Machine Learning vs. Economic Restrictions: Evidence from Stock Return Predictability

Abstract

This paper shows that investments based on deep learning signals extract profitability from difficult-to-arbitrage stocks and during high limits-to-arbitrage market states. In particular, excluding microcaps, distressed stocks, or episodes of high market volatility considerably attenuates profitability. Machine learning-based performance further deteriorates in the presence of reasonable trading costs because of high turnover and extreme positions in the tangency portfolio implied by the pricing kernel. Despite their opaque nature, machine learning methods successfully identify mispriced stocks consistent with most anomalies. Beyond economic restrictions, deep learning signals are profitable in long positions and recent years and command low downside risk. This paper was accepted by Kay Giesecke, finance.

Timing the factor zoo via deep learning: Evidence from China

“…We assume z = 0.5%. Following Gu et al (2020) and Avramov et al (2022), the portfolio choice's monthly turnover at time t can be stated as$$$$ {T}_t=\sum \limits_i\left|{w}_{i,t}‐\frac{w_{it‐1}{r}_{i,t}}{w_{t‐1}{r}_t}\right|, $$$$where $$$ {w}_t $$$ ($$$ {w}_{i,t} $$$) refers to the weight of the wealth invested in assets (element i in $$$ {w}_t $$$) with growth return $$$ {r}_{t+1} $$$ ($$$ {r}_{i,t+1} $$$), $$$ {w}_t=\left(1‐{\alpha}_t,{\alpha}_t\right) $$$ and $$$ {r}_{t+1}=\left({R}_{f,t+1},{R}_{t+1}\right) $$$ corresponding to Equation (5).…”

“…The finding that the quality‐ and profitability‐related factors contribute most to the principal components also echoes their importance in China (Jiang et al, 2018). Meanwhile, with the unique features of the Chinese stock market, for instance, changeable market states and high limit‐to‐arbitrage, we discover the economic channels of factor timing through the mispricing‐based theory, in the spirit of Avramov et al (2022), which contributes to the understanding of cyclical variation in investor behaviours.…”

“…The transaction cost of portfolio choice at time t is c t = z * T t , where z is the transaction cost in the cross section and T t is the turnover at time t. We assume z = 0.5%. Following Gu et al (2020) and Avramov et al (2022), the portfolio choice's monthly turnover at time t can be stated as where w t (w i,t ) refers to the weight of the wealth invested in assets (element i in w t ) with growth return r t+1 (r i,t+1 ), w t = 1 − t , t and r t+1 = R f ,t+1 , R t+1 corresponding to Equation (5). Abbreviations: NN3, three-hidden-layers feed-forward neural network; OLS, ordinary least squares.…”

“…Inspired by Avramov et al (2022), we investigate the profitability sources of deep learning signals based on mispricing.…”

“…Overall, our results suggest that the deep learning method is suitable for China, as it performs well among different subsamples (Leippold et al, 2022). Moreover, the deep learning‐based factor timing signal in China is influenced by mispricing (Avramov et al, 2022).…”