*4.1. Univariate Analysis*

We conducted a univariate analysis to examine Hypothesis 1. First, we computed the debt financing by Equation (1) to choose those firms whose minimum of debt financing in year 0 was 2% of the amount of firm assets in year −1. Next, for those achieving the minimum of debt financing threshold, we then calculated their difference in leverage by Equation (3). The results are summarized in Table 2.

Panel A shows the results of all sample firms. We find that the difference in leverage (Δ *Leverage*) is positive with the mean value of 3.30% and median value of 3.15%. The result is consistent with Hypothesis 1, suggesting that firms would introduce the ADR cross-listing to increase the leverage ratio to a higher level than before cross-listing. Panel B shows the results of the subsample of leverage-increasing firms (those with positive value of Δ *Leverage*). We find that in year −1, the leverage (*Debt*/*Asset*−1) of cross-listing sample firms is slightly lower than that of their industrial rivals; the mean (median) of leverage for sample firms is 19.12% (18.87%), and their rivals' mean (median) is 19.35% (19.56%). However, in year 0, the mean (median) of leverage (*Debt*/*Asset*0) for the leverageincreasing subsample is 22.54% (23.16%), both being higher than their rivals' mean (17.86%) and median (17.95%). Furthermore, their mean and median of Δ *Leverage* is 7.23% and 7.18%, respectively, both being much higher than those of all samples.

Panel C reports the results of the subsample of leverage-decreasing firms (those with negative value of Δ *Leverage*). In year −1, their mean of leverage (*Debt*/*Asset*−1) is 21.04%, and median is 17.32%, both being higher than their rivals' mean (10.21%) and median (10.52%). By contrast, in year 0, their mean (median) of leverage (*Debt*/*Asset*0) is 10.33% (8.95%), both being slightly lower than their rivals' mean (11.32%) and median (11.05%). Furthermore, for the leverage-decreasing subsample, their mean and median of Δ *Leverage* are −4.18% and −4.52%, respectively, both being lower than those of all sample.

Table 2 also compares the number of leverage-increasing firms (141) versus the number of leverage-decreasing firms (74). The result of greater number of leverage-increasing firms in contrast to leverage-decreasing firms suggests that a majority of firms take advantage of the ADR issuing to increase their leverage ratio to above that of their rivals. Overall, these univariate test results support Hypothesis 1.

We provide the differences in R&D expenditures in Table 3. There are three panels: all firms in Panel A, leverage-increasing firms in Panel B, and leverage-decreasing firms in Panel C, respectively. We measured the unmodified change through the difference in R&D expenditures in year [ −1, 0] divided by firm assets in year −1. The industrial modified difference was equal to the unmodified difference in the sample firm minus the mean of the unmodified difference in the rest of firms in an industry. For Hypothesis 2, we report the results of conducting a univariate analysis in Table 3. These values of all sample analyses are found in Panel A. The unmodified values of R&D expenditures and industrial modified values are reported. We find that the mean of the unmodified differences in R&D expenditure increased by 1.88% (*p* < 0.05) in year [−1, 0], increased by 2.54% (*p* < 0.05) in year [−1, 1] and increased by 3.85% (*p* < 0.05) in year [−1, 2], respectively. These results indicate that firms continued to input R&D expenditures for at least 2 years after raising debt financing.



The mean of industrial modified differences in R&D expenditures is positively significant with the value of 0.83% (*p* < 0.05) in year [−1, 0]. The mean of industrial modified differences in R&D expenditures in year [−1, 1] is 0.98% (*p* < 0.05) and in year [−1, 2] is 1.37% (*p* < 0.05), respectively. Overall, we find that the cross-listing firms' R&D expenditure increases industrial rivals in the cross-listing year and the two subsequent years.

In Panel B, we illustrate the results of leverage-increasing firms. The unmodified difference in R&D expenditures of leverage-increasing samples increased significantly by 2.05% (*p* < 0.05) in year [−1, 0]. In year [−1, 1] and year [−1, 2], the means of unmodified differences in R&D expenditures are significant at 2.68% (*p* < 0.05) and 3.83% (*p* < 0.05), respectively. Further, the mean of industrial modified differences in R&D expenditures in year [−1, 0] is 0.88% (*p* < 0.05). This shows that the leverage-increasing firms increased their R&D expenditures higher than those of their industrial rivals. The mean of industrial modified differences in R&D expenditures in year [−1, 1] is also significantly positive at 1.15% (*p* < 0.05) and maintains the significantly positive value of 1.36% (*p* < 0.05) in year [−1, 2]. These results indicate that during the cross-listing period, the leverage-increasing

firms were more aggressive in raising R&D expenditures than their industrial rivals, which is consistent with Hypothesis 2.


**Table 3.** Differences in R&D expenditures.

The significance levels are according to a two-tailed *t* test by Wilcoxon rank test. \* and + are significant at 5 and 10%, respectively.

