Nonlinear Support Vector Machines and Stock Picking
Log in to collectAcademic paper
Strategy in a nutshell
The strategy invests in industrial sector stocks using an SVM classifier, applying liquidity and volatility filters to exclude low-volume, low-price, or illiquid stocks. Volatility-adjusted returns are computed via an exponential moving average, and stocks are sorted into quintiles. Training uses top and bottom quintiles with technical and fundamental features. Each portfolio formation retrains the SVM to adapt to market changes. Ten long (highest SVM) and ten short (lowest SVM) positions are held for 91 days, with 20 new stocks added every 28 days. Portfolios are equally weighted, and filtered stocks remain tradable.
Economic rationale
The strategy leverages technical and fundamental factors (e.g., momentum, accruals) correlated with future performance. The SVM classifier identifies strong signals while ignoring mid-ranking, weakly predictive stocks. Using a reduced dataset improves efficiency without reducing effectiveness. Long holding periods mitigate trading costs, making the strategy economically robust. Combining fundamental insights with optimized data processing drives consistent profitability.