Affordable Care Act and Labor Market
Transcription
Affordable Care Act and Labor Market
Affordable Care Act and Labor Market Hanming Fang University of Pennsylvania CEAR/Huebner Summer Risk Institute July 28-29, 2014 1 / 137 Outline 1 2 Patient Protection and Affordable Care Act (ACA): A Brief Introduction US Employment-Based Health Insurance System and Medical Expenditure I 3 Would the Troubles in the Initial Roll Out of HealthCare.gov last? And Does the Form of Individual Mandate Penalty Matter? I 4 Fang and Gavazza (2011, AER): Dynamic Inefficiencies in an Employment Based Health Insurance System: Theory and Evidence Scheuer and Smetters (2014, NBER Working Paper): Could a Website Really Have Doomed the Health Insurance Exchange? Multiple Equilibria, Initial Conditions and the Constructionof the Fine. Equilibrium Labor Market and Health Insurance Reform I I I Aizawa and Fang (2013): Equilibrium Labor Market Search and Helath Insurance Reform Fang and Shephard (WIP): Joint Household Labor Supply and Health Care Reform Fang, Shephard and Tilly (WIP): Equilibrium Labor Market Search with Endogenous Technology Choice and Health Insurance 2 / 137 I: Patient Protection and Affordable Care Act (ACA): A Brief Introduction 3 / 137 Health Insurance Coverage of the Total Population in the US (2011-12) Source Percentage (%) Employer 48 Other Private 5 Medicaid 16 Medicare 14 Other Public 1 Uninsured 15 Total 100 Source: Kaiser Family Foundation 4 / 137 National Health Expenditure as a Share of GDP: 1961-2011 Share of National Health Expenditures in GDP, 1961-2011 20% 18% 16% 14% 12% 10% 8% 6% 4% 2% 0% 1961 1965 1970 1975 1980 1985 1990 1995 2000 2005 2011 NHE as a Share of GDP SOURCE: Kaiser Family Foundation calculations using NHE data from Centers for Medicare and Medicaid Services, Office of the Actuary, National Health Statistics Group, at http://www.cms.hhs.gov/NationalHealthExpendData/ (see Historical; National Health Expenditures by type of service and source of funds, CY 1960-2011; file nhe2011.zip). Gross Domestic Product data from Bureau of Economic Analysis, at http://bea.gov/national/index.htm#gdp (file gdplev.xls). 5 / 137 0 2.4 11.0 10.9 9.1 9.0 Brazil 8.5 Turkey Indonesia India China Russian Fed. 4.2 6.9 Korea 5.4 6.1 6.4 Poland Estonia Mexico 7.4 7.4 7.0 Hungary 7.9 7.8 Israel Luxembourg ³ 8.4 8.2 Chile Czech Rep. 8.5 Japan South Africa 8.7 Slovak Rep. Australia 9.3 9.2 Finland Total Slovenia 9.5 9.5 9.5 Italy Ireland 17.4 % of GDP Spain 9.6 9.6 9.6 OECD 9.7 Iceland Norway Greece 10.0 9.8 United Kingdom 10.1 Sweden 10.3 Portugal New Zealand Belgium ² Austria 11.4 11.4 Canada 12.0 Switzerland 2 11.8 4 11.6 4.6 6 11.5 8 France 10 Denmark 12 Germany Netherlands ¹ 18 United States National Health Expenditure as a Share of GDP in 2011 In Various Countries 20 Residual 16 14 6 / 137 ACA on Access: Title I: Quality, Affordable Health Care for All Americans prohibits insurance companies from denying coverage based on preexisting conditions Caps out of pocket expenditures Extends dependent coverage for young adults to age 26 Requires full coverage for preventive care and immunization creates individual and small-business insurance exchanges establishes tax subsidies for individuals up to 400% of the federal poverty level (FPL) and employers establishes individual and employer mandates and penalties requires minimum medical loss ratio for insurance companies 7 / 137 ACA on Access: Title II: The Role of Public Programs Expands Medicaid to 133% of FPL (with an across-the-board 5% income disregard) I I I I The federal government would pay 100% of the cost of Medicaid expansion from 2014-2016, declining to 95% (2017), 94% (2018), 93% (2019) and 90% (2020 and beyond) Ruled unconstitutional by the Supreme Court, allowing States to opt out Currently, 26 States and DC have expanded Medicaid; 20 decided not to expand; and 4 are still undecided The expanded Medicaid program is supposed to offer a package that fulfills the requirements of the “essential benefits” that is the basis of the health insurance options in the exchange - this is not as comprehensive as the traditional Medicaid benefits. 8 / 137 Where Do States Stand on Medicaid Expansion? 9 / 137 Health Insurance Exchange Title I, Section 1311-1312: States are permitted to create two health insurance exchanges, one for individuals and one for small businesses of up to 100 employees (called SHOP exchanges, and could be opened to larger companies after 2017). Structure of Marketplace: I I I part of an existing state agency or office (Operated by State); an independent public agency (Quasi-governmental); a non-profit entity (Non-profit). Title I, Section 1333: ACA permits groups of several states to affiliate and form a regional exchange instead of each individual state operating its own exchange (in 2014, no such regional exchange was formed). Beginning in 2016, states can allow insurance companies from another state to offer plans in their state, but not on the exchange. 10 / 137 Health Insurance Plans in the Exchange Four different tiers: bronze, silver, gold and platinum, representing different levels of coverage and costs: I I I I bronze: cover about 60% of total predicted health care costs, with a limit on out-of-pocket expense set at $6,250 for individuals and $12,500 for families. The bronze plan quality as meeting the minimum essential coverage necessary for satisfying individual mandate. Silver: cover 70%; Gold: 80%; Platinum: 90%. Catastrophic coverage option: available only to people up to 30 years of age or those exempt from the individual mandate because they can not afford their health plan 11 / 137 Premium Setting of the Health Insurance Plans in the Exchange Premiums for the different levels of plans will vary by state (including geographic rating area within a state), the way the plans are designed, and the networks and prices insurance companies negotiated with hospitals and physicians; I also by age curve and smoking status (Age: 3:1; Smoking status: 1.5:1) DHHS’ guidance on age curve (Feb. 25, 2013): I I I Children: A single age band for children ages 0 through 20. Adults: One-year age bands for adults ages 21 through 63. Older adults: A single age band for adults ages 64 and older. 12 / 137 Subsidies Two types of subsidies (available only to those who purchase insurance from HIX) I I premium subsidy:refundable and advanceable tax credits to helps people whose income is between 100% and 400% of the FPL to pay for the insurance premium cost sharing subsidy:help poorer individuals and families to pay for the out-of-pocket expenses associated with insurance, such as deductibles and co-pays. 13 / 137 Premium Subsidy (Tax Credits) Eligibility for premium subsidy: I I I Citizens and legal residents in families with incomes between 100% and 400% of poverty who purchase coverage through a health insurance exchange are eligible for a tax credit to reduce the cost of coverage. People eligible for public coverage are not eligible for premium assistance in exchanges. In states without expanded Medicaid coverage, people with incomes less than 100% of poverty will not be eligible for exchange subsidies, while those with incomes at or above poverty will be. People offered coverage through an employer are also not eligible for premium tax creditsunless the employer plan does not have an actuarial value of at least 60% or unless the person’s share of the premium for employer-sponsored insurance exceeds 9.5% of income. 14 / 137 Premium Subsidy (Tax Credits) Level of premium subsidy: I I I based on the premium for the second lowest cost silver plan in the exchange and area where the person is eligible to purchase coverage. A silver plan is a plan that provides the essential benefits and has an actuarial value of 70%. A person who wants to purchase a plan that is more expensive would have to pay the full difference between the cost of the second lowest cost silver plan and the plan that they wish to purchase. 15 / 137 Premium Subsidy (Tax Credits) The amount of the tax credit varies with income such that the premium that the premium a person would have to pay for the second lowest cost silver plan would not exceed a specified percentage of their income (adjusted for family size): Income Level Up to 133% FPL 133-150% FPL 150-200% FPL 200-250% FPL 250-300% FPL 300-400% FPL Premium as a Percent of Income 2% of income 3% of income 4% of income 6.3% of income 8.05% of income 9.5% of income 16 / 137 Cost-Sharing Subsidy Cost-sharing subsidies protect lower income people with health insurance from high out-of-pocket costs at the point of service for essential benefits. Generally, the limits are based on the maximum out-of-pocket limits for Health Savings Account-qualified health plans ($5,950 for single coverage and $11,900 for family coverage in 2010), which will be indexed to the change in the Consumer Price Index until 2014 when the provision takes effect. After 2014, the limits will be indexed to the change in the cost of health insurance. 17 / 137 Cost-Sharing Subsidy People with incomes at or below 400% of poverty have their out-of-pocket liability capped at lower levels, as follows: Income Level 100-200% FPL 200-300% FPL 300-400% FPL Reduction in Out-of-Pocket Liability Two-thirds of the maximum One-half of the maximum One-third of the maximum 18 / 137 States’ Choices of Own Exchange or Federal Exchange State Health Insurance Marketplace Decisions, 2014 VT WA ND MT OR WI SD ID WY UT* CA AZ CO NM IL KS MO OK TX HI MI PA OH IN WV KY AR AL VA CT NJ DE MD MA RI DC NC TN MS AK NY IA NE NV ME NH MN SC GA LA FL State-‐based Marketplace (16 states and DC) Partnership Marketplace (7 states) Federally-‐facilitated Marketplace (27 states) * In Utah, the federal government will run the marketplace for individuals while the state will run the small business, or SHOP, marketplace. SOURCE: State Decisions For Creating Health Insurance Marketplaces, 2014, KFF State Health Facts: h"p://kff.org/health-‐reform/state-‐indicator/health-‐insurance-‐exchanges/. 19 / 137 HIX Competition by State Share on Facebook Post on LinkedIn Tweet on Twitter Health Insurance Exchanges: Individual market competition Now that open enrollment has begun and Qualified Health Plans (QHP) application and rate filing deadlines have passed, details surrounding the competitive landscape of each state’s health insurance exchange (HIX) have emerged. Here’s an analysis on the number of medical carriers competing in states’ individual HIX markets in 2014. Projected number of medical carriers in the individual market per state and breakdown of state-based, state-partnership, and federally-facilitated exchanges as of October 2, 2013 WA (8) ME (2) MT (3) ND (3) VT (2) MN (5) OR (11) ID (4) NH (1) WI (13) SD (3) MI (9) IA (4) NE (4) NV (4) IL (5) UT (6) CA (12) MA (10) RI (2) NY (16) WY (2) CO (10) KS (3) OK (4) NM (5) NJ CT (3) NJ (3) DE (2) VA (5) WV (1) MO (3) KY (5) MD (4) DC (3) NC (2) TN (4) AZ (8) PA (8) OH (12) IN (4) SC (4) AR (3) MS (2) AL (2) GA (5) TX (11) LA (4) AK (2) FL (9) HI (2) Legend: 1-3 carriers 4-6 carriers State-based exchange 7-9 carriers 10-12 carriers State-partnership exchange 13+ carriers Federally-facilitated exchange Additional Notes: • Unless otherwise noted, data represents number of applicants based on QHP or rate filing submissions that are publically available. These are subject to changes and approval by federal and/or state regulators. • Products offered under subsidiaries of the same company are counted as one carrier • National multi-state plans (MSP) BCBS products and state BCBS products are counted as separate carriers (including in NM, MI, KS, and AK. 31 states are expected to have MSP plans but has not released the list. • UT: individual exchange will be facilitated by the federal government; SHOP will be state-based • NM and ID: federal government will help run individual markets in 2014. States will continue to maintain plan management and consumer assistance functions; HHS will operate the IT systems. SHOP will be state-based 20 / 137 Regulations on HIX Minimum Essential Coverage for Qualified Plans: 10 categories of service are specified in the ACA but the States are given discretion over the specific services in each category ACA only permits qualified health plans to be offered in the exchange. To be qualified, a health plan must be accredited and show contracts with hospitals, physicians and other providers, as well as financial resources to operate. Moreover, I I guaranteed issue: no one should be rejected for preexisting conditions and premium can not depend on preexisting conditions guaranteed renewability 21 / 137 Mandates: Individual Mandate Individual Mandate: individuals who do not purchase insurance that meets the minimum essential benefits at the bronze level of coverage will face a “tax”. I I The ”tax” will be phased in: $95 per person in 2014; by 2016, it will be $695 per year per person or 2.5% of the household income, with a max limit of $2,085. The numbers will the indexed to cost of living after 2016. Certain exemptions to individual mandate: certain religious groups, and American Indians 22 / 137 Mandates: Employer Mandate Employer Mandate: the idea is to discourage employers from discontinuing their employees’ health insurance I I I I employers with an average of 50 or more full time workers during the preceding year (or the current year if employer is new) are required to offer ESHI or face a penalty of $2,000 per worker (with an allowance of 30 workers in the calculation of the penalty). IRS regulations point out that: Any group of companies under ”common control”are to be treated as a single company. Common control is defined as the same five or fewer people owning at least 80 percent of the companies 23 / 137 Other Important Elements of ACA Cost Control: I I I Reducing health care prices (e.g. change the payment rates for Medicare Advantage Plans; reduce the annual increase in hospital payments; reduce