Random Forest Models for Time Series and Cross Validation

Question

If one were to build a model using a random forest model that uses lagged variables, for simplicity we'll describe this just using a single feature describing lag 1: $x_{t-1}$.

Which attempts to predict $x_{t}$

Will this model still be subject to standard time series CV rules? I believe the feature vector from one instance to another will be independant and therefore a forecast of rolling origin isn't required and standard K-Fold can apply? Is there any issues that can arise from not doing a rolling origin CV under this context?

Answer 1

Anytime you would use data that you wouldn't have at forecast time while testing then you introduce data leakage. This occurs when you use the lags and you want to forecast more than 1 period for t-1 lag. At that point you will probably use forecasted values whereas if you do standard CV then you will be using actuals.

So use lagged variables, but your CV strategy should be using the predictions. In other words, do time series cross validation.