Firm investment in human health capital

    Firm Investment in Human Health Capital Sara B. Holland PII: DOI: Reference:

S0929-1199(17)30475-3 doi:10.1016/j.jcorpfin.2017.08.003 CORFIN 1243

To appear in:

Journal of Corporate Finance

Received date: Accepted date:

2 August 2017 10 August 2017

Please cite this article as: Holland, Sara B., Firm Investment in Human Health Capital, Journal of Corporate Finance (2017), doi:10.1016/j.jcorpfin.2017.08.003

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

T

ACCEPTED MANUSCRIPT

RI P

Firm Investment in Human Health Capital

SC

Sara B. Holland∗ Terry College of Business University of Georgia

NU

All Comments Most Welcome

MA

August 12, 2017

Abstract

AC CE

PT

ED

In 2005, U.S. employers spent more than $500 billion on health insurance. I argue that firms invest in worker health to mitigate the depreciation in human capital that occurs when workers get sick, which increases the productivity of human and physical capital. Using firm-level health insurance data, I find firms that have higher labor productivity, spend more on research and development, and are larger invest more in health capital. Further, health capital investment positively affects firm value and overall productivity. To identify these effects, I instrument for insurance with state mandates and the number of persons covered by insurance contracts.

Keywords: Human capital; health insurance; intangible assets; investment; market valuation

620 South Lumpkin Street , Athens, GA 30602. Email: [email protected] Many thanks to Nigel Barradale, Paul Gertler, Stu Gillan, Ulrike Malmendier, Jeff Netter, Christine Parlour, Annette Poulsen, Toni Whited, Adam Yonce, and seminar participants at the Haas finance student seminar, Berkeley financial economics group, Cornell University, University of Georgia, University of South Carolina, Texas A&M University, Vanderbilt University, and Vassar College for valuable comments and discussion. Any errors are my own. ∗

1

ACCEPTED MANUSCRIPT

Introduction

T

1

RI P

In 2005, employers in the United States spent more than $500 billion on group health insurance, roughly 7% of total compensation. On average, in large publicly traded firms, health insurance premiums are 12% of capital expenditures, varying from 1% in the energy

SC

industry to 20% in the health care industry. Of firms with positive research and development expenditures, health insurance premiums are on average 54% of R&D, varying from 19% in

NU

the energy industry to 144% in the manufacturing industry.

Why do firms spend so much on the health of their employees, and how does it affect

MA

firm value? I argue that firms use health insurance to invest in their workers’ health capital. Firms use physical and human capital in production, and therefore they too value worker health. Worker health can depreciate. When a worker is sick, he or she is less productive

ED

or even absent. Other workers and machines that depend on the sick worker are then less productive. Firms invest in health capital, mitigating this depreciation and making

PT

complementary human and physical capital more productive. Previous research in finance, however, has not analyzed firms’ expenditures on employee health as an investment in

AC CE

human health capital.

With this paper, I aim to fill a gap in the current literature. My paper complements research that finds investment in physical capital positively affects shareholder value (McConnell and Muscarella (1985), Cho (1998)) by studying investment in health capital. Moreover, this paper contributes to the recent work examining the effect of worker well-being on firm outcomes such as market value (Edmans (2011)) and capital structure (Bae, Kang, and Wang (2011)). Rather than limiting the analysis to the effect of worker health on firm value, I identify why firms invest in worker health – a component of worker well-being – in the first place. Studying health capital investment also contributes to understanding the role of intangible investment, and the factors affecting firms’ utilization of and investment in human capital, as firms move toward more human capital intensive production (Zingales (2000)).

2

ACCEPTED MANUSCRIPT

Health capital is a component of human capital (Mushkin (1962)). Health capital in-

T

vestment expenditures are similar to research and development expenditures because they

RI P

are investments in an intangible asset (Kothari, Laguerre, and Leone (2002) and Lev and Sougiannis (1996)). Kahle and Stulz (2017) document that between 1975 and 2015, a sam-

SC

ple period that spans my study, firms invest less in physical assets and more in research and development, but no one has explored firm investment in intangible worker health.

NU

Empirically, I proxy for health capital investment using health insurance premium data from IRS Form 5500, which I obtained through a Freedom of Information Act request.1 I match these data to firm-level data from Compustat from 1992 to 2005. In a regression of

MA

health insurance premiums per employee on the proposed determinants, I find that firms with higher sales per employee, firms that spend more on research and development, and

ED

firms that have a higher book value of assets invest more in health capital. Health capital investment increases with higher levels of labor productivity, human capital, and physical

PT

capital that firms use in production.

The hypotheses I develop to explore why firms invest in health capital abstract away

AC CE

from features in labor and health markets, but empirically I incorporate the evidence that local labor market conditions and location characteristics can influence firm decisions and outcomes (Kedia and Rajgopal (2009), Pirinsky and Wang (2006)). Additionally, health care costs, contracts, and outcomes vary geographically. Including state fixed effects controls for any strategic geographic considerations by firms, the competitiveness of local insurance markets (Dafny (2010)), and any variation in medical costs across regions (Skinner (2011)). The results are also robust to controlling for whether the insurance plans are subject to collective bargaining agreements and controlling for unobserved heterogeneity in the contracting environment with firm fixed effects. In addition to these static effects, I find that health capital investment is persistent over time and that past firm performance does not drive the level of investment in worker health capital, which suggests that firm value does 1

In the United States health insurance is a means of providing access to these health capital investments (Finkelstein, Taubman, Wright, Bernstein, Gruber, Newhouse, Allen, Baicker, and the Oregon Health Stu (2011)). Currie and Madrian (1999) describe three measures of health that affect labor in the developed world: self-reporting, health limitations on the ability to work, and medical care utilization.

3

ACCEPTED MANUSCRIPT

not drive firms to spend on worker health.

T

Firms ultimately invest in intangible worker health to maximize firm value. I estimate

RI P

the effect by regressing market to book value on health insurance premiums, controlling for variables that may affect firm value. Firms endogenously choose the level of health capital,

SC

so I identify the effect by using two instruments that affect firm value only through their effect on health insurance premiums. First, I use state health insurance mandates, which

NU

are state laws that require insurers to provide certain benefits and lead to higher insurance premiums. Second, I use the firm-level variation in the number of persons covered by the health insurance contracts unrelated to the hiring rate, a measure distinct from the number

MA

of employees and exogenous because firms cannot fully control for the alternative health insurance options available to workers. I discuss in detail the strengths and limitations of

ED

my data and proposed instruments.

I find that a 1% increase in health insurance expenditures increases firm value by 0.05%

PT

to 0.07%, holding other factors constant. The effect varies with industry compensation characteristics. The ratio of market to book value captures this effect because health capital

AC CE

is an intangible asset that the book value of assets does not include. In theory, valuemaximizing firms invest in health up to the point where the marginal benefit is zero. The positive effect on firm value suggests that either it is difficult for the market to value health capital or that managers may underinvest in health capital precisely because it is difficult to value (Lev (2004)).

Firms can also use health expenditures as part of workers’ compensation. Firms offer a total compensation package to workers, and workers are willing to trade some compensation in the form of wages for health benefits because they have utility over wages and health (Gruber and Madrian (2002)). Workers have higher utility when they are healthier because they are happier when they are healthier and because they are more productive and earn higher wages. The stock of health capital increases productivity because it affects the amount of time that workers have to allocate between leisure and market activities. Health is a depreciating asset, and health expenditures can offset this depreciation (Grossman

4

ACCEPTED MANUSCRIPT

(1972)).

T

In my paper, I address two important effects overlooked by the compensation view, in

RI P

which workers have utility over wages and health insurance. First, health capital can affect worker productivity, which directly affects firm profits. Second, when a worker’s human

SC

capital is present and productive, it creates potential positive externalities that make other firm inputs, including human capital and physical capital, more productive.2 In a regres-

NU

sion of total factor productivity on health insurance premiums, health capital investment positively affects overall productivity. By measuring total factor productivity using labor and capital, I show how health capital affects firm value, illustrating the underlying relation

MA

between health capital investment, productivity, and firm value. To better understand a firm’s investment in health capital, consider the following ex-

ED

ample of firms vaccinating workers against influenza. To calculate if offering vaccinations is a positive NPV project for the firm, I use parameters from the Nichol (2001) cost-benefit

PT

analysis of vaccinating healthy workers between the ages of 18 and 64 years. The expected cash flow from vaccinating a worker is the probability of getting the flu (5% to 15%) times

AC CE

the vaccine effectiveness (67%) times the value of a worker per day.3 The cost of the vaccine for one worker is $9.64. A worker misses 2.35 days of work due to the flu.4 When the worker is sick with the flu, he cannot contribute his human capital to the production process, which may also decrease productivity of other workers and of physical assets and lead to changes in real economic activity. For example, McTier and Wald (2013) find that higher incidence of seasonal flu in New York City leads to reduced trading activity and lower volatility. Giving a flu vaccine is positive NPV for workers contributing at least 2

Lisa Brummel from Microsoft Human Resources told the PBS program Frontline, “really you could say we have end to end support for just about every medical issue that you might face during your lifetime...And that’s important to us because we really run the gamut of people who are just out of college all the way through to people who are, you know, well into and toward their retirement ages, so we really have to cover the whole spectrum, and we find that the investment in health care keeps that entire population healthier, coming to work more often, more alert, more productive.” Transcribed from “Sick Around America” Frontline, PBS, originally aired March 31, 2009. 3 The probability that the vaccine is effective is 75% conditional on the vaccine matching the strain of flu circulating that year. Efficacy is 35% conditional on a poor match. The probability of a good match is 80% (Nichol (2001)). Hence, vaccine effectiveness is 0.8(0.75) + 0.2(0.35) = 0.67, and the probability of getting the flu after getting vaccinated is 1.7% to 5%. 4 Nichol (2001) assumes 2 missed days of work and 0.7 days at 50% of normal productivity.

