Multilateral Index Approaches in the Presence of Product‐Specific Price Trends and Data Gaps

The prices of some products respond more strongly to changes in the general price level than others. This study explains why such product‐specific price level elasticities lead to biased time‐product‐dummy (TPD) estimates of price levels. Other popular multilateral index approaches such as the Gini‐Éltető‐Köves‐Szulc (GEKS) and Geary‐Khamis (GK) methods also fail to address this source of bias in inflation measurement. Therefore, this article introduces the NLTPD regression—a nonlinear generalization of the TPD regression. By estimating product‐specific price‐level elasticities, the NLTPD regression substantially mitigates this source of bias. A simulation study and an application to real‐world scanner data compare the performance of the four multilateral index approaches. Except for the rather theoretical case of complete data, the NLTPD regression outperforms the other three approaches, although this improvement comes at the cost of a small finite‐sample bias inherent in nonlinear estimation.