In this thesis we study the predictability and potential for a dynamic weighting strategy to reduce crash risk and improve risk-adjusted returns for the two well-known factor portfolios (1) momentum and (2) a 50/50 combination of the value and momentum factor.
Over the past 90 years, the momentum factor has offered investors a high and statistically significant risk-adjusted return. However, the momentum factor suffers from severe crash risk which makes this portfolio unattractive for investors with some degree of risk aversion. A 50/50 combination portfolio of the momentum and value factor reduces the drawdowns from crashes due to negative correlation, but this combination strategy still has a negative skewness and a high kurtosis. We find that the crashes of the two factor portfolios are partly forecastable. They occur when financial markets are volatile and after negative market returns have been recorded. Furthermore, we find that the volatility of the portfolios is variable over time and can be predicted. For example, high future volatility is forecasted by high volatility in the recent past and by a high GJR-GARCH volatility estimate.
We show that the risk can be managed by using a dynamic weighting strategy based on the forecasts of return and volatility for both the momentum- and the 50/50 combination portfolio. The dynamic strategy substantially reduces the crash risk. The Sharpe ratio more than doubles for the momentum factor portfolio and increases by 70% for the 50/50 combination portfolio. When comparing the two dynamically risk-managed portfolios, the 50/50 combination portfolio reports a higher risk-adjusted return compared to the momentum portfolio. We also show that investors facing shorting restrictions can benefit from a slightly altered version of the dynamic strategy by investing in risk-managed long-only momentum and 50/50 combination portfolios. For both dynamic strategies, the Sharpe ratio more than doubles, and the skewness and kurtosis measures indicate a more normal distribution of returns relative to a constant 1x long-only portfolio.
Our results are robust in sub-sample periods and out-of-sample. In a more recent sample covering the past 40 years we show that the forecasts are easier modelled due to fewer extreme observations, and that the corresponding dynamic strategy risk-adjusted returns are strong. In an out-of-sample expanding window recursive regression, we show that the weights and returns obtained in an out-of-sample context are highly similar to those from an in-sample context. In terms of Sharpe ratio, we show that the shorting-restricted dynamic strategies outperform the market in an out-of-sample context and in all our sub-samples. However, the long/short dynamic factor portfolios are not always outperforming the market.
|Educations||MSc in Applied Economics and Finance, (Graduate Programme) Final Thesis|
|Number of pages||135|
|Supervisors||Lisbeth la Cour|