DO LOCAL MANAGERS GIVE LABOR AN EDGE?
4.4.1 Evidence on Local Managers and Employment
In this section I formally test the main hypothesis that local managers are more labor-friendly than their non-local counterparts.
Evidence on Local Managers and Employment Growth
I begin my tests by estimating Equation (4.1) with industry-adjusted employment growth as the dependent variable. If firms run by local managers are more labor- friendly than those managed by non-locals, then the estimated coefficient on the interaction between the distressed industry dummy and the ex ante local CEO dummy (βL) should be significantly positive.
68The technology firm dummy variable is one if the firm has 3-digit SIC code 737, which are firms in computer programming, data processing, and other computer-related services. This definition follows, that used by Griffin, Harris, Shu, and Topaloglu (2009).
Table 4.3 reports these regression results.69 In column (1) I estimate Equation (4.1) without controlling for within-industry differences in firm-specific characteristics. The estimate ofβL is 0.0285 and is significantly greater than zero at the 5 percent level, indicating that firms run by local managers have significantly greater employment growth following industry shocks than do firms run by their non-local peers. The magnitude of the coefficient indicates that local managers’ effect on employment growth is economically significant as well. This estimated effect is over 40 percent of the average annual employment growth rate in the sample. In terms of jobs, since the average distressed firm has 18,400 employees, this suggests that at the average distressed sample firm, local managers keep roughly 525 more employees working during times of economic distress than do their non-local counterparts.
It is possible that other differences between firms within industries are driving this result. In column (2) I include several firm-specific control variables to control for differences in employment growth rates within industries. Specifically, I control for the lagged logarithm of the number of firm employees, lagged operating performance, lagged stock market returns, and lagged growth opportunities. The estimates indicate that larger firms, firms with superior accounting and stock market performance, and those with higher growth opportunities have higher within industry employment growth rates. After controlling for these factors I find that the estimate onβL falls both in magnitude and in significance to 0.0257 and is significant at the 10 percent level.
It may be important to include several labor-related firm-specific control variables.
In column (3) I add industry adjusted labor productivity and capital intensity to the
69For these regression results and for remainder of the analysis all control variables are industry- adjusted and standard errors of coefficient estimates are White (1980) heteroscedasticity-consistent standard errors, clustered at the firm level. In addition, in order to eliminate the effect of outliers, all outcome variables and control variables are winsorized at the one percent level in both tails using the larger Compustat sample of firms for this procedure whenever possible.
specification in column (2). Both of these measures are included in the wage regression model by Cronqvist et al. (2009) and make economic sense to include when examining employment growth as well. The estimated coefficients on both of these variables are positive and significant at the one percent level and their inclusion increases the adjusted-r-squared from 0.06 to 0.07. With their inclusion, the estimate ofβLis 0.0270 and is significant that the 5 percent level. For the remainder of the paper I refer to the regression model estimated in Column (3) as the “baseline” specification when investigating various employment-related outcome variables.
In columns (4) through (6), I add firm fixed effects to the specifications in columns (1) through (3). Adding firm-level fixed effects controls for firms’ average relative employment growth within the industry. Including these additional controls only strengthens the previous results. Column (6) reports the estimation results of the baseline model with firm fixed effects. In this specification the estimate ofβL indicates that employment growth is 0.0336 higher for firms run by local managers than those run by non-locals. The impact in terms of jobs at the average firm in the distressed sample is that local managers employee roughly 618 more workers than do non-local managers. Overall the evidence in Table 4.3, is supportive of the hypothesis that local managers are more labor-friendly than their non-local counterparts.
Evidence on Local Managers and Layoffs
The results on the relationship between labor growth rates and local managers are consistent with local managers being more labor-friendly than their peers. However, since the identification of the local managers’ behavior comes during times of industry distress, it might be more informative to look at reductions in employment levels. I now investigate the relationship between layoffs and local managers during industry
distress. If local managers are more labor-friendly, then they should be less likely to lay off workers than their peers during difficult times.
