Paying a Premium on your Premium? Consolidation in the U.S. Health Insurance Industry. Leemore Dafny Northwestern University and NBER

Transcription

1 Paying a Preiu on your Preiu? Consolidation in the U.S. Health Insurance Industry Leeore Dafny Northwestern University and NBER Mark Duggan University of Maryland and NBER Subraania Raanarayanan University of California at Los Angeles March 2010 Abstract We exaine whether and to what extent consolidation in the U.S. health insurance industry is leading to higher eployer-sponsored insurance preius. We ake use of a proprietary, panel dataset of eployer-sponsored healthplans enrolling over 10 illion Aericans annually between 1998 and 2006 to explore the relationship between preiu growth and changes in arket concentration. We exploit the differential ipact of a large national erger of two insurance firs across local arkets to estiate the causal effect of concentration on arket-level preius. We estiate real preius increased by 2.8 percentage points (in a typical arket) due to the rise in concentration during our study period. We also find evidence that consolidation facilitates the exercise of onopsonistic power vis a vis physicians, whose absolute eployent and relative earnings decline in its wake. e-ail addresses: l-dafny@kellogg.northwestern.edu, duggan@econ.ud.edu, subbu@anderson.ucla.edu We are grateful for helpful coents by Michael Chernew, Julie Cullen, Roger Feldan, Alan Sorensen, seinar participants at Aerican University, Brown University, Dartouth College, the Departent of Justice, Ohio State University, University of Rochester, University of Michigan, and participants at the Aerican Econoic Association Annual Meetings, International Industrial Organization Conference, the Aerican Society of Health Econoists Conference, the University of British Colubia Suer Industrial Organization Conference, the Searle Center Syposiu on Antitrust Econoics and Policy, and the NBER Suer Institute. We thank Michael Chernew, Jose Guardado, Woolton Lee, and Dennis Scanlon for valuable discussions. Dafny gratefully acknowledges funding fro The Searle Center on Law, Regulation, and Econoic Growth at the Northwestern University School of Law.

2 Although the vast ajority of healthcare expenditures in the U.S. are funneled through the health insurance industry, few researchers have exained whether the industry itself is contributing to rising health insurance preius. This possibility has becoe ever ore salient as consolidations continue in this highly-concentrated sector. In 2001, the Aerican Medical Association (AMA) reported nearly half of the 40 largest Metropolitan Statistical Areas (MSAs) were highly concentrated, using the Horizontal Merger Guidelines cutoff of HHI > 1,800. By 2008, the AMA expanded its annual report to include 314 geographic areas (ainly MSAs), 94 percent of which were found to be highly concentrated. 1 During the sae period (corresponding to data years 2000 and 2006), the average preiu for a faily of four receiving coverage through an eployer rose 81 percent, reaching $11,480 in This study exaines whether there is a causal link between changes in arket concentration and growth in health insurance preius. Fro a theoretical standpoint, both the sign and the agnitude of the effect of concentration on insurance preius are abiguous. On the one hand, increases in arket concentration ay allow health insurers to raise their arkups, leading to higher preius. On the other hand, increases in arket share ay strengthen insurers bargaining positions vis a vis healthcare providers, leading to reduced outlays and lower preius. In addition, there are any potential sources of efficiency gains fro consolidation, including econoies of scale in IT investing and disease anageent progras, which would also reduce costs and optial preius. 3 The net effect on insurance preius is an epirical question. The key challenges to epirically estiating such a link are adequate data and exogenous variation in arket concentration. Coprehensive data on healthplans is extreely difficult to obtain because contracts are custoized for each buyer across any different diensions, renegotiated annually, and considered highly confidential. In addition, preius vary based on the deographics, health risks, and expenditure history (or experience ) of the insured 1 Copetition in Health Insurance: A Coprehensive Study of U.S. Markets, Aerican Medical Association, 2001 and HHI is calculated for the cobined HMO and PPO product arket. Estiates are not strictly coparable over tie due to changes in ethodology and saple selection. For exaple, self-insured HMOs are generally included in 2001 but excluded in Preius include both eployer and eployee contributions. Source: Eployer Health Benefits Suary of Findings, 2000 and 2006, Kaiser Faily Foundation/Health Research and Educational Trust Survey, 3 Rent transfers fro providers to insurers are not efficiency gains, although they ay reduce preius. 1

3 population. Thus, it is difficult to calculate a standardized preiu to enable coparisons across eployers and/or over tie. To address these issues, we ake use of the panel feature of our dataset, which spans the years 1996 through 2008 (inclusive) and includes detailed inforation on the healthplans offered by a saple of large U.S. eployers; over 10 illion Aericans are represented in the saple each year. Our analysis focuses on the growth in average health insurance preius for the sae eployer in a specific arket over tie. This alleviates concerns about tie-invariant unobservable differences in the risk profiles of eployee groups and the characteristics of plans they utilize. We also exploit tie-varying easures such as eployee deographics, the types of plans offered (HMO, POS, etc.), and the generosity of benefit design. After docuenting trends in the level and growth of concentration (as easured by the su of squared arket shares, or HHI) in 139 distinct geographic arkets, we estiate OLS odels of the relationship between preiu growth and concentration levels. We do not find evidence that preius are rising ore quickly in ore concentrated arkets. Although these estiates are useful for descriptive purposes, they do not provide causal estiates of the ipact of arket structure on preius. Differences in HHI across arkets or even changes in HHI within arkets - are likely to be driven by any factors that are not exogenous to preiu growth. These include differences (or changes) in consuer preferences, product offerings and pricing strategies, and the healthcare provider landscape. For exaple, consider a arket with a struggling local econoy. In such a arket, consuers ay flock to low-priced carriers, bringing about an increase in local arket concentration and a siultaneous reduction in average preiu growth. This pattern does not iply consolidations in such a arket would reduce preiu growth, ceteris paribus. To obtain an estiate of the causal ipact of concentration on preiu growth, we exploit sharp and heterogeneous increases in local arket concentration generated by the 1999 erger of two industry giants, Aetna and Prudential Healthcare. Both were national firs, active in ost local insurance arkets, and thus the erger had widespread ipact. However, the pre-erger arket shares of the two firs varied significantly across local arkets, resulting in very different shocks to post-erger copetition. For exaple, in our saple the pre-erger arket shares of Aetna and Prudential in Jacksonville, Florida were 19 and 24 percent, 2

