Gabriel J. Michael / gmichael at gwu dot edu
A previous post discussed Riker’s 2012 article, “Special 301 and Royalty Receipts from U.S. Trade Partners,” published in the International Trade Journal. I mentioned that I was trying to replicate his results. Apart from the problems I discussed before (choice of dependent variable, endogeneity, and limited data coverage), my attempt at replication uncovered additional problems.
First, the article incorrectly claims that Hong Kong was designated a Special 301 priority in every year between 2001 and 2007. In fact, Hong Kong has never been designated a Priority Foreign Country, has never appeared on the Priority Watch List, and has never been subject to Section 306 monitoring. Indeed, Hong Kong is only rarely mentioned in Special 301 reports throughout the years. It appeared on the Watch List in 1997 and 1998, and underwent an out-of-cycle review in 1999.
This error does affect the coefficient estimates in the article, although not significantly. On the other hand, it causes Riker to overstate his estimated dollar value impact of Special 301 designations by $650 million.
Riker uses three sources of data to construct his econometric models: GDP, BEA data on IP royalties, and Special 301 designations. GDP data is widely available from sources such as the World Bank and International Monetary Fund. Both sources yield comparable results. The article does not clearly indicate that the models used inflation adjusted data, but I assumed that they did, although using nominal GDP yields comparable results. The BEA data can be obtained online, although it required some piecing together to get all 33 countries over the time period used in the models. The BEA data appear to report nominal values, so I adjusted them for inflation, though as before, using nominal values yields comparable results. I already had a Special 301 dataset which I have used in other research. Apart from the Hong Kong error in Riker’s article (described above), my Special 301 dataset appears to match the one used in the article.
Riker’s article reports the results of three different models. Each model uses the year-over-year change in logged IP royalty receipts as the dependent variable. Models 1 and 2 use “priority designation” (defined as either being a Priority Foreign Country, being on the Priority Watch List, or being under Section 306 monitoring) as the key explanatory variable. Model 3 separates countries designated in every year from countries designated only in some years. All three models include year fixed effects, and Model 1 also includes country fixed effects. Based on F tests, Riker concludes Model 1 is the most appropriate. For reference, I have included the original regression results reported in the article:
The following table presents the results of my regressions, which attempt to replicate the regressions in the article as closely as possible. As you can see, I use the same panel structure of 33 countries and 7 years, producing 231 observations. My estimates of the effects of changes in GDP were larger than in article, but my estimates of the effects of priority designation closely match the article. Given the errors in the original dataset, updated GDP data, and differing implementations used by various statistical software packages, some discrepancies are expected. Overall, though, I was largely able to replicate Riker’s original findings:
Beyond simply replicating these results, I also wanted to modify the models and see how these modifications affected the results. Importantly, Riker excludes Watch List designations, but does not explain why. Watch List designations might have a less pronounced effect than the more severe Priority Foreign Country and Priority Watch List designations, but if one believes that Special 301 is effective at all, there is little reason to draw the line at the Watch List. Thus, I wanted to include Watch List designations.
Furthermore, two years have elapsed since Riker’s article was published, so I have the benefit of additional data. I wanted to include additional years of data to see if Riker’s original results still held.
The following table presents three modifications of Riker’s original Model 1. In each case, the statistical significance of the Special 301 explanatory variable disappears. Model 1.1 retains the original 231 observation dataset, but redefines the designation variable to include Watch List designations. In doing so, the designation variable becomes insignificant—its effect is indistinguishable from zero. Model 1.2 retains Riker’s original Special 301 variable, but expands the dataset by six years (covering 2000-2012, rather than 2002-2008). Thus, Model 1.2 is identical to Model 1 in the previous two tables; the only difference is that it includes more data. As you can see, the Priority Designation variable is no longer significant. Model 1.3 uses the redefined designation variable of Model 1.1 in conjunction with the expanded dataset; again, the Special 301 variable is insignificant.
To summarize, while I was able to replicate Riker’s results, simply including additional years of data causes his findings of significance to disappear. Likewise, even in his original dataset and models, if Watch List designations are included, the findings of significance disappear.
Ultimately, these results lead me to conclude that Riker’s 2012 article is both theoretically and empirically flawed. It cannot support the conclusion that Special 301 designations are correlated with increased IP royalties from designated countries in subsequent years. Even if it did, this would be an inappropriate measure of Special 301’s effectiveness. In a future post, I plan to test Special 301’s effectiveness using a more appropriate measure: changes in an index of IP rights.
This post is licensed CC BY-SA 4.0, and may be shared and reposted with attribution. Please include a link back to this page, which will contain the most up-to-date version.