Construction of Stock Portfolios Based On K-means Clustering of Continuous Trend Features

Abstract

Optimal portfolio formation to reduce investment risk and increase returns is a concern for investors. There are various problems when investing with portfolio formation. First, it is difficult to select a pool of assets for portfolio formation. When the number of potential assets is relatively large, it will be difficult to select assets that fulfill portfolio formation and appropriate weights. Traditional portfolio theory such as "Markowitz portfolio theory" is only used for the calculation of appropriate weights but cannot be used to automatically select assets from a pool of assets. Secondly, traditional portfolio theory calculates its weights only based on the covariance relationship between different stocks and market data is not taken into account.  Thirdly, the sharpe ratio calculation is used to evaluate investment returns but does not consider risk aversion when stocks go down. Therefore, this thesis aims at portfolio formation based on sustainable trend characteristics. Utilization of k-means clustering is used to group assets, divide different types of asset pools, and calculation of sharpe ratio based on sustainable trend characteristics to avoid downside risk. In addition, it is also combined with the calculation of equal weight for each asset, inverse volatility, risk parity, and Markowitz portfolio theory.