Title: Stock market clustering methods
Authors: Carlos Fidel Selva Ochoa - Universidad Nacional Autonoma de Mexico (Mexico) [presenting]
Abstract: Financial markets have benefited greatly from classification systems like SIC and NAICS as a basis for dissecting the asset universe into comparable groups for analysis, index tracking and forecasting purposes. In the case of equity, the industry classifications in place are based on each company's portrayed business activity, earnings analysis and market perception. Those systems require human intervention, particularly from fundamental analysts, which derives in a resource-consuming process if one considers the data volume of the whole US Stock market. To avoid this manual classification, several distances, similarity and model-based clustering methods are explored for the CRSP database from 2000 to 2020 for the daily returns of the top 3000 stocks by liquidity. The goodness of fit of resulting groups is assessed internally with dispersion and likelihood measures, and externally by backtesting a dollar-neutral mean reversion strategy. For this last measure, the Sharp ratio was greater for some of the proposed methods.