Risk-Managed Momentum Strategy Using Support Vector Machines

Sammanfattning: Investment decisions are difficult to make, given the uncertainty about the future. For the purpose of reducing that uncertainty, I investigate, for one, how the consumer price index and the return on the 3-month US Treasury bill can be used by support vector machines to make monthly directional trend predictions of a value-weighted portfolio of stocks traded at AMEX, NYSE and NASDAQ. For another, I examine to which extent the risk-managed momentum strategy proposed by Barroso and Santa-Clara (2015) can be improved when those predictions are incorporated. I find that the monthly stock market prediction accuracy lies at 61.1 percent over the time horizon from 1965 to 2017. A prediction-based flexible volatility target in the risk-managed momentum strategy achieves an improvement in the higher order moments as well as in the Sharpe ratio which in turn reduces the crash risk of momentum.