The Effect of Chinese Import Tariffs on Innovation of U.S.
Firms
MSc Finance, Track Corporate Finance Master Thesis
Aurelia Mohrmann
Thesis supervisor: Dr. Tolga Caskurlu
University of Amsterdam, Amsterdam Business School June 2018
Statement of Originality
This document is written by Aurelia Mohrmann who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its referenced have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgments
I would like to express my gratitude to my supervisor, Dr. Tolga Caskurlu, for the valuable comments and guidance throughout the process of writing this master thesis.
Abstract
This paper investigates the effect of Chinese import competition on innovation of high- and low-tech firms in the U.S. Using a quasi-natural experiment, I measure changes in firm-level patent data before and after significant reductions of Chinese import tariffs between 1993 and 2012. Based on the Lerner Index, I compare these effects for firms with different degrees of market competition. I find that an import tariff reduction has a positive effect on innovation of high-tech firms. This result suggests that these firms use imports from low-wage countries as intermediate inputs and shift resources towards innovative activities. For low-tech firms in highly competitive markets, I find a negative effect of import competition on innovation. My findings for low-tech firms give support to the Schumpterian hypothesis, which states that an increase in competition decreases profits, resulting in lower incentives to innovate.
Table of Contents 1. Introduction ... 5 2. Literature Review... 8 3. Methodology ... 11 3.1. Causality ... 11 3.2. Research Design ... 12 4. Data ... 15 4.1. Innovation Measure ... 15
4.2. Market Competition Measure ... 16
4.3. Foreign Import Competition Measure ... 16
4.4. Matching Covariates and Controls ... 17
4.5. Treated Firms ... 17
4.6 Matched Firms ... 18
4.7 Descriptive Statistics ... 18
4.8. Innovation around a Tariff Reduction for Treated and Matched Firms ... 20
5. Results... 22
6. Robustness Checks ... 30
6.1. Matching ... 30
6.2. Cutoff for Significant Tariff Reduction ... 30
6.3. 5 Quintiles of Market Competition ... 31
7. Conclusion ... 36
References ... 40
1. Introduction
China’s manufacturing exports grew from 2.3% of total world exports in 1991 to 18.3% of total world exports in 2014, due to significant trade barrier reductions (Autor et al., 2016). The rise in trade with low-wage countries such as China increases foreign import competition for firms in advanced markets (Chakravorty et al., 2017). Although Chinese import competition reduces employment and wages in the U.S. (Acemoglu et al. 2016, Pierce and Schott 2016, Autor et al. 2014, Autor et al. 2013), there is no theoretical or empirical consensus regarding its impact on firm innovation. Innovation is an important determinant of economic growth (Aghion and Howitt, 1992). Therefore, in this paper, I investigate the effect of reductions of Chinese import tariffs on the innovation of high- and low-tech firms in the U.S.
In theory, the relationship between innovation and competition is negative (Schumpeter, 1942), positive (Arrow, 1962) or inverted U-shaped (Aghion et al., 2005). According to the Schumpterian hypothesis, an increase in competition decreases firm profits and lower profits leads to lower incentives to innovate (Schumpeter, 1942). The escape-competition effect theorizes that competition increases incentives to innovate because firms want to out-perform their competitors and have higher profits as a market leader (Arrow, 1962). Aghion et al. (2005) combine these theories into an inverted U-shaped curve. More specifically, when competition is low, the escape-competition effect dominates, resulting in higher levels of innovation when escape-competition increases. However, as competition increases, at some point, the Schumpterian effect dominates since post-innovation rents are no longer higher than pre-post-innovation rents and post-innovation incentives decrease. Recent empirical papers with trade-based measures of competition disagree on whether the relationship is positive (Beneito et al. 2017, Chen et al. 2017, Zhang 2017, Bloom et al. 2016, Correa and Ornaghi 2014, Gorodnichenko et al. 2010), negative (Autor et al. 2017, Nobuaki and Isamu 2017, Liu and Qiu 2016, Hashmi 2013) or inverted U-shaped (Chakravorty et al. 2017, Im et al., 2015, Michiyuki and Shunsake 2013).
I identify significant tariff rate reductions on imports from China as events in which entry costs decrease, resulting in higher competition from China in the U.S. (Frésard, 2010). My analysis includes 677 distinct events occurring between 1993 and 2012. I compare average innovation in the five years after the tariff cut to average innovation in the five years before the tariff cut using a difference-in-difference research design. Matching occurs based on similar firm characteristics
in the year before the event. The only current innovation-related research to use tariff reductions to measure import competition is done by Im et al. (2015). My paper differs from Im et al. (2015)’s research in that I specifically measure tariff reductions of imports from a low-wage country, while Im et al. (2015) look at average worldwide tariffs. Moreover, my paper is different from Im et al. (2015) in that I specifically match treated firms to controls and compare effects in high- and low-tech firms, based on their research and development (R&D) intensity.
I contribute to the theoretical and empirical debate on the effect of competition on innovation by suggesting that a distinction should be made between imports from low-wage and high-wage countries, as well as a distinction between high- and low-tech domestic firms. More specifically, considering imports from a low-wage country, policymakers should reduce tariffs if they want to stimulate innovation in high-tech firms. However, low-tech firms in high competition markets would need to be compensated for the decrease in profits caused by the increase in competition. I find that a significant tariff reduction increases innovation in high-tech firms by 12.7% to 14% and decreases innovation in low-tech firms in highly competitive markets by 19.3% to 22.9%. Understanding the effect of trade on innovation is especially important for policymakers who have direct influence on import tariff rates.
I interpret the effect of import competition on low-tech firms in highly competitive markets using the Schumpterian hypothesis and explain the effect on high-tech firms using the ‘trapped factor theory’ suggested by Bloom et al. (2018). I assume the majority of imports from a low-wage country such as China are considered low-tech compared to U.S. standards. As a result, an increase in competition from abroad reduces profits and innovation incentives for low-tech firms in high competition markets. Autor et al. (2017) and Hashmi (2013) also empirically find a negative relationship between foreign competition and innovation in U.S. firms but they do not look at the level of technology of firms. For high-tech firms, low-tech imports from low-wage countries free up labor that was initially ‘trapped’ in low-tech tasks (trapped factor theory). Consequently, the opportunity cost of high-tech tasks decreases, labor is reallocated towards these tasks, and innovation increases. My findings are also in line with Bernard et al. (2006)’s conclusions that firms alter their mix of products when facing pressure from low-wage country imports. Zhang (2017) empirically documents a positive effect of Chinese imports on innovation in the U.S. but does not use tariff cuts or differentiate between high- and low-tech firms. To the best of my
knowledge, no other paper explains different competition effects by applying different theories to high- and low-tech firms.