4 respectively, versus 11 and 1 percent, respectively, in Las Vegas, Nevada. Holding all else constant, this iplies an increase in post-erger HHI of 892 points in Jacksonville, but only 21 in Las Vegas. (HHI is easured as the su of squared arket shares for each carrier, ultiplied by 10,000.) Focusing on the years iediately surrounding this erger, we exaine the relationship between preiu growth and HHI changes using these predicted changes as instruents for actual changes. The point estiates indicate that rising concentration in local health insurance arkets accounts for a sall share of preiu growth in recent years. Specifically, our instruental variables estiates iply that the ean increase in local arket HHI during raised preius by approxiately 2.77 percent. Given private insurance preius of roughly $850 billion in 2009, if this result is generalizeable the preiu on preius is on the order of $24 billion per year. 4 Although our focus is on the exercise of arket power by insurers in the output arket, consolidation ay also have iportant effects on input prices. Using data on earnings and eployent of healthcare personnel, we exploit the Aetna-Prudential erger to exaine a causal link between concentration and these outcoes. Our analysis indicates that the growth in insurer bargaining power following this consolidation resulted in lower earnings and eployent growth for physicians, and higher earnings and eployent growth for nurses. The paper is organized as follows. Section 1 discusses prior related research. Section 2 describes the data in detail. We exaine the association between local arket concentration and preiu growth in Section 3. In Section 4 we estiate a causal relationship between these two variables using the variation in HHI induced by the Aetna-Prudential erger. Section 5 extends the analysis in Section 4, exaining the pattern of price increases faced by Aetna and Prudential custoers relative to other custoers, and exploring the ipact of the erger-induced changes in HHI on other outcoes of interest such as the percent of enrollees in HMOs. Section 6 describes our analyses of the relationship between changes in concentration and healthcare eployent and earnings. Section 7 concludes. 4 Source: National Health Expenditure Data provided by the Center for Medicare and Medicaid Services; available online at The $850 billion figure underestiates the size of the industry as it excludes revenues fro Medicaid anaged care and Medicare Advantage. 3

5 I. Related Research Our study builds on research fro two distinct streas of literature: studies of the relationship between arket concentration and copetitive outcoes in the epirical industrial organization literature, and studies of the health insurance industry, ainly fro the health services literature. In this section, we suarize the key insights of each, and identify our contributions at the end. A. Price-Concentration Studies in Industrial Organization The structure-conduct-perforance paradig in industrial organization triggered a wave of epirical studies of the relationship between arket concentration and profitability. 5 Using cross-sectional data for a large nuber of industries, any of these studies docuented a positive relationship between profits and concentration. 6 This approach was faously critiqued by Harold Desetz (1973), who argued that the observed relationship could also be explained by differences in efficiency across firs. 7 Subsequent studies focus on price, an outcoe less influenced by this efficiency critique. Recent studies in this literature rely on within-industry variation in concentration and price, priarily by using observations on different geographic arkets. Most docuent higher prices in ore concentrated arkets. Exaples include Morrison and Winston (1990) and Borenstein and Rose (1994) in airlines, Hannan (1992) in banking, and (Cotterill 1986) in grocery retailing. However, uch of this work assues arket structure is exogenously deterined with respect to price. Given any of the sae unobservable factors deterine both, regressions of price on concentration and observable controls likely yield biased estiates. 5 Although our discussion focuses on studies of horizontal consolidation, researchers have also investigated the ipact of vertical consolidation on price (as well as other outcoes). Recent exaples of such studies include Cuellar and Gertler (2005) on physician-hospital integration and Hortacsu and Syverson (2007) on integration in the ceent and ready-ixed concrete industries. 6 See Weiss (1989) for a suary of these early studies. 7 This approach was also criticized on other fronts, particularly on the failure to control for differences in econoic factors across industries, and on the use of accounting easures for profitability. 4

