Skip to content

Repository files navigation

Multi-Dimensional Return Forecasting & Portfolio Management

Overview

This project builds a machine learning-based system to predict short-term stock returns and construct a portfolio using a 6-stock Indian equity universe.

Data

  • Stocks: RELIANCE, HDFCBANK, INFY, M&M, BHARTIARTL, HUL
  • Period: Jan 2020 – Dec 2025
  • Sources: Yahoo Finance, RBI, MOSPI, MoneyControl
  • Final dataset: ~7,400 rows

Features

  • 58 engineered features across technical, momentum, macro, volume, and fundamentals
  • Reduced to top 35 using Mutual Information
  • Macro variables (Crude, CPI, VIX, Yield) were most important

Models

  • LightGBM Regressor
  • XGBoost Regressor
  • LightGBM Classifier
  • Final prediction: average of LightGBM and XGBoost

Validation

  • Walk-forward (expanding window) cross-validation
  • Mean accuracy ~49% (expected for daily returns)

Portfolio Strategy

  • Allocate weights proportional to predicted positive returns
  • Max weight per stock: 40%
  • Negative signals assigned zero weight

Results

Test Period (Sep 2024 – Sep 2025):

  • Return: 4.68%
  • Sharpe: -0.008
  • Sortino: 0.644
  • Max Drawdown: -12.31%

Forward Test (Oct 2025 – Dec 2025):

  • Return: 6.62%
  • Sharpe: 2.47
  • Max Drawdown: -2.35%

Tech Stack

Python, LightGBM, XGBoost, scikit-learn, pandas, numpy, matplotlib

Key Takeaways

  • Macro features dominate short-term return prediction
  • Daily prediction accuracy remains near 50%
  • Signal-based allocation improves risk-adjusted performance

About

Multi-dimensional financial ML framework for stock return forecasting and portfolio management using market, macroeconomic, and fundamental features with walk-forward validation and portfolio-level evaluation.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages