Brosius Flashcards
3 Special Cases for Brosius Least Squares
Chain Ladder (a = 0) BF (b = 1) Budgeted Loss (b = 0)
Claim counts Poisson and d = probability of reporting in first period, what is the Bayes reserve?
R(x) = mu(1 -d)
Negative binomial distribution
E[Y] = r(1-p)/p
Claim counts NegBin(r,p), d is probability of reporting in first period, x is actual reported, what is the Bayes reserve?
R(x) = s(x+r)/(1-s) s = (1-d)(1-p)
Formulas for a and b in Brosius Least Squares
b = [mean(xy) - mean(x)mean(y)]/[mean(x^2)- mean(x)^2]
For ultimate Y and Reported loss X, what is best linear estimate of Y given X?
L(x) = x - E[x] + E[Y]
When is Brosius appropriate/inappropriate
Appropriate: when distribution is same across multiple years
Inappropriate: year to year changes are systematic shifts (inflation, legal environment)
Estimate L(x) using credibility
L = Z(x/d) + (1-Z)(E[Y]) Z = VHM/(VHM + EVPV)
Statistics for Brosius
c = mean(y)/mean(x) Z = b/c d = 1/c