Disproportionate Share Hosptial Payments; competitive bidding for durable medical equipment (DME) Reducing Health Care Utilization: Accountable Care Organization (ACO) instead of fee-for-service. Independent Payment Advisory Board (IPAB) Quality of Care: (electronic health records, etc) Prevention and Health Promotion 24 / 137 II: US Employment-Based Health Insurance System and Medical Expenditure Fang and Gavazza (2011, AER): Dynamic Inefficiencies in an Employment Based Health Insurance System: Theory and Evidence 25 / 137 Fang and Gavazza (2011): Introduction The health care system in the United States differs in at least two stark ways from those of other industrialized countries. The first is its institutional organization: I I I The United States is unique among industrialized nations in that it lacks a national health insurance system. The U.S. health insurance system is a mixture of private and public insurances, with private insurance playing a much more important role than in other industrialized countries. In particular, in the U.S., a private, employment-based system provides insurance to most of the working-age population, while a public program—i.e., Medicare—provides insurance to almost all individuals aged 65 and over. The second difference is its costs: I I The U.S. spends more than twice as much on health care as a fraction of G.D.P. as other developed countries. For example, in 2005, the U.S. and the U.K. spent about 17 and eight percent of their GDP, respectively, on health care. 26 / 137 Introduction This paper investigates the effects of the institutional settings of the U.S. health care system on individuals’ life-cycle medical expenditures. Our premise is that health is a form of general human capital and that health investment—medical expenditures in particular—determines the stock of health. Hence, like all other forms of human capital, health increases labor productivity, thereby affecting the surplus generated in the employment relationship. Thus, current health expenditures are an investment that affects the current and future surplus of the employment relationship. We embed this link between health investment and employment surplus in a frictional labor market, and derive the implications of employee turnover on the employer-employee pair’s incentives to invest in the employee’s health. 27 / 137 Motivating Figure Ave.$Medical$Expenditure$ Workers$in$industries$ $with$Above$Median$Annual$$ Turnover$Rate$ Workers$in$industries$ $with$Below$Median$Annual$$ Turnover$Rate$ Age$ 28 / 137 Explaining the Economic Mechanism We show that employee turnover leads to an inefficiently low level of investment in the employees’ health, and that investment is lower and inefficiencies larger when employee turnover is higher. The reason is that frictions in the labor market imply that part of the surplus from the current investment in the employee’s health accrues to a future employer. Hence, the employer-employee pair does not internalize the full social surplus created by the current investment in the employee’s health. As a result, the pair under-invests in health capital. Further, we show that this inefficiently low level of medical expenditures during the working years increases medical expenditures during retirement, possibly increasing the overall expenditures. 29 / 137 Summary of Empirical Methods and Findings We provide extensive empirical evidence consistent with the predictions of the model using two datasets, the Medical Expenditure Panel Survey (MEPS) and the Health and Retirement Study (HRS). Our empirical model is designed to deal explicitly with two issues that may hinder the identification of the effect of job turnover. I I The first is selection: Workers may select into different jobs due to unobserved characteristics, such as ability, discount factor, risk aversion, etc. that could potentially be correlated with their job turnover. The second is reverse causality: Workers’ health outcome and health expenditures could affect their job turnover. We deal with these issues using panel data to control for fixed and persistent unobservables that could affect selection into different jobs, along with demand-side instruments—i.e., plant closures—that arguably are not affected by reverse causality. 30 / 137 Summary of Empirical Methods and Findings We find that workers with shorter job tenures spend less on health care. However, we find a stark reversal of expenditures among the elderly: Retirees who had longer job tenures spend less on health. The magnitude of our results is considerable. I I Workers with job tenures that are one standard deviation longer have medical expenditures about $660 higher per year. Individuals over 65 whose tenure at their main job is one standard deviation spend about $4,700 less per year on health care. 31 / 137 Magnitude of The Effects: A Back-of-Envelope Calculation Consider the lifetime medical expenditures of two workers whose only difference is their job tenures. Suppose that both individuals work 45 years and then retire for 15 years before dying, but the first individual’s job tenure is one standard deviation longer than the second individual’s. According to our estimates, during their working years, the first individual spent approximately $29,700 more on health care than the second individual did. During retirement, the first individual’s health expenditures are approximately $70,500 lower than the second individual’s. The total difference is around $40,000. This calculation suggests that one additional dollar of health expenditures during working years may lead to about 2.5 dollars of savings in retirement. 32 / 137 Health and Productivity: Literature A key premise of our model is that health is a productive general human capital (Grossman 1972). Several papers establish that increased life expectancy and reduced morbidity increase productivity and output. I I Fogel (1991, 1994) shows how improvements in health affect living standards over time in Europe and in the United States. Several empirical studies document that health has a significant and positive effect on economic growth (e.g., Barro and Sala-i-Martin, 1995; Knowles and Owen, 1995; Bloom, Canning and Sevilla, 2001). At the individual level, many papers find that less healthy individuals are less productive, broadly defined. I I I Haveman et al. (1994) finds that prior health limitations negatively affect work-time and have a significant negative effect on wages. Berkovec and Stern (1991) finds that bad health decreases labor market participation among the elderly. Stern (1996) finds that health limitations on the ability to work have larger effects on individual labor supply than on labor demand. 33 / 137 Health and Productivity: Literature A few recent studies focus on even more-detailed micro-evidence to study the effects of health on worker productivity. I I Nicholson et al. (2006) uses survey data from a sample of establishments and provides direct evidence that the cost or productivity loss associated with missed work is higher than the wage. Davis et al. (2005), using a survey, finds that: “labor time lost due to health reasons represents lost economic output totaling $260 billion per year.” More broadly, an individual’s current health investment can affect his future health costs, and the individual and his current or future employer will need to pay for these future costs. I In an interesting study of diabetes management, Beaulieu et al. (2007) finds that improved diabetes care affords economic benefits to health plans, as well as valuable benefits to people with diabetes. 34 / 137 A Simple Model The model is adapted from Acemoglu and Pischke (1999) frictional labor market. There are two periods with no discounting. Health is a form of general human capital and thus is an input in the production function of the worker. Assume that health is the only input in the production function f (h) , where f (·) is increasing, differentiable and concave. 35 / 137 A Simple Model: Timing All workers are risk neutral, are endowed with an initial stock of health h1 and can invest m1 in health at a unit cost p. Health evolves according to h2 = k (h1 , m1 ) where k is increasing in the stock of health h1 and in the investment in health m1 . In period 2, the worker either stays with the firm at wage w2 (h2 ) or decides to quit and obtains an outside wage of v (h2 ) . The key assumption is that v 0 (h) < f 0 (h) , i.e. wage compression. Exogenous Turnover Rate: With probability q the firm and the worker receives an adverse shock and decide to separate. With probability (1 − q) the continue the productive relationship. Assume that firms compete in the first period by offering a pair of wage and medical consumption {w1 , m1 } to workers, and in equilibrium they make zero profits. 36 / 137 Equilibrium Frictional labor markets imply that the worker receives a lower wage than his marginal productivity, i.e. v (h2 ) < f (h2 ) . Hence, the worker and current employer can share the surplus f (h2 ) − v (h2 ) if they continue the employment relationship. We assume that the surplus is divided according to the Nash Bargaining solution, i.e. the wage w2 (h2 ) is equal to w2 (h2 ) = (1 − β) v (h2 ) + βf (h2 ) where β is the bargaining power of the worker and the outside option of the firm has been normalized to 0. 37 / 137 Equilibrium Firm expected profits in period 2 are π 2 (h2 ) = (1 − q) [f (h2 ) − w2 (h2 )] = (1 − q) (1 − β) [f (h2 ) − v (h2 )] and in the first period are π 1 (h1 ) = f (h1 ) − w1 − pm1 . So firm maximize the sum of profits π 1 (h1 ) + π 2 (h2 ) = f (h1 ) − w1 − pm1 + (1 − q) (1 − β) [f (h2 ) − v (h2 )] by choosing m1 and w1 , subject to the constraint that workers receive as much utility as that offered by other firms U. Competition ensures that U is high enough such that π 1 (h1 ) + π 2 (h2 ) = 0. 38 / 137 Equilibrium The first order condition implies that the optimal m∗1 solves 0 ∂h2 qv (h2 (m∗1 )) + (1 − q) f 0 (h2 (m∗1 )) = p. ∂m1 (1) Proposition A decrease in the turnover rate q increases medical expenditure m∗1 . 39 / 137 Dynamics of Health Expenditure Now add a third period in which the individual is retired. In this third period, an individual receives utility d (h3 ) from his health with d0 (·) > 0. h3 evolves according to: h3 = min k (h2 , m2 ) , h3 (h2 ) . m2 is free for simplicity (think of Medicare). All individuals choose medical expenditures m∗2 so that their health reaches h3 (h2 ): k (h2 , m∗2 ) = h3 (h2 ) . (2) Proposition If ∂h3 /∂h2 < ∂k/∂h2 , then workers in jobs with lower turnover rates have: (i) a higher medical expenditures m∗1 while working; and (ii) a lower medical expenditures m∗2 and better health during retirement. 40 / 137 Data US: 4 sources of data: 1 2 3 4 MEPS 1996-2006; HRS (1996-2002) for retirees (RAND version); Statistics of U.S. Businesses (SUSB); Employment Protection Laws. UK: British Household Panel Survey (BHPS): 1995-2002. 41 / 137 MEPS MEPS: large-scale national survey of health care use, expenditures, sources of payment, and insurance coverage. Two components: the Household Component (HC) and the Insurance Component (IC). I I HC: demographic characteristics, health conditions, health status, use of medical services, charges and source of payments, access to care, satisfaction with care, health insurance coverage, income, employment with 3-digit industry code. IC: health plans from a sample of private and public employers. Number and types of private insurance plans offered (if any), premiums, contributions by employers and employees, and benefits associated with these plans. We use HC. 42 / 137 HRS HRS: We use waves 1996-2002 (focus on retirees). I I I individual employment history. We can reconstruct the tenure at longest reported job with 3-digit industry codes. total medical expenditure in each waves (RAND computes sums for us). 43 / 137 Statistics of US Businesses The Statistics of U.S. Businesses (SUSB) is a dataset extracted from the Business Register, a file of all known single- and multi-establishment employer companies maintained and updated by the U.S. Census Bureau. It provides national and sub-national data on the distribution of economic data by size and industry, reporting the number of establishments, employment, and annual payroll for each geographic-industry-size cell. More importantly, SUSB reports the number of establishments and corresponding employment change for births, deaths, expansions, and contractions by employment size of enterprise, industry, and state. We use data on establishment deaths to construct our instruments for job turnover in the empirical analysis. 44 / 137 Employment Protection Laws During the 1970s and 1980s, the majority of U.S. state courts adopted one or more common-law exceptions to the employment-at-will doctrine that limited employers’ ability to fire employees. Autor, Donohue, and Schwab (2006) presents a detailed dataset of these wrongful-discharge laws prevailing in each state and year for the period from 1972 to 1999, and investigates the effects of these employment protection laws on the labor market. We use data on one protection—i.e., the implied contract exception—to construct our instruments for job turnover in the empirical analysis for retirees. 45 / 137 British Household Panel Survey (BHPS) The British Household Panel Survey (BHPS) is an annual panel survey beginning in 1991, following about 5,500 households and 10,300 individuals drawn from 250 areas of Great Britain. It is a data set with rich individual-level demographic, social and economic variables, as well as detailed information on health-related issues such as number of doctor visits and self-perceived health status. 46 / 137 Summary Statistics MEPS(1995-2005) HRS(1996-2002) Variable Mean Std. Dev. Mean Std. Dev. Medical Expenditure BHPS(1995-2002) Mean Std. Dev. 1,814 1,574 8,327 24,707 ... ... Job Tenure 6.7 4.3 ... ... 6.1 7.2 Longest Job Tenure ... ... 