Unmasking Methodological Flaws in the Research Linking Concentration and Inflation
A recent article published by the Federal Reserve Bank of Boston contends that an increase in concentration in the last two decades is amplifying price increases. The article, authored by Faulk Brauning, Jose Fillat, and Gustavo Joaquim, is one of the latest pieces contributing to the escalating battle that skeptics have been waging to contend that increased market power is leading to higher price markups. However, like other studies, this article faces methodological issues because of poor data choice and usage. The article has three issues: 1) it used Compustat rather than official Census data; 2) it defined industries at the broad 3-digit NAICS level; and 3) it excluded multiple sectors in its analysis of aggregate concentration. As a result, the study’s conclusion that concentration is rising and leading to higher price increases cannot be supported by this analysis. Indeed, using the most recent official Census data at the 6-digit NAICS level, ITIF has found that corporate concentration has not increased. Nor have price markups increased.
The most problematic issue with this analysis is the usage of Compustat data to measure concentration rather than official Census data. This Compustat dataset covers the sales of all publicly traded firms, but not private. As such, Compustat data cannot accurately replicate the results of analyses using official Census data. A past study found that Compustat data is poorly suited for measuring concentration in the manufacturing industries. More recently, the Board of Governors of the Federal Reserve System concluded that Compustat data is not suited for studying concentration in all industries. Their analysis showed that the correlations between the Compustat and the official Census data’s concentration ratios of the top 4 firms is actually quite low, ranging from just 0.1 to 0.2. This is true for data from 2002 to 2017 at the 4-, 5-, and 6-digit NAICS code levels. Moreover, the correlations for HHI, which the Boston Fed article used, are even lower and were, in some cases, negative—correlations ranged from about -0.05 to 0.2. As a result, the study’s conclusion that concentration is rising is not supported because of poor data choice.
Another issue with the study is that it measures industry concentration at the 3-digit NAICS subsector level rather than the more granular 6-digit NAICS industry level. This may have been done because of limited granular industry-level data from Compustat or the Bureau of Labor Statistics’ producer price datasets. Regardless, measuring concentration at the broad subsector level negatively impacts any results because subsectors include multiple industries that do not compete with one another. For example, it is difficult to argue that Furniture and Home Furnishing Retailers (NAICS: 4491) and Electronic and Appliance Retailers (NAICS: 4492) compete with each other. This is because a couch in the first industry group is not a substitute for a blender in the second. Yet, both are grouped under the 3-digit NAICS subsector Furniture, Home Furnishings, Electronics, and Appliance Retailers (NAICS: 449). As a result, the study’s conclusions about rising concentration and its impact on price increases are flawed.
In addition, despite not affecting within-industry concentration measurements, the study excluded multiple subsectors and sectors from its analysis of changes in aggregate concentration. The study concluded that industry concentration increased by about 50 percent from 2005 to 2020, going from an HHI of about 0.095 to 0.14. However, this conclusion comes from an analysis that did not include many industries, including retail (NAICS 44-45), postal service (NAICS 491), utilities (NAICS 22), finance and insurance (NAICS 52), public administration (NAICS 91/92), and those with fewer than two firms in the Compustat dataset. This means that the changes in concentration from these industries were not weighed into the calculations for aggregate concentration. As a result, the conclusion that aggregate concentration is rising is not truly representative of the overall economy’s change in concentration.
As a result of these methodological issues, the Boston Fed study is faulty in its conclusions that 1) concentration is rising and 2) rising concentration is leading to higher price increases. In contrast, an ITIF study analyzing official Census data at the granular 6-digit NAICS level found that concentration only rose about 1 percentage point from 34.3 percent to 35.3 percent in the period between 2002 and 2017, and that the most concentrated industries actually became less concentrated. Meanwhile, another ITIF study found that the correlation between industry concentration levels in 2017 and consumer price increases for 127 industries is a minuscule 0.0096. A final study by ITIF also notes that “markups have increased only slightly in some industries…and have stayed the same overall.” In other words, when official Census data is used, and other methodological issues are addressed, concentration is not rising and leading to higher price increases.