In Table 4.4 I estimate Equation (4.1) where the dependent variable is an industry- adjusted dummy variable that indicates whether the firm lays off workers during the base year. It is adjusted for the percentage of firms within the industry that lays off workers during the base year. As in the previous section, βL is the coefficient on which I draw my inferences. In this case if local managers are more labor friendly, then βL should be negative, indicating that locals are less likely to lay off employees.
I estimate the regression models with the same control variables as those in columns (1) through (3) of Table 4.3 using three alternative definitions of layoffs.
In columns (1) through (3), layoffs are defined as negative employment growth in the base year. In 35.8% of the firm-year observations a layoff occurs under this definition. In all three models, the estimate of βL is negative and significant at the 5 percent significance level. In the baseline specification (in column (3)), the estimate of βL is -0.0665. This suggests that firms run by local managers lay off workers with 0.0665 lower probability than their non-local industry peers during times of poor industry performance. Among the sample of firms experiencing industry distress the probability of a layoff under this definition is 0.524. The estimate of βL suggests that having a local manager reduces this probability by 12.7%.
Columns (4) through (6) report analogous regression results, where layoffs are defined as employment growth less than negative ten percent in the base year. Under this definition, the percentage of firm-year observations in which layoffs occur is approximately 13 percent. The estimate of βL under this definition of layoffs is again significantly negative in all three model specifications, however the significance level is only 10 percent once firm specific control variables are included. The estimate of βL in the baseline model is -0.0417. Although the statistical significance is reduced under
this definition of layoffs, the economic significance is even greater. The percentage of firms in distressed industries that experience a layoff of 10 percent or more of the workforce is 18.8%. The estimated βL, implies that firms run by local managers are 22 percent less likely to make these workforce cuts.
In columns (7) through (9) I investigate large workforce reductions. In these estimations the layoffs are defined as reductions in employment growth of greater than twenty percent. Layoffs of this size only occur in 5.8% of the firm-year observations in the sample. In all three specifications the estimate ofβLis negative, but not significant.
Finding that local managers do not reduce the likelihood of large scale layoffs during times of poor industry performance suggests that either there are limits to the control that managers have in the U.S.70 or that there is little room for discretion for such large restructuring decisions. It is interesting to note that the coefficient estimate on the ex ante local CEO dummy in the baseline specification for these large layoffs is significantly negative and is -0.01. This suggests that in times of normal industry performance firms run by local managers are less likely to lay off more than 20% of the workforce.
The evidence on the relationship between local managers and layoffs is consistent with local managers being more labor-friendly. These managers are less likely to lay off workers at small and intermediate sizes, but when it comes to large-scale layoffs, having local management in place does not reduce the likelihood of these layoffs. This is consistent with the findings of Atanassov and Kim (2009) since in the U.S., investor protections are adequate and labor has little power.
70Atanassov and Kim (2009) find that in countries with weak investor protection and strong labor laws that managers can prevent layoffs and do so by selling assets at the expense of shareholders.
The United States, using their measures is below the median in labor strength indices and at the median in investor protections.
Effects of Other Manager Characteristics
I now investigate the possibility that the ex ante local CEO measure proxies for some other manager characteristic that influences whether firms implement more labor-friendly policies during industry downturns. In Table 4.5, I estimate Equation (4.1) with industry-adjusted layoffs as the dependent variable. The specification used in the tests follows the baseline specification, but includes two additional terms; the manager characteristic being tested and the interaction of that characteristic with the industry distress dummy. All CEO characteristics are measured two years prior to the base year to mitigate concerns of reverse causality. If some other CEO characteristic is driving the relationship between the ex ante local CEO dummy and layoffs, then we should find that the coefficient on the interaction term with the additional CEO characteristic should enter into the regression significantly and should diminish the significance of βL.
Yonker (2009) documents that a high percentage of local CEOs are hired internally.
