Project Overview

Project Overview

This project explores trading strategies using Deep Q-Networks (DQN) applied to historical ETF (1655.T) price data.
It aims to simulate buying and selling decisions to maximize cumulative returns.

Objectives

• Simulate trading behavior based on historical ETF data
• Explore the impact of different neural network architectures and hyperparameters
• Evaluate strategies based on returns, win rates, and statistical significance

Key Features

• Random Search for hyperparameter exploration
• Separate Agents for buying and selling decisions
• Performance Evaluation using metrics such as total reward, win rate, average holding days, and p-value
• Automatic Saving of model weights based on best validation performance

Technologies Used

• Python
• TensorFlow
• scikit-learn
• yfinance
• pandas
• matplotlib
• numpy

Dataset

• Yahoo Finance (1655.T) ETF historical data
• Date range: 2017-09-29 to 2023-03-31