For comparison, we also analyzed the subsample of leverage-decreasing firms. The results are reported in Panel C. Compared to the leverage-increasing firms, these leveragedecreasing firms have a positive mean of unmodified differences in R&D investments in year [ −1, 0], which is insignificant at 0.83%. Similarly, the means of industrial modified differences in R&D expenditures of these leverage-decreasing firms in year [ −1, 0] (0.47%), year [ −1, 1] (0.52%) and year [ −1, 2] (0.62%) are all positive but insignificant.

## *4.2. Regression Analysis*

The prior literature [34,35] indicated that it was necessary to control variables that may influence R&D expenditure, including the growth in firm assets, the amount of R&D expenditures in year −1, market share, and several industrial structure variables. It was noteworthy that since R&D investments were found to be a continuous event that may last for 2–3 years after launch, we used three time-window years [ −1, 0], year [ −1, 1], and year [ −1, 2], respectively, for more robust results [36]. We considered the linear regression model from these studies [2,7,14,15,19] as follows:

Δ *R*& *Dit* = *β*0*t*+*β*1*tDdebti*+*β*2*t*log(*<sup>H</sup> index*)*i* <sup>+</sup>*β*3*tNumber of firms in the industryi* <sup>+</sup>*β*4*tMarket share of sample firmi* <sup>+</sup>*β*5*tIndustrial modified amount of R* & *D*−1*i* <sup>+</sup>*β*6*tIndustrial modified growth ratio of firm assets from year* − 1 *to year ti*+*εit*, (4)

The difference in R&D expenditures of all firms was the dependent variable. The industrial modified R&D expenditures from year −1 to year t for firm *i*, where *t* was equal to 0, 1, and 2 depending on different time windows (i.e., time-window year [ −1, 0], year [−1, 1], and year [ −1, 2]). Ddebt took a value equal to 1 in the case of the increases in leverage, and zero otherwise. We named the logarithm of the Herfindahl index as log (H index). The number of firms in the industry was the total number of firms in the industry. We adopted the industrial modified amount of R&D expenditures in year −1 to control the changes of R&D expenditures in these ADR events. We also used the industrial modified growth ratio of firm assets to control the changes in the growth ratio of firm assets in these ADR events. We also measured the standard errors of these coefficients due to heteroscedasticity [37].

The empirical results of the regression analysis are found in Table 4. First, about the effect of control variables, the coefficients of the industrial modified values of R&D expenditures in year −1 are significantly positive throughout all models (0.26, 0.34, and 0.48, respectively). This means that the effect of industrial modified values of R&D expenditures in year −1 on the R&D growth rate is a significantly positive effect (*p* < 0.01). The coefficients of the industrial modified growth ratio of firm assets are all significantly positive in all models (0.12, 0.14, and 0.15). This means that the effect of the industrial modified growth ratio of firm assets on the R&D growth ratio is also a significantly positive effect (*p* < 0.01). Further, the coefficients about number of firms are significantly positive in year [ −1, 1] (0.03, *p* < 0.05) and year [ −1, 2] (0.04, *p* < 0.05), respectively. However, the coefficient of log (H index) is not significant to support the difference in R&D expenditures. The coefficient of market share is also insignificant to support the difference in R&D expenditures. After controlling for possible impacts of the control variables as suggested in the prior literature, the coefficients of debt are significantly positive in three regressions (0.02, *p* < 0.05; 0.03, *p* < 0.01; and 0.05, *p* < 0.01, respectively). We find these consistent results in Table 3, thereby further supporting Hypotheses 1 and 2. This suggests that firms increased their financing leverage to be aggressive in R&D investments in terms of undertaking more R&D investments relative to their industrial rivals. Overall, in Table 4, the results of regression analyses present that the firms whose financial leverages increased during the cross-listing period conduct more vigorous R&D investment strategies after controlling for other variables.


**Table 4.** Regression analysis: differences in R&D expenditures.

T-statistics are in parentheses. \*\* and \* are significant at 1 and 5%, respectively.

To test if the aggressive R&D investment conducted by cross-listing firms using greater financial leverage push their industrial rivals to also enhance their R&D investment in response as suggested by Hypothesis 3, we conducted a further regression analysis to examine rival firms' reactions to these firms' expanding R&D investments. In this regard, we focused on the results of the 141 subsample of leverage-increasing firms and showed that R&D expenditures significantly increased post cross-listing. However, the 74 leverage-

decreasing subsample firms show the insignificant change in R&D investments around cross-listing.

