When one wants to study consumer demand based on observational data (e.g. household survey data), imposing theoretical restrictions on the coefficients of the demand model seems to be a good a idea. You use information you have from demand theory to fine-tune your model, e.g. imposing a symmetry restriction on coefficients associated to price variables. This theory-driven modelling appraoch seems to be standard way to do demand analysis (see e.g. Deaton and Muellbaur 1980 and all the others). But now suppose you want to let the data speak and don't want to impose a prior structure on your model - at least not in the form of the theoretical restrictions on the coefficients. You may start simple and just regress budget shares of household on log prices of goods and log total expenditure of a household. Or you may go for a more elaborated approach by first searching for the demand function by a machine learning approach as suggested by the data and do inference based on the proposed function in subsequent step.

My question(s):

Is as data-driven model in contrast to a theory-driven one useful? And if yes, in which situations is it useful? What should one keep in mind when using a data-driven approach to demand modelling. (Hope this is clear and not too general. Let me know if any further specifications are needed).

  • $\begingroup$ It does seem quite general; could you possibly narrow the scope and clarify the definition of what exactly you would consider to be "data-driven"? $\endgroup$
    – Giskard
    Commented Mar 18, 2022 at 10:03


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.