It is possible that CEOs who are insiders interact more with their employees than do those who were hired externally. This may be because insiders are likely to have worked alongside their employees at one time. These increased social interactions could cause insiders to put a greater weight on the welfare of their employees. In column (1) of the table I include a dummy variable that is one if the CEO was hired from within the firm71 and the interaction of this variable with the distressed industry dummy. The estimated coefficient on this interaction term is indistinguishable from zero, indicating that insiders are not less likely than outsiders to cut workers during industry downturns. The estimate ofβL is -0.0685 and statistically significant at the 5 percent level. This estimate is virtually unchanged from the earlier estimate in column (3) of Table 4.4. Although insiders are not less likely to lay off workers during bad
71See the appendix for the calculation of internally hired CEOs.
times, the coefficient estimate on the ex ante internal hire dummy is -0.0408 and is significant at the 1 percent level, indicating that during normal times firms run by internally hired CEOs are less likely to lay off workers.
I next investigate if CEOs who are founders of their companies have a differential effect on employment policies. In column (2), I include a dummy variable that is one if the ex ante CEO is the founder of the company and an interaction term between this dummy variable and the industry distress indicator. The regression results show that founders have a similar effect to insiders. While during normal times firms run by founders are less likely to lay off workers, when forced with tough decisions these founders behave no differently than CEOs who are not founders of their companies.
The estimate of βL remains negative and significant at the 5 percent significance level.
In column (3) I test whether CEO tenure affects firms’ actions toward labor.
Longer tenure could be a sign of entrenchment or it could signal how rooted in the community the CEO has become. In both cases longer tenure should be associated with a lower likelihood of laying off workers. The coefficient estimate on the interaction between the natural log of ex ante CEO tenure and the distress dummy indicate that this is the case. The estimate is negative and significant at the 10 percent level.
The estimate indicates that for each additional ten years that a firm’s CEO is in office, the probability of laying off workers during bad times decreases by about 0.065.
Interestingly, the magnitude of the effect of ten years in office is similar to the effect of being local. This could suggest that it takes about ten years for management to effectively “become” local. Importantly, the effect of tenure on layoffs during industry downturns has almost no effect on the estimate ofβL, which is -0.0616 and is significant at the five percent level.
Cronqvist et al. (2009) show that entrenched managers pay their workers more.
They argue that these managers do this in pursuit of the “quiet life.” Another way that
entrenched managers can treat their employees better is by giving those employees greater employment stability. If entrenched managers pursue the quiet life, then they should be less likely to lay off workers. I therefore construct a measure of entrenchment based on the equity holdings of CEOs. Following the findings of Morck, Shleifer, and Vishny (1988), I define a manager to be entrenched if he holds between five and twenty-five percent of his firm’s equity. Approximately 13.5% percent of the CEOs in the sample meet this requirement. In column (4) of the table I include in the model the dummy variable for ex ante CEO entrenchment and the interaction of this dummy variable with the indicator of industry distress. The estimate on the interaction term suggest that during industry distress, firms run by entrenched managers have a 0.0793 lower probability of laying off workers. The magnitude of this effect is larger than the effect of ex ante local CEOs, however the estimate is only significant at the ten percent level. It is important to note that the inclusion of this entrenchment measure has virtually no effect on the magnitude nor the significance of the estimate ofβL.
The results from Table 4.5 show the robustness of the relationship between the ex ante local CEO measure and layoffs to other CEO characteristics. While some other characteristics seem to help explain the relationship between management and labor, none have the combination of statistically reliability and economic importance that is encompassed by ex ante local CEO.
Effects of Firm Location and Unions
To gain insights into why local CEOs are less likely to lay off workers during industry distress, I next investigate how variation in the geographic dispersion of firm operations, demographics of the population around the headquarters, and state unionization membership influence the effect of local managers on layoffs during times of industry distress. In addition to their influence, I also check the robustness of
my results to these factors since it is possible that one of these factors drives the relationship between layoffs and ex ante local CEO. For instance, if firms with unions are more likely to hire local CEOs, then lower layoffs during industry distress may be falsely attributed to the local status of the CEO when it is actually a strong union presence that prevented the layoffs.
