ABSTRACT
Despite the widespread use of economic information to anticipate changes in business conditions, innovation metrics are not considered to be leading indicators. We argue that aggregate trademark data reflect firm-level choices that can help predict business cycles. In addition to establishing the conceptual basis for considering trademarks, our statistical evaluations, using turning point analysis and a novel machine learning method, find that trademark filings for product and service offerings in commercial use outperform many of the conventional leading indicators. Our work suggests that including trademark metrics in composite indexes could improve recession forecasting performance.
Disclaimer
The views expressed are those of the individual authors and do not necessarily reflect the official positions of the Office of the Chief Economist or the U.S. Patent and Trademark Office.
Acknowledgements
The authors would like to thank Mary Denison, the USPTO Commissioner for Trademarks, for suggesting the possibility that trademarks may help anticipate economic activity.
Disclosure statement
No potential conflict of interest was reported by the authors.
Correction Statement
This article has been republished with minor changes. These changes do not impact the academic content of the article.
Notes
1 The Bureau of Economic Analysis of the Department of Commerce decided in 1995 to seek a private organization to produce and disseminate its monthly cyclical indicators – including the leading economic indicators and the widely publicised composite leading index. After a bidding process, The Conference Board was selected to become the custodian of the official composite leading, coincident, and lagging indexes. See https://www.conference-board.org/data/bci/index.cfm?id=2157.
2 At least since Schumpeter (1942), economists have recognized the possibility that innovation may cause business cycles through the process of creative destruction.
3 The USPTO provides daily-updated XML files containing pending and registered trademark applications via https://bulkdata.uspto.gov/data/trademark/dailyxml/applications/.
4 The USPTO Trademark Case Files dataset is available at https://www.uspto.gov/learning-and-resources/electronic-data-products/trademark-case-files-dataset-0.
5 The ten components of the LEI for the United States (as of March 2019) include: average weekly hours, manufacturing; average weekly initial claims for unemployment insurance; manufacturers’ new orders, consumer goods and materials; ISM® index of new orders; manufacturers’ new orders, nondefense capital goods excluding aircraft orders; building permits, new private housing units; stock prices, 500 common stocks; Leading Credit Index; interest rate spread, 10-year Treasury bonds less federal funds; and average consumer expectations for business conditions.
6 See http://www.nber.org/cycles/recessions.html for more information on the NBER recession dating procedure.
7 The BSTS model applies the Kalman filter (Kalman Citation1960; Harvey Citation1989; Scott and Varian Citation2015), which recursively calculates the predictive distribution of the state in t + 1 by the predictive distribution of the state in t using the time-series of from t = 1, …, t-1 combined with the observed value of
. This process uses ‘a standard set of formulas that is logically equivalent to linear regression,’ (Scott and Varian Citation2014). An in-depth discussion of the Kalman filter is beyond the scope of this paper.
8 The classic constant-trend model can be written as .
9 Steven L. Scott first suggested this BSTS model specification in a blog post, see http://www.unofficialgoogledatascience.com/2017/07/fitting-bayesian-structural-time-series.html.
10 The prior probability of inclusion () for each variable i is set to 1/K, where K is equal to the total number of potential leading indicators in our sample. This corresponds to the total number of relevant variables expected in our model.
11 We consider multiple transformations of the trademark indicators included in , but we only report QPS and BSS results for the seasonally-adjusted growth rates as they were predominantly the best performing transformation.
12 The ROC curve effectively classifies whether the probit model predicted recession accurately or generated a false signal at the probability threshold . The ROC curve plots the true positive rate (
) against the false positive rate (
) at different values of w (Lahiri and Yang Citation2015).