Walk Forward Optimization

[eervin123]

In the following example notebook we learn how to split historical data into training and validation groups in order to identify the optimal parameters without blatantly overfitting.

This is a really helpful example. However, I'm stumped on how to complete the next step and create a holistic walk-forward backtest.

For example, the notebook ends with a plot showing the rolling Sharpe ratio for in-sample and out-of-sample periods, but how can we create a full portfolio object for the entire historical time period with the optimal parameters and evaluate it as a portfolio object?

For example,

use the in_best_index to identify the best parameters for each in_sample period and then apply those params to the out_sample periods. I realize we would first have to set some default params for the first in_sample period, but after that, we could have an entire backtested portfolio using the best walk-forward parameters. I'm just not sure how to string it all together. cc @polakowo

wfo_optimal_pf.stats()
wfo_optimal_pf.plot().show()