An unsupervised learning framework for marketneutral portfolio

In this paper, we present a portfolio optimization strategy based on a novel approach in assets clustering on the financial background of the Arbitrage Pricing Theory, a well-known multi-factor model. In particular, our aim is to exploit data analysis tools, such as the techniques of features extraction and feature selection, to group assets that exhibit a significant exposition to the same risk factors. Then, we exploit the clustering to build a market-neutral portfolio and, more in general, an investment methodology that takes into account the peculiarities of the specific market. Finally, we apply our methodology in various case studies, discussing the results obtained and highlighting the strengths and the limits of the proposed strategy.