5.5 Implication 3: Linearity

The forecast incremental losses might not be a linear function (e.g. \(q(w,d) = a\sqrt{c(w,d)}\))

• Test for linearity by making sure the residuals are not a sequence of positive then negative and vice versa

• Do this for each regression (i.e. every \(d\))

• Look at residuals as function of \(c(w,d-1)\)

• Similar to residual test from mack