Evaluating the Importance of Professional Forecasters and Trading Costs with Deep Learning Sorted Portfolios

Abstract

By examining deep learning sorted portfolios from Gu et al. (2018), we investigate the significance of information provided by professional forecasters' consensus (average) estimates as well as the impact of trading costs and constraints. We find that the information supplied by sell-side analysts does not meaningfully impact our deep learning models when controlling for other factors. We also find that trading costs and constraints have a large negative impact on the performance of deep learning sorted portfolios. When accounting for transaction costs, we find that deep learning sorted portfolios no longer outperform the S&P 500, thus serving as a piece of evidence for the Efficient Markets Hypothesis (EMH).