New paper on welfare measurement with non-homothetic preferences. Our paper is inspired by recent excellent work of @XJaravel and @DanialLashkari . We have a different angle on the problem we are excited to share. I think the intuition is pretty neat! econ.ucla.edu/arielb/welfare…

To calculate a "proper" welfare measure (i.e. money-metric), the price index must weigh past price changes using Hicksian (compensated) demand, which is unobservable, rather than observed (Marshallian/uncompensated) demand. This is a problem.

Problem is solved if you match each consumer today with one in the past with same preferences AND same utility level. If such a match exists for each past period, then the matched consumer's expenditures in the past ARE Hicksian demand curve you need! How to find such a consumer?

Our procedure finds these matched households using a simple observation: this is a recursive problem. If you could match households, you could compute their money-metric utility. If you knew their money-metric utility, you could match households. It is a fixed point.

Procedure identifies set of households for whom it is possible to find a matching household in the past. This is important because if there is growth, not all households today may have a match in past (i.e. the richest household today is richer than any household in the past).