6 Recent studies have pursued two distinct approaches to surount this endogeneity issue. One solution relies on a two-step estiation procedure. In the first step, the authors estiate an equilibriu odel predicting the nuber of copeting firs in a arket. This odel is used to generate a correction ter to include in the second-stage regression of price on concentration, uch in the sae way selection correction ters are included in wage regressions (Heckan 1979). Soe recent exaples include Manuszak and Moul (2008), who use this ethod to evaluate the prospective ipact of the Staples-Office Depot erger, and Singh and Zhu (2006), who study auto rental arkets. Mazzeo (2002) extends this approach to account for the ipact of product differentiation by specifically allowing for differences in the copetitive effects of firs with different product characteristics. This approach lends itself to estiating welfare changes and perforing counterfactual experients (such as estiating the effects of a erger), but it requires strong assuptions about the behavior of firs to enable an accurate characterization of arket structure in the first-stage equation. The second solution requires variables that can serve as instruents for arket structure, i.e. easures that are correlated with arket structure but uncorrelated with unobservable factors affecting price. Two of the best-known studies in this vein use lagged arket structure as an instruent for current arket structure: Evans, Froeb and Werden (1993) (airlines) and Davis (2004) (ovie theaters). For exaple, Davis explores the relationship between within-theater variation in pricing and geographic arket structure, using lagged counts of ovie screens owned by own and rival chains within various distances as instruents for their current levels. He finds ownership structure has a statistically significant but econoically sall effect on adission prices charged to consuers. Unfortunately, using lags of an endogenous variable as an instruent is only valid under relatively strong assuptions. We also pursue an instruental variables approach to estiate the causal relationship between arket structure and price. Our instruent consists of arket-specific shocks induced by a large national erger. To the extent these shocks are both correlated with observed changes in arket structure and orthogonal to other deterinants of preiu growth, our estiates will be unbiased. We are unaware of other studies that explicitly use ergers to instruent for changes in arket concentration, although there is certainly a related literature on erger effects. Although this literature is too vast to be suarized here, we note that ost of the reduced-for 5

7 estiates of erger effects suffer fro selection bias. Markets or industries in which ergers occur are unlikely to be randoly selected, or to be ore precise, to be selected in a way that is unrelated to other deterinants of the outcoes of interest. Soe erger analyses contend with this selection proble by exploiting a teporal shock that induces additional ergers, such as Ki and Singhal (1993) on airlines and Berry and Waldfogel (2001) on radio stations, or by using an instruent to predict which institutions erge (Dafny (2009a) on hospital ergers). Rather than exploiting ultiple exogenously-induced ergers, this study exploits a single erger with different ipacts across geographic arkets. We carefully consider whether the erger we exaine generates plausibly exogenous variation in arket concentration. This identification strategy is siilar to that of Gilbert and Hastings (2005), who use an acquisition of a West Coast refinery as a source of exogenous variation in the degree of vertical integration across retail gasoline arkets in 13 West Coast etropolitan areas. B. Studies Focusing on the Health Insurance Industry Several studies published in health econoics or health services journals exaine the relationship between industry structure and insurance price (i.e., preius). Robinson (2004) uses a database of state regulatory filings to study the state-level arket structure of coercial insurance carriers in He finds the largest fir controls at least a third of the arket in alost 40 states in The top 3 insurance firs control over 50 percent of total enrollent in alost all states. Using a variety of other sources, Robinson also docuents a sharp increase in insurer revenues and profits over the tie period There is, however, no attept to establish a causal relationship between these two phenoena. Wholey, Feldan and Christianson (1995) exaine the effects of HMO arket structure (easured by the nuber of HMOs) on HMO preius fro 1988 to Their analysis uses the HMO (which ay be national, regional, or local) as the unit of observation. Preius are estiated as average preiu revenue per eber, and the arket structure facing each HMO is a weighted average of the nuber of copetitors in the geographic arkets in which the HMO is active. The results suggest preius decrease when entry occurs. However, the specifications 6

8 do not include HMO fixed effects, so the results are subject to the usual biases arising fro cross-sectional sources of identification. 8 Key to our study design is a unique, proprietary dataset containing detailed inforation on the healthplans of roughly 10 illion Aericans in every year fro 1998 to 2006, inclusive. This dataset affords us a nuber of advantages over other studies of the industry. It includes the actual preiu charged to every sapled eployer for each healthplan they offer. Several details are available for each healthplan as well, including the identity of the insurance carrier, the plan type, and a suary easure of enrollee deographics. The icro-level data enables us to avoid the noise and error associated with high-level aggregation. We also ake use of geographic arket definitions supplied by the industry, as opposed to arbitrary geographic units that ay correspond poorly to actual arkets. Finally, the panel nature of the dataset perits us to eliinate cross-sectional differences across arkets and eployers as a source of identification for the relationship of interest. Our research copleents recent work by Dafny (2010). Using the sae dataset eployed here, Dafny finds health insurers engage in direct price discriination, charging higher preius to firs with deeper pockets, as easured by operating profits. This evidence of price discriination iplies insurers possess and exercise arket power in soe local arkets. Here, we focus on whether insurers use their arket power to raise preius overall, and by how uch. II. Data Our priary source is the Large Eployer Health Insurance Dataset (LEHID). LEHID contains inforation on all of the healthplans offered by a large and non-rando saple of eployers between 1998 and 2006, inclusive. Descriptive statistics for each year of data are presented in Table 1. LEHID is gathered and aintained by a leading benefits consulting fir, and the eployers included in the dataset have soe past or present affiliation with the fir. The unit of observation is the healthplan-year. A healthplan is defined as a unique cobination of eployer, arket, insurance type, insurance carrier, and plantype, e.g. Copany X s 8 In a related study, Feldan and Wholey (1996) use data on HMOs to estiate the effect of HMO ergers on preius, and find that ergers do not affect HMO preius except in the ost copetitive arkets. 7