Another contribution of my paper to current literature is that I use a quasi-natural experiment of tariff cuts to establish causality between import competition and innovation. Previous research uses the import penetration ratio to measure the effect of changes in foreign competition on wages and employment (Acemoglu et al. 2016, Autor et al. 2014, Autor et al. 2013). The import penetration ratio is the ratio of imports to domestic consumption and relies on the assumption that the increase in imports is driven by China’s internal supply shocks. Even if most of the Chinese imports are supply-driven, if firms use imports from low-wage countries as intermediate inputs, it is likely that innovation is both a cause and an effect of imports1. Simultaneity is one of the main challenges in estimating the effect of foreign competition on innovation. Following from Frésard and Valta (2016), Flammer (2015) and Frésard (2010), this paper offers an exogenous measure for foreign import competition to ensure that an increase in competition causes changes in innovation rather than innovation causing imports to increase.
Finally, this paper contributes to current literature by being the first to use firm-level patent data past 2006 to study the effects of competition on innovation. I thoroughly filter company names stated on patent application forms and match them to a company identifier (PERMNO) using patent data up to 2010. As a result, I increase the number of PERMNO matches from 22.5% to 66.7% of observations for data from 1991 to 2017. I use this data to include firm innovation responses to recent tariff cuts rather than only responses of more than a decade ago.
The outline of the rest of the paper is as follows. In section 2, I summarize existing theories and empirical studies related to innovation and competition and propose hypotheses to contribute to the literature. Section 3 outlines the methodology of the paper including a detailed research design for establishing a causal relationship. This is followed by section 4, an overview of the data and descriptive statistics for the treated and matched firms. In section 5, I interpret the main results of the paper, and verify these results with various robustness checks in section 6. Finally, I conclude with a discussion of the results, limitations, implications and a future research agenda in section 7.
1 The authors use Chinese imports in the U.K. as an instrument for Chinese imports in the U.S. However, imports
must still be supply-driven for the instrument to be exogenous. For example, an increase in innovation in the U.S. might cause U.K. firms to import more from the U.S. and less from China. Therefore, unless imports are only supply-driven, there is a simultaneity bias.
2. Literature Review
Recent research provides insights into the effects of trade liberalisation and the resulting rise of import competition on firms around the globe. Acemoglu et al. (2016), Pierce and Schott (2016) and Autor et al. (2013, 2014) find a negative effect of Chinese import competition on employment and wages in the U.S. Autor et al. (2013) claim that the import competition causes one fourth of the aggregate decline in employment in the manufacturing industries between 1990 and 2007. Moreover, there is evidence of negative effects of foreign competition on sales growth, profitability (Hombert and Matray, 2017) and capital investment (Frésard and Valta, 2016). More specifically, Frésard and Valta (2016) find that firms reduce capital investments by 7.2% when the threat of foreign entrants increases. Although the evidence of these negative effects of trade policy is conclusive, the effect on firm innovation is unclear.
To start understanding the effects of rising import competition on domestic innovation, I review the existing theories. The Schumpterian hypothesis theorizes that an increase in competition decreases firm profits, which results in lower incentives to innovate (Schumpeter, 1942). Theoretical models developed by Aghion and Howitt (1992) and Dasgupta and Stiglitx (1980) both support the Schumpterian hypothesis of a negative monotone relationship between competition and innovation. On the other hand, the escape-competition effect claims that competition increases incentives to innovate in order to out-perform competition and have higher profits as a market leader (Arrow, 1962). Aghion et al. (2005) consider a non-linear relationship and find evidence for an inverted U-shaped relationship between competition and innovation. More specifically, when competition is low, the escape-competition effect dominates, resulting in higher levels of innovation when competition increases. Thus, for low levels of competition, the curve is increasing, at a decreasing rate. However, as competition increases, at some point the Schumpterian effect dominates as post-innovation rents are no longer higher than pre-innovation rents and innovation incentives decrease. This results in the downward-sloping part of the inverted U-shaped relationship. Figure 1 is an illustration of Aghion et al. (2005)’s inverted U-shaped relationship.
Polder and Veldhuizen (2012) and Aghion et al. (2005) document empirical support for the inverted U-shaped relationship using firm data from the U.K. and the Netherlands. They use the Lerner Index (price-cost margin) to measure competition and patents and R&D to measure innovation. While they do not use trade-based measures of competition, Aghion et al. (2005) focus on changes of market competition around policy reforms including trade liberalization. Other researchers find evidence against the inverted U-shaped relationship. For example, using the same dataset as Aghion et al. (2005), Correa (2012) show that the inverted U-shaped relationship is actually a result of positive correlation for the period 1973 to 1982 and a flat, non-significant relationship for the period 1983 to 1994. More importantly, when including trade-based measures of competition such as imports, quotas and tariffs, there is still no empirical consensus on whether the relationship is positive (Beneito et al. 2017, Chen et al. 2017, Zhang 2017, Bloom et al. 2016, Correa and Ornaghi 2014, Gorodnichenko et al. 2010), negative (Autor et al. 2017, Nobuaki and Isamu 2017, Liu and Qiu 2016, Hashmi 2013) or inverted U-shaped (Chakravorty et al. 2017, Im et al., 2015, Michiyuki and Shunsake 2013).
In order to contribute to the debate on the relationship between import competition and innovation, it is important to highlight why competition from abroad may have different effects than domestic competition. Although the Schumpterian and escape-competition hypotheses are important, these theories do not take into consideration additional effects on innovation caused by
Figure 1. Illustration of Aghion et al. (2005)’s inverted U-shaped
relationship between innovation and competition resulting from the escape-competition and Schumpterian effects.
trade liberalisation. Not many studies acknowledge that the disagreement on the relationship is because trading with a low-wage country has different effects on innovation than trading with a high-wage country. Autor et al. (2014) document a negative effect of import competition from a low-wage country on wages and employment, especially for low-wage workers. However, regarding innovation, trading with a low-wage country can have positive and negative effects. On the one hand, imports as a result of offshoring to a country with lower labor costs can reduce the capacity to innovate since production is moved away from R&D labs (Branstetter et al., 2017). On the other hand, firms can use imports from low-wage countries as intermediate inputs to increase their productivity. The increase in productivity is likely to have a positive effect on innovation and affect firms more than the potential negative effect from offshoring.
Nobuaki and Isamu (2017) find an aggregate negative effect of Chinese import competition on Japanese firms, but positive effect for a subsample of globally engaged Japanese firms. This could indicate that these firms are more productive through the use of imports as intermediate inputs. More specifically, Bloom et al. (2018) recently suggested the ‘trapped factor theory’, that low-tech imports from low-wage countries free up labor that was initially ‘trapped’ in low-tech tasks. This causes the opportunity cost of high-tech tasks to decrease and labor to be reallocated towards these tasks. Consequently, the innovation in these firms increases. In other words, since firms cannot compete with low-wage countries on wages for low-tech tasks, they protect themselves by innovating (Hombert and Matray 2017, Zhang 2017). This theory is in line with Bernard et al. (2006)’s findings that firms alter their mix of products when facing pressure from imports from low-wage countries.
