These are important questions I’ve been thinking about recently, and here are some thoughts to share. The answers are open, and I welcome your contribution of thoughts.
When do we need models? In a broad sense, I think every paper needs to have a conceptual framework to guide the analysis and explain the patterns in the data. In the preliminary stage of a piece of research, writing down the exact assumptions and predictions of the model can also help a researcher to organize her thoughts and frame her question better. That said, whether a structural model is needed really depends on the purpose of the paper. If the point of the paper is to propose a new model, then obviously a model is necessary. However, for papers which contribute in estimation methods a review of relevant models is sufficient.
What makes a good model? One of my professors gave this piece of advice: Make the parts of your paper that aren’t the point of the paper as simple as possible, and spend the richness in the part where your innovation is. It is all too easy to come up with complicated assumptions and notations in every dimension of your model, but adding complexity also sacrifices flexibility and robustness. Another piece of advice, given by the same professor, is “don’t let the perfect get in the way of the good”.
A side note: I have found writing skills to be extremely important in economics, or maybe in all research disciplines. One might not need to be an extraordinary writer to succeed in research, but failing to convey one’s ideas smoothly definitely reduces the rate at which they spread and benefit others.