The Chinese University of Hong Kong issued a study that calls for a second look at the effectiveness of generating returns from investment strategies based on machine learning.
The study found that when applying several well-established deep learning methods to the broader markets, superior value-weighted, risk-adjusted returns could be generated – 0.75-1.87 percent.
But in the event of basic exclusions that weigh down benchmark performance – such as microcaps or distressed firms – performance weakens.
«We find that the return predictability of deep learning methods weakens considerably in the presence of standard economic restrictions in empirical finance,» said Si Cheng, assistant professor at the Chinese University of Hong Kong’s (CUHK) Business School's Department of Finance and one of the study's authors.
Disappearing Returns
When microcaps were excluded, adjusted returns fell 62 percent, attributable to small capitalizations and a higher likelihood of low liquidity. Performance also declined when exclusions involved non-rated firms (68 percent) and distressed firms (80 percent).
Exclusions aside, the study also highlighted the need to be able to stomach high transaction costs at levels that may not be applicable to most retail investors.
«Machine learning methods require high turnover and taking extreme stock positions,» Cheng explained. «An average investor would struggle to achieve meaningful alpha after taking transaction costs into account.»
Unable to Tackle «Usual Challenges»
According to Cheng, the findings demonstrate that the limitations of investing based on machine learning strategies are not dissimilar to that of traditional investing.
«The collective evidence shows that most machine learning techniques face the usual challenge of cross-sectional return predictability, and the anomalous return patterns are concentrated in difficult-to-arbitrage stocks and during episodes of high limits to arbitrage,» Cheng explains.
«Therefore, even though machine learning offers unprecedented opportunities to shape our understanding of asset pricing formulations, it is important to consider the common economic restrictions in assessing the success of newly developed methods, and confirm the external validity of machine learning models before applying them to different settings.»
Still Hopeful
But Cheng notes that the findings are «not be taken as evidence against applying machine learning techniques in quantitative investing». On the contrary, she highlighted strengths and takeaways which will be key to the future development of the asset management industry.
They include lower downside risks and continued positive payoffs during crisis periods including the 1987 market crash, the Russian default, the burst of the tech bubble, and the recent financial crisis (3.56 percent monthly value-weighted return versus the benchmark return of 6.91 percent when excluding microcaps). The findings also show stronger capabilities in stock-picking relative to sector rotations.
The study, «Machine Learning versus Economic Restrictions: Evidence from Stock Return Predictability», was co-authored by Cheng alongside Doron Avramov, a professor at IDC Herzliya, and Lior Metzker, a research student at Hebrew University of Jerusalem.