Using the ‘trapped factor theory’, I hypothesize that foreign imports from a low-wage country affect high- and tech firms differently. High-tech firms can use imports from low-wage countries as substitutes for intermediate inputs and reallocate labor to innovation activities. In this case, the Schumpterian effect is unlikely to dominate because the final goods of high-tech firms do not face direct competition since these firms are technologically more advanced than the firms in the low-wage country. The firms cannot compete on wages for the low-tech tasks, thus they substitute these intermediate inputs with the imports and devote more labor towards executing high-tech tasks. As a result, I predict that innovation will increase. Therefore, my first hypothesis is stated as,
Hypothesis 1: A decrease in the tariff rate on imports from a low-wage country increases innovation for high-tech firms.
For low-tech firms, an increase in foreign competition from a low-wage country can affect innovation positively or negatively, depending on the degree of domestic competition. In other words, the final goods of low-tech firms directly compete with the foreign imports, and both the Schumpterian and escape-competition can cause changes in the innovation levels. This results in the same inverted U-shaped relationship between innovation and competition as predicted by Aghion et al. (2005). Based on this theory, I state my second hypothesis as,
Hypothesis 2: A decrease in the tariff rate on imports from a low-wage country increases innovation for low-tech firms in less competitive markets and decreases innovation for low-tech firms in more competitive markets.
3. Methodology
3.1. Causality
It is important that I establish causality between import competition and innovation. Previous research uses the import penetration ratio to measure the effect of changes in foreign competition on wages and employment (Acemoglu et al. 2016, Autor et al. 2014, Autor et al. 2013). The import penetration ratio is the ratio of imports to domestic consumption. The assumption behind using this variable is that the increase in imports is driven by China’s internal supply shocks due to productivity increases, factor accumulation and falling trade costs. To overcome the possibility that correlation between innovation and imports is caused by unobserved domestic demand shocks, the authors use Chinese exports to other high-income countries as an instrument for Chinese exports to the U.S. However, for the instrument to be exogenous, exports to other high-income countries must still be supply-driven. For example, if US imports are instrumented using UK imports, then an increase in innovation in the U.S. might cause U.K. firms to import more from the U.S. and less from China. Therefore, unless imports are only supply-driven, there is a simultaneity bias. Moreover, correlated demand or productivity shocks across high-income
countries threatens the identification strategy. For example, a worldwide economic boom would cause positive correlation between imports and innovation and this would bias the results.
Although these assumptions might not be as problematic when measuring the effect on wages or employment, there are important concerns to consider when measuring the effect on innovation. Although most of the Chinese imports are supply-driven, if theory predicts that firms use imports from low-wage countries as intermediate inputs, it is likely that innovation is both a cause and an effect of imports. Simultaneity is one of the main challenges in estimating the effect of foreign competition on innovation. Following from Frésard and Valta (2016), Flammer (2015) and Frésard (2010), I use an exogenous measure of increased foreign import competition to ensure that increased competition causes changes in innovation rather than innovation causing imports to increase. More specifically, I use significant reductions in import tariff rates, which reduce entry costs of foreign firms, as a measure of increases in foreign competition. The only other innovation-related paper that uses Frésard (2010)’s research design is Im et al. (2015). My paper differs from Im et al. (2015) in that I specifically measure tariff reductions of imports from a low-wage country, while Im et al. (2015) look at average worldwide tariffs. Moreover, my paper is different from Im et al. (2015) in that I specifically match treated firms to controls and compare effects in high- and low-tech firms, based on their R&D intensity.
3.2. Research Design
Based on the research design of Frésard and Valta (2016), Flammer (2015) and Frésard (2010), the baseline specification of this paper is the following difference-in-difference model,
∆"#,%,& = ()*+,%,&+ .#,/,01)2 + 3&+ 4#,%,& (1)
∆"#,%,& is the difference between average innovation in the five years after t and the average innovation in the five years before t, for firm i in industry j 2. Innovation is measured as the natural logarithm of one plus the number of patent applications filed by the firm in that year, since patent counts are often severely skewed to the right (Im et al., 2015). *+,%,& is a dummy variable equal to one for firms in industries that experience a significant tariff cut on imports from China in the previous year (treated group) and zero for all matched firms. I exclude treated and matched firms that experience significant tariff reductions within three years of the event to ensure that innovation
2 The averages are based on all non-missing values of innovation. Observations without innovation data in the year
data is not affected by other events. Since tariff rates fluctuate yearly and changes can be very small and economically insignificant, I consider a decrease in the tariff rate from 5 − 1 to 5 to be significant when it is larger than the three times the average yearly change in tariff rate for that industry from 1989 to 20153. To check the robustness of my results I also use a threshold of two times the industry average and four times the industry average.
As in Frésard and Valta (2016), I match every firm-year observation with a significant tariff cut (treatment) to a control firm based on similar characteristics in the year before the treatment. More specifically, matching, with replacement, occurs by simultaneously minimizing the Mahalanobis distance between size (natural logarithm of assets), investment opportunities (Tobin’s Q), cash holdings, cash flow, and long-term leverage. I explain these covariates in more detail in section 4.4. Mahalanobis distance is the distance between observations in terms of standard deviations of the covariates. As a robustness check, I use propensity score matching instead of nearest-neighbor matching (Mahalanobis).
.#,/,01)2 is a set of time-varying firm characteristics that may affect innovation. These
controls are measured in the year before the event and include firm size, profitability, cash flow to assets, capital intensity and Tobin’s Q. I expect that firm size has a positive effect on innovation due to economies of scale since large firms can spread innovation costs over more units of output (Nobuaki and Isamu, 2017). Profitability should also be positively correlated with innovation as higher return on investment encourages investment and provides the firm with the means to invest in R&D. Similarly, a firm with higher cash flow to assets is more capable of investing in R&D and thus more likely to innovate. I expect capital intensity to have a negative influence on innovation since a lower intensity allows a firm to more easily shift labor resources towards innovative activities. Lastly, I expect that Tobin’s Q has positive correlation with innovation as it measures investment opportunities and can capture potential future innovations. 3& captures year fixed
effects to control for macroeconomic influences such as a recession year or common trends. I do not use firm fixed effects since I already exclude unobserved time-invariant heterogeneity among firms by taking the difference in innovation ( ∆"#,%,&) for each firm specifically. To control for
potential correlation between observations, the error term, 4#,%,&, is clustered at the industry-year
level.
