Model convergence was in part based on starting the minimization process from dispersed values of the maximum likelihood estimates to determine if the estimation routine results in a smaller likelihood. Starting parameters were jittered using the built-in functionality of Stock Synthesis, where you specify a jitter fraction. Here we used a jitter fraction of 0.05 and the jittering was repeated one hundred times. A better, i.e., lower negative log-likelihood, fit was not found. Several models resulted in similar log-likelihood values with little difference in the overall model estimates, indicating a relatively flat likelihood surface around the maximum likelihood estimate. Through the jittering analysis performed here and the estimation of likelihood profiles, we are confident that the base model as presented represents the best fit to the data given the assumptions made.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.