Dear Renger,
I run the CGE model with social accounting matrix data, the software reports
8674 INFEASIBLE (INFES)
SUM 1.2082E+7
MAX 6.9530E+5
MEAN 1392.915
1 UNBOUNDED (UNBND).
02032020 SAM 2007 dã chinh .gms (541 KB)
Can you explain to me why this is happening?
Thank you very much.
Nguyen Xuan
Hi
In order to find the problem, you should proceed as follows (this is something you should always do when working with CGE models).
- Initialize your variables: Set the starting values according to your SAM (usually this means setting the production levels and prices equal to 1). The idea is to fill the equations at the start of the solve with the solution.
- Set the iteration limit of your model to 0 (e.g. mymodel.iterlim = 0);).
- Run the model: If you initialized correctly, Gams will find the solution (the original SAM) without an iteration. If there is an infeasibility: go to your equation listing and look for the infeasibilities. If you made a mistake in the initialization or in your equation you will have an infeasibility.
- If your initialization and your equations are correct, solve the model with your numeraire mutliplied by, for example, 1.1. If your model solves with the same quantities and prices multiplied with 1.1, your model seems to be correct.
Try it out with a smaller model from the Gams library (make an obvious mistake in one equation or initialization and see what happens).
I hope this helps to get you started in finding the problem.
Cheers
Renger
May I ask a few questions about this answer?
I am new to GAMS, and I used a simple CGE model to test these suggestions. I input an incorrect value at an initial point and solved the model (with unlimited iterations). The model found a solution with no infeasibilities, but the values obtained are very different from the regular solution (with all the correct initial values).
Does this mean that I can only find the correct solution by inputting the correct initial values, i.e., solving the model manually? Is the interpretation that the model is complex enough that it does not allow the algorithm to find the global maximum?
Or is there something else I am missing?
Thank you.
The input point influences the solution that is found by a local optimization solver for a model with multiple optima. You say “correct solution”, there are no correct or incorrect solutions, there are just (locally) optimal solutions, where the KKT conditions are fulfilled. If you need the global optimal solution you either need to let the local solver start close to this solution or use a global solver.
Good luck!
-Michael
1 Like