Finance Forecasting and Portfolio Optimization
A quantitative finance system combining time series forecasting with portfolio optimization to support data-driven investment decisions and risk management.
Overview
A quantitative finance system combining time series forecasting with portfolio optimization to support data-driven investment decisions and risk management.
Problem
Individual investors and small funds lack access to sophisticated quantitative tools for portfolio construction and return forecasting. The goal is to build an accessible system that combines ML-based return forecasting with modern portfolio theory for optimal asset allocation.
Dataset
Historical price data for a diversified set of assets including equities, ETFs, and commodities. Features engineered from technical indicators (RSI, MACD, Bollinger Bands), fundamental ratios, and macroeconomic variables.
Architecture
LSTM-based return forecasting model for individual assets. Mean-Variance Optimization (MVO) and Black-Litterman model for portfolio construction, incorporating forecasted returns as views. Risk parity allocation as an alternative strategy.
Training
LSTM trained on rolling 60-day windows with walk-forward validation to prevent look-ahead bias. Portfolio optimization run monthly with rebalancing. Sharpe ratio and maximum drawdown used as primary evaluation metrics.
Results
LSTM forecasts improved directional accuracy to 58% vs 51% for the random walk baseline. Optimized portfolio achieved Sharpe ratio of 1.34 vs 0.89 for equal-weight benchmark over the test period. Maximum drawdown reduced by 22%.