23.8 12.7 ... ... Age 38.9 11.6 75.1 6.8 41.1 12.3 Yrs. of Education 12.9 1.3 11.9 3.2 11.8 2.4 $31,403 $11,317 ... ... £ 28,934 £ 22,129 ... ... 3.34 8.12 ... ... 0.51 0.50 0.51 0.50 0.46 0.49 Income Total Assets/10,000 Male White 0.80 0.10 0.85 0.35 0.96 0.18 Black 0.14 0.10 0.13 0.33 ... ... Married 0.59 0.20 0.59 0.49 0.60 0.48 Family (Household) Size 3.23 0.55 1.94 0.93 3.00 1.32 Union 0.13 0.14 ... ... 0.96 0.29 47 / 137 An Illustrative Comparison Simple illustrative patterns comparing the average medical expenditures of individuals across two one-digit industries—manufacturing and retailing—that exhibit substantial differences in average job tenure. Variable Medical Expenditure Job Tenure Longest Job Tenure MEPS (1995-2005) Manu. Retail. HRS (1998-2004) Manu. Retail. 1,684 1,580 7,363 8,389 8.65 4.91 ... ... ... ... 25.58 18.40 48 / 137 Synthetic Panel Methods We use MEPS data to construct synthetic panels. As in all papers that use synthetic panels, the definition of a cohort is arbitrary. In our case, we are constrained by the sample size of each MEPS survey and by the limited geographic and industry information available in the public version of the MEPS. As a result, we choose to define cohorts by grouping people by sex, decade of birth, one-digit industry, and Census Region. 49 / 137 Synthetic Panel Methods We write the cohort-version of the empirical model as: yjt = β 0 + β T Job Tenurejt + β X Xjt + η rt + ζ j + jt ∆yjt = ρ∆yjt−1 + β∆Zjt − ρβ∆Zjt−1 + ∆η rt − ρ∆η rt−1 + ∆ν jt , I I I I I I I I j now denotes a cohort, for which industry i and region r are fixed over time. yjt is one of the outcomes of interest for cohort j in year t. Job Tenurejt is the average number of years individuals in cohort j have been employed in their current firm. Xjt is now the cohort-average of a large set of control variables. η rt is, as before, a year fixed effect for each region r. ζ j is now a fixed effect for cohort j. jt is an unobservable, autoregressive component with innovation ν jt —i.e., jt = ρjt−1 + ν jt . We use the (lagged) number and the rate of deaths of establishments in industry i and region r, and the number and the rate of workers that lost their jobs due to establishment deaths in industry i and region r from SUSB data as instruments for Job Tenure. 50 / 137 The Relationship Between Workers’ Job Tenure and Medical Expenditures and Doctor Visits Log (Job Tenure) ρ # Obs Panels A: Log Med. Exp. (1) (2) *** 0.686*** 0.698 (0.256) (0.248) 0.244*** ... (0.030) 4652 3930 620 586 B: Fraction Not Visiting Doc. (3) (4) ** -0.108 -0.111** (0.057) (0.057) 0.100*** ... (0.031) 4652 3930 620 586 51 / 137 The Relationship Between Retirees’ Past Job Tenure and Medical Expenditures (A) and Health Status (B) Log (Job Tenure) # Obs Panels A: Log Medical Expenditure (1) -0.746** B: Health Status (2) -0.430*** (0.356) 27,229 10,395 (0.063) 27,229 10,395 52 / 137 Magnitude of the Effects Consider the lifetime expenditures of two workers A and B whose only difference is their job tenures. Suppose that both individuals work for 45 years and then retire for 15 years before dying. Individual A works in a job in which mobility is high, while individual B works in a job in which mobility is low. Let us assume that individual B’s job tenure is one standard deviation higher than that of individual A. 53 / 137 Magnitude of the Effects In MEPS, one standard deviation of log tenure is equal to .52. Multiplying it by the coefficient of log tenure in the MEPS regressions, we obtain .7*.52=36 percent. At the average of MEPS medical expenditures ($1,814), this implies that individual A has expenditures lower than B’s by approximately $660 per year. Now consider both individuals’ medical expenditures during retirement. One standard deviation of log tenure in the HRS data is equal to .77. Multiplying it by the coefficient of log tenure in the HRS regressions, we obtain .74*.77=56 percent. At the average of HRS medical expenditure ($8,327), this implies that individual A has expenditures higher than B’s by approximately $4,700 per year. 54 / 137 Magnitude of the Effects Thus, if individuals A and B work for 45 years and then retire for 15 years, non-discounting their expenditures, we have that, during their working years, individual A’s health expenditures are approximately $29,700 lower than individual B’s. During retirement, individual A’s health expenditures are approximately $70,500 higher than individual B’s. he total difference is around $40,000, a rather large difference. This calculation suggests that one additional dollar of health expenditures during the working years may lead to about 2.5 dollars of savings in retirement. 55 / 137 Doctor Visits in the U.K. Log (Job Tenure) ρ # Obs Panels IV with FE (1) -0.011 (0.019) IV with First Diff. (2) -0.021 (0.021) AB with iid Res. (3) -0.023 (0.078) ... ... ... 93,709 15,931 95,955 15,760 94,015 16,237 AB with AR(1) Res. (4) 0.111 (0.101) 0.130*** (0.007) 75,955 15,760 56 / 137 Using ASVP as Exogenous Proxy for Job Turnover: Workers A: Total Med. Exp. Variables Level Log (1) (2) ∗∗ ASVP Obs. 199.8 (92.2) 13,459 0.22∗∗∗ (0.06) 13,459 B: Doctor Visits Office-Based Physician Visits Visits (3) (4) ∗ 0.046 0.053** (0.026) 13,459 (0.022) 13,459 57 / 137 Using ASVP as Exogenous Proxy for Job Turnover: Retirees Variables ASVP No. of Obs. Total Med. Exp. (1) -1,012.5* Perceived Health Status (2) -0.045** (537.3) 5,583 (0.020) 6,730 58 / 137 Magnitude of the Estimates: ASVP Results Suppose that both individuals A and B work for 45 years and then retire for 15 years before dying. Suppose that individual A’s ASVP is one unit lower than individual B’s. The coefficient of ASVP in the regressions from MEPS data implies that individual A’s expenditures are lower than B’s by $113 per year. The coefficient of ASVP in the regressions from HRS data implies that individual A has higher medical expenditures than individual B by $1,037 per year. Individual A has approximately $5,000 less in medical expenditures per year than individual B when working, but approximately $15,000 more in medical expenditures when retired. Thus one additional dollar of medical expenditures during the working years may lead to about three dollars of savings during retirement. 59 / 137 Falsification Test: Relationship Between Industry ASVP and Doctor Visits and Perceived Health Status for U.K. Workers Variables ASVP No. of Obs. Doctor Visits (1) -0.007 (0.031) 4,926 Perceived Health Status (2) -0.035 (0.028) 4,928 60 / 137 Alternative Hypotheses Good jobs versus bad jobs Several papers document true wage differentials across industries and jobs and a negative correlation between wage differentials and quit rates (e.g., Pencavel, 1970; Krueger and Summers, 1988; Gibbons and Katz, 1992). If workers are less likely to leave “good jobs” than “bad jobs”, then differences between good jobs and bad jobs could imply a positive correlation between job attachment and health expenditures. But our empirical model on workers’ medical expenditures is designed to precisely control for fixed and for persistent unobserved effects that may induce different workers to select into different jobs/industries. The instruments that we employed in the empirical analysis on workers’ medical expenditures exploit demand-side (i.e., firms) variation in turnover across regions and industries, precluding any reverse causality hypothesis based on supply-side (i.e., workers) variation in quits. 61 / 137 Is health more important in jobs that have also higher attachment? The empirical relationship between job attachment and health expenditures could simply be due to the fact that health is more important in industries that have also higher job attachment. However, the evidence from U.K. workers does not substantiate this claim. This difference in health care utilization between U.K. and U.S. is also in stark contrast with many labor-market patterns—i.e., wages and inequality—that are remarkably similar in the two countries. 62 / 137 Selection Based on Discounting? It may be that in high turnover industries wage profiles are flatter (high earlier, lower later) and more myopic people go into these industries, attracted by the higher initial wage. These people are likely to have a different intertemporal discount, i.e. value today much more than tomorrow. However, this explanation is in contrast with current theories of human capital. General human capital steepens wage-tenure profiles because workers must pay, in the form of lower wages, for any training that is general and thus transferable across employers. Conversely, any type of specific human capital flattens wage-tenure profiles because the firm makes a specific investment, but recoups its investment later once the workers are locked in. Crocker and Moran (2003) indeed found that wage profile in high turnover (low specific human capital) industries are steeper. 63 / 137 What about job lock? Selection based on health status. There is a large established literature showing that employment-based health insurance provides inefficiently low separation between mismatched workers and firms. To take this job lock hypothesis to a dynamic setting, we would expect to see that industries with high ASVP, because they are more likely to offer health insurance and more likely to offer better health insurance contracts should be more attractive to workers with worse health: after all, healthy workers benefit less from generous health insurance. In a steady state, then job lock dynamics should lead to a negative relationship between workers’ health and ASVP. However, from the retirees’ regression, we saw that healthier workers were working in low turnover industries. 64 / 137 Could it be a pure wealth effect? If wages are higher in low turnover industries, then wealth effect explains why health expenditure is higher in low turnover industries. Hall and Jones (2006) argue that the growth of health spending is a rational response to the growth of per capita income. Our explanation and Hall and Jones’ are not mutually exclusive. Hall and Jones focus on the growth of expenditure in the last 20 years, we focus on the intertemporal profile of expenditure. However, we believe that the wealth effect cannot fully explain a number of our cross-sectional results: I I our regressions on MEPS data include individuals’ current income and the best proxy for permanent income, i.e. education. in the regressions using HRS data we find exactly the opposite: we find that wealthier retired individuals spend less in health. In summary, we believe that the wealth effect cannot explain the intertemporal patterns in health expenditure that we document. 65 / 137 Conclusion The paper provides a strong link between the institutional features of the U.S. health care market, the incentives to invest in health, and health outcomes. Employment-based health insurance system leads to dynamic inefficiency in health investment over life-cycle due to hold up problem. Single payer system would do better in internalizing the dynamic externalities. We believe that the interaction between private and public provision of medical care in the U.S. might be particularly subject to the dynamic externality we consider. 66 / 137 III. Would the Troubles in the Initial Roll Out of HealthCare.gov last? And Does the Form of Individual Mandate Penalty Matter? Scheuer and Smetters (2014, NBER Working Paper): Could a Website Really Have Doomed the Health Insurance Exchange? Multiple Equilibria, Initial Conditions and the Constructionof the Fine. 67 / 137 Health Insurance Exchange Roll Out Launch of the federally run healthcare.gov website was not smooth Would this have long-run consequences for the functioning of the HIX? Massachusetts: individual mandate penalty is half of the lowest priced Commonwealth Care enrollee premium that could be charged to an individual at the corresponding income level ACA: individual mandate penalty is $695 per year or 2.5% of income, whichever is higher. Will the way the penalty is structured make a difference on how HIX operates? Scheuer and Smetters: ”Could a Website Really Have Doomed the Health Exchanges? Multiple Equilibria, Initial Conditions and the Construction of the Fine” (NBER WP) 68 / 137 Model Suppose that there is a unit measure of consumers with wealth w > 0 and face a potential loss of size 0 < l < w. Consumer differ in the probability π ∈ [0, 1] of the loss, distributed according to CDF H (·) . Consumers are risk averse with concave utility function u (c) . Consider the Akerlof’s model of the insurance market where consumers choose between (i) not purchasing insurance; and (ii) purchase an insurance with a premium p (to be determined in equilibrium) that covers the loss l if it were to occur. The competitive equilibrium consists of a premium p∗ and a critical risk type π ∗ such that u (w − p∗ ) = π ∗ u (w − l) + (1 − π ∗ ) u (w) Z 1 ∗ ∗ [1 − H (π )] p = l πdH (π) π∗ 69 / 137 Equilibrium via Demand and Supply Curves Supply for insurance: Write Γ (π) ≡ lE [Π|Π ≥ π] = l R1 π̃dH (π̃) 1 − H (π) π as the average cost of insuring everyone with risk equal to or greater than π. Demand for insurance: Write the willingness to pay for type-π consumer for insurance Ω (π) as the solution to u (w − Ω) ≡ πu (w − l) + (1 − π) u (w) Ω (π) can be interpreted as an inverse demand curve, and Ω−1 (p) is the marginal buyer when the premium is p. 70 / 137 Equilibrium via Demand and Supply Curves Equilibrium is characterized by any π ∗ such that Ω (π ∗ ) = Γ (π ∗ ) (= p∗ ) Properties of Γ (π) and Ω (π) : I I Γ (·) is continuous and increasing in π; Γ (0) = E [Π] l, Γ (1) = l; Ω (·) is continuous and decreasing in π; Ω (0) = 0 and Ω (1) = l. 71 / 137 Pareto-Ranked Multiple Equilibria Figure 2: Multiple Competitive Equilibria average cost and demand curves are upward sloping, but their shapes are otherwise unrestricted. A simple example is depicted in Figure 3. There are three competitive equilibria in total, namely the one with unraveling located at π ∗ = 1 as well as two additional ∗ ∗ 72 / 137 Dynamics Suppose that at time t, due to whatever reason, π t is the marginal type of buyer in the insurance market; In period t + 1, insurance companies will set pt+1 = Γ (π t ) ; Thus in period t + 1, the new marginal buyer type will be π t+1 = Ω−1 (pt+1 ) = Ω−1 (Γ (π t )) . 