Estimating from historical data requires many assumptions about similarity. Reducing the number of estimated parameters can control model risk.