Could've been worse - going to be there or thereabouts I think.
Q1 - Credibility factors - Bit of head scratching at first, but I didn't really like any of the options. Option A was the best as Z increased with an decrease in sample variance. The others didn't include s at all.
Q2 - General form of run off triangle - Easy 5 marks
Q3 - Bayes' criterion - Another easy 6 marks
Q4 - Posterior distribution - doable, but I messed up solving the parameters of the beta distribution. I ended up with lamba being 3/25 I think. I completed the rest of the question using the values I'd calculated and am hoping for method marks here.
Q5 - Run off triangle - More easy marks
Q6 - ARCH model - Bit of a bummer, that. Didn't even have a scooby. I wrote down some mess about covariances but I don't suppose any examiner in the land will be impressed. 8 marks lost.
Q7 - Monte Carlo simulation - but not as we know it. Didn't know what to do at all, wrote some more mess down, calculated theta to be e - 1 anyway. I didn't have any idea how to link the uniform distribution to theta. Another 7 marks lost I suspect.
Q8 - Ruin theory - just used a binomial tree, easiest ruin theory question I ever saw! Don't see why I shouldn't get all 10 marks here.
Q9 - NCD - I messed up simultaneous equations again. Got pretty close but I'm too stupid to solve them when you give me p, 1-p rather than actual numbers. I got most of the way though, and for part (b) I used my assumed values so I should pick up those marks. Hoping for 7 out of 10 here.
Q10 - Time series - I wasn't strong on the topic. I had a guess at part (a) (ARMA (2,1) anyone?) with some vague reasons. The next parts I could work out the yule-walker equations easily enough, then I tried to substitute in the values from the sample - is this right? Actually I didn't finish the simultaneous eqns for the AR(2) model - didn't have time so I wrote down what my method would've been. Fingers crossed! Then I wrote down the tests that are in the formula book - portmanteau and the other one. I suggested doing a correllelogram too, just because why not?
Q11 - The Long One -
Part (a) - compared the moment generating function replacing lambda with lambda prime = lambda / k. Seemed to work pretty well.
Part (b) - I got somewhere pretty close to the likelihood function but I managed not to have n in the numerator. The method was there, likelihood, differentiate, solve to zero for lamba. Didn't differentiate again to check it was a maximum though! D'oh!!
Part (c) - stuck the figures into the equation, got lamba = 0.0005 I think.
Part (d) - Was this a separate part? Anyway I calculated the expected loss per claim to the insurer in 2006. integrate x(f)x from 0 to 1600 and then 1600 times the probability that x >1600. Came to about 1101 ish. Then ran out of time to calculate the new retention limit. Wrote down my method again though - solve the same equations replacing 1600 with M', setting the sum equal to 1101 ish.
All in all, I think I'll get a mark in mid to high 50s. I hope the rest of you found it an impossibly tough paper and they'll reduce the pass mark!
Last edited by a moderator: Sep 17, 2008