Always use smaller models when you can. Typically, you want to show that
- surprisingly/interestingly, if we combine A,B in a model, there is an interaction that explains C.
Have the smallest model that you need to make your point (that contains A,B). Additional features are irrelevant to make your point and will only distract you and your audience.
There are many models which are famous for being able to model A, B, or similar. If you need A and C, take a model famous for being good at modeling A, and add feature C. If there is nothing clean (any model that has A, also has A2, A3, A4 - which you really don't need), you might be better off writing your own model / simplifying an existing one).
Example
For example,
- A: Monopolistic competition is a nice way to have firms with positive profit margins
- B: The Calvo-fairy is one way of getting sticky prices in an equilibrium model
If you combine A,B (and a few other basic elements), you can interestingly show an environment with room for monetary policy, the standard NK-model.
Say you want to look at the impact of sticky prices onto unemployment. You could take A,B and add sticky wages and monopolistic competition on the labor supply side. Or, you could start with the Diamonds-Mortensen-Pissarides (DMP) model and extend it with B.
The choice of model depends particularly on
- Which is the easier choice (certainly the first one)
- What is the mechanism you're interested in?
If you believe that workers really are setting the wage and there is some stickiness in their ability to do so, (i) is the way to go. If you believe it is important that employment comes after some matching-period, and wages are set through bargaining from both employers and employees, extending DMP would be a better way.
Fundamentally, if you believe that you could use DMP, but it has some clutter that you don't need, get rid of that first, and then add your price-stickiness.