Among the explanations below, which one is not a reason to favor a probability model over a regression-like (e.g., data-mining) model for long-run projections of customer behavior?

1. In a regression-like model, it is necessary (and potentially difficult) to project future values for the independent variables

2. Its often hard to come up with a full set of independent variables to adequately explain the observed behavior

3. Regression-like models are fine for a one-period-ahead prediction, but not beyond that horizon

4. If the observed behavior is viewed in an as if random manner, it would be wrong to put it into a regression-like model as if its deterministically true

5. Regression-like models cant capture non-stationarity, i.e., changes in behavioral propensities over time