To test the hypotheses of the paper, I regress the baseline specification separately for samples of high- and low-tech firms. These samples are based on the classification of the Organization for Economic Co-operation and Development (OECD). More specifically, low-tech firms are those with an R&D intensity up to 3.0% and high-tech firms are those with an R&D intensity above 3.0% (Hirsch-Kreinsen et al., 2006). R&D intensity is measured as R&D expenses divided by turnover. To test hypothesis 1, I regress equation (1)for firms with an R&D intensity above 3.0% and expect () to be positive and significant. () is interpreted as the difference in the
change in innovation between treated and control firms. To test hypothesis 2, I divide the low-tech firms into three groups based on the intensity of the market competition, as in Im et al. (2015). My results are robust to using five groups instead of three. Market competition is measured using the price-cost margin, also known as the Lerner Index. I include dummy variables into the baseline specification to estimate the different effects of tariff reductions on innovation in firms with different degrees of market competition. More specifically, I estimate the following difference-in-difference model to test hypothesis 2,
∆"#,%,& = ()*+,%,& + (8(*+,%,&× ;<=#,&) + (?(*+,%,& × @"A@#,&) + .#,/,01)2 + 3 &
+ 4#,%,& (2)
;<=#,& is a dummy variable equal to one if the level of competition in the firm is below the first tercile of competitiveness (low-competition) and equals zero otherwise4. In other words, ;<=
#,& =
1 for the lowest 33.3% of observations, in terms of market competition. @"A@#,& is a dummy variable equal to one if the level of competition in the firm is above the second tercile of competitiveness (high-competition). The interaction term for firms above the first and below the second tercile is not included since this medium-competition group of firms is considered a reference group. To test hypothesis 2, (8 should be positive and significant while (? should be negative and significant. This would indicate that there is a non-linear relationship between innovation and import competition from a low-wage country for low-tech firms. I interpret (8 as the difference in the change in innovation between treated and control firms in low-competition markets. Similarly, I interpret (? as the difference in the change in innovation between treated and control firms in high-competition markets.
4 I assign firms to terciles based on the average Lerner Index competition measure in the three years before the tariff
4. Data
I collect data for U.S. public firms in the Compustat North America database from 1989 to 2017 and exclude data observations for regulated utility firms (SIC codes 4900 to 4999) and for financial firms (SIC codes 6000-6999). Moreover, I drop firm-year observations with negative or missing assets, sales, employees and capital stock from the dataset. Data from other datasets are merged into the main Compustat North America dataset by year and GVKEY company identifier, SIC or NAICS. I use historical PERMNO-GVKEY links from the Compustat-CRSP Merged database to merge datasets that only report PERMNO.
4.1. Innovation Measure
Innovation is measured as the total number of patent applications in a given year. Firm-year observations for the number of patents are collected using data from Kogan et al. (2017)5 and the United States Patent and Trademark Office (USPTO), whereby year refers to the patent application year. I use the firm-year patent data from 1926 to 2010, provided by Kogan et al. (2017), to identify company (PERMNO) to patent number links. I then merge these links by patent number into the USPTO data. For each firm, the resulting company name to PERMNO link is extrapolated to other years in which PERMNO is missing. Since company names are often spelled differently on applications, the standard words ‘Incorporated’, ‘Corporation’ and ‘Limited’, abbreviations such as ‘Corp’, ‘LLC’ and ‘INC’ and symbols such as a hyphen are removed to increase the chances of a match. The extrapolation increases the number of PERMNO matches from 22.5% to 66.7% of observations for data from 1991 to 2017 (Table A.1 in the appendix).
Within the research of competition and innovation, this paper is the first to use patent application data past 2006. I only include granted patents but use the application date to document the year of innovation since this more accurately reflects the time at which the firm innovates. As a result, the most recent years suffer from lower patent observations since many patents have not been granted but are likely to be granted in the near future. This is often referred to as the truncation problem in patent data. According to the USPTO’s FY 2017 report, it takes about two years on average for a patent to be granted or refused (USPTO, 2017). Therefore, I exclude data from 2015 to 2017.
4.2. Market Competition Measure
I measure the degree of market competition using the cost margin (Lerner Index). The price-cost margin for firm i in year t is calculated based on the following equation from Aghion et al. (2005),
CD#,& =
EFGHI5DJK FHELD5#,& − LDJIJMDIC MEN5#,&
NICGN#,& (3)
I use data for these variables from the Compustat North America database. Financial cost is calculated as the cost of capital of the firm times the capital stock6. The measure of competition is one minus the average price-cost margin of firms in the same four-digit SIC industry.
MEOF#,& = 1 − 1
P%&Q CD#,&
#∈%
(4) I create the competition measure using the entire sample of U.S. firms from the Compustat data and not just the firms with innovation data.
4.3. Foreign Import Competition Measure
I measure an increase in Chinese import competition in the U.S. using reductions in tariff rates on Chinese imports. Import data, originating from the U.S. Census Bureau, is collected from Peter Schott’s database for the years 1989 to 20157. HS-level import data from China, provided by Schott (2008), is aggregated by NAICS industry code and year. Following Frésard and Valta (2016) and Frésard (2010), I calculate the ad valorem tariff rates as the duties collected by U.S. customs divided by the Free-on-Board value of Chinese imports. I determine that a decrease in the tariff rate from 5 − 1 to 5 is significant when it is larger than the three times the average yearly change in tariff for that industry from 1989 to 20158. To check the robustness of my results I also use a threshold of two times the industry average and four times the industry average. Similar to Frésard and Valta (2016), I exclude industries that experience tariff rate increases at least as large as the cut within the following three years to mitigate concerns that reversal of treatment affects the results. Moreover, I exclude observations for industries in which tariff rates are smaller than 1%.
6 As in Aghion et al. (2005), the cost of capital is assumed to be 0.085 for all firms and time periods.
7 Data retrieved from http://faculty.som.yale.edu/peterschott/sub_international.htm.
4.4. Matching Covariates and Controls
I use data from the Compustat North America database for the control variables of my regressions and for the covariates used to match treated firms to control firms. These variables include firm size (natural logarithm of number of employees), profitability measured by the return on investment (earnings before interest and taxes divided by invested capital), cashflow to total assets, capital intensity (natural logarithm of capital, divided by number of employees) and Tobin’s Q. I measure Tobin’s Q as the market value of assets (total assets minus common equity plus market value of equity) to book value of assets.
4.5. Treated Firms
After merging the tariff data with the Compustat data, I identify 201 unique industry-year observations in which a significant tariff cut occurs. This results in a total of 7,966 firm-year observations with significant tariff reductions. In addition to excluding the treatments which are reversed in the following three years, I eliminate events with significant tariff reductions occurring in the three years before and three years after the event to exclude multiple treatments. Omitting the reversal of treatments and multiple treatments mitigates concerns that the results are driven by pre- or post-event interferences. Moreover, similar to Frésard and Valta (2016) and Flammer (2015), I drop observations without innovation data in the year before or the year after the event to conduct the difference-in-difference estimation9. After applying this selection criteria, I am left with 677 unique firm-year observations with significant tariff cuts, covering 132 different industries and occurring between 1993 and 2012. Frésard and Valta (2016) cover a larger time period and identify 1,116 treated firms and Flammer (2015) covers a smaller time period and identifies 254 treated firms. In comparison with these papers, the number of treated firms in my analysis is not unusual. Approximately one fourth of the events occur between 2000 and 2002, due to the U.S.’ conferral of permanent normal trade relations (PNTR) on China in 2000 and China’s entry into the WTO in 2001. Trade reductions in the other years are part of the multilateral trade agreements established by the WTO (Flammer, 2015).
9 I include observations with missing innovation data in the years 5 − 2 to 5 − 5 or 5 + 2 to 5 + 5. I compute the
4.6 Matched Firms
Each of the 677 firm-year observations with treatment is matched to a controlling firm in a different industry with similar characteristics in the year before the tariff cut. Since tariff reductions occur at the industry level, it is not possible to create matches within the same industry. Matching, with replacement, occurs by simultaneously minimizing the Mahalanobis distance between firm size, profitability, cash flow to assets, capital intensity and Tobin’s Q. Mahalanobis distance is the distance between observations in terms of the standard deviations of the covariates. As a robustness check, propensity score matching is done instead of Mahalanobis (nearest-neighbor) matching. The 677 matched firms include 97 different industries.
