The student honored to introduce commencement speaker Jimmy Page during the Berklee School of Music graduation ceremonies last weekend gushed "You are the space between the stars of the universe." As with the 88 piano keys, for data observations in a triangular display to support the actuarial prediction of alternative future realities, creativity is required. Great strides are being made in this area these days.

]]>"Stochastic" means random, I am still thinking what is random in actuarial world,

1. And in that case, when we say MCMC (Markov Chain Monte Carlo), what is switching?

2. Is re-sampling residual a truly stochastic method?

residual= (A-E)/E^.5

A = Incremental Paid Loss

E = Incremental Paid triangle after applying one set of (let us say all year average) LDF on accumulated Paid Loss (So here is ChainLadder1)

After resample this residual triangle (why? Is residual under uniform distribution?), we do another round of development, to get the lower incremental triangle (So the ChainLadder2)

]]>England and Verall already brought us stochastic reserving, although I am not sure that bootstrapping residuals is as robust as MC simulation. Moreover, Reserves are estimated using GLM/GAMs and the range/variability is calculated using MC simulation. Lastly, Bayesian analysis of reserves almost always uses MCMC estimation as well.

England & Verall is already implemented in the `chainladder` package, Glenn Meyers has posted R code for Bayesian reserving in the Actuarial Review, and it shouldn't be too difficult to simulate from a GLM given you have the estimated parameter standard deviations and pairwise correlations (using `mvtnorm` for example). I've used R for multivariate simulation of parameters before, but for pricing, not reserving.

Can I say Chain Ladder looks like: professional guess from consecutive division average over summary, or something? If we use R for triangle analysis, what can we gain?

]]>