Industry Volatility and Workers Demand for Collective Bargaining

Transcription

1 Industry Volatility and Workers Demand for Collective Bargaining Grant Clayton Working Paper Version as of December 31, 2017 Abstract This paper examines how industry volatility affects a worker s decision to unionize. When deciding whether to vote for unionization, workers weigh the potential gain from higher wages with the risks of reduced employment. Using a database of thousands of unionization elections, I exploit the approximately two-month waiting period between when workers petition the National Labor Relations Board to hold an election and when the actual vote occurs. I find that higher industry equity volatility during the waiting period is associated with fewer votes in favor of unionization. Unemployment insurance, which mitigates workers costs of job loss, dampens the response to industry volatility. Volatility heightens workers concerns about job loss as it shifts the riskreturn tradeoff for pivotal voters away from unionization. Kellogg School of Management, g-clayton@kellogg.northwestern.edu. I am grateful to my advisor David Matsa and committee members Meghan Busse, Benjamin Iverson, Artur Raviv, and Paola Sapienza. In addition, I want to thank the attendees of the Kellogg Finance Bag Lunch, who provided feedback on this paper. This project also received helpful comments from Carola Frydman, Jose Liberti, and Brian Melzer. All errors are my own.

2 I. Introduction As stakeholders in the firm, workers compete with investors for rents. One way workers can increase bargaining power to capture those rents is unionization. However, while unionization offers the potential for higher wages, it also involves risk for the worker. This paper asks how workers demand for unionization responds to the level of risk in their industry. Workers choosing to unionize face several risks. First, employers may respond negatively to workers decision to unionize. For example, an employer may close down a newly unionized location or locate new growth opportunities far away from unionized worksites (Holmes, 2013). Second, even without employer action, unionization might endanger the firm by making it less flexible or competitive. For example, unionization may impede the firms ability to respond to negative business shocks (Chen et al., 2011). In the event that unionization results in loss of employment, workers will have to find new employment in their industry. In other words, the workers industry contains the outside option in the event of job loss. Therefore, facing higher risk in the industry, workers are less likely to choose unionization. To empirically measure risk in the industry, I use the volatility of the Fama-French industry portfolio. An important empirical concern is that workers endogenously choose unionization. In other words, workers take into account the level of firm and industry risk at the time that they choose unionization. Moreover, at least at the firm level, risk may endogenously respond to unionization. For example, firms increase debt in response to unionization (Matsa, 2010). Therefore, causal inference about the effect of risk on workers demand for unionization requires a plausibly exogenous change in risk. 2

3 To identify the effect of industry volatility on workers decision to unionize, I exploit the institutional details of the process by which workers vote on unionization. In the United States, workers can petition the National Labor Relations Board (NLRB) for an election to decide whether to be unionized or not. There is a waiting period of around two months between the time of the petition and the time the election is held. I measure the response of the vote to industry volatility during the waiting period. The results show that workers are less likely to vote for unionization when there is higher industry volatility. Furthermore, the availability of state unemployment insurance (UI) dampens this sensitivity. That is, more generous state UI benefits lessen the effect of industry volatility on workers decision to unionize. Moreover, there is a significant asymmetry in the results. Large negative returns in the industry portfolio result in workers being significantly less likely to unionize. In terms of magnitude, negative industry returns in excess of 15 percent during the waiting period results in workers being 10 percent less likely to unionize. On the other hand, large positive industry returns during the waiting period yield no significant effects. The results expand our understanding of labor s bargaining power and the firm s risk. Specifically, this paper s experiment captures workers behavior when they are on the cusp of changing bargaining power. The industry volatility that happens to occur in the waiting period provides a plausibly exogenous source of risk. Therefore, the findings suggest that economic events leading to higher levels of risk may have an important side effect, namely, discouraging labor from pursuing expanded bargaining power. This paper studies unionization, which is an observable example of workers seeking bargaining power. However, the implications of the results could extend to other, less measurable forms of bargaining power. 3