5

ACCEPTED MANUSCRIPT

$122.45 to $40.82 of value to the firm per day (based on the 5% to 15% probability of getting

T

the flu). This valuation is conservative. The prevalence of flu and reduced productivity due

RI P

to illness suggest that for many firms flu shots are positive NPV projects. Moreover, when aggregating to the firm level, these numbers understate the value of vaccination if the

SC

probability of getting the flu is not independent across workers.5

The paper proceeds as follows. The next section describes the data. Section 3 develops

NU

hypotheses and outlines the empirical methodology for investigating firm investment in health capital and presents the main results for what drives firm investment in health capital. Section 4 presents the results from estimating the subsequent effects on shareholder value.

MA

Section 5 shows results for total factor productivity, and Section 6 concludes. An appendix

2

ED

gives more details on the construction of the dataset.

Data description and sample characteristics

PT

This section provides an overview of the data; more details are available in the data appendix. While it would be optimal to have expenditures on all types of health care,

AC CE

extensive data of this type are not available. Data on welfare benefits, however, including medical and dental benefits, are available from the IRS Form 5500. All benefit plans with more than 100 participants are required to file this form every year; and these data are available on request from the Department of Labor. Although the filing requirements limit the sample to larger firms, I focus on publicly traded companies, which will likely meet the requirement. Also, this research examines the dollar amount of health care the manager chooses rather than firms’ binary decisions to offer insurance. I construct the sample by starting with the universe of Compustat firms between 1992 and 2005 and merging with any welfare benefit plan for which the benefit code indicates a health, dental, or vision plan.6 Firms on average file two welfare plans, and I aggregate 5 For example, Nichol, J., Greenberg, and Ehlinger (2009) report that in a survey from a workplace flu study, 45% of participants who had influenza-like illness reported that they thought they had contracted the illness from someone at work. On the other hand, the results may not aggregate up to the firm level if not all workers choose to participate (D’Heilly and Nichol (2004)) or if there are decreasing returns to vaccination. 6 In general, welfare benefit plans can include “medical, dental, life insurance, apprenticeship and training,

6

ACCEPTED MANUSCRIPT

welfare plans at the firm level. Firms in the sample must also have basic financial data that

T

I discuss further below.

RI P

I restrict the sample to firms that have contracts with insurance companies in order to be able to form my proxy for health capital investment.7 These restrictions reduce the

SC

sample to 12,005 firm-year observations for 3,448 firms.

Premiums vary across years. Figure 1 shows the average premiums per year. Premiums

NU

during the 1990s appear to be generally falling, but average premiums in the 2000s appear to be rising.8 This trend is consistent with the overall change in health care spending nationally. Figure 1 also shows the average number of persons covered per year is increasing.

MA

The average number of firms per year is similar across the sample.

Firm investment in worker human health capital The health investment framework

PT

3.1

ED

3

Firms use human capital and physical capital in production. Just as physical capital

AC CE

depreciates, human health capital depreciates. That is, workers can get sick, reducing productivity due to absenteeism or presenteeism (the worker could be on the job but suffering from an illness that makes the worker unproductive). Investing in health mitigates this depreciation and keeps workers present and productive. Firm investment in human health capital has a direct effect on labor through the worker, but it also makes complementary human capital and physical capital more productive. For example, the worker could be part of project with another worker whose productivity is lower when the partner is sick, or the worker may be home nursing a cold and the physical capital remains idle. If the worker is more productive due to an investment in health capital, a worker paid scholarship funds, severance pay, and disability” according the the Form 5500 Instructions. Some plans in the sample include more than one of the health benefits in addition to other welfare benefits such as life insurance or disability, for example. 7 Insurance premium data are not available for firms that self-fund health benefits. Many firms use a combination of self-funding and insurance. All reported results are robust to limiting the sample to firms that exclusively contract with insurers, which reduces the final sample by two-thirds. 8 There is a large jump in average number of firms per year and the average premiums per year between 1998 and 1999. This difference reflects a change in IRS Form 5500 and does not reflect a trend in the data.

7

ACCEPTED MANUSCRIPT

according to the marginal product of his labor should be able to capture this increase in the

T

form of higher wages. The firm can capture some of the benefits because employer-provided

RI P

health care makes complementary human and physical capital more productive. In short, firms invest in human health capital to keep the workers more productive, to keep their

SC

colleagues more productive, and to keep the assets more productive.

The simple framework predicts that the optimal health capital investment increases with

NU

worker productivity, human capital, and physical capital. Firms with higher levels of labor productivity will invest more in health capital because in firms where present workers are more productive in the technology employed by the firm, an investment in health capital

MA

increases the chance the human capital will be able to take advantage of that technology. Firms will invest more in health capital at higher levels of human capital because the

ED

same depreciation in higher levels of human capital reduces production by a larger amount. This model implicitly assumes that the human capital is not easily replaceable. Supporting

PT

this view, Hamermesh and Pfann (1996) reports results from surveys and accounting studies indicating that labor adjustment costs are larger for high-skilled workers. For example, the

AC CE

worker could have firm-specific skills, which are consistent with higher levels of human capital, that are difficult to acquire in a short amount of time (Topel (1991)). Firms will invest more in health capital as the size of physical capital increases because a present and productive worker can utilize those assets. The presence of human capital affects the marginal product of physical capital, so mitigating any depreciation in human capital increases productivity. This effect is important for understanding how the firm can benefit from investing in a worker’s health capital. I test these hypotheses using the following regression:

hist = Xist γ + θt + ψs + ci + uist ,

(1)

where the dependent variable for firm i in state s in year t is health capital expenditures per employee, Xist is a vector of health capital investment determinants, θt is a year fixed

8

ACCEPTED MANUSCRIPT

effect, ψs is a state fixed effect, ci is a firm fixed effect, and uist is an error term. Following

T

the simple framework outlined above, the null hypothesis is that the coefficients on labor

RI P

productivity, human capital, and size are zero. The alternative hypothesis is that they are positive.

SC

I proxy for the health capital investment by taking the log of health insurance premiums per employee. Though this proxy is naturally imperfect, the drawbacks provide additional

NU

opportunities for interpreting the results. First, there is mixed evidence on the effect of health insurance on health outcomes. Levy and Meltzer (2008) review the literature and find that the effect of health insurance depends on the sample under study, raising endo-

MA

geneity concerns. In fact, most studies focus on recipients of Medicaid or Medicare – health insurance programs for low income or elderly populations. This study, however, focuses on

ED

the working age population, a group largely outside the Medicaid and Medicare populations and a group better suited to value the positive externalities in the production process.

PT

Moreover, recent empirical evidence from a randomized experiment indicates insurance provides access to care that improves even short term health outcomes that firms might value

AC CE

more (Finkelstein, Taubman, Wright, Bernstein, Gruber, Newhouse, Allen, Baicker, and the Oregon Health Stu (2011)). Second, firms can invest in health capital in other ways that my proxy does not capture,9 suggesting that my result is a lower bound estimate for the effect of health capital investment on firm value. To test the hypotheses described above, I need to be able to say something about how firms vary according to labor productivity, human capital, and size. To proxy for labor productivity, I use sales scaled by the number of employees following Cronqvist, Heyman, Nilsson, Svaleryd, and Vlachos (2009). If labor is more productive, sales per employee should be higher.10 To proxy for human capital, I use research and development expenses scaled by total assets because there is evidence of wage differentials in R&D intensive 9

For example, Kaiser Family Foundation and Health Research and Educational Trust (2009) report that 20% of large firms have on-site clinics. 10 Wage is another proxy for labor productivity. More productive workers receive higher wages, but wage data are sparse. Ballester, Livnat, and Sinha (2002) report that only 10% of Compustat firms report labor expenses, and the probability of reporting is higher for larger and more labor intensive firms.

9

ACCEPTED MANUSCRIPT

firms. For example, Bartel and Lichtenberg (1987) report that manufacturing industries

T

where capital equipment was newer and R&D was more intensive employ relatively more

RI P

educated workers, and Mincer (1991), looking at the time series of wage differences, finds higher relative wages of better educated workers in R&D intensive industries compared to

SC

other industries. To proxy for size I use the natural log of assets.

While the predictions from the framework described above motivate the first three de-

NU

terminants of health capital investment, other variables outside my framework may affect health capital investment. Firms that are more capital intensive may rely less on productive, high human capital workers in the production process. I include free cash flow because at

MA

poorly governed firms, managers may provide more health benefits because they derive utility from giving perks to workers when firms are doing well (Jensen (1986)). The Form 5500

ED

data also include whether a particular plan is subject to collective bargaining agreements, which may affect firms’ investment in health capital via insurance premiums. I include an

PT

indicator variable that takes the value of 1 if any plan is subject to a collective bargaining agreement and 0 otherwise.

AC CE

Table 1 shows variable definitions and summary statistics for the sample for the determinants of health capital investment. It also shows summary statistics for market to book value and additional firm characteristics related to firm value that I use later in the paper. These values are in line with summary statistics reported in papers on managerial compensation and firm value such as Palia (2001) and Coles, Lemmon, and Meschke (2012). Compared to the Compustat universe, the firms in my sample are larger as measured by total assets, but have lower labor productivity, research and development expenditures, and market to book value.

3.2

Determinants of human health capital investment

Tables 2 shows the results from estimating the regression in Eq. (1) of health capital investment on the theoretically motivated determinants. The proxy for health capital expenditures is the log of premiums scaled by the number of employees. I scale insurance 10

ACCEPTED MANUSCRIPT

premiums before taking the log to ensure that these results and the ones that follow are

T

not spurious due to nonstationarity. All specifications include year fixed effects to capture

RI P

differences in average premiums over time.

