Tensor Factors to Monitor the Co-Movement of Equity Prices

Abstract

We identify a set of features that are related to extremes of price changes of individual equities. Our hypothesis is that these extreme features may be used to isolate co-movements of prices for groups of equities, reflecting systematic risk. The equities are classified within industry sectors and we create a three mode tensor to represent the dataset; the dimensions of the three mode tensor correspond to the equity, the industry sector and the day on which the feature occurred. We use a method for non-negative tensor factorization (NOTF) to identify factors or communities that are composed of multiple equities, and / or industry sectors. Our preliminary results indicate that our NOTF approach has the potential to identify such communities of price related features that may experience co-movement across industry sectors and temporal intervals.