4.7 Descriptive Statistics
Table 1 displays the descriptive statistics of innovation and the control variables for the samples of treated and matched firms in the year before treatment. I measure innovation as the natural logarithm of one plus the number of patent applications. Therefore, the average number of patent applications in the year before treatment is 12.07 for treated firms and 10.25 for control firms. These averages are quite low for the time period 1993 to 2012, suggesting that either tariff cuts only affect low-patenting firms or there are limitations to my patent data. I return to this concern in my discussion of the limitations in section 7. By looking at the standard deviation, minimum and maximum values of the two samples for the various variables, I notice that treated firms vary more in firm size and profitability and include a wider range of investment opportunities (Tobin’s Q) than matched firms. I conduct a two-sample difference-in-means test (t-test) to see if the means of the samples are significantly different for each variable. The p-values of these tests are reported in the last column. Since all p-values are greater than 0.10, ranging from 0.12 to 0.85, I do not reject the null hypothesis of equal means of the treated and matched samples for each variable. Therefore, treated firms would likely have acted as the matched firms if treatment were absent.
Table 1. Descriptive Statistics for Innovation and Control Variables for Treated and Matched U.S. Firms
This table reports descriptive statistics (mean, median, standard deviation, minimum, maximum and observation count) for the control variables and innovation for the 677 treated firms and 677 matched firms (including replacements) in the year before treatment. Treated firms are firms that experience a tariff cut that is at least three times larger than the industry average tariff cut between 1989 and 2015. Treated firms are matched to control firms by simultaneously minimizing the Mahalanobis distance between firm size, profitability, cash flow to assets, capital intensity and Tobin’s Q. Innovation is measured as the natural logarithm of one plus the number of patent applications. The control variables include firm size (natural logarithm of number of employees), profitability measured by the return on investment (earnings before interest and taxes divided by invested capital), cashflow to total assets, capital intensity (natural logarithm of capital, divided by number of employees) and Tobin’s Q. I measure Tobin’s Q as the market value of assets (total assets minus common equity plus market value of equity) to book value of assets. The ‘p-value (t-test)’ column reports the p-value for the difference-in-means test with the null hypothesis of equal means. *, ** and *** denote statistical significance at the 10%, 5% and 1% level, respectively.
Mean Median Standard
deviation Minimum Maximum Observations
p-value (t-test)
Firm size Treated 0.99 0.99 2.02 - 4.42 6.15 677 0.85
Matched 1.09 1.08 1.89 - 3.61 6.09 677
Profitability Treated 0.08 0.12 0.41 - 8.15 1.24 677 0.12
Matched 0.08 0.13 0.35 - 5.79 1.15 677
Cash flow to assets Treated 0.00 0.04 0.20 - 2.30 0.30 677 0.53
Matched 0.01 0.05 0.18 - 2.08 0.33 677
Capital intensity Treated 4.48 4.42 0.71 1.70 6.96 677 0.26
Matched 4.49 4.43 0.67 2.37 7.27 677
Tobin’s Q Treated 2.04 1.56 1.59 0.26 22.2 677 0.73
Matched 1.94 1.55 1.39 0.59 20.0 677
Innovation Treated 2.57 2.30 1.49 0.69 7.99 677 0.17
4.8. Innovation around a Tariff Reduction for Treated and Matched Firms
Figure 2 is an illustration of the average innovation of treated and matched firms from 5 years before (! = −5) until 5 years after (! = 5) a significant tariff rate reduction. The reduction occurs at ! = 0 and ! is event time measured on the x-axis. On the y-axis I measure average innovation as the natural logarithm of one plus the number of patents. Therefore, a value of 2.4 on the y-axis is approximately equal to 10 patents and a value of 3.0 on the y-axis is approximately equal to 19 patents.
The two samples of firms follow a similar innovation trend before the tariff cut. However, in the five years after the event, average innovation increases for the treated firms and stays relatively constant for the matched firms. It is likely that more innovative industries are more often affected by tariff reductions since the innovation level of the treated firms is higher than for matched firms in all years surrounding the event. However, since innovation levels before the event are not significantly different for the samples (see t-test results from Table 1), it is not likely to bias the results in section 5. Section 5 investigates whether the post-event difference in innovation between the samples is significant compared to the pre-event difference in innovation.
2 2.2 2.4 2.6 2.8 3 A ve ra g e I n n o va ti o n -5 -3 -1 1 3 5 Event Time (t) Treated Matched
Figure 2. This figure illustrates the average innovation of treated and matched firms from
5 years before (! = −5) until 5 years after (! = 5) a significant tariff rate reduction (at t = 0). Treated firms are those that experience a tariff cut which is at least three times larger than the industry average tariff cut between 1898 and 2015. Innovation is measured as the natural logarithm of one plus the number of patent applications in a given year.
5. Results
My results indicate that there is strong evidence that a decrease in the tariff rate on imports from a low-wage country increases the innovation in U.S. public firms. I present the regression results for equation (1) in Table 2 for the full sample of firms. All regressions included year fixed effects and robust standard errors clustered at the year-industry level. In columns (1) and (2) the dependent variable is the average innovation in the five years before to five years after the tariff cut (10-year window). '()*,, is a dummy variable equal to one if the firm is in an industry that experienced a
significant tariff reduction in the past year. In column (1), the coefficient on '()*,, is positive and significant, indicating that there is a positive effect of tariff cuts on innovation. When adding the control variables in column (2), the effect decreases slightly and remains significant. In column (3), I include control variables and use the average innovation in the six years before to six years after the event (12-year window) as the dependent variable. In column (4), the regression includes control variables and the dependent variable is the average innovation in the four years before to four years after the event (8-year window). Compared to the 10-year window in column (2), the effect on innovation is slightly higher and more significant for the 12-year window and very similar for the 8-year window. The coefficients and t-statistics for the control variables are reported in Table A.2 in the appendix. (columns (1) to (3)). These results indicate a positive and significant effect of Tobin’s Q on innovation and a negative and significant effect of capital intensity on innovation. I expect these effects since higher investment opportunities (Tobin’s Q) can capture potential future innovations and lower capital intensity means relatively more labor available to shift from production activities to innovation activities. As in Frésard and Valta (2016), I match with replacement and this results in 515 distinct firm matches for the 677 treated firms. This explains why the total number of observations in the results is 1,192 rather than two times 677 (1,354) firms.
The coefficients on '()*,, for the regressions in Table 2 lie between 0.1070 and 0.1168 and are all significant. Although this result seems modest, it is economically significant in relative terms. More precisely, since innovation is measured in terms of natural logarithms, treated firms increase their number of patent applications by about 11.3% to 12.4% more than untreated firms, following a tariff cut on Chinese imports10. Zhang (2017) finds that an increase in the Chinese
import penetration ratio of one standard deviation increases innovation in U.S. firms by 39.2%. Due to the different measures of import competition, it is difficult to directly compare my results to those of Zhang (2017).