The first specification is a pooled OLS regression with the proposed determinants of

SC

health capital investment. More productive labor will generate higher sales per employee. The coefficient on labor productivity is positive and significant, suggesting that firms with

NU

higher labor productivity invest more in worker health. Research and development expenditures proxy for human capital because firms that are more human capital intensive may also invest more in R&D. The coefficient on research and development is positive and sig-

MA

nificant, which supports the prediction that more human capital intensive firms will invest more in the health capital of workers. The sign of the coefficient on firm size is positive and

ED

significant as predicted by the theoretical framework. The coefficient on capital intensity is negative and statistically significant, which is consistent with the theoretical framework.

PT

The coefficient on free cash flow is actually negative significant, indicating that firms do not invest in health when they are flush with cash. I find that a union presence at a firm does

AC CE

not affect the level of health capital investment. The first specification also includes state fixed effects to control for any strategic geographic considerations by firms, the competitiveness of local insurance markets (Dafny (2010)), and any variation in medical costs and practices across regions (Skinner (2011)). Overall, these determinants explain about 30% of the variation in health capital investment. In order to capture any unobserved heterogeneity across firms, the second specification of Table 2 includes industry fixed effects based on two-digit SIC codes. Including industry fixed effects controls for unobserved heterogeneity if it does not vary among firms within industries. Industry might capture other important factors that affect the level of health capital investment. For example, Neal (1995) investigates differences in wages for workers changing jobs within an industry versus changing to another industry and finds evidence that industry specific human capital influences the differential. Industry fixed effects may also control for differences in occupational risk, which may affect health insurance premiums.

11

ACCEPTED MANUSCRIPT

Estimates in the second column show results are robust to the inclusion of industry fixed

T

effects.

RI P

Finally, in the last specification I drop the state fixed effects to be able to include firm fixed effects.11 Firm fixed effects account for any unobserved heterogeneity in the contract-

SC

ing environment that the collective bargaining dummy does not capture. For example, firms bargain with workers, but they also bargain with insurers. Results are robust, though the

NU

coefficient on size is no longer statistically significant.

I do not include any explicit controls for the tax benefits of health insurance that the firm receives. In the investment view of health benefits, the firm uses health capital expenditures

MA

to mitigate depreciation rather than for compensation, so incorporating individual taxes will not change the optimal level of health benefits. In the compensation view, however, firms

ED

offer health insurance because workers can deduct health insurance before paying taxes. The deductibility of health insurance makes one dollar of health insurance cheaper for the

PT

firm than one dollar of wages. More highly compensated workers who face higher marginal tax rates benefit more from the tax deductibility, but it is difficult to estimate taxes faced by

AC CE

individual workers on the firm’s demand for health insurance (Gruber and Lettau (2004)). Labor productivity should be highly correlated with wages, so the results in Table 2 suggest that holding the tax benefits of health insurance constant, firms that have a higher value of physical assets and have higher levels of human capital invest more in the health capital of the worker.

The results on the determinants of health capital investment are consistent with an alternative model related to worker compensation. Consistent with the health capital investment framework, workers may choose to invest in their own health so that they will increase their productivity and earn higher wages. Higher human capital workers at more productive firms will in turn demand more health benefits, inducing a positive correlation between sales per employee and health insurance premiums per employee. As a robustness test, in unreported results I include wages per employee as an additional regressor in Eq. 11 Firms do not change location frequently, making it difficult to include state and firm fixed effects in the same regression.

12

ACCEPTED MANUSCRIPT

(1). Because most firms in Compustat do not report wage data, the number of observations

T

is only 5% of the full sample.12 The results are qualitatively and quantitatively similar to

RI P

the results in Table 2. Holding wages constant, firms that have higher levels of labor productivity, are larger, and have higher levels of human capital invest more in worker health.

SC

This result complements the findings of Decressin, Lane, McCue, and Stinson (2005) who match the publicly available firm-level health insurance data from IRS Form 5500 data to

NU

confidential Census Bureau data on employer and employee characteristics and find that firms offering health benefits have higher rates of productivity and that wages are higher even when accounting for worker characteristics.

MA

As predicted by the motivating framework, the results indicate that firms with higher levels of labor productivity, higher levels of human capital, and that are larger invest more

ED

in worker health capital. Accounting for unobserved state and industry heterogeneity explains a large amount of health capital investment. Firms invest in intangible health capital

PT

in order to maximize firm value. But firms may spend more on employee health when firms perform better. Past firm performance may be an important determinant that the motivat-

AC CE

ing framework fails to incorporate. Pauly (1998) argues that the ‘business perspective’ on health benefits is to provide insurance when it is affordable. To gauge whether profitability affect health insurance expenditures, Table 3 includes return on assets lagged one and two years as a control for past firm performance. In the first column, which includes year and state fixed effects, the coefficients on one and two year lagged return on assets are actually negative and significant. Firms do not invest more in the health capital of the worker when past performance is strong, which is consistent with the findings of Dafny (2010) who concludes that profitable firms do not increase benefits. Including past performance also does not improve the statistical fit. The results are robust to the inclusion of industry or firm fixed effects. Moreover, even after controlling for past firm performance, the effect of the other determinants does not change qualitatively. Firms with past positive performance do not spend more on worker health, suggesting 12

I show later that the results are robust to including industry level wages for my full sample of firms.

13

ACCEPTED MANUSCRIPT

that firms are not quick to adjust health spending based on firm outcomes. To test the

T

persistence of worker health policy, Table 4 includes the past three years of lagged health

RI P

capital investment. The first specification controls for year and state fixed effects. The coefficients are all positive and statistically significant, but the magnitudes indicate that

SC

the effect declines substantially over time. The explanatory power increases to 77%, indicating that past health capital investment does indeed matter for current health capital

NU

investment. These results are consistent with evidence that other firm policies like leverage (Lemmon, Roberts, and Zender (2008)) and board structure (Wintoki, Linck, and Netter (2012)) depend extensively on past firm decisions. The sign and significance of the

MA

remaining determinants remain much the same when accounting for persistent health capital investment, though the magnitudes are lower. The coefficient on labor productivity is

ED

still positive, but loses statistical significance. These results are robust to the inclusion of industry fixed effects and firm fixed effects as shown in the second and third columns.

PT

Evidence of persistence shows that firms do not change health capital investment very much. This finding, taken together with the result that past positive operating performance

AC CE

does not lead to higher levels of health spending, makes it less likely that firms will spend more on health when firms are doing well. In the next section, I explicitly address how investment in intangible worker health affects firm value.

4

Firm investment in health and shareholder value

4.1

Identifying the effect on firm value

Firms choose to offer a level of health benefits to maximize shareholder value. The evidence above suggests that firm level characteristics besides wages affects the level of health capital investment. In this section, I explore the subsequent effect of that investment on firm value. I use the ratio of market to book value as a proxy for value for two reasons. First, it is consistent with the literature.13 Second, a firm’s investment in worker health capital 13

For some examples, see McConnell and Servaes (1990), Morck, Shleifer, and Vishny (1988), Gompers, Ishii, and Metrick (2003), Himmelberg, Hubbard, and Palia (1999), Lang and Stulz (1994), and Palia (2001).

14

ACCEPTED MANUSCRIPT

is an investment is an intangible investment that is treated as an expense for accounting

T

purposes. The market to book value should then capture any effect on firm value.

RI P

The relevant regression is:

(2)

SC

M/Bist = βhist + Xist γ + θt + ψs + ci + ǫist ,

where the dependent variable for firm i in state s in year t is market to book value, hist

NU

is health capital expenditures per employee, Xist is a set of controls, and ǫist is an error term. The empirical analysis draws upon the regression specifications in the managerial

MA

compensation literature and uses similar control variables.14 Firm value and health capital investment are simultaneously determined, so health capital investment is correlated with ǫist in this regression. In order to explore how health

ED

capital affects firm value, ideally one would like to force a firm to spend different exogenous amounts on health care and observe how firm value changes. It is possible, however, to find

PT

instruments that enable the identification of the coefficients of interest. I estimate the effect of health capital investment on firm value by estimating Eq. (2)

AC CE

using two-stage least squares. The first stage of this regression provides more empirical evidence on which variables drive the level of investment in health capital as in Eq. (1) and shows the effect of the instrumental variables on health capital investment. 4.1.1

Instrumental variables for health insurance premiums

The total amount that firms spend on insurance is the premiums. The premium per person is a function of a number of variables such as the number of persons covered, whether the contracts are experience rated or not,15 whether the coverage is for families or individuals, whether the plan is part of a Health Maintenance Organization (HMO) or Preferred Provider Organization (PPO), state insurance laws, demographics, industry, human capital, 14

See for example Palia (2001). With an experience rated contract, insurers base “the premium on the prior or current claims of a group” (Morrisey (2008)), as opposed to community rating, which may lower premiums. 15

15

ACCEPTED MANUSCRIPT

and bargaining power of the firm. Any of these variables that are not correlated with the

T

residual in the firm value regression would be an appropriate instrument. I use state health

RI P

insurance mandates and the number of persons covered unrelated to the hiring rate. Using two instruments increases the efficiency of the second stage coefficient estimates. In order

SC

to use these proposed variables as instruments, the correlation between the instruments and the error term in Eq. (2) should be zero.

