I have attempted to model chaotic nonlinear timeseries such as the closing stock prices of S&P 500 companies, using echo state networks, a variant of reservoir computing. In a way, it shows how much more efficient ESNs are compared to lstms or regression when it comes to dynamic systems. I have compared the efficiency of three different approaches for the same -
The code for the ESN model has been obtained from clemens korndörfer's repo as it was not available on the keras library I have set the hyperparameters as per optimality and efficiency