To investigate the effect of competition, the firms in each rival group were separated into two categories: (1) the differences in R&D expenditures of firms from year −1 to year t were over the average in the industry; (2) these differences R&D expenditures of firms from year −1 to year t were under the industrial average. The former category was characterized by relatively aggressive firms, and the latter category comprises relatively passive firms. We examined the financing effect of these firms' financial leverage on their industrial peers and whether this effect makes these rivals aggressive or passive in R&D investments. Specifically, the logistic regression was estimated from these studies [2,7,14,15,19,38] as follows:

*Aggressiveit* = *β*0*t* + *β*1*tIndustrial modified leveragei* + *β*2*tMarket sharei* <sup>+</sup>*β*3*tIndustrial modified amount of R*&*D*−1*i* <sup>+</sup>*β*4*tIndustrial modified growth ratio of firm assets from year* − 1 *to year ti*+ *εit*, (5)

The dependent variable took a value equal to 1 in the case of the differences in R&D investments of rival firms being greater than the average of industrial differences in R&D investments, and zero otherwise. We measured the industrial modified leverage through the unmodified amount minus the average of the unmodified amount in the industry. The other variables had the same definitions in the above equations. The four-digit SIC codes of rival firms and those of the sample firms were the same. To explain the rivals' reactions to these sample firms, we excluded the financing-events if two or more rival firms raised their debt financing in the same year.

We present the probability of rival firms taking aggressive R&D investments in Table 5. The result of logistic regression indicates that the industrial modified leverage of the sample firms is significantly and positively related to the level of rival firms' aggressiveness in these three periods, year [−1, 0], year [−1, 1], and year [−1, 2], thereby supporting Hypothesis 3. The impacts of both initial levels of R&D expenditure and the growth ratios of firm assets are also significantly negative throughout all models (*p* < 0.01). The effects of market share, by contrast, are insignificant in the three models. Overall, the results in Table 5 demonstrate a significant intra-industry contagion effect suggesting that the aggressive R&D financing strategies of cross-listing firms (i.e., a significant increase in R&D investments by using greater financial leverage) push their industrial rivals to also enhance R&D investment in response.


**Table 5.** Logistic regressions of rival firms taking aggressive R&D investments.

T-statistics are in parentheses. \*\*\*, \*\*, and \* are significant at 1, 5, and 10%, respectively.

### *4.3. Robustness Testing*

We adopted an instrumental variable approach to examine the robustness of our findings. It provided the evidence to support the hypotheses that changes in financial leverage lead to changes in R&D competition not contaminated by endogeneity bias [39]. To implement this process, we first identified these related instrument variables to test the propensity of raising financing. In this aspect, the decision of firms' debt financing was obviously influenced by the financial leverage of industrial peers in the same industry [40]. Thus, it was a good proxy for firms to adopt the average of industrial leverage as the difference in leverage. Second, this study constructed a competitive strategy measure (CSM) to run the model on the basis of [41]. It was defined as the response of the change in marginal profits relative to output, compared with a change in competitor outputs. This concept measured the effect of differences in quantity on marginal profit in the industry. We used the model of instrumental variables in accordance with these studies [2,7,14,15,19,38,41] to measure the relationship between R&D and leverage as follows:

*R*&*Dit* = *β*0*t*+*β*1*tPredicted probability of increasing leveragei* <sup>+</sup>*β*2*t*log(*<sup>H</sup> index*)*i*+*β*3*tNumber of firms in the industryi* <sup>+</sup>*β*4*tMarket share of sample firmi*+*β*5*tIndustrial modified amount of R*&*D*−1*i* <sup>+</sup>*β*6*tIndustrial modified growth ratio of firm assets from year* − 1 *to year ti*+*εit*, (6)

We defined industrial modified difference in R&D investments as the dependent variable. The industrial modified difference was equal to the unmodified change of the sample firm minus the average in the unmodified change of the others in the industry. The unmodified difference was equal to the change in R&D expenditures from year [−1, 0] divided by firm assets in year −1. In particular, the predicted probability of increasing leverage was from the logistic regression according to [41] as follows:

$$Ddebt\_{it} = \beta\_{0t} + \beta\_{1t} \\ \text{Average in leverage}\_i + \beta\_{2t} \\ \text{CSM}\_i + \varepsilon\_{it} \tag{7}$$

Ddebt took a value equal to 1 in case of increases in leverage, and zero otherwise. We named the average industrial leverage as average in leverage. CSM was the proxy of strategic market competition. These standard errors of the coefficient were estimated in this procedure due to heteroscedasticity [37].

The parameter estimates of Equation (7) are as follows: The coefficient of average in leverage is significantly positive, and it is related to the probability of raising leverage (*p* < 0.01). The coefficient of CSM is also significantly negative (*p* < 0.01). This is consistent with the concept of strategic market competition.

We present the results of instrumental variable analysis in Table 6. The coefficients about the predicted probability of increasing leverage are significantly positive in these three models. The results provide strong evidence to support the contention that the manager likely raises the level of debt financing to be consistent with the increase invested in R&D.




**Table 6.** *Cont.*

T-statistics are in parentheses. \*\* and \* are significant at 1 and 5%, respectively.