NU

The first instrumental variable is the number of health insurance mandates by state and the year of enactment compiled by the Blue Cross and Blue Shield Association.16 . These laws require insurers to cover certain benefits, such as contraceptives, certain providers,

MA

such as marriage therapists, and certain persons, such as dependents. The mandates vary across time and across states. For example, all 50 states, plus the District of Columbia,

ED

mandated minimum maternity stay between 1995 and 1997, but only 23 states mandate coverage of contraceptives, with Ohio adopting in 1991 and Massachusetts and New York

PT

adopting in 2002.

Figure 2 shows the intensity of mandates across time and states. For each year, I group

AC CE

states into quartiles according to the number of mandates. The light blue states have the fewest mandates (below the 25th percentile) and the dark blue states have the highest number of mandates (above the 75th percentile). The top panel shows the distribution for 1992. The bottom panel shows the distribution for 2002 (the last year for which we have mandate data). While some states remain on one side of the distribution over time, others shift. For example, Idaho is in the first quartile (low mandates) in 1992 and 2002, and Maryland is in the fourth quartile (high mandates) in 1992 and 2002. On the other hand, Virginia shifts from the second to the fourth quartile between 1992 and 2002, and Louisiana shifts from the third to the first quartile. Each mandated benefit has a different effect on premiums. The Council for Affordable Health Insurance reports that most mandates increase premiums by less than one percent, but several can increase premiums by five to ten percent. The mandate variable is the sum of the number of mandates in that state per 16

I thank Amanda Kowalski for providing me with this data.

16

ACCEPTED MANUSCRIPT

year. I merge this data with the firm headquarters state identifier in Compustat. Many

RI P

limitation should bias against finding any effect on premiums.

T

firms will have establishments in states other than the Compustat identifier, but this data

All specifications include state fixed effects to allay the concern that unobserved hetero-

SC

geneity across states (for example, time-invariant differences in the political environment) is correlated with both the number of mandates benefits and firm value. Additionally, man-

NU

dating insurers to offer certain benefits may generate positive externalities that affect firm value. Including state, industry, and state by year fixed effects will proxy for many of these market differences.

MA

While mandates are convincingly exogenous, there are two issues that may make interpreting this instrument difficult. First, if firms contract with an insurer already providing

ED

more than the mandated benefits, then there will be no effect of the mandates. Second, if providing these benefits positively affects firm value and the firm did not offer these bene-

PT

fits, then the effect of these mandates is to increase firm value. On the other hand, if the mandates require benefits that the firm does not value, then the effect of these mandates is

AC CE

to decrease firm value. In this case, we should see firms scaling back coverage or opting to self-insure, but Gruber (1994b) provides evidence that self-insured firms also provide benefits mandated by states. Hence, this scenario is likely not a major concern in the current setting.

Because health insurance mandates is a state level variable, I also incorporate a firm level instrument: the number of persons covered scaled by the number of employees. Persons covered is the number of employees who participate in the insurance plan, possibly including dependents.17 Importantly, this instrument is not the same as the number of employees at the firm. The number of persons covered is outside the scope of the firm’s decision making ability because the firm cannot control the household’s outside option.18 For instance, if 17

The data on persons covered are somewhat ambiguous. The instructions for Form 5500 explicitly state that the number of participants should exclude dependents, but the instructions for Schedule A of Form 5500 do not address persons covered. There is some evidence that benefits managers interpret this field as including participating employees and dependents (Brien and Panis (2011)). The ambiguity, however, does not invalidate any argument for using this measure as an instrument. 18 Firms can in theory design contracts to affect the number of persons covered (Dranovea, Spiera, and

17

ACCEPTED MANUSCRIPT

the spouse of a worker has ‘better’ health insurance, the worker and his household will

T

not participate in the firm’s program (but the firm still benefits from the outside option).

RI P

Alternatively, the spouse’s outside option may be ‘worse’ than the worker’s health insurance through the company, and so the worker and his household will participate in the firm’s

SC

program. Finally, the household may choose both options. Empirically, the decision of the worker will be outside of the firm’s maximization problem, making the number of

NU

persons covered a reasonable instrument because the number of persons covered will provide exogenous variation in health expenditures.19

The number of persons covered can change for several of reasons. First, a newly hired

MA

employee enrolls in the plan, possibly with dependents. Second, an employee drops out of the plan because he leaves the firm, he takes an outside insurance option, or he deems the

ED

coverage unaffordable. Third, a currently enrolled employee has a family event and enrolls (marriage, birth) or drops (divorce, death) coverage. Overall, the number of persons covered

PT

may be associated with unobservables in the value regression that would affect employee events such as entering or exiting a firm. To control for how the number of employees affects

AC CE

the number of persons covered by insurance contracts, I construct a hiring rate variable for each firm as

employeest −employeest−1 employeest−1

following Bazdresch, Belo, and Lin (2014). I regress

the number of persons covered per employee on the hiring rate and use the residuals from this linear projection to capture the variation in number of persons covered not related to employee turnover.

Using persons covered as an instrument mitigates the concern that more valuable firms invest more in health (that is, the simultaneity-induced endogeneity). More persons covered should increase health investment, putting us closer to the ideal experiment of forcing a firm to spend different exogenous amounts on health care, as long as higher value does not lead to an increase in persons covered. I argue that it does not by considering the two ways it can. First, firms that are more valuable may hire more workers, but I only exploit the change Baker (2000)) but not perfectly. 19 Cutler (2003) shows that the reduction in employer-provided insurance over the 1990s is due to employees’ decision to take up the insurance rather than a decline in the number of firms offering insurance.

18

ACCEPTED MANUSCRIPT

in persons covered that is orthogonal, and hence unrelated, to the hiring rate. Second,

T

firms that are more valuable may cover more workers or more dependents or offer more

RI P

generous plans. I cannot observe the details of the health benefit contracts in my data. I can, however, observe the number of plans, and the most likely way that firms will offer new

SC

and more generous benefits is to offer more plans. I find negative and statistically significant coefficient in a regression of total numbers of plans on persons covered unrelated to hiring

NU

and including year and state fixed effects. The results suggest that firms do not offer new and more generous benefits when persons covered increases. In short, using persons covered unrelated to the hiring rate makes it less likely that higher firm value leads to higher health

Health capital investment positively affects shareholder value

ED

4.2

MA

investment.

Table 5 show results from an instrumental variables regression of firm value on health

PT

capital investment. All specifications include the determinants of health capital investments from Eq. (1) as controls. The regressions also control for advertising expenditures and

AC CE

leverage, which previous research suggests affect firm value (Palia (2001)). The first column of Table 5 shows the first stage regression that includes instruments to identify the effect of health capital investment on firm value. The first instrument is the variation in number of persons covered that is orthogonal to the hiring rate, and the second instrument is the number of insurance mandates. The variation in persons covered and state insurance mandates are positively correlated with premiums per employee, and both are statistically significant. The F-statistic from a test of whether all coefficients from the first stage are jointly different from zero is 179.93. The p-value from a test of the overidentifying restrictions is 0.28, so it is not possible to reject the null hypothesis that the instruments are exogenous. The coefficients on the other controls in the first stage regression are consistent with the determinants of health capital investment in Table 2. Additionally, the negative and statistically significant coefficient on leverage might indicate that firms that are close to 19

ACCEPTED MANUSCRIPT

financial distress reduce health capital investment. Because the previous section implies

T

that health and human capital are complements, however, this result may also indicate

RI P

that firms with more human capital at risk have lower leverage due to bankruptcy costs (Berk, Stanton, and Zechner (2010)).

SC

The second column in Table 5 shows the second stage estimates from a two-stage least squares regression of market to book on log health insurance premiums per employee as

NU

in Eq. (2). The coefficient on insurance premiums is positive and statistically significant, indicating that health capital investment positively affects shareholder value. The second specification includes industry fixed effects based on two-digit SIC codes to

MA

control for unobserved heterogeneity in the contracting environment. First stage results are very similar. The second stage firm value results indicate that the partial effect of health

ED

capital investment is positive, and the magnitude of the effect is higher.20 Including state fixed effects is important for explaining variation in the geography of

PT

health spending (Skinner (2011)), local labor market conditions that affect firm policies toward labor (Kedia and Rajgopal (2009)), and firm performance (Pirinsky and Wang

AC CE

(2006)). It is also necessary because I use state level insurance mandates as an instrument for health capital investment. Furthermore, local labor and health market characteristics may also affect variation in persons covered. These state fixed effects, however, are constant over time by construction, but it is not possible, for example, to rule out that time-varying unobserved heterogeneity in the state political environment affects insurance mandates. Moreover, state-specific trends in persons covered might be correlated with unobserved shocks to firm value and productivity. To control for state-specific time trends, the third specification of Table 5 include state by year fixed effects. The sample ends at 2002 for this specification because the health insurance mandate data I use do not vary after 2002. In the first stage regression with state by year fixed effects, the coefficients on persons 20

I do not include firm fixed effects given the importance of state fixed effects both for explaining variation in health capital investment and for using a state level instrument. In unreported results, the coefficient on insurance premiums in the second stage regression is positive, but it is not statistically significant. This result suggests cross firm variation drives the positive effect of health capital investment on firm value, which is not surprising given that the results in Table 4 indicate that health capital investment is persistent.

20

ACCEPTED MANUSCRIPT

covered and mandates are qualitatively similar, but the coefficient on mandates is now sta-

T

tistically insignificant. In the second stage regression, the effect of health capital investment

RI P

is again positive and statistically significant.

