No productivity internal link—too soon to collect data and productivity maintained post eBay

Filippo, Northwestern University Kellogg School of Management; Timothy, Boston University Questrom School of Business and NBER. “Patent policy and American innovation after eBay: An empirical examination.” Volume 48, Issue 5, June 2019, Pages 1271-1281. LJS

U.S. productivity growth has been sluggish in recent years compared to late 1990s, and there is a robust ongoing debate among¶ economists about the causes of this slowdown (Syverson, 2017). It is¶ possible that critics of eBay attribute the broad U.S. productivity growth slowdown to changes in patent policy. In our view, that inference would be misplaced for several reasons. First, it is too soon: the¶ historical evidence suggests that it takes decades for significant inventions to have an impact on aggregate productivity (e.g. David, 1990;¶ Solow, 1987). Because patent policy impacts the early stages of the¶ innovation process, any impacts of post-eBay patent policy are not¶ likely to have influenced the productivity statistics.29 Second, if patent policy were the mechanism driving the productivity slowdown, we¶ should see it in the R&D and patent data. But we do not.¶ Nevertheless, as a final step in our analysis, we examine industry level links between patenting and productivity. Our approach compares¶ industries that do and do not use patents, before and after 2006. If¶ patents are a significant driver of innovation in patenting industries,¶ and if the eBay decision harmed incentives to innovate in those industries, then we expect to find slower productivity growth in those¶ industries after 2006. As with our previous analysis, we might fail to¶ measure the impact of the patent-policy changes because of unmeasured changes in industries over time that lead to faster relative¶ productivity growth in patenting industries. However, this statement is¶ just another way of saying that unmeasured factors were equally important to changes in patent policy.¶ Our productivity data come from the NBER-CES manufacturing industry productivity database. Although the NBER data contain productivity information for 473 NAICS 6-digit industries, after merging¶ them to the USPTO industry-level patent statistics, we are left with a¶ balanced panel of seventeen NAICS 3 and 4-digit manufacturing industries from 2000 through 2011.30 We use a four-factor productivity¶ index that accounts for capital, production and non-production¶ workers, and materials.¶ Fig. 7 plots the TFP index for Computers (NAICS 334), Pharmaceuticals (NAICS 3254) and other manufacturing industries. Each index¶ is normalized to equal one in 2006, the year of the eBay decision. The¶ figure shows that computer-related manufacturing industries experienced faster TFP growth both before and after the eBay decision than¶ Pharmaceuticals or the other manufacturing sectors in our data. This is¶ a common feature of many productivity time-series and is partly due to¶ rapid quality improvements associated with Moore’s law (as well as the¶ difficulties of accounting for them).¶ To check for a systematic relationship between patenting and post eBay productivity trends, we use a difference-in-difference regression¶ based on the following specification:¶ TFPit = αi + λt + β1 * ln(Patentsi) * PostEbayt + εit¶ where TFPit is the log of four-factor TFP for industry I in year t; αi are a¶ set of industry fixed effects; λt are a set of year effects that capture¶ common economy-wide productivity trends; ln(Patentsi) measures the¶ patent-intensity of industry i; and PostEbayt is an indicator variable that¶ equals one for 2006 and all subsequent years. The main effects of patent-intensity and the eBay decision are absorbed by the industry and¶ year fixed effects respectively.31 The coefficient β1 measures changes in¶ the TFP growth rate for patent-intensive industries relative to less patent-intensive industries following the 2006 eBay decision.¶ To address the differential pre-eBay trends observed in Fig. 7, we¶ consider two robustness checks. First, we estimate a model where each¶ industry has its own liner productivity trend. Specifically, we interact¶ each industry fixed effect with a time trend (αi*t). In that specification,¶ β1 captures any deviation from trend following the eBay decision for¶ more patent-intensive sectors. Our second robustness test drops all¶ observations from the “Computer and electronic product manufacturing” sector (NAICS 334), where hedonic adjustments play a particularly important role in productivity measurement. The results are¶ presented in Table 2.¶ Results in the first column show that total factor productivity¶ growth accelerated after 2006 for more patent-intensive industries. This¶ is inconsistent with the idea that changes in patent-policy during that¶ period acted as a drag on innovation. In column 2, we add industry specific trends to the model. Although this reduces the size of β1, the¶ coefficient remains positive and statistically significant. Columns 3 and¶ 4 are based on the same specification as columns 1 and 2, respectively,¶ while dropping observations from the computer-manufacturing sector¶ from the estimation sample. In both cases, we continue to find that TFP¶ growth increased for more patent-intensive industries following the¶ 2006 eBay decision.¶ We do not wish to make too much of these models. Productivity is¶ notoriously difficult to measure, and changes slowly in response to a¶ wide variety of factors. However, this is precisely the reason to avoid¶ thinking that broader changes in productivity over the last decade are¶ necessarily linked to changes in patent policy. Our analysis simply¶ shows that if there is any statistically discernable link between the eBay¶ decision and subsequent short-run productivity growth, the data suggest a positive correlation between innovation and recent changes in¶ U.S. patent policy.