I find evidence for hypothesis 1 using the results in Table 3, which presents the estimations of equation (1) for high-tech firms (columns (1) to (4)) and low-tech firms (columns (5) to (8)). All regressions included year fixed effects and robust standard errors clustered at the year-industry level. As in Table 2, the results include several different specifications. In columns (1), (2), (5) and (6), the dependent variable is the average innovation in the five years before to five years after the tariff cut (10-year window). Columns (3) and (7) include a 12-year window and columns (4) and (8) have an 8-year window. Controls are included in all regressions except for columns (1) and (5). For high-tech firms, firm size and Tobin’s Q have a positive and significant effect on innovation, as expected, while the coefficients on the controls are insignificant for the low-tech firms (Table A.2 in the appendix, columns (4) to (8)). For high-tech firms, the coefficient on '()*,, is positive, significant and stable when I include controls or use a smaller or larger window of years for the dependent variable. Compared to the results for all firms in Table 2, the coefficient on '()*,, is higher for high-tech firms. For low-tech firms, the coefficient on '()*,, is smaller and insignificant, regardless of the specification. I infer that the positive effect of tariff cuts on innovation observed in Table 2 is driven by the positive effect of tariff cuts on innovation in high-tech firms. More precisely, treated high-high-tech firms increase their number of patent applications by about 12.7% to 14.0% more than untreated firms, following a tariff cut on Chinese imports. I conclude that this result provides evidence for hypothesis 1, that a decrease in the tariff rate on imports from a low-wage country increases the innovation of high-tech firms.
Table 2. The Effect of a Tariff Reduction on Chinese Imports on Innovation of U.S. Firms
This table presents the effect of a reduction of a tariff rate on Chinese imports (treatment) on innovation of U.S. firms. Equation (1) is regressed for the entire sample of firms. "#$%,' is a dummy variable equal to one if the firm is in an industry that experienced a significant tariff reduction in the
past year. The dependent variable in columns (1) and (2) is the change in average innovation in the 5 years before to 5 years after the event. Column (3) measures the change in average innovation in the 6 years before to 6 years after the event. Column (4) measures the change in average innovation in the 4 years before to 4 years after the event. Innovation is measured as the natural logarithm of one plus the number of patent applications. Controls are not included in the regression for column (1) and are included in columns (2) to (4) (the coefficients and t-statistics of controls are reported in Table A.2). All regressions include year fixed effects and standard errors are clustered at the year-industry level. T-statistics are reported in parentheses and *, ** and *** denote statistical significance at the 10%, 5% and 1% level, respectively.
All Firms (1) ∆)-5/+5 (2) ∆)-5/+5 (3) ∆)-6/+6 (4) ∆)-4/+4 "#$%,' 0.1122** (2.60) 0.1070** (2.48) 0.1168*** (2.61) 0.1075** (2.60)
Controls no yes yes yes
Year FE yes yes yes yes
Adj. R+ 0.0842 0.1032 0.1037 0.1003
Table 3. The Effect of a Tariff Reduction on Chinese Imports on Innovation of High- and Low-Tech U.S. Firms
This table presents the effect of a reduction of a tariff rate on Chinese imports (treatment) on innovation of high-tech firms (columns (1) to (4)) and low-tech firms (columns (5) to (8)). The results are based on equation (1). High-tech firms are those with a R&D intensity above 3%, and low-tech firms are those with a R&D intensity below 3%. "#$%,' is a dummy variable equal to one if the firm is in an industry that experienced a significant
tariff reduction in the past year. The dependent variable in columns (1), (2), (5) and (6) is the change in average innovation in the 5 years before to 5 years after the event. Columns (3) and (6) measure the change in average innovation in the 6 years before to 6 years after the event. Columns (4) and (8) measure the change in average innovation in the 4 years before to 4 years after the event. Innovation is measured as the natural logarithm of one plus the number of patent applications. Controls are not included in the regression for columns (1) and (5) and are included in columns (2) to (4) and columns (6) to (8) (the coefficients and t-statistics of controls are reported in Table A.2). All regressions include year fixed effects and standard errors are clustered at the year-industry level. T-statistics are reported in parentheses and *, ** and *** denote statistical significance at the 10%, 5% and 1% level, respectively.
High-tech Firms Low-tech Firms
(1) ∆)-5/+5 (2) ∆)-5/+5 (3) ∆)-6/+6 (4) ∆)-4/+4 (5) ∆)-5/+5 (6) ∆)-5/+5 (7) ∆)-6/+6 (8) ∆)-4/+4 "#$%,' 0.1283** (2.18) 0.1228** (2.11) 0.1197** (1.98) 0.1313** (2.33) 0.0520 (0.92) 0.0551 (0.97) 0.0822 (1.40) 0.0452 (0.82)
Controls no yes yes yes no yes yes yes
Year FE yes yes yes yes yes yes yes yes
Adj. R+ 0.0825 0.1198 0.1219 0.1178 0.0987 0.0967 0.0969 0.0835
The results in Table 4 suggest that there is no evidence of a non-linear relationship between innovation and competition for the full sample of firms. I regress equation (2) with several specifications. All regressions included year fixed effects and robust standard errors clustered at the year-industry level. Control variables are included in columns (2) to (4). The coefficients and t-statistics for the control variables are reported in Table A.3 in the appendix (columns (1) to (3)). Similar to my results in Table 1, capital intensity is negatively related to innovation and Tobin’s Q is positively related to innovation. In columns (1) and (2), the dependent variable is the average innovation in the five years before to five years after the tariff cut. In column (3), the dependent variable is the average innovation in the six years before to six years after the event. In column (4), the dependent variable is the average innovation in the four years before to four years after the
event. A positive and significant value of the coefficient on
"#$%,' × )*+,,' and negative and significant value of the coefficient on "#$%,'× -./-,,' would indicate a non-linear relationship between competition and innovation such as the inverted-U shaped relationship suggested by Aghion et al. (2005). I find, however, that when including all firms, both 01 and 02 are insignificant, indicating that the degree of market competition does not matter for the effect of tariffs on innovation. This result holds for all specifications in Table 4. The only other existing research using innovation and tariff cuts for different degrees of market competition is done by Im et al. (2015), who do find evidence for the inverted U-shaped curve when using U.S. data. Following a tariff cut, Im et al. (2015) document an increase in the market value of innovation of 3% in terms of excess returns for firms with low market competition and a decrease in market value of 3.4% for firms with high market competition. The inconsistency between our results is due to different measures of tariffs, different matching methods and different measures of innovation.