Overall, in regressions that account for unobserved heterogeneity across states, health

SC

capital investment positively affects shareholder value. On average, a 1% increase in health insurance expenditures increases firm value by 0.05% to 0.07%, holding other factors con-

NU

stant. The effect on market to book captures the effect on firm value because investment in health capital is investment in an intangible asset, but the book value of assets does not capture this investment.21 The positive effect on firm value suggests that either it is difficult

MA

for the market to value or that managers may underinvest in health capital precisely because it is difficult to value (Lev (2004)). The magnitude of the effect is economically small,

ED

however, suggesting that firms may not deviate drastically from optimal health capital investment policy in contrast to the conventional wisdom that health expenditures reduce

Firm investment in health and industry compensation characteristics

AC CE

4.3

PT

firm value (Semenova and Kelton (2008)).

Previous studies in economics consider health exclusively as a component of compensation. In largely omitting wages from the analysis thus far, I do not mean to suggest that health insurance cannot be a part of compensation, only that it may not be the only model for firms. In this section I consider the health capital investment framework in conjunction with the more traditional compensating wage differentials approach in which workers are willing to accept lower wages in exchange for health insurance. As noted earlier, wage data in Compustat are sparse.22 To capture wages for the broader cross section in this paper, I measure industry level wage using wages and salaries per fulltime equivalent employee from the BEA NIPA tables23 and match the data to the firm level 21

Chan, Lakonishok, and Sougiannis (2001) make a similar point regarding firm expenditures on research and development. 22 When the baseline regressions include labor and related expenses per employee from Compustat, the results are similar but the number of firm-year observations drops to 679, about 5% of the sample. 23 U.S. Bureau of Economic Analysis, Table 6. Income and Employment by Industry (accessed June 10,

21

ACCEPTED MANUSCRIPT

data using SIC codes for 1992 to 2000 and NAICS codes for 2001 to 2005.24 Table 6 shows

T

results from including the industry wage control. Each specification includes year, state,

RI P

and industry fixed effects. The first specification shows the first and second stage results. In the first stage regression, the coefficient on industry wages is positive and statistically

SC

significant, confirming the previous finding in the labor literature of a positive correlation between wages and health insurance. Additionally, labor productivity, size, and research

NU

and development expenditures continue to have a positive and statistically significant effect on health, suggesting a complementary role for the the investment framework advanced here and the role of health insurance as compensation. In the second stage, the effect of health

MA

capital investment on firm value is very similar in magnitude to those reported in Table 5, indicating the firm value results are robust to controlling for industry level wages.

ED

In addition to controlling for wages, the effect of investment in intangible health capital on firm value might depend on the wage level of the firm. Differences could occur for several

PT

reasons. Under the health capital investment framework I propose, firms with higher human capital workers benefit more from the investment in health, but these firms with higher wage

AC CE

workers might also be the type at which managers find it more difficult to value health. High wage firms may also benefit more by compensating workers with more health benefits because the tax deductibility makes it cheaper to with health insurance relative to cash. On the other hand, in a compensating wage differentials framework, firms that pay higher wages may be more able to pass on health costs to employees, bringing them closer to optimal health expenditures. In the next two specifications of Table 6, I split the sample into low industry wage and high industry wage firms. Looking at the first stage health regression for low industry wage firms in specification (2), the coefficient on the mandates variable is now positive and statistically significant at the 5% level, suggesting that the effectiveness of this instrument may be sensitive to the sample of firms. For example, low wage firms may spend more on 2015). 24 Matching yields about 60 industry level observations per year depending on whether I used SIC codes or NAICS codes.

22

ACCEPTED MANUSCRIPT

health insurance with an increase in mandates if they are more likely to provide basic plans

T

to workers.25 In the second stage market to book regression for low industry wage firms,

RI P

insurance premiums have a positive and statistically significant effect on firm value. The third specification in Table 6 shows the first and second stage regressions for high industry

SC

wage firms. The first stage results are similar to the full sample. For the second stage, the positive and statistically significant effect on health is consistent with the hypothesis

NU

that firms that find it difficult to value investment in intangible health may not be investing enough. However, the strong positive effect of health on the value of low wage firms suggests that high wage firms do not benefit exclusively from using health insurance to compensate

MA

workers cheaply. These results suggest that exclusively considering health insurance as a

Total factor productivity

5.1

Total factor productivity and firm value

PT

5

ED

component of wages cannot fully explain firm behavior.

AC CE

In the simple framework described above, firms invest in health capital to mitigate depreciation in human capital and make labor and physical capital more productive. I explore the effect of health capital investment on total factor productivity as a natural analog to understanding the effect of health capital investment on shareholder value. I first estimate total factor productivity from a Cobb-Douglas model by regressing the log of deflated sales on log of number of employees and log of capital stock. To control for possible endogeneity, I estimate the model using firm fixed effects. The residuals from this regression are estimates of total factor productivity. One can interpret total factor productivity as the deviation from mean productivity. If investment in health capital mitigates depreciation, then in the absence of the health capital investment, the deviation from average efficiency should be negative when a shock causes the human capital to be absent from the production process. Similarly, a health capital 25

Gruber (1994b) find that self-funding firms voluntarily offer certain benefits more often than firms that are subject to state mandates.

23

ACCEPTED MANUSCRIPT

investment should cause a corresponding positive deviation in average efficiency. A change

T

in total factor productivity captures the effect on both labor and physical capital, and the

RI P

comparative statics discussed above indicate that health capital should affect both, either directly or indirectly.

SC

I then examine the effect of health capital investment on total factor productivity using the following regression:

NU

T F Pist = βT F P hist + Xist γ + θt + ψs + ci + υist ,

(3)

MA

where the dependent variable for firm i in state s in year t is total factor productivity, hist is log of health insurance premiums per employee, Xist is a set of controls, and υist is an error term. The null hypothesis is that the marginal effect of another dollar spent on health

ED

care is zero.

The total factor productivity regression helps reinforce the effect on firm value. While

PT

the instrumental variables proposed in the firm value section meet the requirements for exogenous variation, they are not ideal for interpreting the results. In light of this potential

AC CE

drawback, estimating the effect of health capital investment on total factor productivity provides a way to judge the consistency of the results. Because unobserved variables may drive both health capital investment and productivity, I use the same instruments for health insurance premiums as in the firm value regressions to estimate Eq. (3).

5.2

Firm investment in health and productivity

Table 7 shows results from estimating a regression of total factor productivity on a proxy for health capital investment described by Eq. (3). Each specification includes year, state, and industry fixed effects as well as the same controls as the market to book value regressions. In the first column, the coefficient on log of insurance premiums per employee is positive and significant, which is consistent with the assumption in the theoretical framework that health capital investment positively affects productivity. In effect, investment per

24

ACCEPTED MANUSCRIPT

employee has a positive and significant effect on total factor productivity.

T

When estimating total factor productivity, firms do not vary according to productivity

RI P

because the coefficients are the same for all firms in the sample. I re-estimate total factor productivity as the residual from a regression of the log of deflated sales on log of number

SC

of employees and log of capital stock by industry using the five Fama and French industries (available on Professor French’s website). The second column of Table 7 shows results

NU

from regressions of health capital investment on total factor productivity by industry. The results are nearly identical to the results in the first column in both sign and magnitude. Health capital investment positively affects total factor productivity as in the theoretical

MA

framework.

In sum, the total factor productivity results provide evidence for why health capital

ED

investment positively affects firm value. These tests reinforce the assumption of the theoretical framework. Additionally, these results provide simple evidence of a link between

Conclusion

AC CE

6

PT

productivity and firm value.

I propose a framework in which firms invest in human health capital–an intangible investment treated as an expense. I argue that firms mitigate depreciation in a worker’s human capital by investing in health to increase the productivity of human and physical capital. Firms may also use health benefits to compensate workers, but the health capital investment view acknowledges that because firms use worker human capital in production, they also value worker health. I find evidence that firms that have higher levels of labor productivity and spend more on research and development, a proxy for human capital, invest more in health capital. These results imply that human capital and health capital are complements and that we should see a positive relation between wages and health insurance. There is additional evidence that larger firms, as measure by total assets, spend more on health capital even

25

ACCEPTED MANUSCRIPT

when controlling for the number of employees. Results from an instrumental variables

T

regression indicate that the marginal effect of health capital investment on shareholder

RI P

value is positive and statistically significant. It is not possible to reject the hypothesis that health capital investment positively affects firm value. To identify the effect, I instrument

SC

for health insurance premiums with a time series of state level health insurance mandates and the number of persons covered by health insurance contracts.

NU

The ratio of market to book value reflects the value of this investment because health capital is an intangible investment treated as an accounting expense. Hence, we expect to observe this effect in market value rather than book value. Furthermore, health capital

MA

investment positively affects productivity, a connection between firm value and productivity that is often absent in the literature.

ED

Finally, this research provides useful information for policymakers. As policymakers attempt to confront both the rising cost of health entitlement programs and the large

PT

number of Americans without health insurance, the role of the firm is key: sixty percent of Americans receive health insurance through employers. Understanding firm motives

AC CE

is important in any health care reform because it affects the number of people covered and because firm incentives and actions are going to affect changing health costs. If the investment in health capital is a value enhancing project for at least some firms, then the role of the employer in health insurance will not disappear. Even with the ongoing implementation of the Affordable Care Act, many private firms continue to offer health benefits to employees. Furthermore, the incentives of the firm to maintain productive workers while minimizing costs may be important to informing policy goals in the face of changing health costs.

