Portfolio Optimisation Using Machine-Learning Return Forecasts and Risk Parity

A robo-advisor that turns an 8-question risk-profile questionnaire into a risk-tolerance score (0-100) and one of five profiles with a target volatility of 5, 8, 11, 14 or 17 % a year, and invests in a risk-parity (equal risk contribution) portfolio of the 49 US industry portfolios of the Kenneth French Data Library plus T-bills; the equity share is the target volatility divided by a volatility estimate. An own two-state hidden Markov model of the market since 1926, re-fitted every January on past data only, detects calm and turbulent regimes, and the regime-aware allocation is compared out of sample 1990-2026 with a static share and classical 12-month volatility targeting. A second part tests machine-learning return forecasts (ridge, elastic net, gradient boosting, neural network, ensemble) against the historical mean in mean-variance portfolios, next to minimum variance, risk parity, the market and 1/N. A web application advises clients and manages proposals; an Excel workbook, a base R cross-check and Power BI files are included.

Category: Finance

Master of Business Administration Project MBA02 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.