4 This paper contributes new evidence to the literature relating finance and unionization. The evidence on the relationship between firm risk and unionization is mixed. For example, Chen et al. (2011) found a positive association between industry equity returns and industry unionization rates, attributing the higher return to a compensation for the added risks that unions impose on firms. In contrast, Lee and Mas (2012) find significantly lower, long-run returns at firms where workers unionize, compared to similar firms. Other research has established that firms increase debt as a strategic response to unionization threats (Bronars and Deere (1991), Matsa (2010)). A side effect of increasing debt is that the firm s equity becomes riskier. The literature in this area mitigates the endogeneity of workers selecting unionization through instrumental variables (Chen et al., 2011), regression discontinuity (Lee and Mas, 2012), and legal heterogeneity (Matsa, 2010). In other words, the literature attempts to use as-if random or exogenous differences in union power to explain firm risk or capital structure. In contrast, this paper explores the effect of risk on workers selection process. This paper continues as follows. Section II describes the data and empirical approach. Section III presents the results. Section IV concludes. II. Data and Empirical Approach Workers can petition to vote on unionization in NLRB representation elections. Over 80 percent of these elections involve non-unionized workers deciding whether or not to unionize (i.e., certify union representation), while the rest involve unionized workers deciding whether or not to abandon unionization (i.e., decertify union representation). The certi- 4

5 fication process begins with a petition, which must show the support of at least 30 percent of the workers. After gaining the required signatures, the workers submit their petition, and a waiting period begins of around two months before the election. Then, if workers vote to unionize, the particular union winning the election will be certified as representing the workers. Unionization election data for the years 1978 through 2010 come from two sources. First, for the years 1978 through 1999, I use a union election dataset from Thomas Holmes (Holmes, 2006). For the years 2000 through 2010, I use raw election XML files obtained online from the NLRB. 1 Industry portfolio returns for the forty-nine Fama-French industries come from Ken French s website. 2 Annual state UI data was provided by David Matsa. 3 The unionization election data provides results from over fifty thousand elections. For each election, the data indicate the type of employee (e.g., clerical, trucker), SIC industry code, size of the bargaining unit, union, state, petition date, election date, and election result. I use elections that had at least 10 voting workers and came after a waiting period of no more than four months. The vast majority occur within about two months, but there can be delays due to disputes. Table I below gives summary statistics. 1 Accessed on data.gov. 2 Accessed at library.html 3 For other uses of UI data in finance, see Agrawal and Matsa (2013) and Hsu et al. (2017). 5

6 Table I: Summary Statistics Mean Median Std Dev 25th Pctle 75th Pctle Observations Waiting Period (days) ,809 Industry Return ,809 Industry Volatility ,809 Votes Cast ,809 Union Victory ,809 Teamsters ,809 AFL-CIO ,809 Log UI Benefits ,809 Union representation election data cover the years and come from Holmes (2006) and the NLRB raw XML files. Waiting period is the number of days from the petition date to the election date. Industry Return is the return on the Fama-French industry portfolio during the waiting period. Industry Volatility is the standard deviation of the daily returns during the waiting period. Union Victory is an indicator variable recording whether the workers voted to be unionized or not. Log UI Benefits is the natural logarithm of the maximum unemployment insurance benefit offered by a given state in a given year. Industry return is the return on the Fama-French industry portfolio during the waiting period, that is, from the petition date to the election date. Industry volatility is the standard deviation of the daily returns during the waiting period. Recall the main focus of the paper: how workers considering collective bargaining power respond to volatility in their industry. Workers would consider the prevailing risks in their industry when deciding whether or not to petition the NLRB for an election. The goal here is to capture their reaction when unforeseen changes occur in the industry. Therefore, I exploit the approximately two-month waiting period between petition and election. The empirical approach is to test how the unionization election outcome responds to industry returns during the waiting period. The motivation for the experiment is that the workers cannot foresee the returns that will occur during the two months following petition. In other words, the experiment involves observing groups of workers that are very similar. The only difference is that most groups vote on unionization following normal waiting periods, while others vote following large positive or negative returns in their industry during the waiting 6