26

ACCEPTED MANUSCRIPT

Data Appendix

T

7

RI P

I obtain data from IRS Form 5500 from the Department of Labor, which is available to anyone who files a Freedom of Information Act Request. This form is an annual report of employee benefit plans including pension benefits, welfare benefits, and fringe benefits. All

SC

plans subject to Employee Retirement Income Security Act (ERISA) must file this form. Firms file Form 5500 data on a per plan basis. I first match Form 5500 data to Compustat

NU

data using an identifier. There were several changes in the form between 1998 and 1999. Until 1999, firms submitted a CUSIP number with Form 5500. For 1991 to 1998, I matched

MA

filings to Compustat using CUSIP. After 1999, firms no longer reported a CUSIP, so I matched Form 5500 data to Compustat using the Employer Identification Number. In both cases, I handchecked the data to eliminate any false positives. This merge resulted

ED

in approximately 40,000 to 50,000 observations per year. The only exception is for 2006 which had half as many observations. This is most likely because I filed the request in

PT

2007 before all forms for year 2006 were filed. I then restricted the data to welfare benefit plans, which include benefits such as medical, dental, life insurance, apprenticeship and

AC CE

training, scholarship funds, severance pay, and disability. This step reduces the number of observations by about 25%. I then restrict the sample to plans with medical, dental, and vision plans. This reduces the sample to 4,000 to 9,000 observations per year. (Additionally, this reduces the sample for 1991 to 44 observations.) There are more observations following the change in the form starting in 1999 because small and large plans are consolidated on the same form. If firms use contracts with insurers for welfare plan funding and benefit arrangements, they must file Schedule A. After restricting the sample to welfare benefit plans with medical, dental, and vision benefits, I merge these plans with Schedule A data. Schedule A contains information about the insurance contract, including coverage and fees. For welfare benefit contracts, it also has detailed information about experience-rated and nonexperience-rated contracts, including premiums. A single welfare benefit plan may have more than one

27

ACCEPTED MANUSCRIPT

Schedule A. I then aggregate all Schedule A at the Form 5500 level. I subsequently aggregate

T

Form 5500 data at the firm level. This aggregation leaves 44,237 firm year observations for

RI P

7,136 firms from 1991 to 2006.

From this data set, I then drop observations based on data availability. I require obser-

SC

vations to have basic financial data as well as health insurance data. I drop any observation that does not have data on sales, number of employees covered, or the number of persons

NU

covered by insurance contracts. I also drop data from 1991 and 2006 because of data limitations described above. These reduction lead to 30,795 firm year observations for 5,744 firms from 1992 to 2005.

MA

Health insurance premiums are the sum of experience rate premiums received by the insurer plus nonexperience rated premiums. I drop outliers for health insurance premiums

ED

and leverage by dropping any observations below the 1st percentile and above the 99th percentile. I exclude financial firms (with primary standard industrial classification (SIC)

PT

codes between 6000 and 6999). This step leaves 22,855 firm year observations for 4,485 observations from 1992 to 2005. Most plans are filed by single employers or a controlled

AC CE

group of corporations.26 I further restrict data to plans that are either filed by a single employer or the plan of a controlled group of corporations. In the regressions, I also only use data for firms that are single employer plans or the plan of a controlled group of corporations. As detailed in Table 1, I use research and development expenditures and advertising expenditures scaled by total assets as control variables. Because there are many missing observations in Compustat, and I do not want to bias results by only using those firms which report these variables, I set any missing observations to zero following Palia (2001).

26 A controlled group of corporations is two or more corporations that are component members of a ‘parent-subsidiary’ controlled group, a ‘brother-sister’ controlled group or a ‘combined control’ group.

28

ACCEPTED MANUSCRIPT

References

RI P

T

Bae, K.-H., J.-K. Kang, and J. Wang (2011): “Employee treatment and firm leverage: A test of the stakeholder theory of capital structure,” Journal of Financial Economics, 100, 130–153.

SC

Ballester, M., J. Livnat, and N. Sinha (2002): “Labor Costs and Investments in Human Capital,” Journal of Accounting, Auditing, and Finance, 17, 351–373. Bartel, A., and F. R. Lichtenberg (1987): “The comparative advantage of educated workers in implementing new technology,” Review of Economics and Statistics, 69, 1–11.

NU

Bazdresch, S., F. Belo, and X. Lin (2014): “Labor hiring, investment, and stock return predictability in the cross section,” Journal of Political Economy, 122, 129–177.

MA

Berk, J. B., R. Stanton, and J. Zechner (2010): “Human Capital, Bankruptcy and Capital Structure,” Journal of Finance, 65, 891–926.

ED

Brien, M. J., and C. W. Panis (2011): “Large Group Health Plans Study,” Commissioned report by Deloitte Financial Advisory Services LLP and Advanced Analytical Consulting Group, Inc. for U.S. Department of Labor. Chan, L. K. C., J. Lakonishok, and T. Sougiannis (2001): “The stock market valuation of research and development expenditures,” Journal of Finance, 56, 2431–2456.

PT

Cho, M.-H. (1998): “Ownership structure, investment, and the corporate value: an empirical analysis,” Journal of Financial Economics, 47, 103–121.

AC CE

Coles, J. L., M. L. Lemmon, and F. Meschke (2012): “Structural Models and Endogeneity in Corporate Finance: The Link between Managerial Ownership and Corporate Performance,” Journal of Financial Economics, 103, 149–168. Cronqvist, H., F. Heyman, M. Nilsson, H. Svaleryd, and J. Vlachos (2009): “Do Entrenched Managers Pay Their Workers More?,” Journal of Finance, 64, 309–339. Currie, J., and B. C. Madrian (1999): “Health, health insurance, and the labor market,” Handbook of Labor Economics, 3, 3309–3416. Cutler, D. (2003): “Employee Costs and the Decline in Health Insurance Coverage,” Frontiers in Health Policy Research, 6, 1–27. Dafny, L. S. (2010): “Are Health Insurance Markets Competitive?,” American Economic Review, 100, 1399–1431. Decressin, A., J. Lane, K. McCue, and M. H. Stinson (2005): “Employer-Provided Benefit Plans, Workforce Composition and Firm Outcomes,” U.S. Census Bureau, Longitudinal Employer - Household Dynamics Technical Paper No TP-2005-01. D’Heilly, S. J., and K. L. Nichol (2004): “Work-site-based influenza vaccination in healthcare and non-healthcare settings,” Infection Control and Hospital Epidemiology, 25, 941–945. 29

ACCEPTED MANUSCRIPT

T

Dranovea, D., K. E. Spiera, and L. Baker (2000): “‘Competition’ among employers offering health insurance,” Journal of Health Economics, 10, 121–140.

RI P

Edmans, A. (2011): “Does the stock market fully value intangibles? Employee satisfaction and equity prices,” Journal of Financial Economics, 101, 621–640.

SC

Finkelstein, A., S. Taubman, B. Wright, M. Bernstein, J. Gruber, J. P. Newhouse, H. Allen, K. Baicker, and the Oregon Health Study Group (2011): “The Oregon Health Insurance Experiment: Evidence from the First Year,” NBER Working Paper 17190.

NU

Gompers, P., J. Ishii, and A. Metrick (2003): “Corporate Governance and Equity Prices,” Quarterly Journal of Economics, 118, 107–155.

MA

Grossman, M. (1972): “On the concept of health capital and the demand for health,” Journal of Political Economy, 80, 223–255. Gruber, J. (1994b): “State-mandated benefits and employer-provided health insurance,” Journal of Public Economics, 55, 433–464.

ED

Gruber, J., and M. Lettau (2004): “How elastic is the firm’s demand for health insurance?,” Journal of Public Economics, 88, 173–1293.

PT

Gruber, J., and B. C. Madrian (2002): “Health insurance, labor supply, and job mobility: A critical review of the literature,” NBER Working Paper 8817. Hamermesh, D., and G. A. Pfann (1996): “Adjustment costs in factor demand,” Journal of Economic Literature, 34, 1264–1292.

AC CE

Himmelberg, C. P., R. G. Hubbard, and D. Palia (1999): “Understanding the Determinants of Managerial Ownership and the Link Between Ownership and Performance,” Journal of Financial Economics, 53, 353–384. Jensen, M. (1986): “Agency costs of free cash flow, corporate finance, and takeovers,” American Economic Review, 76, 323–329. Kahle, K. M., and R. M. Stulz (2017): “Is the U.S. public corporation in trouble?,” Journal of Economic Perspectives, 31, 67–88. Kaiser Family Foundation, and Health Research and Educational Trust (2009): Employer Health Benefits Annual Survey. Kedia, S., and S. Rajgopal (2009): “Neighborhood matters: The impact of location on broad based stock option plans,” Journal of Financial Economics, 92, 109–127. Kothari, S. P., T. E. Laguerre, and A. J. Leone (2002): “Capitalization versus Expensing: Evidence on the Uncertainty of Future Earnings from Capital Expenditures versus R&D Outlays,” Review of Accounting Studies, 7, 355–382. Lang, L. H., and R. M. Stulz (1994): “Tobin’s q, Corporate Diversification, and Firm Performance,” Journal of Political Economy, 102, 1248–1280. 30

ACCEPTED MANUSCRIPT

RI P

T

Lemmon, M. L., M. R. Roberts, and J. F. Zender (2008): “Back to the Beginning: Persistence and the Cross-Section of Corporate Capital Structure,” Journal of Finance, 63, 1575–1608. Lev, B. (2004): “Sharpening the intangibles edge,” Harvard Business Review, pp. 109–116.

SC

Lev, B., and T. Sougiannis (1996): “The Capitalization, Amortization, and ValueRelevance of R&D,” Journal of Accouting and Economics, 21, 107–138. Levy, H., and R. Meltzer (2008): “The Impact of Health Insurance on Health,” Annual Review of Public Health, 29, 399–409.

NU

McConnell, J. J., and C. J. Muscarella (1985): “Corporate capital expenditure decisions and the market value of the firm,” Journal of Financial Economics, 14, 399– 422.

MA

McConnell, J. J., and H. Servaes (1990): “Additional evidence on equity ownership and corporate value,” Journal of Financial Economics, 27, 595–612.

