What drives Income Inequality at the Firm - Level? Literature Review based on Recent Trends in Germany and the U.S.

Bachelor Thesis , 2017 , 52 Pages , Grade: 1,0

Economics - Finance

This thesis will concern research on causes of income inequality, it asks the following question: What drives income inequality at the firm-level? More precisely this would entail the questions: What influences the development of market earnings inequality between firms, understood as establishments? What influences the development of market earnings inequality between firms, understood as distinct corporate units?

I start my thesis with two recent articles that adress these questions and that employ a similar methodology to different countries. The first one is a paper by David Card, J¨org Heining and Patrick Kline Card et al. (CHK), the second one is by Jae Song, David J. Price, Fatih Guvenen, Nicholas Bloom and Till von Wachter Song et al. (SPG). CHK is concerned with firms as establishments in Germany and SPG with firms as corporate units in the U.S. Both articles are concerned with more than just between-firm inequality. For brevities sake the parts on their other research concerns will be mentioned, but not as in-depth as the parts that concern between-firm inequality.

Contents

1 Introduction: Conceptualizing Inequality Research

2 Data and Conceptual Overview
2.1 Summary of Main Points
2.1.1 Card, Heining and Kline - Basic Findings
2.1.2 Song, Price, Guvenen, Bloom and von Wachter - Basic Findings
2.2 Data and Concepts
2.2.1 TimeFrame
2.2.2 DataSource
2.2.3 Earnings Measure
2.2.4 FirmMeasure
2.2.5 Summary of Data and Concepts
2.2.6 Potential Issues

3 Income, Income Inequality and its Drivers
3.1 Descriptive Statistics on Earnings Inequality
3.1.1 Development of overall Earnings Inequality in Germany by CHK
3.1.2 Development of Overall Earnings Inequality in the U.S. by SPG
3.2 Main Analysis: What factors drive Earnings Inequality?
3.2.1 OLS Estimation of Earnings Function
3.2.2 DecompositionofVariance
3.2.3 Results of these Decompositions in both Articles . . . . . . . .
3.3 What drives Income Inequality at the Firm-Level? . . . . . . . . . . .
3.3.1 ComponentsofInterest
3.3.2 TheoreticalMechanism
3.3.3 Five Potential Factors driving Inequality at the Firm-Level . .

4 Conclusion: The Influences at the Firm-Level

Bibliography

5 Appendix

1 Introduction: Conceptualizing Inequality Re­search

Voters in western societies demand, that politicians reduce income inequality.¹ The notion that the rich are getting richer and everybody else lags behind thus is politi­cally relevant and requires the attention of policymakers. Income inequality is thus one of the top research areas in economics. This area is manifold.

First, economists tackle questions about the causes and consequences of income inequality. How income inequality impacts economic growth (e.g., Forbes (2000) or Barro (2000)) or how it impacts health (e.g., Deaton and Paxson (1998)), are part of the literature on its consequences. The literature about causes is manifold with regards to its proposed explanations. They include among others human capital, technology, institutions or demographics.

Secondly, methodologically the subject is also diverse. Income is many things. Economists study market income (wages, salaries), pre- and post-tax incomes or compensation (earnings and benefits) (e.g., Pierce (2001), Sorkin (2015), Kaest- ner and Lubotsky (2016)). But one can also measure inequality differently, using percentile-gaps, variances or the well-known gini-coefficient, to name just a few.

Thirdly, not only can the main concepts of income and income inequality vary, but also the level of analysis. Earlier studies for example (going back to Kuznets (1955)) analyze at the country-level. Later one's follow up on cross-country analy­ses by utilizing panel data also on the country-level. This level of analysis provides more robust effects, since they controlled unobserved heterogeneity between coun­tries, but the level still leaves out the large variation within countries. Since datasets grew, researchers unterake also within-country panel studies on regional differences (e.g., Frank (2009)), taking advantage of a natural experimental setting, delivered by institutional variation. Industry-wide differences in wage setting are also studied (e.g., Katz et al. (1989)).