9 Chicago-area fully-insured Aetna HMO. We now discuss each of the coponents that jointly identify this unit of observation in turn. The full dataset includes observations fro 813 eployers. Eployers ay enter or exit the saple at any tie. The edian nuber of years an eployer is present in the saple is two. One-quarter of eployers appear in the saple for 4 or ore years. A non-trivial nuber of eployers reappear after exiting. Most eployers are large, ulti-site, publicly-traded firs, such as those included on the Fortune 1000 list. The leading industries represented include anufacturing (110 eployers), finance (101), and consuer products (73), although nonprofit and governent sectors are also represented (43 in the governent/education category). Geographic arkets are defined by the source using 3-digit zipcodes. According to the data source, the 139 arkets reflect the geographic boundaries typically used by insurance carriers when quoting prices. Large etropolitan areas are separate arkets, and nonetropolitan areas are luped together within state boundaries, (e.g., New Mexico Albuquerque and New Mexico except Albuquerque ) 9 To atch county-level data to these arkets, we allocate zipcodes within the arkets to counties, and use zipcode population data to weight the county data appropriately when aggregating to the arket level. The two county-year easures we use are the uneployent rate (fro the Bureau of Labor Statistics), and the average Medicare costs per capita (known as the AAPCC, fro the Center for Medicare and Medicaid Services). We also calculate the general, acute-care hospital HHI at the arket-year level using hospital-year data on the nuber of beds for all general hospitals included in the Aerican Hospital Association Annual Surveys of Hospitals. To create this easure, we assign hospitals with the sae syste ID to a coon owner. The saple includes both fully-insured and self-insured plans. As these ters suggest, the forer is classical insurance in which the insured pays the carrier to bear the risk of realized healthcare outlays. Many large eployers choose to self-insure, outsourcing benefits anageent and/or clais adinistration but paying realized costs of care. Such eployers can spread risk across large pools of enrollees, and ay purchase stop-loss insurance to liit their reaining exposure. Per ERISA (the Eployee Retireent Act of 1974), these plans are also 9 There is only one arket that crosses state boundaries, Massachusetts Southern and Rhode Island. A few rural areas of the U.S. are explicitly excluded. A ap of the arkets is available in Dafny (2010). 8

10 exept fro state regulations (such as specific benefit andates) and state insurance preiu taxes. Even though insurance carriers do not bear the risks associated with edical expenditures under self-insurance, their role in adinistering clais and especially in negotiating provider discounts affords the the scope to potentially ipose arkups via steeper charges. (Selfinsured contracts typically include soe cobination of fees per eployer, per eployee, and/or per clai.) In our saple, the fraction of plans that are fully-insured declines fro 45 to 20 percent between 1998 and The decline is soewhat less precipitous when calculated using the fraction of enrollees 42 to 25 percent but clearly reains an iportant phenoenon in the data. The reasons for this decline are the subject of a current research project. Here we note the decline is not particular to our data source: it has been corroborated in the Kaiser Faily Foundation/Health Retireent Education Trust Annual Survey of Eployer Benefits and the Medical Expenditure Panel Survey-Insurance Coponent (MEPS-IC), and appears to be especially pronounced aong the very largest firs. 10 We revisit this issue again when discussing our unit of analysis below the eployer-arket-year cell. Each fir that adinisters any plan in the data is labeled an insurance carrier. 11 During the entire study period, there are 357 carriers that serve at least one eployer, and 195 that serve 5 or ore. The saller carriers tend to be local or regional firs, or soeties third party adinistrators who pay clais and contract with another fir to rent its network of providers and associated discounts. The industry is highly concentrated and becoing ore so over tie. Figure 1 presents the four-fir concentration ratio for the nation as a whole, estiated using the LEHID saple. This easure increased fro an ipressive 58 percent in 1998 to 79 percent in As we illustrate in the following section, concentration ratios within local arkets - arguably where ost of the copetition takes place - are uch higher We are grateful to Kosali Sion for tabulating the MEPS-IC data to investigate this trend. 11 Blue Cross and Blue Shield (BC/BS) affiliates are all assigned the sae carrier ID. (Note: both Wellpoint and Anthe (before it was acquired by Wellpoint) own BC/BS affiliates, so they also have the BC/BS carrier ID. Given we calculate concentration within each arket, and there are only a handful of arkets in which BC/BS affiliates coplete, the unifor coding of these affiliates is unlikely to be consequential for our analysis. 12 The notable exception is the arket for ultisite eployers interested in a unifor plan across all sites. Our data do not include an identifier for jointly-negotiated plans. 9