ED

McTier, Brian C., T. Y., and J. K. Wald (2013): “Do stock markets catch the flu?,” Journal of Financial and Quantitative Analysis, 48, 979–1000. Mincer, J. (1991): “Human capital technology and the wage structure - What does the time series show?,” NBER Working Paper 3581.

PT

Morck, R., A. Shleifer, and R. W. Vishny (1988): “Management ownership and market valuation: an empirical analysis,” Journal of Financial Economics, 20, 293–315.

AC CE

Morrisey, M. A. (2008): Health Insurance. Health Administration Press. Mushkin, S. J. (1962): “Health as an investment,” Journal of Political Economy, 70(5), 129–157. Neal, D. (1995): “Industry-specific human capital: Evidence from displaced workers,” Journal of Labor Economics, 13(4), 653–677. Nichol, K. L. (2001): “Cost-benefit analysis of a strategy to vaccinate healthy working adults against influenza,” Archives of Internal Medicine, 161, 749–759. Nichol, K. L., D. S. J., M. E. Greenberg, and E. Ehlinger (2009): “Burden of influenza-like illness and effectiveness of influenza vaccination among working adults aged 50-64 years,” Clinical Infectious Diseases, 48, 292–298. Palia, D. (2001): “The endogeneity of managerial compensation in firm valuation: A solution,” Review of Financial Studies, 14(3), 735–764. Pauly, M. V. (1998): Health Benefits at Work: An Economic and Political Analysis of Employment-Based Health Insurance. University of Michigan Press. Pirinsky, C., and Q. Wang (2006): “Does corporate headquarters location matter for stock returns?,” Journal of Finance, 61, 1991–2015. 31

ACCEPTED MANUSCRIPT

RI P

T

Semenova, A., and S. Kelton (2008): “Are Rising Health Care Costs Reducing U.S. Global Competitiveness,” Working Paper, Center for Full Employment and Price Stability. Skinner, J. (2011): “Causes and Consequences of Regional Variations in Health Care,” Handbook of Health Economics, 2, 45–93.

SC

Topel, R. (1991): “Specific capital, mobility, and wages: Wages rise with job seniority,” Journal of Political Economy, 99, 145–176.

NU

Wintoki, M. B., J. S. Linck, and J. M. Netter (2012): “Endogeneity and the Dynamics of Internal Corporate Governance,” Journal of Financial Economics, 105, 581–606.

AC CE

PT

ED

MA

Zingales, L. (2000): “In search of new foundations,” Journal of Finance, 55, 1623–1653.

32

ACCEPTED MANUSCRIPT

9,000

MA

8,000

6,000

ED

5,000

3,000 2,000

12,000

10,000

8,000

6,000

PT

4,000

4,000

2,000

AC CE

Thousands of dollars

7,000

14,000

1,000 0

0

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 Year

33

Number

NU

SC

RI P

T

Figure 1: Summary Health Benefit Plan Data. This figure shows information on welfare benefit plans by year. The dotted red line shows annual mean premiums in thousands of dollars (left axis). Premiums is the annual mean of sum of the amount of experience rated premiums received and nonexperience rated premiums from Schedule A of IRS Form 5500. The solid black line shows mean number of firms (right axis). The dashed blue line shows persons covered (right axis). Persons Covered is the total number of persons covered summed across all insurance contracts.

ACCEPTED MANUSCRIPT

AC CE

PT

ED

MA

NU

SC

RI P

T

Figure 2: Mandates Across the States. This table shows the prevalence of mandates across states in 1992 and 2002. Mandate data is only available through 2002. States in Quartile 1 are in the lowest quartile. States in Quartile 4 are the highest quartile. The top panel shows the distribution of mandates by state in 1992 by quartile. The 25th, 50th, and 75th percentiles in 1992 are 13, 17, and 21. The bottom panel shows the distribution of mandates by state in 2002 by quartile. The 25th, 50th, and 75th percentiles in 2002 are 23, 28, and 34.

34

RI P

T

ACCEPTED MANUSCRIPT

SC

Table 1: Summary Statistics for Firm Characteristics The table shows summary statistics for firm characteristics. The sample includes 4,233 firms from 1992 to 2005. All data are from Compustat. Standard Deviation 0.40

Firm-Year Observations 12,359

Logarithm of total assets

0.35

0.62

12,359

Research and development expenses scaled by total assets

0.06

0.12

12,359

0.27

0.22

12,359

Operating profits scaled by total assets

0.08

0.20

12,359

Collectively Bargained

1 for firms with a collectively bargained health benefit plan and 0 otherwise

0.07

0.26

12,359

Market to Book Value

(Total assets - book value of equity + market value of equity)/(Total assets)

2.01

1.51

12,359

Leverage

Long-term debt scaled by long term debt plus market value of equity

0.19

0.23

12,359

Advertising

Advertising expenses scaled by total assets

0.02

0.06

12,359

NU

Mean 0.25

Description Sales scaled by number of employees

Size Research and Development

ED

Capital stock scaled by total assets

AC CE

PT

Capital Intensity

Free Cash Flow

MA

Variable Labor Productivity

35

RI P

T

ACCEPTED MANUSCRIPT

Table 2: Health Capital Investment

=

Xist γ + θt + ψs + ci + uist

NU

hist

SC

This table shows results from the following regression of health capital investment expenditures on proposed determinants:

MA

where the dependent variable for firm i in state s in year t is log insurance premiums per employee, Xist is matrix of proposed determinants, and uist is an error term. Robust standard errors clustered by firm are shown in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level, respectively. Each specification is from 1992 to 2005. The sample is restricted to single employer and controlled group plans. 0.267∗∗∗ (0.060)

0.328∗∗∗ (0.064)

0.290∗∗∗ (0.056)

0.333∗∗∗ (0.046)

0.245∗∗∗ (0.046)

0.079 (0.052)

2.327∗∗∗ (0.256)

1.612∗∗∗ (0.247)

0.402∗ (0.212)

-1.062∗∗∗ (0.131)

-0.214 (0.143)

-0.052 (0.192)

Free Cash Flow

-0.759∗∗∗ (0.111)

-0.841∗∗∗ (0.103)

-0.184∗∗ (0.083)

Collectively Bargained

0.092 (0.089)

-0.132 (0.085)

-0.056 (0.075)

R2

0.31

0.38

0.15

Number of Observations Year Fixed Effects State Fixed Effects Industry Fixed Effects Firm Fixed Effects

12,359 Yes Yes No No

12,359 Yes Yes Yes No

12,359 Yes No No Yes

Size

AC CE

Capital Intensity

PT

Research and Development

ED

Labor Productivity

36

RI P

T

ACCEPTED MANUSCRIPT

Table 3: Past Firm Performance and Health Capital Investment

hist

=

SC

This table shows results from the following regression of health capital investment expenditures on proposed determinants: β1 roaist−1 + β2 roaist−2 + Xist γ + θt + ψs + ci + uist

NU

where the dependent variable for firm i in state s in year t is log insurance premiums per employee, Xist is matrix of proposed determinants, and uist is an error term. Robust standard errors clustered by firm are shown in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level, respectively. Each specification is from 1992 to 2005. The sample is restricted to single employer and controlled group plans. -0.269∗∗∗ (0.076)

-0.216∗∗∗ (0.070)

-0.063∗ (0.037)

-0.297∗∗∗ (0.064)

-0.275∗∗∗ (0.060)

0.015 (0.036)

0.152∗∗ (0.072)

0.198∗∗∗ (0.071)

0.345∗∗∗ (0.097)

0.414∗∗∗ (0.066)

0.363∗∗∗ (0.062)

0.123 (0.092)

Research and Development

2.737∗∗∗ (0.405)

2.123∗∗∗ (0.375)

0.381 (0.283)

Capital Intensity

-1.270∗∗∗ (0.174)

-1.069∗∗∗ (0.168)

-0.234 (0.287)

Free Cash Flow

-0.425∗ (0.233)

-0.502∗∗ (0.212)

0.093 (0.121)

Collectively Bargained

-0.060 (0.107)

-0.013 (0.105)

-0.075 (0.115)

R2 Number of Observations Year Fixed Effects State Fixed Effects Industry Fixed Effects Firm Fixed Effects

0.27 6,446 Yes Yes No No

0.32 6,446 Yes Yes Yes No

0.17 6,446 Yes No No Yes

MA

Return on Assetst−1

ED

Return on Assetst−2

PT

Labor Productivity

AC CE

Size

37

T

ACCEPTED MANUSCRIPT

RI P

Table 4: Persistence of Health Capital Investment

This table shows results from the following regression of health capital investment expenditures on proposed determinants: =

β1 hist−1 + β2 hist−2 + β3 hist−3 + Xist γ + θt + ψs + ci + uist

SC

hist

NU

where the dependent variable for firm i in state s in year t is log insurance premiums per employee, Xist is matrix of proposed determinants, and uist is an error term. Robust standard errors clustered by firm are shown in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level, respectively. Each specification is from 1992 to 2005. The sample is restricted to single employer and controlled group plans. 0.593∗∗∗ (0.020)

0.571∗∗∗ (0.021)

0.183∗∗∗ (0.025)

0.191∗∗∗ (0.020)

0.184∗∗∗ (0.020)

0.026 (0.021)

0.084∗∗∗ (0.014)

0.085∗∗∗ (0.014)

-0.009 (0.018)

0.081 (0.058)

0.091 (0.066)

0.171 (0.131)

0.062∗∗∗ (0.023)

0.069∗∗ (0.029)

0.192∗ (0.106)

Research and Development

0.415∗∗ (0.167)

0.304 (0.193)

0.500∗ (0.286)

Capital Intensity

-0.226∗∗∗ (0.065)

-0.108 (0.083)

