Can you Deep Learn the Stock Market? "Honestly," no

You can find many examples of Deep Neural Network (DNN) models that successfully forecast the stock market. Typically, these models are using a very short time frequency. As variables inputs, these DDN models use a number of other stock indices that correlate with the S&P 500. They often use autoregressive variables (most recent S&P 500 levels). The mentioned high-frequency trading DNN models use covariates, or variables that are absent any explanatory logic besides being correlated with the S&P 500 (or whatever stock they predict). Let's step back and differentiate between covariates and explanatory variables because this is at the essence of my effort to Deep Learn the stock market "honestly." The mentioned "successful" high-frequency trading DNN models use covariates, or variables that are absent any true exogenous explanatory logic regarding the behavior of the S&P 500. Stating that the S&P 500 moves in tandem with the Nikkei 225 is not explanatory per se. It just exploits a tautological correlation.