Using the results from Table 5, I conclude that there is partial evidence for hypothesis 2, which states that a decrease in the tariff rate on imports from a low-wage country increases innovation in low-tech firms in less competitive markets and decreases innovation for low-tech firms in more competitive markets. Table 5 presents the regression estimates of equation (1) for high-tech firms (columns (1) to (4)) and low-tech firms (columns (5) to (8)). All regressions included year fixed effects and robust standard errors clustered at the year-industry level. Similar to the specifications in Table 3, the dependent variable in columns (1), (2), (5) and (6) is the average innovation using a 10-year window. Columns (3) and (7) have a 12-year window and columns (4)
and (8) have an 8-year window. Controls are included in all regressions except in columns (1) and (5). For high-tech firms, firm size and Tobin’s Q have positive and significant effects on innovation, as expected, while the coefficients on the controls are insignificant for the low-tech firms (Table A.3 in the appendix, columns (4) to (9)). For the sample of high-tech firms, there is also no evidence of an asymmetric relationship. More specifically, the coefficients on "#$%,' × )*+,,' and "#$%,'× -./-,,' are both positive and significant in all specifications. This
result supports hypothesis 1 since it indicates that an increase in import competition from a wage increases innovation in high-tech firms, irrespective of whether the firm is in a high- or low-competition market. Although the results for low-tech firms are consistently insignificant in Table 3, an interesting result arises in Table 5. In columns (6) and (7), I find an insignificant coefficient on "#$%,'× )*+,,' and a negative and significant coefficient on "#$%,'× -./-,,'. While I find
no evidence for the positive effect of tariff cuts on innovation of low-tech firms in low-competition markets, I do find evidence for the negative effect of tariff cuts on innovation of low-tech firms in high-competition markets. Therefore, I find partial evidence for hypothesis 2 and conclude that the Schumpterian effects dominates for low-tech firms in high-competition markets.
According to the Schumpterian effect, the increase in competition decreases firm profits, which results in lower incentives to innovate. It may take some time for the foreign competition to enter, decrease the profits and drive down firms’ innovation incentives. This may explain why the effect is insignificant when I measure innovation up to four years after the tariff cut (column (8)), becomes significant and more negative as I increase the innovation data window to five years (column (6)) and six years (column (7)) after the event. The coefficients on "#$%,'× -./-,,' for columns (6) and 7) lie between -0.2139 and -0.2604. This effect is economically quite significant. More specifically, low-tech firms in high-competition markets decrease their number of patent applications by about 19.3% to 22.9% more than untreated low-tech firms in high-competition markets, following a tariff cut on Chinese imports. Hashmi (2013) also empirically supports the Schumpterian hypothesis and documents a mildly negative relationship using U.S. data. My research differs in that Hashmi (2013) uses source-weighted industry exchange rates instead of tariff cuts and does not look separately at high- and low-tech firms.
Table 4. The Effect of a Tariff Reduction on Chinese Imports on Innovation of U.S. Firms in High- and Low-Competition Markets
This table presents the effect of a reduction of a tariff rate on Chinese imports (treatment) on innovation of U.S. firms with different degrees of market competition. The results are based on equation (2). !"#$,& is a dummy variable equal to one if the firm is in an industry that experienced a significant tariff reduction in the past year. Based on the Lerner Index, '()*,& is a dummy variable equal to one for firms in a low-competition market and +,-+*,& is a dummy variable equal to one for firms is in a high-competition market. The dependent variable in columns (1) and (2) is
the change in average innovation in the 5 years before to 5 years after the event. Column (3) measures the change in average innovation in the 6 years before to 6 years after the event. Column (4) measures the change in average innovation in the 4 years before to 4 years after the event. Innovation is measured as the natural logarithm of one plus the number of patent applications. Controls are not included in the regression for column (1) and are included in columns (2) to (4) (the coefficients and t-statistics of controls are reported in Table A.3). All regressions include year fixed effects and standard errors are clustered at the year-industry level. T-statistics are reported in parentheses and *, ** and *** denote statistical significance at the 10%, 5% and 1% level, respectively.
All Firms (1) ∆,-5/+5 (2) ∆,-5/+5 (3) ∆,-6/+6 (4) ∆,-4/+4 !"#$,& 0.0732 (1.28) 0.0587 (1.03) 0.0759 (1.29) 0.0549 (1.01) !"#$,&× '()*,& 0.0373 (0.57) 0.0468 (0.72) 0.0406 (0.60) 0.0596 (0.96) !"#$,&× +,-+*,& 0.0809 (1.00) 0.1006 (1.24) 0.0844 (1.02) 0.1030 (1.32)
Controls no yes yes yes
Year FE yes yes yes yes
Adj. R2 0.0838 0.1034 0.1033 0.1007
Table 5. Effect of a Tariff Reduction on Chinese Imports on Innovation of High- and Low-Tech U.S. Firms in High- and-Low Competition Markets
This table presents the effect of a reduction of a tariff rate on Chinese imports (treatment) on innovation of high-tech firms (columns (1) to (4)) and low-tech firms (columns (5) to (8)) with different degrees of market competition. The results are based on equation (2). High-tech firms are those with a R&D intensity above 3%, and low-tech firms are those with a R&D intensity below 3%.!"#$,& is a dummy variable equal to one if the firm
is in an industry that experienced a significant tariff reduction in the past year. Based on the Lerner Index, '()*,& is a dummy variable equal to one for firms in a low-competition market and +,-+*,& is a dummy variable equal to one for firms is in a high-competition market. The dependent variable in columns (1), (2), (5) and (6) is the change in average innovation in the 5 years before to 5 years after the event. Columns (3) and (6) measure the change in average innovation in the 6 years before to 6 years after the event. Columns (4) and (8) measure the change in average innovation in the 4 years before to 4 years after the event. Innovation is measured as the natural logarithm of one plus the number of patent applications. Controls are not included in the regression for columns (1) and (5) and are included in columns (2) to (4) and columns (6) to (8) (the coefficients and t-statistics of controls are reported in Table A.3). All regressions include year fixed effects and standard errors are clustered at the year-industry level. T-statistics are reported in parentheses and *, ** and *** denote statistical significance at the 10%, 5% and 1% level, respectively.
High-tech Firms Low-tech Firms
(1) ∆,-5/+5 (2) ∆,-5/+5 (3) ∆,-6/+6 (4) ∆,-4/+4 (5) ∆,-5/+5 (6) ∆,-5/+5 (7) ∆,-6/+6 (8) ∆,-4/+4 !"#$,& 0.0082 (0.11) - 0.0102 (- 0.13) - 0.0136 (- 0.17) 0.0070 (0.09) 0.1098 (1.34) 0.1091 (1.33) 0.1542* (1.84) 0.0784 (0.98) !"#$,&× '()*,& 0.2332** (2.30) 0.1892* (1.87) 0.1980* (1.91) 0.1924* (1.93) - 0.0391 (- 0.43) - 0.0179 (- 0.19) -0.0369 (-0.39) 0.0085 (0.09) !"#$,&× +,-+*,& 0.1763* (1.74) 0.2234** (2.20) 0.2206** (2.10) 0.2021** (2.04) - 0.1905 (- 1.64) - 0.2139* (- 1.87) - 0.2604** (- 2.24) - 0.1673 (- 1.49)
Controls no yes yes yes no yes yes yes
Year FE yes yes yes yes yes yes yes yes
Adj. R2 0.0871 0.1259 0.1275 0.1232 0.1019 0.1021 0.1054 0.0862
6. Robustness Checks
In this section, I discuss several robustness checks to test the consistency of my results. The main results in Tables 2 to 5 barely change when I include control variables and alter the number of years surrounding the event to measure innovation changes. Therefore, in Tables 6 to 8, I focus on the robustness of the results for specifications that include the control variables and use a 10-year window surrounding the event. Coefficients and t-statistics for the control variables for Tables 6 and 7 are reported in Tables A.4 and A.5 in the appendix. My results are robust to alterations, including a different matching method, different cutoffs for whether I consider a tariff reduction significant or not and to using quintiles of the Lerner Index instead of terciles when assigning the firms to groups of market competition.