The unit of analysis in recent years became smaller, in part because merged employer­employee datasets became more available. These allow for a disentanglement of firm- and individual-specific-effects (Card et al., 2016b, p. 2). A growing liter­ature on earnings inequality is concerned, how it relates to firms (e.g., Card et al. (2013b), Song et al. (2016), Barth et al. (2016), Gruetter and Lalive (2009)). Methodologically this literature often roots in Abowd et al. (1999), who proposed and implemented an additive separable earnings function, in order to understand earnings outcomes. This literature is nevertheless still manifold, since in different analyses firms are differently understood. These concepts include - in rising order of aggregation - ”establishments” (e.g., Card et al. (2013b) or Barth et al. (2016)), ”corporate units” (e.g., Engbom and Moser (2017) or Song et al. (2016)) and ”com- panies” (e.g., Mueller et al. (2015)). Depending on these different understandings, results of these studies could yield in part diverging results for the same country.

Fourthly, differences in the datasets used are also important. Some studies have found differences in developments of inequality over time, using representa­tive datasets like the U.S. census or administrative data (Social Security System), for example Spletzer (2014) and (Abowd et al., 2017, p. 4). Here researchers of­ten acknowledge trade-offs. The advantage of administrative data, which is usually uncensored at the top of the earnings distribution, is weighted against the represen­tative selection of the census data.

This thesis will concern research on causes of income inequality, it asks the following question: What drives income inequality at the firm-level?

More precisely this would entail the questions: What influences the development of market earnings inequality between firms, understood as establishments? What influences the development of market earnings inequality between firms, understood as distinct corporate units?

As could be derived from the literature overview, one has to be clear about ones own definitions. In the following I will use corporate unit, if I neither mean the whole corporation, nor the plant (as in Song et al. (2016)). For localized production facilities I will use the term establishment (as in Card et al. (2013b)). If both are meant I will use firm. What is meant by earnings? Earnings are all forms of monetary compensation by the employer to the employee and thus exclude for example income from savings, healtcare contributions by employer or governmental programs.

Why does this question matter? First, staying at the topic of policy makers: Those who are concerned with mitigation of inequality may need to find out, how it is actually caused. If increases in income inequality are driven mostly by companies, then policies that might be designed to mitigate, may actually hurt the policymakers objective.

For instance: The law may require companies to disclose details of their compen­sation schemes in order to tackle gender based income inequality. This may cause increased regulatory burdens, which may threaten to put smaller companies out of business or may prohibit smaller ones from market entry. For example Dreher and Gassebner (2013) studies the mediating effect of corruption on the effect of regula­ tions on market-entry. For 43 countries they find a statistically significant negative effect of procedures required to start a business and of minimum capital require­ments (Dreher and Gassebner, 2013, pp. 420-422). This plausibly translates into a larger market for larger businesses and generates rents for employers and employees as well. Research widely acknowledges distributional effects of rents (Card et al., 2016b, Appendix, Table 1).

Secondly, on a more practical note: If inequality is mainly influenced at the firm-level, then one's own economic outlook depends more on the firm he works for. Selecting the right job than as compared to acquiring skills becomes more impor­tant. Anecdotal evidence suggests that one's own first steps on the labor market depend on one's contacts to businesses. It is no accident that universities offer usually job-presentations and job-fares for prospective graduates. Thus applicants might be well advised to shift more attention to increasing personal connections as opposed to skill acquisition, if income inequality is heavily influenced by the firm one is working for.

I start my thesis with two recent articles that adress these questions and that employ a similar methodology to different countries. The first one is a paper by David Card, Jorg Heining and Patrick Kline Card et al. (2013b) (CHK), the second one is by Jae Song, David J. Price, Fatih Guvenen, Nicholas Bloom and Till von Wachter Song et al. (2016) (SPG).

CHK is concerned with firms as establishments in Germany and SPG with firms as corporate units in the U.S.. Both articles are concerned with more than just between-firm inequality. For brevities sake the parts on their other research con­cerns will be mentioned, but not as in-depth as the parts that concern between-firm inequality.

In chapter 2 I present the main questions of the articles and their answers and draw on their data and concepts and offer some potential channels that would influ­ence their main results. Chapter 3 is the main part of the thesis. There I will present an overview of the development of inequality in both papers. Then I will present their empirical model and how it leads to a decomposition of earnings inequality. I will then take the most relevant decomposed components and ask, what might have influenced them in the past thirty years in both countries. Here I answer my ques­tion on the drivers of inequality at the firm-level. Chapter 4 of the thesis concludes, summarizing the points made before and presenting an outlook on possible future research on the subject.