73 / 137 Dynamics Figure 3: Dynamics and Equilibrium Stability π1∗ and to the right of π2∗ , whereas the dynamics imply falling premiums and more indienrolling otherwise. dynamics have been documented in website states If viduals the problems in theEvidence initial for rollsuch out of the HealthCare.gov ∗ , then which, before the ACA, which placed restrictions on adjusting premiums based on has induced only the most risky types to enroll, i.e., if π 0 > πage 2 and preexisting conditions. Writing about the New Jersey Individual Health Coverage the above dynamics will lead to unraveling of the HIX. Program (IHCP) started in 1993, Monheit et al. (2004) found dynamics similar to those 74 / 137 Empirical Implications In order to test whether the disruptions in the first year of the operation of the HIX will lead to unravelling of the market, we will need to create the empirical analogs of the average cost curve and the inverse demand curve as in the above figure. The analysis can be generalized to account for richer consumer heterogeneity: I loss, if occurs, is drawn from distribution Fl , for example The difficulty is how to empirically separate the risk type of an individual from the random loss? Do insurers cross subsidize different age groups? They are not forced to because DHHS allows age bracket to be each age from 21-64 in the premium setting. In a competitive equilibrium, cross-subsidization across age groups could not occur. So then, why the common perception that somehow 18-34 years old males signing up in HIX is crucial for the success of ObamaCare? 75 / 137 Do the Forms of Mandate Penalty Matter? ACA absolute fine: fixed fine f if not having insurance and subsidy s if purchasing insurance. The supply side average cost function Γ (π) is not affected by f and s; The demand side is now characterized by Ω̂ (π; s, f ) which is implicitly defined by: u w − Ω̂ + s = πu (w − f − l) + (1 − π) u (w − f ) 76 / 137 Do the Forms of Mandate Penalty Matter? Relative to Ω (·) , Ω̂ shifts up in a parallel manner with s, and shifts up by f. Figure 7: Enforcement of a Mandate through a Fine f this outcome may be both inefficient and politically challenging. Indeed, the peculiar na77 / 137 Do the Forms of Mandate Penalty Matter? MA relative fine: u w − Ω̃ = πu (w − l − kΓ (π)) + (1 − π) u (w − kΓ (π)) where kΓ (π) = kp is the fine proportional to premium If f under ACA fine is set to be equal to kΓ (π̂ ∗1 ) , then Ω̃ (·) is counter-clockwise rotation of Ω̂ (·) pivoted at π̂ ∗1 . 78 / 137 Do the Forms of Mandate Penalty Matter? Figure 8: Relative versus absolute fine with kΓ(π̂1∗ ) = f by (w − Ω̃) = πu (w − l − kΓ(πfrom )) + (1 − π )u(w − kΓ(π )).19 (7) best The interval uof initial conditions which convergence to the equilibrium occurs the fine wider than demand under for the The benefit of the relativeunder fine is that therelative fine value — and,is hence, consumers’ insurance — automatically increase as the market unravels towards a bad stable equilibabsolute fine. rium. This outcome occurs even if we choose k such that kΓ(π̂ ∗ ) = f , so the relative and 79 / 137 IV: Equilibrium Labor Market and Health Insurance Reform Aizawa and Fang (2013): Equilibrium Labor Market Search and Helath Insurance Reform Fang and Shephard (WIP): Joint Household Labor Supply and Health Care Reform Fang, Shephard and Tilly (WIP): Equilibrium Labor Market Search with Endogenous Technology Choice and Health Insurance 80 / 137 Aizawa and Fang (2013): Introduction Affordable Care Act represents the most significant reforms to the U.S. health insurance and health care market since the establishment of Medicare in 1965. There are many provisions in the ACA; some of the most significant changes started taking effects from January 2014. 81 / 137 Major Components of ACA (Individual Mandate) All individuals must hold health insurance or face a penalty of $695 or 2.5 percent of income, whichever is higher; (Employer Mandate) Employers with more than 50 employees must provide health insurance or pay a fine of $2,000 per worker each year if they do not offer health insurance. (Insurance Exchanges) State-based health insurance exchanges will be established where the uninsured and those employed without insurance can purchase insurance from the exchange where premium will be based on community rating. (Premium Subsidies) Subsidies will be provided to individuals and families whose income is between the 133% and 400% of the federal poverty level. Individuals with income below 133% will receive Medicaid. 82 / 137 Goal of This Paper ... is to understand how the health care reform will affect the health insurance and labor markets. Would the ACA significantly reduce the uninsured rate? Would more employers be offering health insurance to their employees? How would the reform affect wage, health, productivity, employment, and employer size distributions? What is the impact on total health expenditures and on government budget? 83 / 137 Goal of the Paper We are also interested in several counterfactual policies, e.g., I I I How would the remainder of the ACA perform, had the individual mandate been struck down by the Supreme Court? What would happen if the current tax exemption status of employer-provided insurance premium is eliminated? Can we identify alternative reforms that can improve welfare relative to the ACA? 84 / 137 Labor Market and Health Insurance Market To address these questions, it is important to have an equilibrium model that integrates the labor and health insurance market (e.g., Dey and Flinn 2005). The U.S. is unique among industrialized nations in that it lacks a national health insurance system and most of the working age populations obtain health insurance coverage through their employers. There have been many well-documented connections between firm sizes, wages, health insurance offerings and worker turnovers: employer size annual wage annual worker separation rate all employers with HI w/o HI 24.43 $25,863.72 0.163 33.89 $29,077.49 0.158 8.83 $20,560.4 0.173 Workers in firms that offer health insurance also tend to have better self-reported health: 95.36% (HI) vs. 93.89% (No HI) are Healthy in our data. 85 / 137 In this paper ... We present and empirically implement an equilibrium labor search model where employers make decisions to offer health insurance. I I we incorporate health and health insurance to Burdett and Mortensen’s (1998) model with heterogenous firm productivity. wage, insurance provision, employment, employer size, and worker’s health status are endogenously determined. Use structural estimates to assess the impact the ACA on health insurance and labor market outcomes. 86 / 137 The Model: Worker Ex ante homogenous except health. Preference: risk averse. Health status: {healthy, unhealthy} Health insurance status: {uninsured, insured} Health insurance has two effects: 1 2 insure medical expenditure shocks. affect the law of motion for health status. Health insurance is only available through employers. Given the offer distribution of compensation (wage, health insurance provision), both unemployed and employed individuals decide whether to accept a new offer, if any. 87 / 137 The Model: Employer Ex ante heterogenous with respect to productivity p. Production technology: I I linear with labor inputs; an unhealthy worker produces d fraction of output, d ≤ 1. Choose wage and health insurance coverage to maximize the steady state profit flow subject to the constraint that all workers in the same firm are equally treated (HIPAA). In each firm that offers health insurance, the health insurance premium is set to cover the total expected medical expenditure by its workforce, plus a fixed administrative cost C. 88 / 137 Worker’s Preference and Health Utility function: u(c) = − exp(−γc). Health status: h ∈ {H, U }; Health insurance status: x ∈ {0, 1}. Medical Expenditure: I prob of a medical shock: Pr(m > 0|h, x) = Φ(α0 + β 0 1 {h = U } + γ 0 x), I (3) conditional on a medical shock, medical expenditure is drawn from: m| (h, x) ∼ exp (αm + β m 1 {h = U } + γ m x + hx ) , where hx ∼ N (0, σ 2hx ) and iid across time periods. Health status follows Markov process, which depends on insurance status x: πx = π xHH π xHU π xU H π xU U , where π xU H = 1 − π xHH and π xHU = 1 − π xU U . 89 / 137 Worker’s Problem: Expected Flow Utility A worker’s flow utility with income y and insurance status x is: u (T (y)) if x = 1 vh (y, x) = Em̃0 [u T (y) − m̃0h ] if x = 0, h where T (y) is after tax income: T (y) = τ 0 + τ 1 y (1+τ 2 ) , 1 + τ2 where τ 0 > 0, τ 1 > 0, τ 2 < 0. 90 / 137 Unemployed Worker’s Problem Let F (w, x) be the job offer distribution that each worker faces. It is endogenously determined in an equilibrium. The value function of the unemployed with health status h, Uh , is: Uh = vh (b, 0) 1−ρ Z +βEh0 |(h,0) λu max{Vh0 (w, x), Uh0 }dF (w, x) + (1 − λu )Uh0 Job acceptance decision: let wxh be Vh (wxh , x) = Uh . An unemployed worker accepts an job offer (w, x) if w ≥ wxh . 91 / 137 Employed Worker’s Problem The value function of the employed: Vh (w, x) = vh (w, x) 1−ρ R (1 − δ)Eh0 |(h,x) R max{Vh0 (w̃, x̃), Vh0 (w, x), Uh0 }dF(w̃, x̃) +βλe +δEh0 |(h,x) max {Uh0 , Vh0 (w̃, x̃)} dF (w̃, x̃) (1 − δ)Eh0 |(h,x) [max {Uh0 , Vh0 (w, x)}] +β(1 − λe ) . +δEh0 |(h,x) [Uh0 ] 92 / 137 Employed Worker’s Problem Job-to-job switching decision: sxh (·, ·) ; Job quitting decision: q xh . 93 / 137 No HI HI wage Unemployment Steady State Condition Worker distribution is characterized by (uh , exh , Gxh (w)) where I I I uh is the measure of unemployed workers with health status h; exh is the measure of employed workers with health status h and health insurance status x; Gxh (w) is the fraction of employed workers with health status h working on jobs with insurance status x and wage below w; ghx (w) is the associated density. We require that worker distribution must satisfy the steady state conditions: Given F (w, x), 1 2 3 the inflow and outflow of uh are equalized. the of exh ghx (w) are equalized. P inflow and outflow 0 1 h∈{U,H} (uh + eh + eh ) = M. 94 / 137 Employer’s Problem An employer draws the health insurance offering preference shock σ f , which is persistent over time. max{Π0 (p), Π1 (p) + σ f }, max (p − w0 ) nH (w0 , 0) + (pd − w0 ) nU (w0 , 0) w0 p − w1 − m1H n H (w1 , 1) − C. Π1 (p) = max Π (w1 , 1) ≡ + pd − w1 − m1U nU (w1 , 1) {w1 } Π0 (p) = Assuming that follows i.i.d. Type-I extreme value distribution, the fraction of employers offering health insurance among those with productivity p is ∆ (p) = exp( Πσ1 (p) ) f exp( Πσ1 (p) ) + exp( Πσ0 (p) ) f f . (4) 95 / 137 The Definition of Equilibrium A is Dsteady state equilibrium E x x x x wh , sh (·, ·) , q h , (uh , eh , Gxh (w)) , (wx (p) , ∆ (p)) , F (w, x) s. t. (Worker Optimization) Given F (w, x) , for each x x x (h, x) ∈ {U, H} × {0, 1} , wh , sh (·, ·) , q h solves worker’s optimization problem. (Steady State Worker Distribution) Given wxh , sxh (·, ·) , q xh and F (w, x) , (uh , exh , Gxh (w)) satisfies the steady state flow conditions. (employer Optimization) Given F (w, x) and the steady state employee sizes implied by (uh , exh , Gxh (w)), (w0 (p) , w1 (p) , ∆ (p)) solves employer’s optimization problem. (Equilibrium Consistency) F (w, x) must satisfy: Z p 1(w1 (p) < w)∆(p)dΓ(p), F (w, 1) = p Z F (w, 0) = p p 1(w0 (p) < w) [1 − ∆(p)] dΓ(p). 96 / 137 Why Are Large Firms More Likely to Offer Health Insurance? Statistics Frac. of Unhealthy in SS Frac. of Unhealthy (New Hires) One-Period Ahead Nine-Period Ahead J-to-J Transition for Healthy J-to-J Transition for Unhealthy Low-Prod. Firms High-Prod Firms HI No HI HI No HI 0.0494 0.096 0.037 0.107 Adverse Selection Effect 0.080 0.074 0.051 0.050 Health Insurance Effect on Health 0.067 0.084 0.046 0.067 0.038 0.109 0.037 0.107 Retention Effect 0.109 0.126 8.29E-9 4.03E-14 0.104 0.126 8.29E-9 5.91E-5 97 / 137 Data Sets Worker-side Data I I 1996 Panel of Survey of Income and Program Participation (SIPP 1996): labor market dynamics, wage, health insurance, and health variables. 1997-1999 Panels of Medical Expenditure Panel Survey (MEPS 1997-1999): medical expenditure, health, and health insurance. Employer-side Data: I 1997 Robert Wood Johnson Foundation Employer Health Insurance Survey (RWJ-EHI 1997): employer size distribution, health insurance coverage, and wage. 98 / 137 Sample Selection for SIPP and MEPS We restrict the samples which satisfy the following criteria: I I I I I I Men, aged between 26-46 at most high school graduates do not attend in school, military service, and any government welfare program (AFDC, WIC, Food Stamps) do not work as a self-employed or in public agency. are not covered by other sources (Medicaid, individual insurance, and spouse insurance). wage is between 3-97 percentiles. Sample size for SIPP is 5,309. Sample size for MEPS 1997-1999 are 4,815. 99 / 137 Sample Selection for RWJ-EHI 1997 Sample selection: I I belong to private sector. at least 3 workers. The sample size is 19,089. 