Below are some tips for anyone interested in applying the
Matlab routines used in our project.

   1. Most of the time should be spent finding a good 
      starting point for the maximum likelihood estimation.

      a. Use means (e.g., average ratio of market consumption 
         to output) to determine share parameters.

      b. Manually adjust subsets of parameters to see if
         the likelihood is near-flat for some -- fix those
         initially.

      c. For all candidate parameter vectors, plot the
         innovations to see if they look like white noise.

      d. Work up to estimating the full set of parameters
         gradually -- it is difficult to make progress
         in a very high dimensional parameter space.

   2. One problem that arose in our project was the trends
      in tax rates which caused near unit-root processes.
      (This is why we do checks with HP-filtered data.)
      I have built in penalty functions in the main routines.
      Users should make sure that the parameters are set 
      appropriately for their own examples.

   3. Always ask, Is there something first-order that my 
      model is missing?   If there is, it should be obvious
      because the MLE routines will pull you away from 
      estimates that are typically used in the literature.
      In our first attempts, we did not have taxes in the
      model -- and the mismatch of theory and data was 
      evident in the innovations we plotted.

Ellen McGrattan