2 Data and Conceptual Overview

Here I first present the main questions of the articles and their answers. Second I draw on their data and concepts and offer some potential channels that could affect their results and follow up with a short discussion about the merits of these channels.

2.1 Summary of Main Points

The summary of the main points includes one part for each paper and attempts to give a brief overview of their results.

2.1.1 Card, Heining and Kline - Basic Findings

The main research question in CHK is as follows: How are establishment-specific wage premiums influencing the rising earnings inequality in Germany between 1985 and 2009? In order to answer this they estimate an earnings function with OLS for different time periods. This function includes worker- and establishment-time- variant and time-invariant fixed effects (Card et al., 2013b, p. 986). The authors find three empirical trends:

- Rising inequality of establishment-specific wage premiums (Card et al., 2013b, p. 1000)
- Rising inequality of person-specific wage premiums
- Rising sorting of individuals to establishments. This basically is grouping of high earning individuals to high paying establishments.

Furthermore, the authors show how the three trends explain the rising differ­ences between occupational, vocational, age and gender groups (Card et al., 2013b, p. 1002). Here the results differ depending on the group-level.

In the closing of their paper the authors present an explanation for why establishment­specific wage premiums grew more dispersed over time, namely a decline in the number of establishments, which are covered by collective bargaining agreements (Card et al., 2013b, pp. 1008-1009).

2.1.2 Song, Price, Guvenen, Bloom and von Wachter - Basic Findings

SPG tackle four different questions in their paper concerning earnings inequality in the U.S. between 1981 and 2013:

- First, how is earnings inequality developing between corporate units?
- Secondly, how is earnings inequality developing within corporate units?
- Thirdly, what explains the rise in earnings inequality between corporate units?
- Fourthly, what explains the rise in earnings inequality within corporate units?

Similarly to CHK the authors estimate the impact of corporate-unit-specific pre­miums and person-specific premiums on individual earnings, in order to derive a decomposition of the rise in earnings variance (Song et al., 2016, p. 22).

Their results for the first two questions display a rising trend in between- and within-firm inequality. Rising between-firm inequality explains around two thirds of the overall increase in earnings inequality and rising within-firm inequality explains one third (Song et al., 2016, pp. 13-14). Answers to their third and fourth question are more complicated. Their basic decomposition of variances confirms a major role of rising sorting, of rising variance of person-specific premiums and most notably an absence of rising variance of corporate-unit-specific wage premiums (Song et al., 2016, Table C.3).

By decomposing the variance even further, they still find a major role of rising sorting and an absence of rising variance of corporate-unit-specific wage premiums, but the rise of the variance of the person-specific premium is split. This split reveals that the variance of the average individual wage component at corporate units is rising and almost as fast as the other component of the individual component (Song et al., 2016, Table 2). This is a novel component of the variance decomposition literature, which they call segregation.

Within-firm inequality is mainly increasing in larger firms. It is driven by in­creases in inequality mainly at the top and at the bottom (Song et al., 2016, p. 38). The authors undertake a lot of robustness checks, whether the findings on the top and bottom hold for different geographical, educational, industrial, gender, age or company groups (Song et al., 2016, p. 49-54). These findings are by and large supported within every group.

2.2 Data and Concepts

Here I want to structure the comparison of data and concepts along the following points: Time frame, data source, earnings measure and firm measure. For each point I intend to summarize the basics for each paper. In 2.2.6 I want to propose some issues regarding concepts and data.

2.2.1 Time Frame

CHK study the period between 1985 and 2009 for West German earnings inequality and analyze yearly data on earnings. The estimation of the wage parameters is implemented for seven- and eight-year intervals. Therefore they split the overall period in four in part overlapping intervals (Card et al., 2013b, p. 1006).

SPG study the period between 1978 and 2013 for U.S. American earnings in­equality and also analyze yearly data on earnings. They estimate the parameters of the earnings function in intervals containing multiple years. Because of potential measurement problems, they start the analysis in 1980 (Song et al., 2016, p. 8). Here they split the period between 1980 and 2013 in five non-overlapping intervals (Song et al., 2016, p. 40).