7 period. Table II below reports the worker types in elections that occurred after large industry volatility, compared to other elections. Table II: Worker Types and Industry Shocks No Shock During Large Returns During Worker Type Waiting Period Waiting Period Clerical Craft Department Guard Healthcare Industrial Other Professional or Technical Trucker Total For each NLRB representation election, the data reports the type of workers in that bargaining unit. The types are given by the NLRB. The table above reports the frequency of each worker type, as a percentage of total elections. The distribution is reported for the two different groups in the experiment. First, the table shows those elections that experienced no large industry returns (positive or negative) during the waiting period. Second, the table shows the elections that occurred after large returns. Large returns are defined as those seen in 5 percent of elections, roughly positive or negative 15 percent industry returns during the waiting period. The distributions shown in Table II provide evidence that the same types of voters (e.g, workers) are voting after large industry returns and normal industry returns. This makes sense, as it seems improbable that workers at a given worksite could accurately forecast two-month ahead industry returns and endogenously time their elections. Therefore, I now present the results of the experiment, where I observe how the voting behavior responds to large industry shocks during the waiting period. 7

8 III. Results This section presents the results of the experiment described in Section II. The goal is to relate the election outcome to the industry returns during the waiting period, controlling for known election characteristics. The main regression specification is given below: union victory = β 1 industry shock + β 3 election controls + Fixed Effects + ε The dependent variable is a binary indicator of the union winning the election, i.e., the workers choosing to unionize or keep their union. Election controls include categorical variables indicating the size of the bargaining unit, state, worker type, petition type (i.e., certification or decertification), Fama-French industry, and union (e.g., Teamsters). Fixed effects include state, industry, year, and industry year. I use two main approaches for the industry shock variable. First, I use an indicator variable of a large downward or upward movement in the industry portfolio. I define a large industry return as a waiting period return having magnitude in the top 5 percent of the distribution, which translates to a return of approximately positive or negative 15 percent. Second, I use the industry volatility itself, i.e., the standard deviation of the waiting period returns. Results are given below in Table III. 8

9 Table III: Election Result Responds to Waiting Period Returns Union Victory Union Victory Union Victory Large Negative Return 0.047*** (0.015) Large Positive Return (0.011) Volatility 1.18** (0.482) Major Unions: AFL-CIO (0.014) (0.014) (0.014) Teamsters 0.089*** 0.089*** 0.089*** (0.019) (0.019) (0.019) Constant 1.30*** 1.29*** 1.28*** (0.043) (0.042) (0.043) Election Controls Yes Yes Yes Fixed Effects Yes Yes Yes Adj R-Squared Observations 51,809 51,809 51,809 The dependent variable is an indicator of whether the voters chose unionization in an election. The Fama-French industry returns are measured for the waiting period between the petition date and election date. A large return is one with absolute magnitude in the top 5 percent of waiting period returns, roughly positive or negative 15 percent over a two-month period. Volatility is the standard deviation of the daily returns during the waiting period. The Teamsters and AFL-CIO make up a majority of the elections, and the regression includes indicators for them. Fixed effects include state, industry, year, and industry year. Election controls include worker type, bargaining unit size, and petition type. Errors are clustered at the industry level. 9

10 The results in Table III suggest that workers are less likely to unionize when large downward volatility occurs in their industry. My hypothesis is that the marginal worker is concerned that unionizing (or maintaining a union) could result in job loss. To test this, I run additional regressions including the de-meaned state UI benefits, measured as the natural log of the maximum UI benefit that a given state allows. I include the UI benefit as a regressor and interact it with the industry shock measure. 10

