The adaptive markets hypothesis through the lens of machine learning

The adaptive markets hypothesis (AMH) postulates that markets are not stationary environments permanently in equilibrium as the efficient market hypothesis (EMH) posits, but evolve over time. However, testing this requires tools that capture the inherent complexity of financial markets. Therefore, in this paper we introduce an approach that uses the Long Short-Term Memory (LSTM) model to capture changes in the structure of the economic-financial environment. We apply the methodology to the U.S. equity market represented by the S&P500 for 40 years of daily data. With this tool we will demonstrate that, even with non-heuristic approaches such as the LSTM model, the learning and adaptation process after a change in the environment is slow, which is consistent with the AMH.