Lift chart
A chart that ranks policyholders into bands of predicted premium and compares, in each band, actual losses with the prediction, the most telling tool for showing that a new tariff separates risks better than the old one.
A lift chart orders the policyholders of a test sample by the premium or probability a model predicts, groups them into equal-sized bands, often deciles, and shows for each band the observed and predicted losses. A good model produces a steeply rising curve, the first bands having losses well below average and the last well above, with observed values close to predicted ones. The ratio between the losses of the riskiest band and the least risky band measures the lift. The so-called double lift variant compares two models: policyholders are ranked by the ratio of the premiums proposed by the new and old models, and in each band one looks at which of the two better predicted actual losses. It is the most convincing tool for management or a regulator, because it shows concretely where the new tariff corrects the old one and which policyholders will rightly see their premium rise or fall. It also highlights adverse selection risk: the policyholders the old tariff overcharged are those a better-segmented competitor will come and take.
Textbook case: on a double lift between the old and new home tariffs, the band where the new model proposes a 20 percent lower premium shows a loss ratio of 45 percent under the old tariff: these policyholders were overcharged, and a better-segmented competitor would have attracted them.
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.