11 The plan types, ordered fro ost to least restrictive in ters of provider choice, are Health Maintenance Organization (HMO), Point of Service (POS), Preferred Provider Organization (PPO), and Indenity. HMOs and POS plans control utilization of care through priary-care physicians ( gatekeepers ). Only in-network providers are covered by HMOs, while POS plans provide soe coverage for out-of-network providers (once the gatekeeper has approved the service in question). PPOs engage in less utilization anageent, and like POS plans, typically cover out-of-network care at a reduced rate. Finally, indenity plans are traditional fee-for-service arrangeents in which benefits do not depend on the network status of the provider. As Table 1 reveals, the coposition of plan types fluctuated during the study period, with a clear resurgence of PPOs toward the end of the study period. In addition to the eleents that jointly define a plan, we have the following variables: preiu, deographic factor, plan design factor, and nuber of enrollees. Preiu is expressed as an average aount per enrollee (i.e. a covered eployee); it therefore increases with the average faily size of enrollees in a given plan. Preiu cobines eployer and eployee contributions, and for self-insured plans it is a projection of expected costs per enrollee (including estiated adinistrative fees paid to an insurance carrier). These projections ay include a partial risk preiu if the eployer purchases stop-loss coverage; whether stop-loss coverage is purchased is not captured in the data. Because the forecasts are used for budgeting and to establish eployee preiu contributions, they are carefully developed and vetted. Eployers often hire outside actuaries and benefits experts (such as our source) to assist in forulating accurate projections. Deographic factor is a easure that reflects faily size, age, and gender coposition of enrollees in a given plan. Plan design factor captures the generosity of benefits within a particular carrier-plan type, with an ephasis on the degree of coinsurance and copays. Both factors are calculated by the source, and the forulae were not disclosed. The nuber of enrollees in each plan refers to the nuber of enrolled eployees, i.e. it does not reflect dependents. The total nuber of enrollees in all LEHID plans averages 4.7 illion per year. Given an average faily size above 2, this iplies over 10 illion Aericans are represented in the saple in a typical year. 10

12 As noted above, we perfor ost analyses using data aggregated to the eployerarket-year level. Table 2 presents descriptive statistics for this unit of observation. Because our priary outcoe is growth in health insurance preius (in order to avoid cross-sectional identification of the coefficients of interest), aggregating the data to the eployer-arket-year level enables us to use a uch larger proportion of the data. With the healthplan-level data, growth in preiu is undefined when an eployer terinates a particular plan. Analogously, new plans can only enter into the analysis after ultiple observations are available. Changes to plan offerings are quite coon in our data. Moreover, changes in arket concentration ay affect the insurance carriers and plan types chosen by eployers, so we do not want a priori to eliinate this substitution fro our saple. 13 Given this aggregation, both fully and self-insured plans ust be included together in the analysis to ensure the set of eployees represented over tie is the sae (but for hiring and attrition, of course). To the extent that self-insured plans are less subject to arkups, the estiates will understate the effects of concentration on fully-insured preius. However, for the saple of large eployers we observe, self and full-insurance are substitutes, hence pooling the plan types yields the ost accurate estiate of preiu growth in the large group arket. We use the penetration of self-insurance in each eployer-arket-year cell as a control variable in ost specifications, and as an outcoe easure in Section 5 ( Extensions ). Before proceeding to the analyses, we offer soe rearks regarding the representativeness of the LEHID data. As previously stated, LEHID consists priarily of large, ultisite eployers. In Appendix Figure 1, we copare annual preiu growth observed in LEHID with annual preiu growth for all firs, as reported by the Kaiser Faily Foundation (KFF) and the Health Research and Educational Trust (HRET). 14 The KFF/HRET Annual Survey of Eployer-Sponsored Health Benefits is specifically designed to yield nationallyrepresentative data. Although we would not expect preiu levels to be siilar across the two saples, if growth rates are siilar this would suggest the results of our study are applicable to a broader saple of eployers because all specifications rely on preiu growth over tie. The 13 As an exaple of the frequency with which this occurs, consider eployer-arket pairs that are present in both 1999 (the year of the Aetna-Prudential erger) and More than half of the plans offered by these firs in 1999 are no longer present in 2002, either because the eployer switched to different carriers or because it changed the type of plan with the sae carrier. 14 The KFF/HRET survey randoly selects public and private eployers to obtain nationally-representative statistics for eployer-sponsored health insurance; approxiately 2000 eployers respond each year. The icro data are not publicly available, nor is the saple designed to provide estiates at the arket level. 11

13 graph reveals that the trends in both saples are very siilar over tie. Dafny (2010) reports that the ratio of sapled enrollees to total insured lives (available at the county-level fro the US Census of 2000) varies little across geographic arkets. In the appendix, we describe our efforts to copare the LEHID-based estiates of arket structure with those obtained by other researchers using the proprietary InterStudy database, specifically Scanlon, Chernew, and Lee (2006). Scanlon et al (2008) use these data to show that increased levels of HMO copetition do not lead to increases in plan quality. InterStudy reports soe enrollent and preiu figures at the insurer and MSA level, but for reasons outlined in the Appendix, it is not an ideal source for our purposes. III. Is Preiu Growth Correlated with Local Market Concentration? In this section, we exaine the relationship between preiu growth and local arket concentration. We begin by describing the distribution of arket-level HHI and how this has changed over tie. Next, we estiate OLS regressions relating preiu growth at the eployer-arket level to the corresponding arket HHI. We include arket fixed effects in our odels, so that we identify the coefficient of interest using changes in within-arket HHI. The richness of the data also perits us to control for iportant tie-varying differences (such as the percent of enrollees in HMOs and the degree of copays). Although interesting as a descriptive exercise, this analysis does not yield estiates of the ipact of changes in arket structure on preiu growth, as changes in arket structure are unlikely to be exogenous. In Section IV, we estiate this causal relationship by using the Aetna-Prudential erger to construct an instruent for arket concentration. A. Market Structure of Large Group Insurance Markets, During our 9-year study period, the average arket-level HHI (estiated using our saple, on a scale fro 0 to 10,000) increased fro 2,286 to 2,984. Using the categorization fro the Horizontal Merger Guidelines, the fraction of arkets falling into the top highly concentrated category (HHI > 1,800) rose fro 68 to 99 percent. Thus, our data confir the conclusions of the well-publicized reports issued by the Aerican Medical Association: local health insurance arkets are highly concentrated and becoing ore so over tie 12