6.1. Matching
I find that my results are robust to altering the method used to match every treated firm to a control firm. I use propensity score matching instead of Mahalanobis (nearest-neighbor) matching. While the Mahalanobis matching method minimizes the distance between observations in terms of the standard deviations of the covariates (nearest-neighbor), the propensity score matching procedure bases matches on the results of a logistic regression and assesses the likelihood of an observation being in the control or treated group. In Tables 6 and 7, columns (1), (4) and (7), I check the robustness of the results from Tables 2 to 5 and use the same covariates as before11. I find that the coefficients of the regressions have the same sign as in the main results and are also significant when the main results are significant. In fact, for some of the significant results, the level of significance increases.
6.2. Cutoff for Significant Tariff Reduction
My results are also consistent when I change the cutoff for determining whether a tariff reduction is significant or not. In Tables 2 to 5, tariff reductions are considered significant when they are at least three times larger than the absolute value of the average tariff reduction in the industry
11 In Table 6 columns (1), (4) and (7), I check the results from Table 2 column (2) and Table (3) columns (2) and (6),
respectively. In Table 7 columns (1), (4) and (7), I check the results from Table 4 column (2) and Table 5 columns (2) and (6), respectively.
between 1989 and 2015. In Tables 6 and 7, columns (2), (5) and (8), I check the robustness of these results using tariff reduction which are at least four times larger than the industry average12. The results are robust to this change in cutoff. The reason the number of observations increases is not because more tariff cuts are identified but because there are more distinct matched firms (less replacement during the matching process). All regressions have the same sign as the main results. In most cases, the significance level increases, except for Table 7, column (5), when significance is slightly lower.
In Tables 6 and 7, columns (3), (6) and (9), I check the robustness of these results using tariff reduction which are at least two times larger than the industry average13. In this case, the increase in observations is because more tariff reductions are included as significant events since the cutoff is now lower. Again, all regressions have the same sign as the main results and in most cases, the significance level increases.
6.3. 5 Quintiles of Market Competition
Lastly, the main results are consistent when I use different percentiles to divide the data into groups with different degrees of market competition. In my main results (Tables 4 and 5), I assign the data into three different terciles, based on the Lerner Index. !"#$,& is a dummy variable equal to one if the level of competition in the firm is below the first tercile of competitiveness (low-competition) and equals zero otherwise. In other words, !"#$,& = 1 for 33.3% of the firms which have the lowest degree of market competition, and *+,*$,& = 1 for 33.3% of the firms which have the
highest degree of market competition. To test the robustness of the results in Tables 4 and 5, I alternatively assign the data into five quintiles, based the Lerner Index, and create a dummy variable for each group (-1$,&, -2$,&, -3$,&, -4$,& and -5$,&). In this case, -1$,& = 1 for 20% of the
firms which have the lowest degree of market competition, -2$,& = 1 for the next 20% of the firms
with slightly higher degree of market competition and so on. I assign firms to quintiles based on the average competition in the three years before the tariff cut. The regression is the same as in
12 In Table 6 columns (2), (5) and (8), I check the results from Table 2 column (2) and Table (3) columns (2) and (6),
respectively. In Table 7 columns (2), (5) and (8), I check the results from Table 4 column (2) and Table 5 columns (2) and (6), respectively.
13 In Table 6 columns (3), (6) and (9), I check the results from Table 2 column (2) and Table (3) columns (2) and (6),
respectively. In Table 7 columns (3), (6) and (9), I check the results from Table 4 column (2) and Table 5 columns (2) and (6), respectively.
equation (2) except that I now include four interaction terms instead of two. The interaction term for firms in the third quintile is not included in the regression since this medium-competition group of firms is considered a reference group, as in Im et al. (2015). Columns (1), (2) and (3) of Table 8 are robustness checks for Table 4 column (2) and Table 5 columns (2) and (6), respectively.
The results are similar to my main results. When I include all firms, none of the interaction terms are significant. As before, column (3) indicates that innovation does not significantly impact low-tech firms, except for firms in high-competition markets. More specifically, tariff cuts decrease innovation for the 20% of firms with the highest degree of market competition. However, this effect is only significant at the 10% level. However, for the high-tech firms, all of the interaction terms are positive and significant. Interestingly, the coefficients for 3456,& × -2$,&, 3456,&× -4$,& and 3456,&× -5$,& gradually increase in magnitude and significance as the
degree of market competition increases. This suggests that the positive effect of tariff cuts on innovation is stronger for high-tech firms in high-competition markets, compared to firms in markets with less competition.
Table 6. Robustness of the Effect of a Tariff Reduction on Chinese Imports on Innovation of High- and Low-Tech U.S. Firms
This table checks the validity of the effect of a tariff rate reduction (treatment) on innovation of all firms (columns (1) to (3)), high-tech firms (columns (4) to (6)) and low-tech firms (columns (7) to (9)). The results are based on equation (1). !"#$,& is a dummy variable equal to one if the firm is in an industry that experienced a significant tariff reduction in the past year. High-tech firms are those with a R&D intensity above 3%, and low-tech firms are those with a R&D intensity below 3%. The dependent variable is the natural logarithm of one plus the number of patent applications. Columns (1), (4) and (7) use propensity score matching instead of Mahalanobis matching. Columns (2), (5) and (8) include only tariff reductions which are 4 times larger than the industry average and columns (3), (6) and (9) include only tariff reductions which are 2 times larger than the industry average. Controls are included, but not reported in all regressions. All regressions include year fixed effects and standard errors are clustered at the year-industry level. T-statistics are reported in parentheses and *, ** and *** denote statistical significance at the 10%, 5% and 1% level, respectively.
All Firms High-tech Firms Low-tech Firms
(1) PSM (2) Cut >4x (3) Cut >2x (4) PSM (5) Cut >4x (6) Cut >2x (7) PSM (8) Cut >4x (9) Cut >2x !"#$,& 0.1269*** (2.83) 0.1210*** (2.93) 0.1318*** (3.16) 0.1553*** (2.78) 0.1187** (2.19) 0.1894*** (3.57) 0.0468 (0.78) 0.0776 (1.39) 0.0565 (1.01)
Controls yes yes yes yes yes yes yes yes yes
Year FE yes yes yes yes yes yes yes yes yes
Adj. R( 0.0795 0.0944 0.1048 0.1129 0.1148 0.1295 0.0555 0.0895 0.0892
