Why Do the Rerun?

The reasons for carrying out the Rerun are summarised below. The overview is in points 1 to 3 below. We explain points 1 and 2 later.

1. The closure used in the Policy run is often quite different from the closure for the same year of the Base.

2. Even if there are no extra shocks, changing the closure can change the numerical results slightly.

3. You dont want such differences to contaminate the reported effects of the Policy shocks. So, when calculating the effects of the Policy shocks, it is best to compute the differences between simulations using the same closure, namely between years of the Policy and the corresponding years of the Rerun.

[If you computed the difference between the Policy results and the Base case results, these differences would also include the numerical effects due to the closure changes.]

Peter Dixon and Maureen Rimmer summarise these reasons very neatly in the final paragraph on page 197 of the MONASH Model book.

Details

1. Base and Policy Closures are Often Different. In the Base, you often set exogenously the values of variables which are naturally endogenous and which will be endogenous in the Policy runs.

For example, you may set volumes and prices for agricultural and mineral exports exogenously in the Base case because these variables have been forecast by government departments, and you want out Base case to reflect these forecasts. However, you will want these prices and quantities to be endogenous in the Policy run since you will want to be able to report the effects of the Policy shocks on these variables.

2. Closure changes affect numerical results. When you carry out the Rerun, you are effectively repeating the same year of the Base but with a different closure. The values of all variables which are exogenous in the Policy closure (this is the closure used for the Rerun) are taken exactly from the results of these variables in the same year of the Base. So, theoretically, the numerical results from the Rerun should be the same as those from the same year of the Base. However they are different in practice for two technical reasons.

The first reason is because you are only solving the underlying equations approximately. [For example, you may use just 4 Euler steps.] While you might expect exactly the same results from the Rerun and the Base if you solved the model exactly in each case, you must expect different approximations when you only solve it approximately. Hence you should expect (and find) numerical differences for this reason.

The second reason is that some of the variables are naturally path-dependent. That is, their values depend on the path taken from the pre-simulation solution to the post-simulation solution. The most important examples are the Divisia price and quantity indices used in models like ORANI-G and GTAP. They are path-dependent because the linearised equations defining them involve shares whose values change along the solution path. There are no simple levels formulas for these linear variables. But the path depends on the closure. Hence you should expect (and find) numerical differences for this reason.

Is the Rerun Always Necessary?

The Rerun is not necessary if you have the same closures in the Base and Policy.

Even if you have different Policy and Base closures, the Rerun may not be necessary if the effects of the Policy shocks are very large. Then it probably does not matter if the small numerical differences between the Base and the Rerun are included in the results you report.

URL of this topic: www.copsmodels.com/webhelp/rundynam/hc_whyrerun.htm

Link to full GEMPACK Manual

Link to GEMPACK homepage