11 Table IV: UI Dampens Sensitivity to Industry Returns Union Victory Union Victory Union Victory Large Negative Return 0.047*** (0.014) Large Negative Return UI 0.074* (0.039) Large Positive Return (0.010) Large Positive Return UI (0.032) Volatility 1.239*** (0.454) Volatility UI 2.009* (1.055) UI (0.032) (0.033) (0.026) Constant 1.299*** 1.302*** 1.297*** (0.045) (0.038) (0.039) Election Controls Yes Yes Yes Fixed Effects Yes Yes Yes Adj R-Squared Observations 51,809 51,809 51,809 The dependent variable is an indicator of whether the voters chose unionization in an election. The Fama-French industry returns are measured for the waiting period between the petition date and election date. A large return is one with absolute magnitude in the top 5 percent of waiting period returns, roughly positive or negative 15 percent over a two-month period. Volatility is the standard deviation of the daily returns during the waiting period. UI is the demeaned natural logarithm of the maximum state unemployment insurance benefit offered by a state in a given year. Fixed effects include state, industry, year, and industry year. Election controls include worker type, bargaining unit size, union, and petition type. Errors are clustered at the industry level. 11

12 The results in Table IV support the hypothesis that the workers shift away from unionization is related to job loss. In particular, more generous UI benefits provide workers with added insurance. In turn, their decision to unionize or not is less sensitive to negative events in their industry. Intuitively, there is no effect from UI benefits for upward, positive industry shocks. An interesting question is whether the change in outcome is due to the pivotal workers changing their minds, or instead, due to participation from workers who would not otherwise have participated. In other words, large industry returns might be shaking voters out of complacency or inattention. Therefore, I use as response variable the participation rate of eligible workers in the bargaining unit. Results are shown in Table V. 12

13 Table V: Participation and Volatility Election Election Election Participation Participation Participation Large Negative Return (0.002) Large Positive Return 0.007*** (0.002) Volatility (0.084) Major Unions: AFL-CIO (0.004) (0.004) (0.004) Teamsters 0.017*** 0.017*** 0.017*** (0.004) (0.004) (0.004) Constant 0.971*** 0.970*** 0.970*** (0.014) (0.013) (0.014) Election Controls Yes Yes Yes Fixed Effects Yes Yes Yes Adj R-Squared Observations 51,809 51,809 51,809 The dependent variable is the percentage of eligible voters who voted in the election. The Fama-French industry returns are measured for the waiting period between the petition date and election date. A large return is one with absolute magnitude in the top 5 percent of waiting period returns, roughly positive or negative 15 percent over a two-month period. Volatility is the standard deviation of the daily returns during the waiting period. The Teamsters and AFL-CIO make up a majority of the elections, and the regression includes indicators for them. Fixed effects include state, industry, year, and industry year. Election controls include worker type, bargaining unit size, and petition type. Errors are clustered at the industry level. 13

14 As shown in Table V, the results suggest that the change in outcome is not due to inattention during normal times. That is, workers participate in about the same rates in volatile and peaceful times. Moreover, participation is typically very high, on average over 90 percent. Therefore, it appears that the same workers are voting after a shock as in normal times, but they are voting differently. Finally, I conduct a falsification test using the industry returns in the two months before the petition. If they are acting rationally, then the petitioners take the industry conditions into account before deciding the time of petition. The workers considering a petition could wait for conditions to improve before petitioning. Note that my data only shows those workers who went ahead with a petition. Assuming petitioners are trying to maximize the chance of winning the election, I expect that the returns before an observed petition should not have a significant impact on the election outcome. As shown in Table VI, the election result is not affected by preperiod returns. 14

