How the Gradient of a Model Tells Us Which Questions the Data Can Answer

My first technical post on the work I am doing at Macrocosm is up on the company blog.

How the gradient of a model tells us which questions the data can answer

Most people treat simulators as black boxes: change the inputs, see what comes out. But if you built the thing, you have more to work with than that. In the post I show how the gradients of an economic model tell you which questions it can actually answer, before you spend any time collecting data or compute time trying to fit it. I then apply it to a well-known financial market model.

The model is Brock and Hommes (1998), the standard heterogeneous-agent asset pricing model, with its two trader types: one trading on value, one following trends. The same price series turns out to pin down the effect of a transaction tax on volatility while saying essentially nothing about the trend followers. Same model, same data, and one policy question is answerable where the other is not.

It is joint work with colleagues at Macrocosm.