Multilateral Methods, Substitution Bias, and Chain Drift: Some Empirical Comparisons

Abstract

Multilateral methods are increasingly being used in the computation of official price indexes such as the consumer price index (CPI). This reflects the growing use of scanner data by statistical agencies and the fact that fixed‐based or chained price comparisons perform poorly in this context. A range of multilateral approaches have been pursued by different statistical agencies. Yet, it can be shown that some of these methods, at least theoretically, could suffer from substitution bias. We investigate this as well as the drivers of chain drift. We adopt a simulation‐based approach using actual scanner data prices with the corresponding quantities being generated assuming constant elasticity of substitution (CES) preferences. We find that most methods systematically deviate from the exact CES benchmark index. This is even the case for the superlative index methods, which should not exhibit substitution bias. Interestingly, we also find significant chain drift even in the exact CES indexes. We argue this reflects life cycle pricing and particularly run‐out sales at the end of a product's life.