Flirting with Models Replacing linear factors with a non-linear, characteristic approach in quant equity
Jan 9, 2023
AI Snips
Chapters
Transcript
Episode notes
Anomalies Are Characteristics Not Risk Factors
- Anomalies like value and momentum are characteristics, not priced risk factors, because many anomaly-based hedges with negative expected returns don't exist.
- Daniel and Titman plus Viswanathan's thesis show characteristics, not covariances, drive returns across many anomalies globally.
The Factor Zoo Actually Matters Globally
- There are 100+ meaningful anomalies; many deliver significant Fama–French three-factor alphas globally, so they are unlikely to be data mining.
- Viswanathan found 44 of 86 anomalies produced significant FF3 alphas from 1995–2018 across markets.
Build Characteristic Driven Models Not New Factors
- Treat anomalies as predictive features about firm value and use rich feature sets rather than building more linear factors.
- Use price, valuation, accounting, news, and alternative signals so models can learn interactions and timing of price incorporation.