I conduct the tests in a similar fashion to earlier tests. To test the robustness of the relationship between ex ante local CEO and layoffs, I include the robustness variable as well as its interaction with the industry distress dummy. To test the influence of the variable on my main result, I create a triple interaction variable between the influence (robustness) variable, ex ante local CEO, and the industry distress dummy.
If local managers’ effect on layoffs in times of industry distress differs significantly across samples split by the influence variable, then the coefficient estimate on this triple interaction term will be significantly estimated.
Table 4.6 reports the results of these tests. As pointed out earlier, Compustat only provides geographic location at the firm level, so my results rely on the assumption that layoffs occur close to the corporate headquarters. If firms are less geographically dispersed, then this is a better assumption. I therefore create two dummy variables that attempt to measure the geographic dispersion of the firm with the data I have available. The first is a dummy variable that is equal to one if the firm has foreign segment sales greater than zero. The second measures whether the firm has only one business segment. Using the methodology previously described, I include these variables as interactions and triple interactions in columns (1) through (4) of the table.
I do not find a significant effect from either of these measures during times of industry distress and in neither of the robustness regressions do either of these variables affect the estimates of βL. However, the one significant effect that I do find is that during
times of normal industry performance single segment firms are less likely to lay off workers.
I next turn to demographic measures in the area of the corporate headquarters.
Landier et al. (2009) find that firms are more likely to lay off workers in divisions further from the corporate headquarters. Additionally, they find that this effect is limited to firms headquartered in less populated counties. They therefore attribute their findings to managers taking more labor-friendly actions for the private benefit of improved social interactions. If local managers are less likely to lay off workers for the same reasons, then layoffs by local managers should be more likely in less populated areas. If local managers are more likely to be hired in less populated areas then including an interaction between sparsely populated areas and industry distress should wipe out any effect of local managers on layoffs. I measure less populated areas in two ways. The first follows Landier et al. (2009), where I split the sample by the median of the total population in the county of the firm headquarters according the 2000 U.S. Census and the second splits the sample by the median of the percentage of the population living in rural areas in the county in which the firm is headquartered. The results of the robustness and interaction regressions for these measures are reported in columns (5) through (8) of the table. Neither of these variables materially effects the estimates ofβL in the robustness regressions (columns (5) and (7)). The results on the triple interaction term are puzzling. Using the small town measure of Landier et al. (2009), I find that local managers in small towns are more likely to lay off workers during times of industry distress than local managers in larger towns. The coefficient estimate suggests that local managers in small towns behave no differently than non-local managers and that local managers in larger areas have a 0.1210 lower probability of laying off workers than non-locals. This means local managers in more populated areas are 23 percent (12.1/52.4) less likely to lay off workers during industry
downturns than their peers. Why might this be the case? Although this finding seems puzzling on the surface, it could be that industry distress in small towns is a socially acceptable excuse for reducing the workforce and that managers use times of industry distress to make needed reductions.
I next investigate the robustness and influence of the results to variation in unionization. Although I do not have data on unionization at the firm level, if certain industries tend to be more unionized than others, then the industry adjustment in my methodology should control for this fact. However, unions are also more prevalent in certain areas of the country. I therefore create a dummy variable based on the percentage of the population in the state of the firm’s headquarters that are members of unions. If local managers are more likely to form alliances with unions than non- local managers, then I should find that the triple interaction term between states with strong union membership, ex ante local CEO, and industry distress should be significantly negative. If local managers are less likely to lay off workers during industry downturns because firms with unions tend to hire local managers, then the interaction between strong unions and industry distress should push the estimate ofβL toward zero. The results of the robustness regression are displayed in column (9) and show that the estimate on βL remains negative and significant at the five percent level. In addition, the results show that firms in states with strong union presence are less likely to lay off workers during times of industry distress than those located in states with lower union membership. The estimated coefficient on this interaction term is -0.0577 and is significant at the 10 percent level. Column (10) shows the results of the regression testing the influence of state unionization on the relationship between ex ante local CEOs and layoffs. The estimated coefficient on the interactions term is negative but not significantly so (t-stat of 1.48), which shows that there is not a