kaybenleroll / carinsurance_pricing Goto Github PK

In 2.2.1 Iterating on Claim Size Prediction we can see:

ggplot(claimdata_dt) +
    geom_line(aes(x = claim_amount, y = claim_amount), size = 0.25) +
    geom_point(aes(x = claim_amount, y = exp(predict(model6, type = 'response')))
              ,size = 0.4, alpha = 0.2) +
    expand_limits(y = 0) +
    xlab("Claim Amount") +
    ylab("Predicted Amount")

we have a prediction for lognormal:

exp(predict(model6, type = 'response'))

If we do the mean of de predicted values probably the result will be much less than the empirical mean. After reading some literature I think that it should be:

sigma <- summary(model)$sigma
exp(predict(model))*exp(sigma^2/2)

What do you think?

Hello,

in 3.2 Simulation Using Standard Errors for Parameters there is a function called run_mc_sim and I see that inside the function:

as.vector(coef(arm::sim(claim_model_05, n.sim = 1)))

is always 0. When I use n.sim = 1 always all coef are equal to 0.

Is there something wrong?