Clustering i momentumstrategier: En undersøgelse af anvendelsen af Machine Learning som delkomponent i en momentum-investeringsstrategi

Abstrakt

In this thesis we take a starting point in the two momentum investment strategies provided in Jegadeesh og Titman (1993) and Moskowitz og Grinblatt (1999). We demonstrate a further extension of the latter, by implementing alternative, Machine Learning based classifications to replace and to be compared with the established classification provided in GICS and SIC. Furthermore we investigate the performance of these investment strategies from a returns and downside risk perspective, and compare these. We find that the industry momentum strategy is not able to beat the stock momentum strategy in the period 1963-1995, despite this being a central conclusion in Moskowitz og Grinblatt (1999). Furthermore we show that an industry momentum strategy based on the GICS classification can in fact outperform the stock momentum strategy in the period 1995-2020. At the same time we conclude that using a classification which is more contemporary, has a significant positive impact on the performance of the industry momentum strategy. Finally, we provide evidence that cluster based momentum strategies are able to generate significant returns and Sharpe Ratios, with downside risk measures relatively similar to the ones achieved by the strategies proposed in the literature in the period 1995-2020. However, we also find that our implementations of the cluster momentum strategies are not able to beat the performance of the GICS based industry momentum strategy, nor the stock momentum strategy in the same period of time.