14 Figure 2 presents histogras of the arket-level changes in HHI, separately for , , and The biggest increases occurred during the second half of the study period, but sizeable increases are present in the first half as well. Between 1998 and 2002, 53 percent of arkets experienced increases in HHI of 100 points or ore, and 25 percent saw increases of 500+ points. The corresponding figures for 2002 to 2006 are 78 and 53 percent, respectively. The Merger Guidelines provide a helpful frae of reference for interpreting these changes. According to the Guidelines, ergers resulting in an increase of 100+ points are presued likely to create or enhance arket power or facilitate its exercise. 15 There is wide variation in the agnitude of changes in HHI across arkets, notwithstanding the fact that ost are positive. The reasons for these changes in HHI (apart fro saple coposition, which we discuss below), can be subdivided into structural (related to entry, exit, and consolidation of insurance carriers) and non-structural sources. Using data on fully-insured HMOs only, Scanlon et al (2006) report that 61 to 65 percent of the variation in HHI between 1998 and 2002 is attributable to changes in arket structure. Structural changes (priarily due to consolidation or exit) are also iportant in our saple: the ean nuber of carriers per arket declined fro 18.9 in 1998 to 9.6 in Figure 3 contains histogras for changes in the nuber of carriers. Between 1998 and 2002 the odal net loss is 1 to 3 carriers; the corresponding range for 2002 to 2006 is 4 to 6 carriers. Of course, neither structural nor non-structural sources of changes in HHI can be presued exogenous to other deterinants of preiu growth. Exit and consolidation of carriers ay be ipacted by expectations of preiu growth, and consuer preferences siultaneously deterine arket shares and preiu growth. B. OLS Estiates of the Relationship between Market Structure and Preius To explore the relationship between preiu growth and arket concentration, we being by estiating equations of the following for: 15 Horizontal Merger Guidelines, Federal Trade Coission and Departent of Justice, issued in 1992 and revised in Accessed at 16 As the data on HHI suggests, any of these carriers are quite sall. This is due to the presence of any sall self-insured plan adinistrators, particularly in the earlier part of the study period. Soe of these adinistrators ay not be active participants in a given arket, i.e. they rent networks fro other carriers so as to offer a particular client a consistent plan across all geographies. 13

15 ( 1) Δ ln( preiu) et = α + βhhi, t 1 + X t-1ϑ + ΔC [ + ς ][ + ωδplan type shares e et et + τ + λ t + ϑδplan design et ] + ε Stated in words, we odel preiu growth between year t and year t-1 for a given eployer e in arket as a function of lagged arket characteristics (including HHI), conteporaneous changes in observable characteristics of the insured population (such as deographics), and year and arket fixed effects. Market characteristics are lagged by one year because preius are set prospectively, i.e. preius for 2006 are deterined in In addition to HHI, the et. arket-year covariates (denoted X t-1 ) include the uneployent rate (to capture local econoic conditions), the log of per-capita Medicare costs (to capture trends in healthcare utilization), and the general, acute-care hospital Herfindahl index (to capture concentration in the provider arket, which could independently lead to preiu increases). Note these characteristics are included in level for (rather than first differences) to allow for a delayed response to changes. In contrast, we anticipate concurrent preiu responses to changes in characteristics easured at the eployer-arket-year level ( Δ Cet ), specifically deographic factor (which is doinated by changes in the average nuber of dependents who sign up for coverage) and the percentage of enrollees in self-insured plans. The year fixed effects capture national changes in preiu growth, and the arket fixed effects capture differences in average growth across arkets. The inclusion of these fixed effects eliinates general tie-series and cross-sectional differences in concentration as sources of identifying variation for β. Results are presented in coluns 1 through 3 of Table 3. In all odels we discuss, HHI is easured on a scale fro 0 to 1, and standard errors are clustered by arket. The first colun corresponds to the baseline specification, which excludes the bracketed ters in equation (1). Colun 2 adds eployer fixed effects, which will affect the coefficient on HHI if eployers with particularly high or low average preiu growth are systeatically located in arkets with particularly high or low growth in HHI. Colun 3 introduces controls for changes in the generosity of plans, naely the change in the percent of enrollees in each plan type (excluding POS, the oitted category), and the change in plan design. Relaxing constraints on provider choice and utilization (i.e. oving toward PPOs) should be associated with higher preius. 14

16 Increases in plan design should also result in higher preius. Because substitution across plan types and odifications to plan design ay constitute a response to changes in HHI, controlling for these ters is akin to using a Laspeyres price index as a dependent variable, i.e. using the change in price for a fixed product type and design. The OLS estiates reveal no significant association between concentration levels and preiu growth, and the estiates change little upon inclusion of additional controls. 17 However, we can only ake causal inferences using this odel if within-arket variation in insurer concentration is uncorrelated with other unobserved deterinants of preius, and if variation in preiu growth does not induce variation in concentration. There are good reasons to doubt the validity of these assuptions. Exits or ergers of carriers (and hence increases in HHI) ay be ore likely in arkets where preiu growth is expected to be low. Nonstructural changes in HHI ay also generate a downward bias in the HHI coefficient. For exaple, if eployers in arkets with di econoic prospects substitute toward a low-priced Walart-style carrier, HHI will increase while preius decrease. Indeed, ost plausible sources of endogeneity suggest the OLS coefficient will be downward-biased. Hence in the section that follows we pursue an instruental variables approach. IV. Do Increases in Local Market Concentration Cause Increases in Preius? In this section, we attept to estiate the causal effect of changes in arket concentration on preiu growth by exploiting shocks to arket concentration produced by ergers and acquisitions (M&A). Because M&A activity in local or regional arkets ay itself be otivated by expected trends in preiu growth, we considered only large, non-local ergers as candidates for this analysis. We also ruled out ergers with insufficient pre or post periods (e.g. Aetna and NYLCare in 1998), few overlapping arkets, or very sall shares in our saple for one of the erging parties (e.g. United Healthcare and MAMSI). Only one erger reained: the Aetna-Prudential erger of Post-erger, the new fir (known as Aetna ) 17 For the ost part, the coefficient estiates on the arket-level control variables are statistically insignificant. The coefficient estiates on the eployer-arket controls are highly significant, and all have the expected signs. For exaple, a shift fro 100% enrollent in POS plans (the oitted category) to 100% enrollent in HMO plans is associated with a 5 percent decline in preius. 15