15 Table VI: Election Result Not Affected by Pre-period Returns Union Victory Union Victory Union Victory Large Negative Return (0.023) Large Positive Return (0.011) Volatility (0.703) Major Unions: AFL-CIO (0.014) (0.014) (0.014) Teamsters 0.089*** 0.089*** 0.089*** (0.019) (0.019) (0.019) Constant 1.264*** 1.262*** 1.266*** (0.042) (0.043) (0.046) Election Controls Yes Yes Yes Fixed Effects Yes Yes Yes Adj R-Squared Observations 51,809 51,809 51,809 The dependent variable is an indicator of whether the voters chose unionization in an election. The Fama-French industry returns are measured for the pre-period, that is, the two months before petition. A large return is one with absolute magnitude in the top 5 percent of returns, roughly positive or negative 15 percent over a two-month period. Volatility is the standard deviation of the daily returns during the pre-period. The Teamsters and AFL-CIO make up a majority of the elections, and the regression includes indicators for them. Fixed effects include state, industry, year, and industry year. Election controls include worker type, bargaining unit size, and petition type. Errors are clustered at the industry level. 15

16 In summary, the results above show two main findings. First, industry volatility especially downward volatility causes workers to be less likely to vote for unionization. Second, the presence of unemployment insurance dampens this response. In other words, when workers have more protection in the event of job loss, they are less responsive to industry volatility. These findings support the hypothesis that higher risk in an industry heightens workers concerns about job loss, shifting their preferences away from unionization. The key economic implication is that financial volatility impacts workers decisions regarding bargaining power. Note also that unionization is a medium-term to long-term decision. Once a work site is unionized, it tends to stay unionized for years. Therefore, the results show how a sudden shock in the industry during the short waiting period causes workers to back away from bargaining power that could impact their wages for years to come. Hence, the micro-level results in this paper suggest that macro-level volatile events may have significant effects on workers ability to capture profits. Moreover, the evidence shows that the presence of social insurance (e.g., unemployment insurance) makes workers less concerned about volatility when considering bargaining power. Hence, a more generous social safety net results in workers who are more willing to pursue additional bargaining power, even in the face of negative economic conditions. IV. Conclusion This paper examined how industry volatility impacts workers demand for collective bargaining. Using industry volatility during the waiting period between the petition for unionization and the actual election, I find that higher industry volatility causes workers to be less likely 16

17 to select unionization. This is especially true for downside volatility. I hypothesize that this response is due to worker concerns about potential job loss or other negative results of unionization. Supporting this hypothesis, I find that the presence of state unemployment insurance mitigates the workers sensitivity to industry volatility. The evidence suggests that economic events increasing risk in an industry or the economy as a whole may have the effect of discouraging workers from pursuing additional bargaining power. Workers with less bargaining power will be less able to extract rents through increased wages. Hence, this paper provides evidence that financial volatility affects how firm profits are shared between labor and capital. V. References Agrawal, Ashwini, and David Matsa, 2013, Labor Unemployment Risk and Corporate Financing Decisions, Journal of Financial Economics 108, Bronars, Stephen G., and Donald R. Deere, 1991, The Threat of Unionization, the Use of Debt, and the Preservation of Shareholder Wealth, The Quarterly Journal of Economics 106, Chen, Huafeng (Jason), Marcin Kacperczyk, and Hernn Ortiz-Molina, 2011, Labor Unions, Operating Flexibility, and the Cost of Equity, The Journal of Financial and Quantitative Analysis 46, Holmes, Thomas J., 2006, Geographic Spillover of Unionism, Federal Reserve Bank of Minneapolis Staff Report. 17

18 Holmes, Thomas J., 2013, New Manufacturing Investment and Unions, Federal Reserve Bank of Minneapolis Economic Policy Paper. Hsu, Joanne, David Matsa, and Brian Melzer, 2017, Unemployment Insurance as a Housing Market Stabilizer, American Economic Review, forthcoming. Lee, David S., and Alexandre Mas, 2012, Long-Run Impacts of Unions on Firms: New Evidence from Financial Markets, , The Quarterly Journal of Economics 127, Matsa, David A., 2010, Capital Structure as a Strategic Variable: Evidence from Collective Bargaining, The Journal of Finance 65,