Imputed price indices via matrix completion

This paper explores matrix completion as an alternative to impute missing entries of a price tableau to compute a given price index formula. Missing prices are usually imputed by means of predictions from estimated hedonic regression models. These models require the collection of a significative set of covariates and a set of assumptions to be respected. On the other hand, matrix completion is able to work free from all of that. In this paper, we show how matrix completion can be used to impute missing entries when computing (bilateral) price index formulas in a temporal framework. We provide two applications: one to housing in four metropolitan areas in Seul and the other to cars. Each application is compared to traditional hedonic regression imputation.