Forest Behind the Trees (joint with Markus Pelger and J ason Zhu)
Abstract:
Sorting-based strategy of building portfolios has been a default empiri cal approach in asset pricing for creating both test assets and factor-mimi cking returns. One of the natural limitations of this technique, however, i s its inability to adequately reflect the information contained in more tha n 2 characteristics and their interaction. Yet recent advances in empirical asset pricing have repeatedly highlighted the importance of the latter, e. g. Freyberger et al (2017), Kozak et al (2018). We propose to analyze the e ffect of a large number of characteristics on expected stock returns with t he machine learning technique known as random forest. As an ensemble learni ng method for classification, the new approach is particularly well-suited for building composite cross-sections of portfolios that reflect the rich c onditional information contained in a large number of characteristics simul taneously, and can be viewed as a natural generalization of the conventiona l sorting-based strategies. We build decision trees for various sets of sto ck-specific characteristics, and demonstrate that the new approach is able to create cross-sections that a) reflect the information in a joint conditi onal distribution of characteristics, b) are challenging to price based on the conventional models, even when pitted against the tradable factors base d on the underlying characteristics, and c) imply investment strategies tha t achieve yearly out-of-sample Sharpe ratios above 2.
DTSTAMP:20260405T192557Z DTSTART:20181011T163000Z DTEND:20181011T180000Z SEQUENCE:0 TRANSP:OPAQUE END:VEVENT END:VCALENDAR