The impact of tones of executive communication on firm risk-taking: Evidence from performance volatility and acquisition spending

Abstract

Although current research has demonstrated that executive verbal communication could shape shareholders’ expectancy and responses, the hazards of executive communication and firms’ follow-up responses are largely neglected. Based on expectancy violation theory, we explore how different levels of managerial tone trigger a firm’s risk-taking to avoid violating shareholder expectancy. Using a computer-aided approach to identify managerial tones, our empirical study based on Chinese listed firms indicates that managers tend to take more risks (illustrated by higher performance volatility and acquisition spending) after delivering high-level (optimistic) or low-level (pessimistic) linguistic tones at an earnings communication conference. The results are robust by employing several endogeneity checks. We also identify a mediating role of shareholder reactions and the moderating role of firm prominence. These findings contribute to the executive communication literature by suggesting firms adopting risky strategies in response to shareholder reactions led by managerial tone.

Appendix

This Appendix contains information about the sample and additional analyses supporting the robustness of the results. The following paragraphs provide some background about these analyses to facilitate navigation through the explanations, tables, and figures.

Appendix A. Sample

Appendix A presents details about the sample, including the year and industry distribution.

Tables 6 and 7

Table 6 The sample distribution by Year

Table 7 The sample distribution by Industry

Appendix B. Robustness tests

Appendix B provides regression results of robustness tests of alternative measures and alternative panel regression models, showing that our baseline results are still robust.

Tables 8 and 9

Table 8 Robustness regression results

Table 9 Results of the U–shape robustness test

Appendix C. Fixed-effects 2SLS model

Appendix C provides both first-stage and second-stage results of fixed-effect 2SLS model. The results show that using provincial average tone and its square term as instruments, the regression results are still in line with our baseline models.

Table 10 presents the fixed-effects 2SLS panel regression results of the effect of managerial tone on earnings volatility and acquisition spending. Columns (1) and (2) present the first-stage regression results, in which we account for the endogeneity of managerial tone and its squared term. The Sanderson and Windmeijer (2016) under-identification statistic (i.e., SW χ²) is significant in both two first-stage models. The significances (i.e., SW χ2 = 62.78, p = 0.000; SW χ2 = 50.10, p =0.000) suggest that the model-specific endogenous regressor is not under-identified. Moreover, the weak identification statistic of Sanderson and Windmeijer (2016) is also significant in two first-stage models (i.e., SW F = 612.07, p = 0.000; SW F =49.58, p = 0.000), suggesting that the model-specific endogenous regressor is not weakly identified.

Column (3) and Column (6) respectively present the second-stage regression results with dependent variable of earnings volatility and acquisition spending. The Anderson canon. Corr. LM under-identification statistic is significant in Columns (3) and (6) (i.e., Anderson LM = 36.12, p <0.01; Anderson LM = 28.67, p <0.01), suggesting that the excluded instruments are sufficiently correlated with the endogenous variables and not under-identified (Anderson, 1951). Weak identification test statistics of the Cragg–Donald Wald F are above Stock–Yogo’s critical values at 10% of maximal IV size (CD Wald F = 17.96, SY critical value (10%) = 7.03; CD Wald F = 14.26, SY critical value (10%) = 7.03), confirming the weak instrument validity of the instrument variables (Cragg & Donald, 1993; Stock & Yogo, 2002). Together, diagnostic statistics suggest that the excluded instruments are sufficiently efficient. As the results show in Columns (3) and (6), the 2SLS model using provincial average tone and its squared term as instruments yielded similar results to those obtained from the main models.

Table 10

Table 10 Results of fixed-effects 2SLS model

Appendix D. GPSM model

Appendix D provides detailed procedures and results of the GPSM model.

In the robustness analysis, this study specifies several models that correct for observable self-selection bias. For these models, Appendix D outlines the technical details of the PSM method and shows the results. Appendix D also tests the balancing properties to provide support that matching and bias reduction were successful.

According to the technical procedure of GPSM in business literature (Hock & Raithel, 2020), we model the conditional distribution of the treatments given the pretreatment covariates. The treatment variable is managerial tone, which ranges from -1 to 1 and thus does not follow the normal distribution. We use the Fractional Logit model to modify the density function for estimation as Guardabascioistat and Ventura (2014) proposed. For a few values smaller than 0, the treatment variable is winsored to ensure that the values meet the Fractional Logit model range requirements. The pretreatment covariates are all control variables used in the study, including the year and industry dummies.

We inspected the tests of balancing property after propensity score matching. This analysis decides all observations into four treatment intervals via quartiles and assesses the extent to which each covariate is balanced after GPS adjustment, i.e., the probability that managerial tone in either interval does not depend on the value of each covariate. As shown in Table 11, the balancing property is satisfied at a level lower than 0.01 according to a standard two-sided t-test. In only three cases, there is strong evidence against the balancing property. The conclusion is that the adjustment for the GPS creates a balanced sample.

Tables 11 and 12

Table 11 Results of GPSM balancing property tests

Table 12 Results of dose response function

We then constructed the conditional distribution of earnings volatility given the propensity score, managerial tone, and square terms. Table 12 reports the regression results for the dose response function, as also shown in Fig. 3(a). Furthermore, Fig. 3(b) shows the resulting treatment effect function. For ease of interpretation, positively-sloped (negatively-sloped) portions of the dose response functions indicate a positive (negative) marginal contribution from managerial tone to earnings volatility, which is represented in the treatment effect function when the function is above (below) zero. Therefore, the dose response function and treatment function highlight the generally U-shaped relationship between managerial tone and corporate risk-taking, further alleviating concerns that endogeneity might bias our results.

Fig. 3

Fig. 3

a & b Dose response function and treatment effect function of GPSM model