DO LOCAL MANAGERS GIVE LABOR AN EDGE?
4.2 Hypothesis & Empirical Methodology
I investigate the effect of manager heterogeneity on the real economy by testing whether firms run by local managers are more likely to take labor-friendly actions than those run by non-local managers during industry downturns. The empirical methodology that I employ is similar to that of Opler and Titman (1994).58 The basic idea is that negative shocks to industries induce managers in those industries to make decisions about how to “weather the storm.” If heterogeneity in managers affects corporate decisions, then managers who are more labor-friendly will be more likely to make decisions that benefit their employees during these downturns. Specifically, when managers are forced to make cuts, I expect that local managers will be less likely than non-locals to cut employment and will be more likely to make cuts in other areas, such as investment spending and payouts to shareholders. Of course, this test suffers from a joint hypothesis problem. If I fail to reject the null hypothesis that local and non-local managers make similar decisions during industry downturns, then this is not evidence that manager heterogeneity does not affect the real economy because it could be that manager geographic origin does not measure a characteristic that is correlated with labor friendliness. The existence of this problem only biases my tests against finding that differences in managers affect real economic outcomes.
The empirical strategy is then to investigate how the variation in the geographic origins of firms’ managers relates to variation in corporate outcomes within industry groups using industry shocks as a mechanism for identifying a causal relationship. The main outcome variable tested is employment growth and more specifically negative employment growth, or layoffs. In addition to this main outcome variable, I also investigate measures of firm payouts, financing, investment, performance, value, and
58Opler and Titman (1994) use their methodology to identify the effect of financial distress on corporate performance, but their methodology can be used to investigate how other firm characteristics affect corporate decisions during economic distress.
labor productivity to gain insights into why managers with differing geographic origins are more likely to implement more (or less) labor-friendly policies.
I refer to the year in which the outcome variables are measured as the base year.
Industry economic distress and firm-level control variables are measured one year prior to the base year. In the spirit of Opler and Titman (1994), I identify industries in economic distress as those industries (by 3-digit SIC code) whose median sales growth is negative. This definition of industry distress is less stringent than that used by Opler and Titman (1994). I leave whether my measure actually captures industry distress as an empirical issue.59 Identifying distress by industry instead of by firm has the benefit of reducing concerns that the incumbent manager is the cause of the distress or more generally it avoids the endogeneity problem to the extent that distress is measured at the industry level and is not endogenously driven by firm-specific decisions. The local status of the manager is measured two years prior to the base year in order to mitigate concerns of reverse causality. Doing so, however, creates the additional problem that the manager running the firm in the base year may be different from the manager two years prior, whose geographic origin is actually measured.60
Since I am interested in testing whether differences in manager characteristics can explain the variation in corporate outcomes and decisions for similar firms under similar conditions, a natural empirical framework to conduct my tests is using a panel regression model with industry-time fixed effects. However, the sample of firms for which I have data on CEO geographic origin is small, so estimating a panel regression of this nature suffers not only from a dimensionality problem, but also from the fact that
59Opler and Titman (1994) include the additional requirement that the distressed industry’s median stock market return is less than -30 percent. They find that roughly three percent of their sample meets these criteria. I employ a broader definition of industry distress, due to the smaller sample of firms in this study.
60Yonker (2009) shows firms that hire local CEOs persistently do so, which reduces this concern.
In addition, in unreported results I estimate the baseline model on layoffs using the measure of local CEO at a one-year lag and measured in the base year and I find no difference in the results.
the time-industry adjustments will be imprecisely estimated. One potential solution to this problem is to broaden the industry definitions, but because my identification comes from industry shocks, following this path creates the problem that the industry shocks may not be relevant to the firms identified as economically distressed. An alternative solution is making the industry-time adjustment with a larger sample of firms than those for which I have data on CEO geographic origin. This creates more precise estimates of the industry-time adjustments. This is the approach I take.
Specifically, I estimate,
yij,t−yˆ¯j,t =α+δDistressj,t−1 +γLocali,t−2+βLDistressj,t−1×Locali,t−2
+ Γ(Xij,t−1−Xˆ¯ij,t−1) +ij,t,
(4.1)
where yij,t is the outcome variable for firm i in industry j during year t, ˆy¯j,t is the industry-time adjustment to the outcome variable estimated using the larger sample of firms, α is a constant, Distressj,t−1 is an indicator variable that is one if the firm’s industry is in distress at time t−1, Locali,t−2 is an indicator variable that is one if firm i’s CEO is local at time t−2,Xij,t−1 is a vector of firm-level control variables measured at time t−1, and ˆX¯ij,t−1 is a vector of industry-time adjustments to the control variables estimated using the larger sample of firms. Note that if there are systematic differences between the sample used in the estimation if Equation (4.1) and the sample used in the estimation of the industry-time adjustments, then the adjusted variables will not have a mean of zero as they do in an industry-time fixed effects specification. This problem is alleviated however by the inclusion of the constant term and the industry distress dummy, which adjust for the difference between the outcome variables of the two samples on average and during industry distress, respectively.
The within industry difference in the outcome variable during normal times attributed to local managers is estimated by γ, while the estimate of βL is interpreted as the
within industry difference in the outcome variable attributed to local management during times of poor industry performance. Since the main question addressed by this specification is whether industry downturns induce local managers to make different decisions from their industry peers, the main conclusions drawn in this paper focus on the estimate of βL.
In addition to the baseline regression specification in Equation (4.1), I also estimate this equation with two additional modifications. In several tests I also include firm-level fixed effects to the regressions. The inclusion of firm-level fixed effects controls for the firms’ average ranking within the industry for the outcome variable. For example, including fixed effects when investigating employment growth controls for whether the firm typically has above (or below) average employment growth relative to its industry peers. Although it makes sense to include these effects when investigating outcome variables such as employment growth or investment behavior, when estimating infrequent decision behavior the inclusion of firm-fixed effects is not relevant for a short time-series. In other specifications I include lagged outcome variables in addition to firm fixed effects. When this is done, the estimation procedure is similar to a difference-in-difference specification, where the persistence of the outcome variable is estimated as opposed to assumed to be one. I use this specification to investigate changes in firm outcomes. In the next section I explain how I construct the sample and the data used in the estimation of Equation (4.1).