MPA favours three modelling methods:
– LASSO (least absolute shrinkage and selection operator) and ‘Ridge Regression’ (also called Tikhonov Regularisation) which are favoured over multivariate linear regression as these techniques are more effective in reducing model complexity, reduce over-fitting and are also helpful with dealing with colinearity in the drivers.
– Multivariate Adaptive Regression Splines which help to deal with non-linear relationships and pinch points of the individual drivers
– Classification and Regression Trees which help to subset the data into different regions that may exhibit different behaviour (e.g. very low stocks or very high y-o-y IP).