering the intersection of machine learning and quantitative finance — building the infrastructure to compute technical indicators, backtest trading strategies against a market simulator, and compare rule-based, supervised, and reinforcement-learning approaches on the same data.

Approach

Implemented a market simulator that models order execution, commissions, and market impact. Built a manual rule-based strategy from technical indicators (Bollinger Bands, RSI, momentum), then trained a random-forest learner and a Q-learning agent on the same indicator set, using in-sample and out-of-sample splits to guard against overfitting.

Result

The reinforcement-learning strategy outperformed the manual and supervised baselines on out-of-sample data, at the cost of longer training time and more sensitivity to hyperparameters.