100 / 137 Summary Statistics: SIPP Variable Fraction of Insured Among Employed Workers Average (4-Month) Wages for Employed Workers ... for insured employees ... for uninsured employees Fraction of Unemployed Workers Fraction of Healthy Workers ... among insured workers ... among uninsured workers Mean 0.7619 0.8538 0.9240 0.6187 0.0318 0.9511 0.9536 0.9389 Std. Dev. 0.4260 0.3532 0.3462 0.2750 0.1758 0.2177 0.2103 0.2398 101 / 137 Summary Statistics: RWJ-EHI Variable Name Average Establishment Size ... for those that Offer Health Insurance ... for those that Do Not Offer Health Insurance Health Insurance Coverage Rate ... for those with less than 50 workers ... for those with 50 or more workers Average Annual Wage Compensation, in $10,000 ... for those that Offer Health Insurance ... for those that Do Not Offer Health Insurance Mean 19.92 30.08 6.95 0.56 0.53 0.95 2.53 2.92 2.03 Std. Dev. 133.40 177.24 11.03 0.50 0.50 0.23 2.44 2.50 2.27 102 / 137 Two-Step Estimation Strategy In First Step, we estimate parameters of the medical expenditure distributions hα0 , β 0 , γ 0 , αm , β m , γ m , σ hx i, as well as the health transitions π without explicitly using the model. In Second Step, we estimate the remaining parameters by Generalized Method of Moments (Imbens and Lancaster, 1994), where moments are constructed from: I I likelihood of worker-side labor market transitions. firm-side characteristics (size distribution, coverage rate...). 103 / 137 First Step We estimate the parameters in medical expenditure distributions hα0 , β 0 , γ 0 , αm , β m , γ m , σ hx i by GMM using the MEPS. We estimate the parameters in health transition matrix, π 1HH , π 1U U , π 0HH , and π 0U U , using SIPP 1996 based on maximum likelihood. We calibrate some of other parameters: I I I discount factor β = 0.99; exogenous retirement rate ρ = 0.001 (from mortality rate); (1+τ 2 ) Parameterization of after-tax income, T (y) = τ 0 + τ 1 y1+τ 2 (from Kaplan’s 2011). 104 / 137 Second Step Estimate θ = [θ1 θ2 ] where θ1 = (λu , λe , δ, γ, µ, b) and θ2 = (C, d, M, µp , σ p , σ f ) by minimum distance estimation: min g(θ)0 Ωg(θ) where " g(θ) = I I P i ∂ log(Li (θ)) ∂θ s − E[s; θ] # , L(θ1 ) is likelihood of workers’ labor market transitions. E[s; θ] is other firm-side moments. 105 / 137 Firm-Side Moments Mean establishment size; Fraction of firms less than 50 workers; Mean size of establishments that offer health insurance; Mean size of establishment that do not offer health insurance; Health insurance coverage rate; Health insurance coverage rate among employers with more than 50 workers; Health insurance coverage rate among employers with less than 50 workers; Average wages of firms with less than 50 workers; Average wages of firms with more than 50 workers. 106 / 137 Likelihood Components from Worker-Side Labor Market Transitions An unemployed workers can transition to a job (w̃, x) after l periods; An employed workers at job (w, x) can experienece one of the four job transitions after l periods, I I I I [Event “Job Loss”] the individual experienced a job loss at period l + 1; [Event “Switch 1”] the individual transitioned to a job (w̃, x0 ) such that x0 = x and the accepted wage is w̃ > w; [Event “Switch 2”] the individual transitioned to a job (w̃, x0 ) such that x0 = x and the accepted wage is w̃ < w; [Event “Switch 3”] the individual transitioned to a job (w̃, x0 ) such that x0 6= x and the accepted wage is w̃. 107 / 137 Details in the Second Step 1 Initialize a guess of θ; 2 Given the guess, solve equilibrium numerically, by using the numerical algorithm. Obtain the offer distribution F̂ (w, x) from the equilibrium. 3 We will then use F̂ (w, x) and other parameters to evaluate the moments. 108 / 137 First-Step Estimates 109 / 137 Parameter Estimate Std. Err. Panel A: Medical Expenditure Parameters in Eq. (4) α0 -1.0909 (0.0446) β0 0.5247 (0.0723) γ0 0.5787 (0.0747) αm -4.4222 (0.3099) βm 1.6262 (0.3268) γm 0.7227 (0.3867) αp -1.0909 (0.0446) σ H1 1.4783 (0.0662) σ H0 1.9895 (0.1235) σU 1 1.3584 (0.0919) σU 0 1.3193 (0.0173) Panel B: Health Transition Parameters in Eq. (5) π 1HH π 0HH π 1U U π 0U U 0.9865 (0.0023) 0.9689 (0.0058) 0.7294 (0.0310) 0.7587 (0.0365) Table 7: Parameter Estimate from Step 1. Note: Standard errors are in parentheses. The unit of medical expenditure is $10,000. 99.30%. Panel B reports our estimates of parameters θ2 ≡ (C, d, M, µp , σ p ). We find that the productivity of a worker in bad health, d, is only 0.3386, implying that there is a significant amount of productivity loss from bad health. This seems plausible because we categorize only those whose self-reported health is “Poor” or “Fair” as unhealthy. Moreover, we find that the fixed administration cost of offering health insurance is about $730 per four month, (equivalent to about $2,190 per year). In order to fit the average firm size, our estimate of M, the ratio between workers and firms, is about 18.92. This estimate is smaller than the average establishment size of 19.92 reported in Table 5 because in our model some low-productivity firms do not attract any workers in equilibrium. We also estimated that the scale and shape parameters of the lognormal productivity distribution are respectively -0.5860 and 0.4043, which implies that the mean (4-month) productivity of firms is about 0.6149 (i.e. $6149). The fact that the mean accepted four-month wage in our sample is 0.8538 (see Table 3) is largely due to the fact more productive firms attract more workers in the steady state as our model implies, but also due to the fact that a fraction of the low-productivity firms are not able to attract any workers. in equilibrium (i.e. they are inactive.)39 39 One advantage of postulating a parametric productivity distribution is that it allows us to potentially capture how counterfactual policies might have affected the set of active firms. If we estimated firms’ productivity non-parametrically from 35 Parameter Estimate Std. Err. Panel A: Medical Expenditure Parameters in Eq. (4) α0 -1.0909 (0.0446) β0 0.5247 (0.0723) γ0 0.5787 (0.0747) αm -4.4222 (0.3099) βm 1.6262 (0.3268) γm 0.7227 (0.3867) αp -1.0909 (0.0446) σ H1 1.4783 (0.0662) σ H0 1.9895 (0.1235) σU 1 1.3584 (0.0919) σU 0 1.3193 (0.0173) Panel B: Health Transition Parameters in Eq. (5) π 1HH π 0HH π 1U U π 0U U 0.9865 (0.0023) 0.9689 (0.0058) 0.7294 (0.0310) 0.7587 (0.0365) Table 7: Parameter Estimate from Step 1. Note: Standard errors are in parentheses. The unit of medical expenditure is $10,000. 99.30%. Panel B reports our estimates of parameters θ2 ≡ (C, d, M, µp , σ p ). We find that the productivity of a worker in bad health, d, is only 0.3386, implying that there is a significant amount of productivity loss from bad health. This seems plausible because we categorize only those whose self-reported health is “Poor” or “Fair” as unhealthy. Moreover, we find that the fixed administration cost of offering health insurance is about $730 per four month, (equivalent to about $2,190 per year). In order to fit the average firm size, our estimate of M, the ratio between workers and firms, is about 18.92. This estimate is smaller than the average establishment size of 19.92 reported in Table 5 because in our model some low-productivity firms do not attract any workers in equilibrium. We also estimated that the scale and shape parameters of the lognormal productivity distribution are respectively -0.5860 and 0.4043, which implies that the mean (4-month) productivity of firms is about 0.6149 (i.e. $6149). The fact that the mean accepted four-month wage in our sample is 0.8538 (see Table 3) is largely due to the fact more productive firms attract more workers in the steady state as our model implies, but also due to the fact that a fraction of the low-productivity firms are not able to attract any workers. in equilibrium (i.e. they are inactive.)39 39 One advantage of postulating a parametric productivity distribution is that it allows us to potentially capture how counterfactual policies might have affected the set of active firms. If we estimated firms’ productivity non-parametrically from 35 Parameter Estimate Std. Err. Panel A: Medical Expenditure Parameters in Eq. (4) α0 -1.0909 (0.0446) β0 0.5247 (0.0723) γ0 0.5787 (0.0747) αm -4.4222 (0.3099) βm 1.6262 (0.3268) γm 0.7227 (0.3867) αp -1.0909 (0.0446) σ H1 1.4783 (0.0662) σ H0 1.9895 (0.1235) σU 1 1.3584 (0.0919) σU 0 1.3193 (0.0173) Panel B: Health Transition Parameters in Eq. (5) π 1HH π 0HH π 1U U π 0U U 0.9865 (0.0023) 0.9689 (0.0058) 0.7294 (0.0310) 0.7587 (0.0365) Table 7: Parameter Estimate from Step 1. Note: Standard errors are in parentheses. The unit of medical expenditure is $10,000. 99.30%. Panel B reports our estimates of parameters θ2 ≡ (C, d, M, µp , σ p ). We find that the productivity of a worker in bad health, d, is only 0.3386, implying that there is a significant amount of productivity loss from bad health. This seems plausible because we categorize only those whose self-reported health is “Poor” or “Fair” as unhealthy. Moreover, we find that the fixed administration cost of offering health insurance is about $730 per four month, (equivalent to about $2,190 per year). In order to fit the average firm size, our estimate of M, the ratio between workers and firms, is about 18.92. This estimate is smaller than the average establishment size of 19.92 reported in Table 5 because in our model some low-productivity firms do not attract any workers in equilibrium. We also estimated that the scale and shape parameters of the lognormal productivity distribution are respectively -0.5860 and 0.4043, which implies that the mean (4-month) productivity of firms is about 0.6149 (i.e. $6149). The fact that the mean accepted four-month wage in our sample is 0.8538 (see Table 3) is largely due to the fact more productive firms attract more workers in the steady state as our model implies, but also due to the fact that a fraction of the low-productivity firms are not able to attract any workers. in equilibrium (i.e. they are inactive.)39 39 One advantage of postulating a parametric productivity distribution is that it allows us to potentially capture how counterfactual policies might have affected the set of active firms. If we estimated firms’ productivity non-parametrically from 35 Second-Step Estimates 110 / 137 Parameter Estimates Std. Err. Panel A: Parameters in θ1 ≡ (λu , λe , δ, γ, µ, b) CARA Coefficient (γ) 0.4915 (0.0051) Unemployment Income (b) 0.0137 (0.0002) Offer Arrival Rate for the Unemployed (λu ) 0.4340 (0.0112) Offer Arrival Rate for the Employed (λe ) 0.2680 (0.0038) Probability of Exogenous Match Destruction (δ) 0.0179 (0.0003) Fraction of New Born Workers that are Healthy (µH ) 0.9930 (0.0156) Productivity of a Worker in Bad Health (d) 0.3386 (0.0063) Fixed Administrative Cost of Insurance in $10,000 (C) 0.0730 (0.0063) Panel B: Parameters in θ2 ≡ (C, d, M, µp , σ p , σ f ) Total Measure of Workers Relative to Firms (M ) 18.8920 (8.7940) -0.5680 (0.0031) Shape Parameter of Firms’ Lognormal Productivity Distribution (σ p ) 0.4043 (0.0036) Scale Parameter of Choice Specific Shock to EHI offering (σ f ) 0.2397 (0.0025) Scale Parameter of Firms’ Lognormal Productivity Distribution µp Table 8: Parameter Estimate from Step 2 7.2 Within-Sample Goodness of Fit In this section, we examine the within-sample goodness of fit of our estimates by simulating the equilibrium of our estimated model and compare with the model predictions to their data counterparts. Worker-Side Goodness of Fit. Table 9 reports the model fits for medical expenditure in the first step. It shows that our parameter estimates fit the data on the means (in Panel A) and variances (in Panel B) of medical expenditure conditional on health and health insurance status very well; moreover, in Panel C we show that we accurately replicate the fraction of individuals with zero medical expenditures conditional on health and health insurance status. Table 10 reports the model fit for the worker-side moments. It shows that the model fit really well for cross section worker distribution in terms of health, health status, health insurance, wage, and employment distribution. Figure 1 plots the distribution of workers’ accepted wages by health insurance status. It shows that our model is able to capture the overall patterns reasonably well, but it predicts a much more concentrated wage distribution than what is in the data, especially among workers who have health insurance from their employers. Employer-Side Goodness of Fit. Table 11 compares the model’s predictions of the targetted employerside moments listed in Section 6.2.2 with those in the data. With the exception of the average wage of workers’ productivity, we would not have been able to examine this margin. 36 Parameter Estimates Std. Err. Panel A: Parameters in θ1 ≡ (λu , λe , δ, γ, µ, b) CARA Coefficient (γ) 0.4915 (0.0051) Unemployment Income (b) 0.0137 (0.0002) Offer Arrival Rate for the Unemployed (λu ) 0.4340 (0.0112) Offer Arrival Rate for the Employed (λe ) 0.2680 (0.0038) Probability of Exogenous Match Destruction (δ) 0.0179 (0.0003) Fraction of New Born Workers that are Healthy (µH ) 0.9930 (0.0156) Productivity of a Worker in Bad Health (d) 0.3386 (0.0063) Fixed Administrative Cost of Insurance in $10,000 (C) 0.0730 (0.0063) Panel B: Parameters in θ2 ≡ (C, d, M, µp , σ p , σ f ) Total Measure of Workers Relative to Firms (M ) 18.8920 (8.7940) -0.5680 (0.0031) Shape Parameter of Firms’ Lognormal Productivity Distribution (σ p ) 0.4043 (0.0036) Scale Parameter of Choice Specific Shock to EHI offering (σ f ) 0.2397 (0.0025) Scale Parameter of Firms’ Lognormal Productivity Distribution µp Table 8: Parameter Estimate from Step 2 7.2 Within-Sample Goodness of Fit In this section, we examine the within-sample goodness of fit of our estimates by simulating the equilibrium of our estimated model and compare with the model predictions to their data counterparts. Worker-Side Goodness of Fit. Table 9 reports the model fits for medical expenditure in the first step. It shows that our parameter estimates fit the data on the means (in Panel A) and variances (in Panel B) of medical expenditure conditional on health and health insurance status very well; moreover, in Panel C we show that we accurately replicate the fraction of individuals with zero medical expenditures conditional on health and health insurance status. Table 10 reports the model fit for the worker-side moments. It shows that the model fit really well for cross section worker distribution in terms of health, health status, health insurance, wage, and employment distribution. Figure 1 plots the distribution of workers’ accepted wages by health insurance status. It shows that our model is able to capture the overall patterns reasonably well, but it predicts a much more concentrated wage distribution than what is in the data, especially among workers who have health insurance from their employers. Employer-Side Goodness of Fit. Table 11 compares the model’s predictions of the targetted employerside moments listed in Section 6.2.2 with those in the data. With the exception of the average wage of workers’ productivity, we would not have been able to examine this margin. 