Modelling addiction in life-cycle models: Revisiting the treatment of latent stocks and other unobservables

Dynamic modeling of demand for goods whose cumulated stocks enter an intertemporal utility function as latent variables, is discussed. The issues include: how represent addiction, how handle unobserved expectations and changing plans, how deal with 'dynamic inconsistency'? Arguments are put forth to give all optimizing conditions attention, not only those in which all variables are observable. If the latter, fairly common, 'limited information-reduced dimension' strategy is pursued, problems are shown to arise in attempting to identify coefficients of the preference structure and to test for addictive stocks. Examples, based on quadratic utility functions, illustrate the main points and challenge the validity of testing the 'rational addiction' hypothesis, by using linear, single-equation autoregressive models, as suggested by Becker, Grossman, and Murphy (1994) and adopted in several following studies.