2.2.2 Data Source

CHK use a dataset compiled by the Institute for Employment Research (IEB) based on the German social security system. In the data earnings are based on total earn­ings. The data is censored at the social security maximum (”Beitragsbemessungs- grenze”) (Card et al., 2013b, pp. 973-974). Therefore CHK have to ”guess” the distribution of wages above the threshold. For that purpose they use an imputation procedure based on Tobit models (Card et al., 2013b, p. 976).²

They restrict their sample along the following lines (Card et al., 2013b, pp. 973­975):

- First, they include full-time employees, whereby full-time is based on the no­tification of employment, sent to the social security administration by the employer.³
- Secondly, only male individuals between 20 and 60 are included.⁴
- Thirdly, they exclude governmental workers and self-employed.

SPG use the official Master Earnings File (MEF), administered by the social security administration (SSA) in the U.S.. The database contains demographic and earnings information for every individual with a social security number and is comparable with regards to its dynamics to representative databases (Song et al., 2016, p. 47).⁵ More explicitly they use two different samples: One for their first two research questions and a second one for their second two research questions, where the inequality increase is decomposed in more detail. The second sample is the one that concerns the main part of this thesis.

Their first sample contains male and female employees and is restricted along four lines:

- First, they only include full-time employed individuals, where they define full­time as earning the minimum-wage for 40 hours for at least a quarter/ a year.
- Secondly, only individuals between 20 and 60 are included.
- Thirdly, only firms- (and correspondingly individuals working at those firms) with more than 20 employees are considered.
- Fourthly, educational and governmental sectors are excluded (Song et al., 2016, p. 11).

Their second sample is restricted along the first two lines as before, but crucially they lift the firm-threshold and also the sectoral restriction. Because of computation concerns the sample is then restricted to males only (Song et al., 2016, p. 23).

The largest connected set ⁶ for each study includes millions of firms and millions of persons per year. Around 95% of all workers in each period are in the largest connected set (see (Card et al., 2013b, p. 977) and (Song et al., 2016, Table C.1)).

2.2.3 Earnings Measure

CHK calculate daily wages based on yearly earnings reported by the employer to the social security system (Card et al., 2013b, p. 973). They assign only the job to an individual, which generates the most earnings per year, and base their calculations for the daily wages only on this one job.

SPG use yearly earnings, which comprise all forms of remuneration that is di­rectly reported to the Internal Revenue Service by the employer, which is not cen­sored (Song et al., 2016, p. 7). They again assign the most important job with regards to revenue to each individual.

Abbildung in dieser Leseprobe nicht enthalten

Table 1. Summary of Data and Concepts

2.2.4 Firm Measure

As a measure for a firm CHK take the establishment identifiers of the IEB. These identifiers are not necessarily singular establishments or production facilities, but are on a very local level (Card et al., 2013b, p. 974). Production facilities have the same identifier, if they are in the same industry and in the same municipality (Bundesagentur, 2017b).

SPG take as a measure of corporate unit the employer identification number (EIN), mentioned on the tax files. They are more aggregated than the identifiers in the paper by CHK, but on a lower level than a full corporation.

2.2.5 Summary of Data and Concepts

Table 1 summarizes data and concepts for both articles, for SPG it displays the criteria for the second sample. Both samples are very similar with regards to their restrictions, however there remain differences in the definitions of full-time and the earnings measure.

2.2.6 Potential Issues

Certain caveats arise from this choices. First, both articles focus on earnings as the dependent variable. This is widely done in the literature, however with regards to compensation of workers, it may not reflect reality in full. Several authors mention how non-monetary compensation of workers is relevant for the analysis of levels or development of compensation inequality (e.g., Pierce (2001) or (Kaestner and Lubotsky, 2016, pp. 64-66)).⁷ It is however less clear in what direction the inclusion of employer-provided benefits in the analysis of SPG or CHK would change their respective findings.

Monaco and Pierce (2015) find that the inclusion of a variety of employer benefits increases the inequality increase in the U.S. between 2007 and 2014 (Monaco and Pierce, 2015, p. 2), an analysis using a broad range of benefits finds a similar trend for the period between 1981 and 2007 in the U.S. (Pierce, 2001, p. 1520). Contrasting these analyses is (Burkhauser et al., 2013, p. 787), who find for the period between 1995 and 2008 a reduction in the inequality increases, once either employer-provided or government-provided healthcare is included.

I think based on these points in the literature, it is reasonable to refrain from the employer-sponsored benefits, since the direction of a bias on the development of inequality is unclear. I will come back to benefits however in the answer to my research question.