36 Parameter Estimates Std. Err. Panel A: Parameters in θ1 ≡ (λu , λe , δ, γ, µ, b) CARA Coefficient (γ) 0.4915 (0.0051) Unemployment Income (b) 0.0137 (0.0002) Offer Arrival Rate for the Unemployed (λu ) 0.4340 (0.0112) Offer Arrival Rate for the Employed (λe ) 0.2680 (0.0038) Probability of Exogenous Match Destruction (δ) 0.0179 (0.0003) Fraction of New Born Workers that are Healthy (µH ) 0.9930 (0.0156) Productivity of a Worker in Bad Health (d) 0.3386 (0.0063) Fixed Administrative Cost of Insurance in $10,000 (C) 0.0730 (0.0063) Panel B: Parameters in θ2 ≡ (C, d, M, µp , σ p , σ f ) Total Measure of Workers Relative to Firms (M ) 18.8920 (8.7940) -0.5680 (0.0031) Shape Parameter of Firms’ Lognormal Productivity Distribution (σ p ) 0.4043 (0.0036) Scale Parameter of Choice Specific Shock to EHI offering (σ f ) 0.2397 (0.0025) Scale Parameter of Firms’ Lognormal Productivity Distribution µp Table 8: Parameter Estimate from Step 2 7.2 Within-Sample Goodness of Fit In this section, we examine the within-sample goodness of fit of our estimates by simulating the equilibrium of our estimated model and compare with the model predictions to their data counterparts. Worker-Side Goodness of Fit. Table 9 reports the model fits for medical expenditure in the first step. It shows that our parameter estimates fit the data on the means (in Panel A) and variances (in Panel B) of medical expenditure conditional on health and health insurance status very well; moreover, in Panel C we show that we accurately replicate the fraction of individuals with zero medical expenditures conditional on health and health insurance status. Table 10 reports the model fit for the worker-side moments. It shows that the model fit really well for cross section worker distribution in terms of health, health status, health insurance, wage, and employment distribution. Figure 1 plots the distribution of workers’ accepted wages by health insurance status. It shows that our model is able to capture the overall patterns reasonably well, but it predicts a much more concentrated wage distribution than what is in the data, especially among workers who have health insurance from their employers. Employer-Side Goodness of Fit. Table 11 compares the model’s predictions of the targetted employerside moments listed in Section 6.2.2 with those in the data. With the exception of the average wage of workers’ productivity, we would not have been able to examine this margin. 36 Parameter Estimates Std. Err. Panel A: Parameters in θ1 ≡ (λu , λe , δ, γ, µ, b) CARA Coefficient (γ) 0.4915 (0.0051) Unemployment Income (b) 0.0137 (0.0002) Offer Arrival Rate for the Unemployed (λu ) 0.4340 (0.0112) Offer Arrival Rate for the Employed (λe ) 0.2680 (0.0038) Probability of Exogenous Match Destruction (δ) 0.0179 (0.0003) Fraction of New Born Workers that are Healthy (µH ) 0.9930 (0.0156) Productivity of a Worker in Bad Health (d) 0.3386 (0.0063) Fixed Administrative Cost of Insurance in $10,000 (C) 0.0730 (0.0063) Panel B: Parameters in θ2 ≡ (C, d, M, µp , σ p , σ f ) Total Measure of Workers Relative to Firms (M ) 18.8920 (8.7940) -0.5680 (0.0031) Shape Parameter of Firms’ Lognormal Productivity Distribution (σ p ) 0.4043 (0.0036) Scale Parameter of Choice Specific Shock to EHI offering (σ f ) 0.2397 (0.0025) Scale Parameter of Firms’ Lognormal Productivity Distribution µp Table 8: Parameter Estimate from Step 2 7.2 Within-Sample Goodness of Fit In this section, we examine the within-sample goodness of fit of our estimates by simulating the equilibrium of our estimated model and compare with the model predictions to their data counterparts. Worker-Side Goodness of Fit. Table 9 reports the model fits for medical expenditure in the first step. It shows that our parameter estimates fit the data on the means (in Panel A) and variances (in Panel B) of medical expenditure conditional on health and health insurance status very well; moreover, in Panel C we show that we accurately replicate the fraction of individuals with zero medical expenditures conditional on health and health insurance status. Table 10 reports the model fit for the worker-side moments. It shows that the model fit really well for cross section worker distribution in terms of health, health status, health insurance, wage, and employment distribution. Figure 1 plots the distribution of workers’ accepted wages by health insurance status. It shows that our model is able to capture the overall patterns reasonably well, but it predicts a much more concentrated wage distribution than what is in the data, especially among workers who have health insurance from their employers. Employer-Side Goodness of Fit. Table 11 compares the model’s predictions of the targetted employerside moments listed in Section 6.2.2 with those in the data. With the exception of the average wage of workers’ productivity, we would not have been able to examine this margin. 36 Parameter Estimates Std. Err. Panel A: Parameters in θ1 ≡ (λu , λe , δ, γ, µ, b) CARA Coefficient (γ) 0.4915 (0.0051) Unemployment Income (b) 0.0137 (0.0002) Offer Arrival Rate for the Unemployed (λu ) 0.4340 (0.0112) Offer Arrival Rate for the Employed (λe ) 0.2680 (0.0038) Probability of Exogenous Match Destruction (δ) 0.0179 (0.0003) Fraction of New Born Workers that are Healthy (µH ) 0.9930 (0.0156) Productivity of a Worker in Bad Health (d) 0.3386 (0.0063) Fixed Administrative Cost of Insurance in $10,000 (C) 0.0730 (0.0063) Panel B: Parameters in θ2 ≡ (C, d, M, µp , σ p , σ f ) Total Measure of Workers Relative to Firms (M ) 18.8920 (8.7940) -0.5680 (0.0031) Shape Parameter of Firms’ Lognormal Productivity Distribution (σ p ) 0.4043 (0.0036) Scale Parameter of Choice Specific Shock to EHI offering (σ f ) 0.2397 (0.0025) Scale Parameter of Firms’ Lognormal Productivity Distribution µp Table 8: Parameter Estimate from Step 2 7.2 Within-Sample Goodness of Fit In this section, we examine the within-sample goodness of fit of our estimates by simulating the equilibrium of our estimated model and compare with the model predictions to their data counterparts. Worker-Side Goodness of Fit. Table 9 reports the model fits for medical expenditure in the first step. It shows that our parameter estimates fit the data on the means (in Panel A) and variances (in Panel B) of medical expenditure conditional on health and health insurance status very well; moreover, in Panel C we show that we accurately replicate the fraction of individuals with zero medical expenditures conditional on health and health insurance status. Table 10 reports the model fit for the worker-side moments. It shows that the model fit really well for cross section worker distribution in terms of health, health status, health insurance, wage, and employment distribution. Figure 1 plots the distribution of workers’ accepted wages by health insurance status. It shows that our model is able to capture the overall patterns reasonably well, but it predicts a much more concentrated wage distribution than what is in the data, especially among workers who have health insurance from their employers. Employer-Side Goodness of Fit. Table 11 compares the model’s predictions of the targetted employerside moments listed in Section 6.2.2 with those in the data. With the exception of the average wage of workers’ productivity, we would not have been able to examine this margin. 36 Parameter Estimates Std. Err. Panel A: Parameters in θ1 ≡ (λu , λe , δ, γ, µ, b) CARA Coefficient (γ) 0.4915 (0.0051) Unemployment Income (b) 0.0137 (0.0002) Offer Arrival Rate for the Unemployed (λu ) 0.4340 (0.0112) Offer Arrival Rate for the Employed (λe ) 0.2680 (0.0038) Probability of Exogenous Match Destruction (δ) 0.0179 (0.0003) Fraction of New Born Workers that are Healthy (µH ) 0.9930 (0.0156) Productivity of a Worker in Bad Health (d) 0.3386 (0.0063) Fixed Administrative Cost of Insurance in $10,000 (C) 0.0730 (0.0063) Panel B: Parameters in θ2 ≡ (C, d, M, µp , σ p , σ f ) Total Measure of Workers Relative to Firms (M ) 18.8920 (8.7940) -0.5680 (0.0031) Shape Parameter of Firms’ Lognormal Productivity Distribution (σ p ) 0.4043 (0.0036) Scale Parameter of Choice Specific Shock to EHI offering (σ f ) 0.2397 (0.0025) Scale Parameter of Firms’ Lognormal Productivity Distribution µp Table 8: Parameter Estimate from Step 2 7.2 Within-Sample Goodness of Fit In this section, we examine the within-sample goodness of fit of our estimates by simulating the equilibrium of our estimated model and compare with the model predictions to their data counterparts. Worker-Side Goodness of Fit. Table 9 reports the model fits for medical expenditure in the first step. It shows that our parameter estimates fit the data on the means (in Panel A) and variances (in Panel B) of medical expenditure conditional on health and health insurance status very well; moreover, in Panel C we show that we accurately replicate the fraction of individuals with zero medical expenditures conditional on health and health insurance status. Table 10 reports the model fit for the worker-side moments. It shows that the model fit really well for cross section worker distribution in terms of health, health status, health insurance, wage, and employment distribution. Figure 1 plots the distribution of workers’ accepted wages by health insurance status. It shows that our model is able to capture the overall patterns reasonably well, but it predicts a much more concentrated wage distribution than what is in the data, especially among workers who have health insurance from their employers. Employer-Side Goodness of Fit. Table 11 compares the model’s predictions of the targetted employerside moments listed in Section 6.2.2 with those in the data. With the exception of the average wage of workers’ productivity, we would not have been able to examine this margin. 36 Data Model Panel A: Mean Annual Medical Expenditure Healthy & insured 0.0672 0.0673 Healthy & uninsured 0.0365 0.0359 Unhealthy & insured 0.4804 0.4794 Unhealthy & uninsured 0.1249 0.1249 Panel B: Variance of Annual Medical Expenditure Healthy & insured 0.0393 0.0392 Healthy & uninsured 0.1601 0.1601 Unhealthy & insured 0.8084 0.8084 Unhealthy & uninsured 0.0856 0.0856 Panel C: Fraction with Zero Medical Expenditure Healthy & insured 0.3324 0.3368 Healthy & uninsured 0.6458 0.6413 Unhealthy & insured 0.1290 0.1213 Unhealthy & uninsured 0.3600 0.3646 Table 9: Fit for Medical Expenditure Distributions: Model vs. Data. Moments Data Model Fraction of individuals who are unemployed and healthy 0.0314 0.0301 Fraction of individuals who are unemployed and unhealthy 0.0040 0.0021 Fraction of individuals who are employed, healthy and have health insurance 0.7009 0.7667 Fraction of individuals who are employed, unhealthy and have health insurance 0.0340 0.0319 Fraction of individuals who are employed, healthy and do not have health insurance 0.2156 0.1525 Fraction of individuals who are employed, unhealthy and do not have health insurance 0.0140 0.0167 Mean wage ($10,000) 0.8538 0.8501 Mean wage with health insurance ($10,000) 0.9240 0.8986 Mean wage without health insurance ($10,000) 0.6187 0.6211 Mean medical expenditure ($10,000) 0.0266 0.0253 Table 10: Worker-Side Moments in the Labor Market: Model vs. Data. 37 Data Model Panel A: Mean Annual Medical Expenditure Healthy & insured 0.0672 0.0673 Healthy & uninsured 0.0365 0.0359 Unhealthy & insured 0.4804 0.4794 Unhealthy & uninsured 0.1249 0.1249 Panel B: Variance of Annual Medical Expenditure Healthy & insured 0.0393 0.0392 Healthy & uninsured 0.1601 0.1601 Unhealthy & insured 0.8084 0.8084 Unhealthy & uninsured 0.0856 0.0856 Panel C: Fraction with Zero Medical Expenditure Healthy & insured 0.3324 0.3368 Healthy & uninsured 0.6458 0.6413 Unhealthy & insured 0.1290 0.1213 Unhealthy & uninsured 0.3600 0.3646 Table 9: Fit for Medical Expenditure Distributions: Model vs. Data. Moments Data Model Fraction of individuals who are unemployed and healthy 0.0314 0.0301 Fraction of individuals who are unemployed and unhealthy 0.0040 0.0021 Fraction of individuals who are employed, healthy and have health insurance 0.7009 0.7667 Fraction of individuals who are employed, unhealthy and have health insurance 0.0340 0.0319 Fraction of individuals who are employed, healthy and do not have health insurance 0.2156 0.1525 Fraction of individuals who are employed, unhealthy and do not have health insurance 0.0140 0.0167 Mean wage ($10,000) 0.8538 0.8501 Mean wage with health insurance ($10,000) 0.9240 0.8986 Mean wage without health insurance ($10,000) 0.6187 0.6211 Mean medical expenditure ($10,000) 0.0266 0.0253 Table 10: Worker-Side Moments in the Labor Market: Model vs. Data. 37 Moments Data Model Mean establishment size 19.92 18.5239 Fraction of firms less than 50 workers 0.93 0.9026 Mean size of establishments that offer health insurance 30.08 27.0368 Mean size of establishment that do not offer health insurance 6.95 7.2363 Health insurance coverage rate 0.56 0.5581 Health insurance coverage rate among employers with less than 50 workers 0.53 0.5200 Health insurance coverage rate among employers with more than 50 workers 0.95 0.9113 Average wages of firms with less than 50 workers 0.84 0.4129 Average wages of firms with more than 50 workers 0.92 0.9563 Table 11: Employer-Side Moments: Model vs. Data. 8 Counterfactual Experiments In this section, we use our estimated model to conduct counterfactual policy experiments and evaluate the impact of the Affordable Care Act and its various components. For the ACA, we consider a stylized version which incorporates its main components as mentioned in the introduction: first, all individuals are required to have health insurance or have to pay a penalty; second, all firms with more than 50 workers are required to offer health insurance, or have to pay a penalty; third, we introduce a health insurance exchange where individuals can purchase health insurance at community rated premium; fourth, the participants in health insurance exchange can obtain income-based subsidies. The introduction of health insurance exchange where individuals can purchase health insurance if they are unemployed or if their employers do not offer them represents a substantial departure from our benchmark model because premium in exchange will be endogenously determined. As a result, we will first describe how we extend and analyze our benchmark model to incorporate the health insurance exchange. 