17 was widely reported to be the national s largest insurer, covering 21 illion individuals. 18 Iportantly, and as we describe in detail below, there was substantial overlap in the local arket participation of Aetna and Prudential prior to the erger, generating the potential for sizeable post-erger changes in local arket concentration. Our analysis is subdivided into four sections. First, we discuss the context for the erger, paying special attention to whether the tiing was affected by anticipated, arketspecific changes in preiu growth trends. Second, we estiate the ipact of the erger on arket concentration (the first stage analysis). In so doing, we docuent the range of preerger arket shares for Aetna and Prudential, as well as the degree of pre-erger overlap. Third, we perfor a reduced-for analysis, in which we exaine the ipact of the erger on preiu growth. Fourth, we cobine these analyses to produce our estiate of the causal ipact of concentration on preius. A. The Aetna-Prudential Merger of 1999 In Deceber 1998 Aetna Inc. announced its intention to purchase Prudential Health Care (hereafter Prudential) for $1 billion. Prudential had been publicly searching for an acquirer since at least October of the year prior; it was widely reported to be losing oney and its parent fir, Prudential Insurance Copany of Aerica, had decided to exit the health insurance business. Iportantly, Aetna was an unlikely suitor, as it had recently closed another $1 billion acquisition (of NYLCare), and had publicly stated that future acquisitions would not occur for at least a year. 19 In announcing the deal, Aetna s CEO claied Prudential had ade an offer we can t refuse. 20 The deal closed in July 1999, after Aetna signed a consent decree to address concerns raised by the Departent of Justice (DOJ). DOJ alleged that after the erger, Aetna would have a arket share for fully-insured HMOs of 63 percent (in Houston), and 42 percent (in Dallas). As a precondition to approve the erger, it required the divestent of all Houston and Dallasarea plans Aetna had acquired in the 1998 NYLCare purchase. 18 Sanders, Alain L., Will the Aetna-Prudential Merger Hurt the Patient? TIME agazine, June 22, Freudenhei, Milt, Aetna to Buy Prudential s Health Care Business for $1 Billion, The New York Ties, Deceber 11, 1998, Section C, page Ibid 16

18 According to industry analysts, Aetna s acquisition of Prudential was part of a strategic bet on the long-ter viability of anaged care. Originally focused on providing fee-for-service plans to large, self-insured eployers, Aetna gabled on the rising popularity of HMOs with the 1996 purchase of U.S. Healthcare, which offered fully-insured HMOs to sall groups. The acquisitions of NYLCare (New York Life s healthcare unit) and Prudential soon followed; anaged plans were also the doinant segent for these units. At its peak (after the Prudential acquisition in 1999), the fir covered 21 illion lives. However, enrollent fell rapidly thereafter, plateauing at 13 illion in A 2004 article in Health Affairs declared Aetna the poster child for the aspirations and failures of anaged care. This history provides two iportant insights for our analysis. First, the Aetna-Prudential erger does not appear to raise ex ante concerns about endogeneity. There is no anecdotal evidence indicating that the erger disproportionately affected arkets that were experiencing low (or high) preiu growth. We corroborate this conjecture epirically below, by exaining whether preiu growth in the pre-erger period was systeatically different in arkets where both firs had significant pre-erger overlap. We also exaine whether the consent decree in Texas successfully neutralized the effect of the erger in these arkets; to the extent it did, we anticipate a different preiu response in the affected arkets. Second, our estiates will rely on a erger whose effect was short-lived, and ay therefore understate the effect of typical consolidations in the industry. 22 B. The Effect of the Aetna-Prudential Merger on Market Concentration In our saple fro 1999, Aetna and Prudential were the third and fifth largest insurers in ters of the nuber of enrollees. All 139 arkets included plans offered by both firs. There was significant variation across arkets, however, in the pre-erger shares of each fir. We hypothesize that arkets served by both firs experienced increases in arket concentration 22 To the extent that Aetna and Prudential offered different products prior to the erger, the preiu effects would be saller than we would expect fro a erger between ore siilar firs. However, in our saple the proportion of anaged care plans (HMOs and POS plans) is siilar for Aetna and Prudential prior to the erger. 22 To the extent that Aetna and Prudential offered different products prior to the erger, the preiu effects would be saller than we would expect fro a erger between ore siilar firs. However, in our saple the proportion of anaged care plans (HMOs and POS plans) is siilar for Aetna and Prudential prior to the erger. 17