A second problem is the focus on different earnings measures: CHK calculate daily earnings by dividing yearly earnings through the length of the job-spell. SPG use yearly earnings. SPG mention the potential effect of differences in hours worked on the variance of earnings (Song et al., 2016, p. 8). Both papers start with yearly earnings, but hours worked tend to be higher in the U.S. OECDstat (2017). Since 1985 average hours in full-time jobs in Germany dropped from 40.8 to 39.9 and in the U.S. it increased from 41.1 to 41.5. Yearly hours on the full-time job would then have changed accordingly.

Ceteris parisbus this would have caused on average an increase in daily earnings in the CHK study and a decrease in the U.S., since yearly earnings would have been divided by less in Germany as time went by and by more in the U.S.. However, this alone would not change the direction of the development of inequality per se, however it would lead to different relative inequality between both countries, depending on, whether earnings is measured as yearly earnings or hourly earnings.

For effects of the earnings measure on inequality trends, one would have to look at the changes in hours worked by income bracket. Or more concrete: Maybe high earners work way more hours in 2013 than low earners and maybe the differences is more dramatic in 2013 than in 1980? Then inequality based on yearly earnings is measured in a biased way. To adress this possibility, one needs to study, how inequality, either measured through hourly wages or yearly earnings, has changed. Table 5 depicts a comparison of the development of one inequality measure: the 90th to 10th percentile ratio.

The 90th to 10th ratio of hourly earnings for male employees grew from 3.6 in 1979 to 4.8 in 2009. The 90th to 10th ratio for yearly earnings grew from 7.8 in 1981 to 10.6 in 2013 (Song et al., 2016, p. 12). The percentage changes in these measures of inequality over time point to an increase of 33% in the hourly wage inequality and an increase of 36% for the yearly wage inequality over a similar period. These similarities suggest a negligible influence of hours worked on the analysis of the determinants of earnings inequality.

Put simply: the trend in inequality for annual earnings and hourly earnings is similar in the U.S., thus the focus on annual earnings is not as worrisome.⁸ Following up this point it is reasonable to assume that the calculation of daily wages in Ger­many is also valid and the development of daily wages resembles the development of hourly wages.

3 Income, Income Inequality and its Drivers

In order to answer my research question one has to consider first the steps of the argument, namely how estimates of earnings components are obtained and then fol­lowed over time. This development yields the firm-related components of inequality.

First, I intend to present some descriptive findings for both articles. Second, I follow the above mentioned process to get the firm-related components of inequality and their development over time. Third, I will answer my research question on the drivers of the identified firm-related components.

3.1 Descriptive Statistics on Earnings Inequality

I first start with descriptive findings of CHK for Germany and follow up with the findings of SPG for the U.S..

[...]

---

¹ Representative polls consistently rank social justice/ income inequality/ social inequality among the most important issues facing countries (e.g, Wichmann (2017))

² CHK argue that the procedure is valid. For a subset of male German workers, for example, who have a very low censoring rate, the trend in inequality does not change, using the imputation procedure and the explanatory power does not change using varying degrees of censoring (Card et al., 2013a, pp. 3-4).

³ The employer sents the notification to the social security system. It contains four statuses of employment, two of which concern full-time employment (Bundesagentur, 2017a, p. 15).

⁴ They however use women to check their results.

⁵ This is an important comparison. (Abowd et al., 2017, p. 4) find SSA-data to be noisy, based on possible fraudulent behaviour by claimants and therefore restrict their sample of SSA-numbers accordingly. Another differences concerns the level of variance. This is larger for each year in the MEF, since it is not censored at the top (or the bottom) (Song et al., 2016, p. 10)

⁶ I will come back to this term later in the chapter on the main empirical model.

⁷ Kaestner and Lubotsky (2016) however don't asses how inequality of pre-tax family income changes, once employer sponsored healthcare is included. They only assess, how post-tax inequality changes, once employer healthcare is included.

⁸ However there is a striking difference in levels of inequality for each year between hourly wages and yearly earnings.

• College: LMU Munich
• Grade: 1,0
• Author: Martin Schaller (Author)
• Publication Year: 2017
• Pages: 52
• Catalog Number: V1270709
• ISBN (PDF): 9783346715074
• Language: English
• Tags: Income Inequality, Firm Level