• Flexible Entry/Exit Adjustment for Price Indices (Job Market Paper) link

This paper proposes a new method for constructing an entry/exit adjusted price index. I utilize a translog demand system with mild restrictions that remains tractable when there are many goods and when products are allowed to enter and exit. In particular, I impose no restrictions on the level or distribution of own-price effects while mild restrictions are placed on cross-price effects. This restricted translog yields a sufficient statistic for the price index that only requires calibrating one parameter per product (avoiding a curse of dimensionality) and I show how the relevant parameter, a modified semi-elasticity, is stable even as product availability adjusts over time. To estimate this semi-elasticity parameter without imposing ex-ante restrictions on the distribution of own-price effects present in the data, I adapt the generalized random forest method of Athey et. al. (2019) to my panel data setting. As an application, I apply this method to the ready-to-eat cereal market using Nielsen Consumer Panel data. I find a novel asymmetry between entering and exiting goods that is not admissible in standard CES-based techniques; specifically, I find entering goods are relatively inelastically demanded (on entry) while exiting goods on elastically demanded (on exit). This asymmetry boosts the net gains from entry and exit, partially offsetting the effect of switching from unbounded reservation prices (CES) to finite reservation prices (translog).

  • Is heteroskedasticity-based identification robust to parameter heterogeneity?

This paper shows that the heteroskedasticity-identified double-difference (HIDD) estimation strategy performs poorly if products with different underlying parameters are improperly grouped together in estimation. Unlike for OLS and linear IV, HIDD estimation is not consistent for an “average” parameter value when the data generating process has heterogenous parameters. Using simulated data, I show that even with small differences in the underlying parameter(s) the HIDD estimator may behave erratically, yielding point estimates well outside the range of the underlying product-specific parameters. To evaluate the empirical importance of these findings, I evaluate industry-level point estimates for U.S. HS4 level trade data, a common setting for HIDD estimation, relative to partial identification results that rely on strictly weaker conditions. In almost all industries, I find products (import source countries) where the industry-level point estimate is outside of the product-specific bounds for some products. In addition, I show that conditional on the point estimate for the industry supply elasticity there is substantial heterogeneity in the implied product-specific demand elasticity. While previous work has suggested heteroskedasticity-based double difference estimators may be downward biased, I document suggestive evidence that the industry-level point estimates are higher than the product-specific values.