0.035 (0.309)

Free Cash Flow

-0.242∗∗ (0.112)

-0.266∗∗ (0.115)

-0.036 (0.157)

Collectively Bargained

-0.118∗∗ (0.048)

-0.144∗∗∗ (0.052)

0.003 (0.127)

R2 Number of Observations Year Fixed Effects State Fixed Effects Industry Fixed Effects Firm Fixed Effects

0.77 4,600 Yes Yes No No

0.78 4,600 Yes Yes Yes No

0.23 4,600 Yes No No Yes

MA

Insurance Premiumst−1

Insurance Premiumst−3

PT

Labor Productivity

ED

Insurance Premiumst−2

AC CE

Size

38

ACCEPTED MANUSCRIPT

T

Table 5: Health Capital Investment and Firm Value

hist M/Bist

RI P

This table shows results from a two stage least squares regression of a proxy for firm value on health capital expenditures and other controls: =

Xist γ + δ1 mandatesst + δ2 personscoveredist + θt + ψs + ci + uist

=

βhist + Xist γ + θt + ψs + ci + ǫist ,

MA

NU

SC

where the dependent variable for firm i in state s in year t is market to book, hist is log insurance premiums per employee, Xist is a set of controls, and ǫist is an error term. In each specification, the left column is the first stage of the instrumental variables regression. The right column is the second stage regression of market to book on explanatory variables. The endogenous variable is log of insurance premiums per employee. The instruments are the number of insurance mandates in the state, the number of persons covered by insurance premiums, and the remaining exogenous variables. Robust standard errors clustered by firm are shown in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level, respectively. Specifications (1) and (2) are from 1992 to 2005, and specification (3) is from 1992 to 2002. (1)

h 0.784∗∗∗ (0.017) 0.014∗ (0.008)

ED

Persons Covered Mandates

Labor Productivity

AC CE

Size

PT

Insurance Premiums

Research and Development Capital Intensity Free Cash Flow

Collectively Bargained Leverage Advertising

R2 F-statistic p-value for test E[ǫist zist ] = 0 Number of Observations Year Fixed Effects State Fixed Effects Industry Fixed Effects State by Year Fixed Effects

0.103∗∗∗ (0.035) 0.144∗∗∗ (0.030) 1.013∗∗∗ (0.181) -0.489∗∗∗ (0.099) -0.609∗∗∗ (0.089) 0.070 (0.067) -0.282∗∗∗ (0.074) -0.950∗∗ (0.384) 0.49 179.93 0.28 12,359 Yes Yes No No

M/B

0.056∗∗∗ (0.021) -0.145∗∗∗ (0.033) 0.122∗∗∗ (0.031) 4.558∗∗∗ (0.437) -0.321∗∗∗ (0.085) 2.033∗∗∗ (0.202) -0.098∗∗ (0.039) -1.527∗∗∗ (0.074) 1.023∗∗∗ (0.396)

39

(2) h 0.741∗∗∗ (0.017) 0.014∗ (0.007)

0.133∗∗∗ (0.037) 0.111∗∗∗ (0.031) 0.702∗∗∗ (0.181) -0.170 (0.113) -0.645∗∗∗ (0.086) -0.042 (0.066) -0.244∗∗∗ (0.075) -0.305 (0.357) 0.42 154.30 0.19 12,359 Yes Yes Yes No

(3) M/B

0.070∗∗∗ (0.023) -0.052 (0.035) 0.049 (0.030) 4.039∗∗∗ (0.439) -0.443∗∗∗ (0.117) 2.075∗∗∗ (0.198) -0.038 (0.038) -1.479∗∗∗ (0.073) 0.854∗∗ (0.396)

h 0.719∗∗∗ (0.018) 0.012 (0.010)

0.197∗∗∗ (0.057) 0.116∗∗∗ (0.034) 0.623∗∗∗ (0.215) -0.160 (0.122) -0.654∗∗∗ (0.095) 0.014 (0.073) -0.253∗∗∗ (0.080) -0.407 (0.340) 0.36 197.57 0.14 8,547 No No Yes Yes

M/B

0.051∗ (0.026) -0.053 (0.052) 0.109∗∗ (0.049) 3.971∗∗∗ (0.387) -0.371∗∗∗ (0.131) 2.069∗∗∗ (0.196) -0.020 (0.041) -1.421∗∗∗ (0.074) 0.517 (0.352)

ACCEPTED MANUSCRIPT

Table 6: Health Capital Investment and Firm Value: Industry Wage

RI P

T

This table shows results from a two stage least squares regression of a proxy for firm value on health capital expenditures and other controls: hist

=

Xist γ + δ1 mandatesst + δ2 personscoveredist + θt + ψs + ci + uist

M/Bist

=

βhist + Xist γ + θt + ψs + ci + ǫist ,

MA

(1)

h 0.735∗∗∗ (0.017) 0.012 (0.007)

Persons Covered

ED

Mandates

AC CE

Size

0.133∗∗∗ (0.036) 0.099∗∗∗ (0.033) 0.644∗∗∗ (0.179) -0.182 (0.114) -0.644∗∗∗ (0.087) -0.059 (0.065) -0.253∗∗∗ (0.076) -0.269 (0.355) 0.287∗∗∗ (0.080)

PT

Insurance Premiums Labor Productivity

NU

SC

where the dependent variable for firm i in state s in year t is market to book, hist is log insurance premiums per employee, Xist is a set of controls, and ǫist is an error term. In each specification, the left column is the first stage of the instrumental variables regression. The right column is the second stage regression of market to book on explanatory variables. Industry wage is wages and salaries per full-full time employee. Specification (1) is the full sample, specification (2) is low industry wage firms, and specification (3) is high industry wage firms. Robust standard errors clustered by firm are shown in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level, respectively. Each specification is from 1992 to 2005.

Research and Development Capital Intensity Free Cash Flow

Collectively Bargained Leverage Advertising Industry Wages

R2 F-statistic p-value for test E[ǫist zist ] = 0 Number of Observations Year Fixed Effects State Fixed Effects Industry Fixed Effects

M/B

0.072∗∗∗ (0.024) -0.054 (0.037) 0.108∗∗ (0.043) 4.080∗∗∗ (0.442) -0.448∗∗∗ (0.120) 2.067∗∗∗ (0.201) -0.032 (0.039) -1.488∗∗∗ (0.075) 0.845∗∗ (0.397) -0.256∗∗∗ (0.092)

0.42 146.51 0.23 12,097 Yes Yes Yes

40

(2) Low Wage h M/B 0.695∗∗∗ (0.021) 0.022∗∗ (0.009) 0.079∗∗ (0.033) 0.261∗∗ -0.063 (0.103) (0.061) 0.099∗∗ 0.130∗∗ (0.047) (0.063) 0.965∗∗∗ 3.534∗∗∗ (0.330) (0.602) -0.042 -0.652∗∗∗ (0.155) (0.161) -0.597∗∗∗ 2.397∗∗∗ (0.125) (0.258) -0.03 -0.004 (0.089) (0.049) -0.299∗∗∗ -1.335∗∗∗ (0.100) (0.085) -0.058 1.162∗∗∗ (0.344) (0.428) 0.081 0.459∗∗ (0.205) (0.193)

(3) High Wage h M/B 0.783∗∗∗ (0.023) 0.014 (0.012) 0.059∗ (0.033) 0.059 -0.048 (0.041) (0.049) 0.087∗∗ 0.093∗ (0.041) (0.055) 0.544∗∗∗ 3.987∗∗∗ (0.190) (0.554) -0.355∗∗ -0.204 (0.150) (0.182) -0.653∗∗∗ 1.826∗∗∗ (0.105) (0.277) -0.102 -0.106∗ (0.087) (0.063) -0.180∗ -1.678∗∗∗ (0.108) (0.133) -1.762∗∗ 0.078 (0.809) (0.746) 0.346∗∗ -0.382∗ (0.171) (0.204)

0.39 72.74 0.61 6,053 Yes Yes Yes

0.41 74.68 0.01 6,044 Yes Yes Yes

RI P

T

ACCEPTED MANUSCRIPT

Table 7: Total Factor Productivity

SC

This table shows results from the following two stage least squares regression of total factor productivity on a proxy for health capital investment and size controls:

NU

T F Pist = βT F P hist + Xist γ + θt + ψs + ci + υist ,

MA

where the dependent variable for firm i in state s in year t is total factor productivity, hist is log health insurance premiums per employee, Xist is a set of controls, and υist is an error term. The instruments are the number of insurance mandates in the state and the variation in the log of number of persons covered orthogonal to the hiring rate, and the remaining exogenous variables. Robust standard errors clustered by firm are shown in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% level, respectively. Each specification is from 1992 to 2005. The sample is restricted to single employer and controlled group plans.

PT

Labor Productivity Size

TFP 0.020∗∗∗ (0.005) 0.221∗∗∗ (0.044) -0.010 (0.015) 0.090 (0.072) -0.215∗∗∗ (0.027) 0.280∗∗∗ (0.040) 0.011 (0.009) 0.009 (0.017) 0.107 (0.068) 0.10 12,191 Yes Yes Yes

ED

Insurance Premiums

AC CE

Research and Development Capital Intensity Free Cash Flow

Collectively Bargained Leverage

Advertising R2 Number of Observations Year Fixed Effects State Fixed Effects Industry Fixed Effects

41

TFP by Industry 0.022∗∗∗ (0.004) 0.217∗∗∗ (0.044) -0.016 (0.015) 0.095 (0.071) -0.203∗∗∗ (0.026) 0.267∗∗∗ (0.039) 0.009 (0.010) 0.004 (0.017) 0.114∗ (0.069) 0.09 12,191 Yes Yes Yes