Too Long; Didn't Read
Creating a powerful predictive algorithm usually involves a certain amount of hyperparameter optimization. This involves tuning a model’s parameters to maximize a certain objective function, such as the Sharpe Ratio in finance. Bayesian optimization is by far the most popular method for tuning hyperparameters. We will use a stochastic oscillator momentum indicator as an example to illustrate the potential problems one can encounter when developing and tuning a trading algorithm on 30 minute bitcoin prices from Coinbase. The overfitting problem of a biased model in this scenario concerns tuning a. model until the parameters specifically pick one or multiple highly successful trade(s) that maximize the objective function.