8.1 Model for the Counterfactual Experiments We provide a brief explanation of the main changes in the economic environment, as well as the definition of equilibrium, for the model used in our counterfactual experiments. The Main Change in Individuals’ Environment. We now assume individuals who are not offered health insurance by their employers and those who are unemployed can purchase individual insurance from the health insurance exchange. We assume that the insurance purchased from the exchange is similar to those offered by the employers in that it also fully insures medical expenditure risk. Thus in the extended 39 Wage distribution of workers with HI 1.4 Data Model 1.2 1 0.8 0.6 0.4 0.2 0 0 0.2 0.4 0.6 0.8 wage 1 1.2 1.4 Wage distribution of workers without HI 1.5 Data Model 1 0.5 0 0 0.2 0.4 0.6 0.8 wage 1 1.2 1.4 Size distribution of establishment 0.035 Data Model 0.03 0.025 0.02 0.015 0.01 0.005 0 0 50 100 150 200 250 employer size 300 350 400 Size distribution of establishment offering HI 0.03 Data Model 0.025 0.02 0.015 0.01 0.005 0 0 50 100 150 200 250 employer size 300 350 400 Size distribution of establishment NOT offering HI 0.04 Data Model 0.035 0.03 0.025 0.02 0.015 0.01 0.005 0 0 50 100 150 200 250 employer size 300 350 400 Counterfactual for ACA Insurance Exchange (EX): I I I Health insurance effect on health between ESHI and individual health insurance is the same. Premium in the insurance exchange is determined via community rating; Set loading factor ξ to 25% (ACA stipulates that the medical loss ratio should at least 80%). Individual Mandate (IM): ACA PW (y) = max {0.025 × (y − TFT 2011) , $695} , (5) 111 / 137 Counterfactual for ACA Employer Mandate (EM): If n ≥ 50, PEACA (n) = (n − 30) × $2, 000. (6) Premium Subsidies (Sub): I I I I If an individual’s income is at 133% of the FPL, his contribution to the premium is equal to 3.5% of his income; When an individual’s income is at FPL400, his premium contribution is set to be 9.5% of the income. When his income is below FPL133, he will receive insurance with zero premium contribution. If his income is above FPL400, he is no longer eligible for premium subsidies. 112 / 137 Main Results from Counterfactual Experiments 113 / 137 Benchmark (1) Frac. of firms offering HI ...if firm size is less than 50 ...if firm size is 50 or more Frac. of firms with less than 50 workers Average labor productivity Firm’s profit Frac. of firms in operation 0.5581 0.5200 0.9113 0.9026 1.1300 0.4717 0.9872 Uninsured rate Frac. of emp. workers with HI from ESHI Frac. of emp. workers with HI from EX Average wage ... with health insurance ... without health insurance Unemployment rate Frac. of healthy workers ... among uninsured ... among insured through ESHI ... among insured through EX Frac. of emp. workers in firms with 50+ workers Average worker utility (CEV) 0.2012 0.8253 0.8501 0.8986 0.6211 0.0322 0.9494 0.9017 0.9600 0.6142 0.6152 Average tax expenditure to ESHI Subsidies to exchange purchases Revenue from penalties Average health expenditure Average premium in ESHI Premium in exchange 0.0084 0.0253 0.0306 - ACA EX+Sub+EM EX+Sub+IM (2) (3) (4) Panel A: Effects on the Firm Side 0.5486 0.5494 0.5531 0.5039 0.5012 0.5111 0.9993 0.9988 0.9506 0.9097 0.9031 0.9043 1.1299 1.1309 1.1349 0.4937 0.4809 0.4765 0.9872 0.9872 0.9872 Panel B: Effects on the Worker Side 0.0727 0.1218 0.0644 0.8284 0.8259 0.8364 0.0965 0.0482 0.0971 0.8449 0.8482 0.8526 0.8934 0.9002 0.9019 0.6132 0.6021 0.6014 0.0320 0.0322 0.0322 0.9592 0.9558 0.9598 1.0000 1.0000 1.0000 0.9636 0.9628 0.9636 0.8890 0.7170 0.8984 0.5940 0.6129 0.6096 0.6133 0.6164 0.6184 Panel C: Effects on Expenditures 0.0083 0.0083 0.0084 0.0034 0.0038 0.0032 0.0010 0.00002 0.0009 0.0273 0.0272 0.0273 0.0301 0.0302 0.0300 0.0439 0.0595 0.0427 Table 12: Counterfactual Policy Experiments: Evaluation of the ACA and its Two Variations. 38 Benchmark (1) Frac. of firms offering HI ...if firm size is less than 50 ...if firm size is 50 or more Frac. of firms with less than 50 workers Average labor productivity Firm’s profit Frac. of firms in operation 0.5581 0.5200 0.9113 0.9026 1.1300 0.4717 0.9872 Uninsured rate Frac. of emp. workers with HI from ESHI Frac. of emp. workers with HI from EX Average wage ... with health insurance ... without health insurance Unemployment rate Frac. of healthy workers ... among uninsured ... among insured through ESHI ... among insured through EX Frac. of emp. workers in firms with 50+ workers Average worker utility (CEV) 0.2012 0.8253 0.8501 0.8986 0.6211 0.0322 0.9494 0.9017 0.9600 0.6142 0.6152 Average tax expenditure to ESHI Subsidies to exchange purchases Revenue from penalties Average health expenditure Average premium in ESHI Premium in exchange 0.0084 0.0253 0.0306 - ACA EX+Sub+EM EX+Sub+IM (2) (3) (4) Panel A: Effects on the Firm Side 0.5486 0.5494 0.5531 0.5039 0.5012 0.5111 0.9993 0.9988 0.9506 0.9097 0.9031 0.9043 1.1299 1.1309 1.1349 0.4937 0.4809 0.4765 0.9872 0.9872 0.9872 Panel B: Effects on the Worker Side 0.0727 0.1218 0.0644 0.8284 0.8259 0.8364 0.0965 0.0482 0.0971 0.8449 0.8482 0.8526 0.8934 0.9002 0.9019 0.6132 0.6021 0.6014 0.0320 0.0322 0.0322 0.9592 0.9558 0.9598 1.0000 1.0000 1.0000 0.9636 0.9628 0.9636 0.8890 0.7170 0.8984 0.5940 0.6129 0.6096 0.6133 0.6164 0.6184 Panel C: Effects on Expenditures 0.0083 0.0083 0.0084 0.0034 0.0038 0.0032 0.0010 0.00002 0.0009 0.0273 0.0272 0.0273 0.0301 0.0302 0.0300 0.0439 0.0595 0.0427 Table 12: Counterfactual Policy Experiments: Evaluation of the ACA and its Two Variations. 38 Benchmark (1) Frac. of firms offering HI ...if firm size is less than 50 ...if firm size is 50 or more Frac. of firms with less than 50 workers Average labor productivity Firm’s profit Frac. of firms in operation 0.5581 0.5200 0.9113 0.9026 1.1300 0.4717 0.9872 Uninsured rate Frac. of emp. workers with HI from ESHI Frac. of emp. workers with HI from EX Average wage ... with health insurance ... without health insurance Unemployment rate Frac. of healthy workers ... among uninsured ... among insured through ESHI ... among insured through EX Frac. of emp. workers in firms with 50+ workers Average worker utility (CEV) 0.2012 0.8253 0.8501 0.8986 0.6211 0.0322 0.9494 0.9017 0.9600 0.6142 0.6152 Average tax expenditure to ESHI Subsidies to exchange purchases Revenue from penalties Average health expenditure Average premium in ESHI Premium in exchange 0.0084 0.0253 0.0306 - ACA EX+Sub+EM EX+Sub+IM (2) (3) (4) Panel A: Effects on the Firm Side 0.5486 0.5494 0.5531 0.5039 0.5012 0.5111 0.9993 0.9988 0.9506 0.9097 0.9031 0.9043 1.1299 1.1309 1.1349 0.4937 0.4809 0.4765 0.9872 0.9872 0.9872 Panel B: Effects on the Worker Side 0.0727 0.1218 0.0644 0.8284 0.8259 0.8364 0.0965 0.0482 0.0971 0.8449 0.8482 0.8526 0.8934 0.9002 0.9019 0.6132 0.6021 0.6014 0.0320 0.0322 0.0322 0.9592 0.9558 0.9598 1.0000 1.0000 1.0000 0.9636 0.9628 0.9636 0.8890 0.7170 0.8984 0.5940 0.6129 0.6096 0.6133 0.6164 0.6184 Panel C: Effects on Expenditures 0.0083 0.0083 0.0084 0.0034 0.0038 0.0032 0.0010 0.00002 0.0009 0.0273 0.0272 0.0273 0.0301 0.0302 0.0300 0.0439 0.0595 0.0427 Table 12: Counterfactual Policy Experiments: Evaluation of the ACA and its Two Variations. 38 Frac. of firms offering HI ...if firm size is less than 50 ...if firm size is 50 or more Frac. of firms with less than 50 workers Average labor productivity Firm’s profit Frac. of firms in operation Uninsured rate Frac. of emp. workers with HI from ESHI Frac. of emp. workers with HI from EX Average wage ... with health insurance ... without health insurance Unemployment rate Frac. of healthy workers ... among uninsured ... among insured through ESHI ... among insured through exchange Frac. of emp. workers in firms with 50+ workers Average worker utility (CEV) Tax expenditure to ESHI Subsidies to exchange purchases Tax revenue from penalties Average health expenditure Average premium in ESHI Premium in Exchange Benchmark ACA Exempt No Exempt Exempt No Exempt (1) (2) (3) (4) Panel A: Effects on the Firm Side 0.5581 0.5419 0.5486 0.5290 0.5200 0.5053 0.5039 0.4889 0.9113 0.8834 0.9993 0.9837 0.9026 0.9031 0.9097 0.9188 1.1300 1.1287 1.1299 1.1133 0.4717 0.4692 0.4937 0.4995 0.9872 0.9872 0.9872 0.9919 Panel B: Effects on the Worker Side 0.2012 0.2339 0.0727 0.0915 0.8253 0.7916 0.8284 0.7871 0.0965 0.1170 0.8501 0.8510 0.8449 0.8401 0.8986 0.9072 0.8934 0.8983 0.6211 0.6374 0.6132 0.6336 0.0322 0.0320 0.0320 0.0353 0.9494 0.9470 0.9592 0.9557 0.9017 0.9007 1.0000 1.0000 0.9600 0.9597 0.9636 0.9627 0.8890 0.8798 0.6142 0.6127 0.5940 0.5698 0.6152 0.6077 0.6133 0.6028 Panel C: Effects on Expenditures 0.0084 0.0083 0.0034 0.0044 0.0010 0.0014 0.0253 0.0249 0.0273 0.0275 0.0306 0.0307 0.0301 0.0302 0.0439 0.0466 Table 14: Counterfactual Policy Experiments: Evaluating the Effects of Eliminating the Tax Exemption for EHI Premium under the Benchmark and the ACA. 44 EX+ IM1 (1) Frac. of firms offering HI ...if firm size is less than 50 ...if firm size is 50 or more Frac. of firms with less than 50 workers Average labor productivity Firm’s profit Frac. of firms in operation 0.5293 0.5071 0.7400 0.9047 1.1388 0.4689 0.9919 Uninsured rate Frac. of emp. workers with HI from ESHI Frac. of emp. workers with HI from EX Average wage ... with health insurance ... without health insurance Unemployment rate Frac. of healthy workers ... among uninsured ... among insured through ESHI ... among insured through EX Frac. of emp. workers in firms with 50+ workers Average worker utility (CEV) 0.0000 0.6948 0.3052 0.8633 0.9019 0.7755 0.0318 0.9644 0.9643 0.9644 0.6093 0.6175 Average tax expenditure to ESHI Subsidies to exchange purchases Revenue from penalties Average health expenditure Average premium in ESHI Premium in exchange 0.0070 0.00000 0.0273 0.0304 0.0341 No ESHI Sub2 +EX MA Sub+EX Sub+EX 3 +EM Reform +IM +IM4 (2) (3) (4) (5) Panel A: Effects on the Firm Side 0.5286 0.5509 0.4779 0.5083 0.9972 0.9529 0.9025 0.9041 0.5192 0.6958 1.1212 1.1280 1.2185 1.2169 0.4995 0.4893 3.0564 1.6809 0.9898 0.9872 0.2065 0.3264 Panel B: Effects on the Worker Side 0.0752 0.0529 0.5896 0.0000 0.8061 0.8470 0.1162 0.0983 0.2929 1.0000 0.8426 0.8449 0.8179 0.8909 0.9006 0.8949 0.6121 0.5787 0.8909 0.0320 0.0322 0.1662 0.1090 0.9591 0.9606 0.9229 0.9644 1.0000 1.0000 1.0000 0.9636 0.9638 0.8994 0.9106 0.7146 0.9643 0.6150 0.6103 0.7827 0.7080 0.6142 0.6146 0.4866 0.5630 Panel C: Effects on Expenditures 0.0081 0.0085 0.0051 0.0029 0.0179 0.0057 0.00004 0.0011 0.0096 0.00000 0.0273 0.0273 0.0269 0.0273 0.0300 0.0300 0.0429 0.0411 0.0603 0.0342 Table 15: Counterfactual Policy Experiments: Evaluation of Alternative Policy Arrangements. Notes: (1). Individual mandate penalty in Column (1) is set to be 15 times as large as the $695 specified in the ACA; (2). In Column (2), the individual mandate is eliminated, but besides the income based premium subsidies as specified under the ACA, any employed worker who chooses to purchase health insurance from the exchange receives a $135 flat subsidy; (3). In Column (3) we assume that the individual mandate penalty is the same as that in the ACA; the rest follows the MA reform rules; (4). In Column (5), the individual mandate penalty is assumed to be 2.5% of income, or $1,390, whichever is higher. 46 Conclusion We presented a structural estimation of an equilibrium labor market search model with endogenous health insurance provision. The implementation of the full version of the ACA would significantly reduced the uninsured rate from 20.12% in the benchmark economy to 7.27%. This large reduction of the uninsured rate is mainly driven by low-wage workers participating in the insurance exchange with their premium supported by the income-based subsidies. 114 / 137 Conclusion We find that the ACA would also have achieved significant reduction in the uninsured rate even if its individual mandate component were removed: the uninsured rate would be 12.18%. If the subsidies were removed from the ACA, the insurance exchange will suffer from severe adverse selection problem, resulting in a much more modest reduction in the uninsured rate to 17.14 − 17.28%. Interestingly, we find that the current version of ACA without employer mandate ismore efficient than the one with employer mandate. We also find that eliminating the tax exemption for employer-sponsored health insurance (ESHI) premium both under the benchmark and under the ACA would increase uninsured rates both under the benchmark and under the ACA, but quite modestly. 115 / 137 Major Changes in the Current Revision Include female workers in the integrated labor market (though not joint with spouses) Introduce worker heterogeneity in value from unemployment Include three health states (Poor/Fair, Good/Very Good, Excellent) 116 / 137 Fang and Shephard (WIP): Introduction In the United States, most of the workers obtained health insurance from either their own employers, or if they are married, from their spouses’ employers. An important, yet under explored aspect of the employer-sponsored health insurance system is that employers typically offer insurance not only to their own employees, but also the employees’ spouses (and dependent children as well). 