Testing Methodology

5.1 Testing Methodology

5.1.1 Out-of-Sample Validation

We implemented a leave-one-out cross-validation approach across annual periods from 2022 to 2024. For each test year, the model ensemble was trained on the remaining two years, ensuring complete separation between training and testing data.

The process can be formalized as follows:

This methodology ensures that each prediction used in our backtesting analysis is genuinely out-of-sample, while maximizing the utilization of available training data.