Currently I am trying to find a prediction interval for $q_{x}(t)$ in the following model:

\begin{align} q_{x}(t) &= 1 - \exp(-\mu_{x}(t))\\ \ln(\mu_{x}(t)) & = \ln(\mu_{x}^{EU}(t)) + \ln(\mu_{x}^{NL}(t)) \\ \ln(\mu_{x}^{EU}(t)) &= A_{x} + B_{x}K_{t} \\ \ln(\mu^{NL}_{x}(t)) &= \alpha_{x} + \beta_{x} \kappa_{t} \end{align} where \begin{align} K_{t+1} &= K_{t} + \theta + \epsilon_{t+1} \\ \kappa_{t+1} &= a\kappa_{t} + \delta_{t+1} \end{align}

The error terms $(\epsilon_{t},\delta_{t})$ are assumed to be independent and are drawn from a bi variate normal distribution with mean $(0,0)$ and covariance matrix $\mathbf C = \begin{bmatrix} \sigma^{2}_{\epsilon} & \rho \sigma_{\epsilon} \sigma_{\delta}& \\ \rho \sigma_{\epsilon} \sigma_{\delta} & \sigma^{2}_{\delta} \end{bmatrix}$

Basically, this is an extension of the Lee-Carter model that allows for coherent forecasting of sub-populations. From an earlier question ( Maximum Likelihood in a time series multi-population model) , I have estimated the parameters necessary to compute prediction intervals.

The problem is that I would like to have a prediction interval for $q_{x}(t)$. If this were a standard lee carter model I know I can get a prediction interval by taking a confidence interval for the AR(1) process and then plug in the resulting limits into the expression for $\mu_{x}(t)$ and $q_{x}(t)$. However when I tried doing this for the multi population model the confidence bands seemed to be too small. Is there perhaps a different way to get the prediction intervals for the bivariate case?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.