117 / 137 Introduction To the best of our understanding, the spousal health insurance benefit is not required by any existing law. Indeed, as have been reported in the news several large employers, e.g. the United Parcel Services and the University of Virginia, have decided to modify their health insurance plans not to automatically include spousal benefits in response to the Affordable Act Act. In this project, we aim to present and implement an equilibrium model where household members jointly search for employment opportunities in the labor market and firms make decisions regarding wage and health insurance offerings. We will use the estimated model to evaluate the equilibrium responses to labor market responses to the Affordable Care Act, with a particular focus on how the ACA would impact married couples’ employment options, decisions and welfare. 118 / 137 Coverage Status and Sources of Health Insurance for Married Couples MALE MALE Panel A: Both Employed FEMALE Own ESHI Spousal ESHI Other Ins. 0.247 0.364 0.017 0.178 0.004 0.000 0.018 0.001 0.047 0.020 0.002 0.003 Uncovered 0.016 0.001 0.002 0.082 Panel B: Husband Employed, Wife Not Employed FEMALE Own ESHI Spousal ESHI Other Ins. Own ESHI 0.026 0.528 0.031 Spousal ESHI 0.006 0.006 0.000 Other Ins. 0.002 0.000 0.067 Uncovered 0.001 0.002 0.008 Uncovered 0.062 0.002 0.004 0.255 Own ESHI Spousal ESHI Other Ins. Uncovered 119 / 137 Coverage Status and Sources of Health Insurance for Married Couples MALE MALE Panel C: Husband Not Employed, Wife Employed FEMALE Own ESHI Spousal ESHI Other Ins. Own ESHI 0.043 0.039 0.003 Spousal ESHI 0.357 0.006 0.000 Other Ins. 0.066 0.008 0.113 Uncovered 0.072 0.008 0.012 Uncovered 0.003 0.002 0.013 0.257 Panel D: Both Not Employed FEMALE Own ESHI Spousal ESHI Other Ins. Own ESHI 0.006 0.084 0.014 Spousal ESHI 0.017 0.003 0.000 Other Ins. 0.006 0.000 0.181 Uncovered 0.006 0.008 0.030 Uncovered 0.021 0.003 0.011 0.610 120 / 137 ACA and Firms’ Incentives to Offer Spousal Health Insurance Benefits One reason is related to a provision in the ACA that, from 2014, employers are required to pay an annual fee of about $65 per covered life. The second, and potentially more important reason, is related to how the ACA specifies the tax credits if individuals purchase health insurance from the health insurance exchange. 121 / 137 ACA and Firms’ Incentives to Offer Spousal Health Insurance Benefits First of all, eligible individuals can receive tax credit only if they purchase health insurance from their states’ health insurance marketplace. Importantly, however, the eligibility for the health insurance tax credits depends on two criterion. The first is income. Only individuals and families who make between 138 percent and 400 percent of the federal poverty level (FPL) are eligible for a tax credit. The second criterion is that the individual does not have access to affordable health insurance through their employer or another government program. To the extent that, net of tax credits, the spouse of an employee can get similar insurance from the health insurance exchange less than the full cost of spousal insurance offered by the employer, the spouse would have preferred that the employer did not offer spousal insurance benefits. The same could even happen for the employees themselves. 122 / 137 ACA and Firms’ Incentives to Offer Spousal Health Insurance Benefits A third reason that ACA can fundamentally change firms’ decisions to offer spousal health insurance benefits to their employees is simply because of the availability of health insurance from a regulated health insurance exchange. Prior to ACA, the individual private insurance market was dysfunctional due to adverse selection, and as a result employers have strong incentives to provide spousal insurance benefits to their workers because otherwise it would be close to impossible for the non-working spouses to obtain insurance elsewhere. 123 / 137 ACA and Firms’ Incentives to Offer Spousal Health Insurance Benefits If health insurance exchange established under the ACA operates as well as it is intended, the employees would no longer value the spousal insurance benefits as much. Firms benefit from improving the health of their employees because healthy workers are more productive, so they may still have incentive to offer health insurance to their workers; but they know they do not directly benefit from the improved health of the spouses of their employees, especially if the mobility decisions of their employees are now less dependent on whether spousal insurance benefits are offered. 124 / 137 Survey Evidence There has already been some early evidence that employers are responding to the ACA in their offerings of benefit plans. According to a survey conducted by Towers Watson National Business Group on Health titled “Employer Survey on Purchasing Value in Health Care” (2013): I I I 18 percent of surveyed firms either have already or are planning to require spouses to purchase health insurance through their employer plan before enrolling in their health plan; 12 percent of the respondent firms either have already or are planning to exclude spouses from enrolling in their health plan when similar coverage is available through their own employer; and 5 percent are planning to completely eliminate spousal coverage (page 19). 125 / 137 Our Research Question Do these short-run responses by the firms represent a longer-term trend? Our proposed research will help uncovering the mechanisms underlying firms’ decisions regarding the offerings of their health insurance benefits packages. We plan to estimate the model using MEPS and SIPP data and use the estimated model to conduct counterfactual experiments to evaluate the new labor market equilibrium under the ACA, as well as variations to the ACA. A structurally estimated model of joint household labor market search would also allow us to evaluate the job lock hypothesis, particularly how much spousal health insurance benefits affect job lock. 126 / 137 Model: Overview In the equilibrium model that we develop to study health care reform, we consider a frictional labour market where workers receive job offers when both employed and non-employed. These job offers are characterized by a wage rate, together with a vector of insurance offerings that the worker is able to select from. The vector of insurance offers corresponds to the different combinations of employee and spousal coverage, and the associated insurance premiums. The worker side of the economy is populated by both singles and couples, both of whom differ in observed and unobserved dimensions. These risk averse households make a series of job mobility and insurance take-up decisions. In the context of couple households, the acceptance and insurance decision of any one adult will in general depend upon the state of their spouse, thus generating potentially rich behavior and dynamics. 127 / 137 Model: Overview Health and health insurance impact the labour market through several channels. Workers recognize that their decision to purchase insurance (when available) not only insures their household against medical expenditure risk, but also influences how their health statuses evolve. There is also a direct impact on the demand side of the market, since health affects how productive workers are in their job. The provision of health insurance (and the structure of the compensation package more generally) is an equilibrium object, that emerges as an outcome of a non-cooperative game between heterogeneous firms. The reforms to the health care market that we study here have a direct impact on the incentive for firms to provide health insurance, and also change the value that households place on different compensation packages. 128 / 137 Modeling Firms’ Insurance Offering Options 1 No health insurance (I = 0). Workers therefore receive pre-tax monetary compensation equal to the wage w. A worker in such a job may still be insured if they are covered by their spouse’s insurance. 2 Employee only insurance (I = 1). Insurance is offered, but it does not extend coverage to spouses. Workers decide whether to decline insurance (i = 0) and receive pre-tax monetary compensation w (since r(0; w, I) = 0), or to purchase the insurance (i = 1) at the premium r(1; w, 1) (which is a pre-tax deduction). 3 Employee and spouse insurance (I = 2). Insurance is offered, and made available to both the employee and their spouse. Again, workers decide whether to decline insurance (i = 0) and receive w, to purchase insurance for the employee only (i = 1) at premium r(1; w, 2), or to purchase insurance for the employee and spouse (i = 2) at premium r(2; w, 2). 129 / 137 Qualifying Events and Open Enrollment Period Employees are not able to change insurance coverage options freely during a job spell. In particular, workers are not able to change coverage in response to a changes in health. There are two ways that coverage may be changed. 1 2 it may be changed in response to a qualifying event, which (given the absence of family transitions in our model) is associated with either adult starting a new job, or entering the non-employment pool. coverage may be changed when an open enrollment event occurs. We model an open enrollment period by assuming it takes place at some exogenous rate η > 0, which then allows the household to reoptimize over the set of insurance options. 130 / 137 Fang, Shephard, and Tilly (WIP): ACA and Firm Behavior So far, we have limited firms’ responses to ACA to their designs of compensation packages; ACA may also change firms’ incentives in their choice of production technology Would firms decide to use more skill-biased technology in response to ACA, which may be interpreted as a regulation on the labor market that can have differential impact on the cost of hiring different types of labor? This paper aims to understand the important determinants of the strength of firm response in this dimension. 131 / 137 Model Firms produce output with a production function that is multiplicative in the firm’s productivity type and the effective units of labor employed. The effectiveness of a given worker depends on the worker’s skill type, k ∈ {H, L}, the worker’s endogenous health status q ∈ {q1 , . . . , qQ }, and the firm’s costly technology choice, j ∈ {1, 2}. We use subscripts for characteristics related to the worker, and superscripts for characteristics related to the firm. 132 / 137 Model: Technology and Insurance Choice The column vector Yqj describes the effectiveness of labour employed when workers are of health status q and employed in a firm with technology j, and is defined as: " # j Y Hq Yqj = . j YLq At some cost, firms also make a health insurance offering decision I ∈ {IN S, IN S}. This does not affect productively, as given by Yqj , but impacts both the steady state size and composition of its workforce. Without loss of generality, we refer to the sector of the firm, as comprising its joint technology and insurance offering decision, and index this by s ∈ {[j = 1, I = IN S], [j = 1, I = IN S], [j = 2, I = IN S], [j = 2, I = IN S]}. 133 / 137 Model: Firms The measure of workers of H and L workers with health status q in sector s employed by a firm is given by the column vector Lsq : s s ) LHq (wH s s Lq (w ) = . s) LsLq (wL s , w s ]0 are the wage offers (which, by assumption, do where ws = [wH L not vary with current health status q). Excluding any potential fixed sector and insurance offering costs, the profit flow of a productivity p firm that is active in sector s is X 0 π s (ws , p) = pYqs − ws − msq Lsq (ws ) q where msq = [msHq , msLq ]0 is the vector of expected medical expenditure (as faced by the firm) per-worker. For sectors s such that I = IN S, all elements of this vector are zero, with workers facing the full cost. 134 / 137 Model: Labor Market The labor market is subject to search frictions. There is search off and on the job. The total rate at which unemployed and employed workers receive job offers depends on their current labor force status (u or e), their skill type (H or L), and their health status q. These are respectively denoted λeHq , λuHq , λeLq and λuLq . The rate at which workers meet firms in a particular sector are given by these arrival rates multiplied by the fraction of firms active in a particular sector. The rates at which jobs are exogenously destroyed is given by δ sHq and δ sLq . The offer arrival and destruction rates for type L workers are defined analogously. 135 / 137 Firms’ Problem Firms are exogenously endowed with a productivity type p that is distributed according to a distribution denoted Γ(p), with lower bound p and upper bound p̄. A sector s firm takes the aggregate labor supply functions Lsq (ws ) as given. The sector s firm chooses a pair of wages ws to maximize its steady state payoffs. The maximization problem of a sector s firm is as follows: + * X s s s 0 s s s (7) π (p) = max pYq − w − mq Lq (w ) s w q subject to Lsq (ws ) being described by the flow equations below, and expected medical expenditure per-worker given by: Z 0 Z msq = [m − z s (m)] · dMHq (m), [m − z s (m)] · dMLq (m) 136 / 137 Firms’ Problem Firm choose to be active in the sector that leads to higher steady state profits. They are choosing amongst four possible sectors conditional upon the (persistent) sector specific shocks. The flow cost in sector s is given by cs + cs , where cs is common to all firms, and cs is distributed independently across firms and sectors according to a Type-I extreme value distribution. The probability that a productivity p firm will be active in sector s is therefore given by: i h s s exp π (p)−c σc h s0 Pr(p firm in sector s) = P 0 i. π (p)−cs exp s0 σc where σ c is the scale parameter. 137 / 137