1. While the walk forward method is not wrong by itself, it suffers from some weaknesses:
- The walk forward only tests over a single scenario (the historical path), meaning that each observation only has one forecast.
- The results can be biased by the specific behavior of the data in the training sample.
- E.g., training a model for the S&P500 in the mostly-booming period 2010–2019 and testing its performance in the 2007–2008 crisis period.
- However, there may be leakage of information from the test to the training sample that may generate misleading results.
- We only get one forecast per observation, and we only obtain an interpretation of how a strategy would have performed under hypothetical paths.
2.