Project BALD

BALD (Bayesian Analysis of Loss Development) is a Bayesian model for developing aggregate loss triangles in property & casualty insurance. This actuarial model makes use of a heteroskedastic and skewed log-t likelihood with endogenous degrees of freedom, employs model averaging by means of Reversible Jump MCMC (Markov Chain Monte Carlo simulation), and accommodates a structural break in the path of the consumption of benefits. The model solves the identification problem (of having three dimensions applied to a two-dimensional data matrix) by means of centering the calendar year effect on a chosen rate of inflation. This centering approach allows for incorporating expert information in the calendar year effect and obviates the need for supplying information on exposure. In an accompanying vignette, the model is applied to two widely studied General Liability and Auto Bodily Injury Liability loss triangles.

Literature

Schmid, Frank A., “Robust Loss Development Using MCMC,” Working Paper, 2010

Schmid, Frank, “The Workers Compensation Tails,” Variance 6, 48-77, 2012