0

RESULTS OF PRODUCT-PRICE STUDIES: Robustness to Industry Selection and Weighting 5

Posted by Connie R. Aponte on June 18, 2014 in RESULTS OF PRODUCT-PRICE STUDIES |

international-trading
At the very least, when possible the direction of bias introduced by “bad” data should be considered before excluding data. For example, SSh claim that reported computer prices do not adequately control for quality improvements. Stated another way, they argue that the reported price decline for computers, in absolute value, understates the true quality-adjusted price decline for computers. This suggests that the reported price decline in computers is biased upwards towards zero. The solution that SS use of dummying out the computer industry actually reinforces this bias rather than mitigating it. Rather than using information about the direction of bias, the dummy variable effectively sets the price change for the computers to zero when estimating the cross-industry relationship between price changes and relative employment. To control for the bias introduced by computers, therefore, the results without the computer dummy are arguably better than the results with the computer dummy.

Related to the issue of industry selection is the issue of industry weighting. The logic of the SS theorem suggests that empirically all industries should be weighted equally. The link from product prices to factor prices relies on the existence of industries, not their sizes. Thanks to the assumption of perfect interindustry factor mobility, as long as an industry has positive output its product price can and does affect factor prices in every industry. That is, as long as an industry has some positive output it accounts for one of the rows in matrix equations (1) and (2) regardless of how large the industry is. A product-price change in even the smallest industry is qualitatively just as important as a product price in the largest industry. This suggests that any data analysis should weight all industries equally.

Given the theoretical preference for equally weighted industries, weighting data differently probably requires some overriding empirical justification. For example, as BC suggest (fn. 29, p. 22) one might weight larger industries more heavily if smaller industries had poorer quality data. Another reason to weight might be that many smaller sectors are residual categories of a wide range of products.

Tags: , ,

Copyright © 2013 - 2019 Blog About Business And Finance. All rights reserved

Home | Site Map | Contacts