19 iediately following the erger of Aetna and Prudential, and that these increases varied by the pre-erger shares of the two firs. Specifically, for every arket we calculate siulated HHI change ( si ΔHHI ) as follows: (2) si ΔHHI = 2 [ ] 2 2 [ Aetna 1999 share + Pru 1999 share ] ( Aetna 1999 share ) + ( Pru 1999 share ) = 2* Aetna 1999 share * Pru 1999 share si Δ HHI represents the erger-induced increase in arket s HHI that would have occurred fro 1999 to 2000 absent any other changes in carriers arket shares. For exaple, if Aetna and Prudential were two of four firs with equal arket share in 1999, si ΔHHI would equal = (0.5) 2 ((0.25) 2 + (0.25) 2 ) or 2*0.25*0.25. Figure 4 provides detail on the actual distribution of si ΔHHI. We propose to use si ΔHHI * post t, where post is an indicator variable for the posterger years in the saple, as an instruent for HHI. To evaluate this instruent, we estiate the following equation using arket-year level data, excluding observations for the state of Texas: (3) HHI α + λ + τ + βsi ΔHHI * τ + ε t = t t t The vectors denoted by λ and τ t represent a full set of arket and year fixed effects, respectively. By interacting si ΔHHI with separate duies for each year (except 1998, the oitted category), this odel investigates the possibility that trends in arket concentration ay have been different prior to the erger in arkets differentially ipacted by the erger. The erger was effectively cleared in July 1999, when the Departent of Justice subitted its Proposed Final Judgent. Given insurance preius are set a few onths prior to the start of the calendar year, the ipact of the erger should becoe apparent in 2000 or later. The estiated coefficients will also deterine the appropriate study period for our analysis. 18

20 Figure 5 graphs the coefficient estiates on the yearly interactions with si Δ HHI, together with the 95% confidence intervals. The saple includes data fro 1998 to Estiates are presented in nuerical for in colun 1 of Table 4. Relative to the oitted interaction ter, si ΔHHI * ( year == 1998), only the interactions with indicators for 2000 and 2001 are statistically significant. The coefficient estiate for β in 1999 is sall and negative (-0.10), whereas estiates for β in 2000 and 2001 are large (0.49 and 0.46, respectively) and significant at p<0.05. These coefficients are significantly saller than 1, iplying eployers substituted away fro Aetna and Prudential in the wake of the erger. In addition, there is likely attenuation bias due to easureent error. The coefficient estiates on β in 2002 are 2003 are both noisy and negative. These estiates reveal that the effect of the erger on arket concentration declined sharply after This finding is consistent with reports fro industry experts. According to a 2004 Health Affairs article by Jaes Robinson, [G]ossip speculates [Aetna] would be lucky to still have 30,000 of the 5 illion it acquired fro Prudential. Given the results in Figure 4, we focus our attention on the period fro Next, we estiate a parsionious odel that replaces the individual year interactions with a single post indicator that takes a value of one during 2000 and 2001: = α + λ + τ t + siδhhi * postt + 1 * t t ε t ( 4) HHIt β0 [ β siδhhi post * Texas ] + [ ψpost * Texas ] +. After estiating the baseline odel (which excludes the ters in brackets), we add the six Texas arkets to the saple and include a triple-interaction, si Δ HHI * post * Texas, to explore whether the post-erger ipact of siδ HHI differs in these arkets. We then add the ter post t * Texas to control for average changes in Texas as copared to other states during the post-period, although it ay be difficult to separately identify the coefficient on the two Texas interactions because there are only 6 Texas arkets and two post years. t 19

21 on The results are displayed in Coluns 2 and 3 of Table 4. As anticipated, the coefficient si Δ HHI * post is statistically significant: 0.52, with a standard error of The results t in Coluns 3 and 4 show that the federal governent achieved its objective of neutralizing the erger s effect on arket concentration in Texas arkets: the triple-interaction ter for Texas arkets is negative and statistically significant in both specifications, and copletely offsets the ipact of the erger: in both odels, we cannot reject the hypothesis that the su of the relevant double and triple-interaction ters is equal to zero. Observations fro Texas ay therefore constitute a useful coparison group for our later analyses involving health insurance preius and related outcoes of interest. To suarize, we find the erger of Aetna and Prudential resulted in significant increases in arket concentration, the agnitudes of which are directly related to the degree of overlap in arket shares prior to the erger. The erger-induced shocks to local concentration dissipated quickly, with no lingering effect by We therefore focus our analyses on the early years of our saple, duly noting the shortcoings of this narrow window of observation. Last, we find no effect of the erger on concentration within Texas arkets, where the DOJ consent decree ore than offset the predicted effects of the erger on arket structure. Observations fro Texas will therefore constitute a useful coparison group in our analyses involving health insurance preius and related outcoes of interest. C. The Effect of the Aetna-Prudential Merger on Health Insurance Preius To investigate the effect of the erger-induced increases in local arket concentration on plan preius, we estiate odels of the following for: ( 5) Δ ln( preiu) = α + κ siδhhi * post et + X t-1 [ + ς ][ + ωδplan type shares e [ + κ siδhhi * post * Texas ][ + γpost * Texas ] + ε. 1 0 ϑ + ΔC et + τ + λ t t t et + ϑδplan design t et ] et In light of the results fro the preceding section, we focus on the period between 1998 and 2002 (i.e. annual preiu growth fro , , , and ). Note that in 20