IS THE MEDICAL LOSS RATIO A GOOD TARGET MEASURE FOR REGULATION IN THE INDIVIDUAL MARKET FOR HEALTH INSURANCE?

Authors

  • Pinar Karaca-Mandic,

    Corresponding author
    1. Division of Health Policy and Management, School of Public Health, University of Minnesota, Minneapolis, MN, USA
    • Correspondence to: Division of Health Policy and Management, School of Public Health, University of Minnesota, Minneapolis, MN, USA 55455, USA. E-mail: pkmandic@umn.edu

    Search for more papers by this author
  • Jean M. Abraham,

    1. Division of Health Policy and Management, School of Public Health, University of Minnesota, Minneapolis, MN, USA
    Search for more papers by this author
  • Kosali Simon

    1. School of Public and Environmental Affairs, Indiana University, Bloomington, IN, USA
    Search for more papers by this author

ABSTRACT

Effective January 1, 2011, individual market health insurers must meet a minimum medical loss ratio (MLR) of 80%. This law aims to encourage ‘productive’ forms of competition by increasing the proportion of premium dollars spent on clinical benefits. To date, very little is known about the performance of firms in the individual health insurance market, including how MLRs are related to insurer and market characteristics. The MLR comprises one component of the price–cost margin, a traditional gauge of market power; the other component is percent of premiums spent on administrative expenses. We use data from the National Association of Insurance Commissioners (2001–2009) to evaluate whether the MLR is a good target measure for regulation by comparing the two components of the price–cost margin between markets that are more competitive versus those that are not, accounting for firm and market characteristics. We find that insurers with monopoly power have lower MLRs. Moreover, we find no evidence suggesting that insurers' administrative expenses are lower in more concentrated insurance markets. Thus, our results are largely consistent with the interpretation that the MLR could serve as a target measure of market power in regulating the individual market for health insurance but with notable limited ability to capture product and firm heterogeneity. Copyright © 2013 John Wiley & Sons, Ltd.

1 INTRODUCTION

Currently, an estimated 7.1% of the non-elderly U.S. population obtains health insurance through the individual market.1 Provisions of the Affordable Care Act (ACA), including the 2014 opening of exchanges, the availability of subsidized private insurance to lower-income families, and the individual mandate, are expected to dramatically expand this market. By 2016, the Congressional Budget Office projects that half of those expected to gain coverage through the ACA expansion will do so through the individual market and that this market will serve an estimated 17% of the non-elderly population (Congressional Budget Office, March 2012).

The ACA also alters the regulatory environment of insurers. One provision that has gained considerable attention is the establishment of federal minimum medical loss ratios (MLRs) for firms that sell health insurance. An MLR represents the proportion of a health insurer's premium revenue that is paid out for clinical services, captured primarily by medical claims. The complement of an insurer's MLR provides an estimate of its costs for general administration (e.g., claims adjudication, medical management, underwriting, marketing, and broker commissions) as well as profits.

Effective January 1, 2011, insurers must meet a minimum MLR of 80% in the individual market (Federal Register, 2010). Non-compliant insurers must provide a rebate to enrollees reflecting the premium revenue corresponding to the difference between its actual MLR and the minimum threshold. Thus, by construction, this provision will not fall short of its intended target, because compliance will be achieved ex-post through rebates if insurers fall short of compliance ex-ante through premium and claims adjustments.

The federal MLR law is similar to several state-level regulations, but its requirements are generally stronger. A primary motivation for state minimum MLR requirements was to ensure that premiums overwhelmingly reflect costs associated with enrollees' receipt of clinical services, rather than excess profitability or administrative costs (Robinson, 1997). During U.S. Congressional Hearings from the 2009–2010 sessions, proponents of regulation also made multiple statements suggesting the presence of market power among insurers and excessive pricing motivated their actions. For example, the opening statement of Senator Harkin referred to ‘insurance companies jacking up premiums simply because they can’ (Congressional Hearing, CHRG, April 20, 2010a, 2010b). Other quotes include ‘industry has consolidated to the point that it is now dominated by a cartel of large for-profit insurers’ (CHRG, June 24, 2009, Statement of Wendell Potter, Former Health Insurance Executive, Philadelphia, PA); ‘too many healthcare decisions have been made behind closed doors, with industry profits—not the patient's best interests—in mind’ (CHRG, July 16, 2009, Statement of Hon. John D. Rockefeller IV, U.S. Senator from West Virginia); ‘when I see profits [referring to profits of insurance companies] increasing like that, that tells me those markets aren't working, that tells me those people have market power’ ( CHRG, July 16, 2009, Statement of David Balto, Senior Fellow, Center for American Progress Action Fund); and ‘the real question is why should we tolerate the fragmented, highly profitable, administratively wasteful private health insurance industry any longer?’ (CHRG, June 23, 24 & 25, 2009, Statement of Sidney M. Wolfe, M.D.).

Within economics, the classic argument for regulation is that it corrects market failures. The aforementioned statements suggest that in the context of MLR regulation, the presumed market failure is insurer market power as reflected by excessive profits. Lerner proposed the price–cost margin as a measure of evidence of market power (Lerner, 1934). Because MLRs are mechanically related to the price–cost margin (as explained later), the regulation could be viewed as directly attempting to limit the ability of insurers to exercise market power.

However, there are two major complications to this line of reasoning. The first is that the price–cost margin is an imperfect measure of market power. Its primary shortcoming in this context is that it ignores product heterogeneity and does not take into account that some firms may have justifiably high price–cost margins because they offer superior products (Hay, 1991; Elzinga, 1989). Similarly, certain firm characteristics (non-profit status, firm size, and business tenure) may be inherently related to the price–cost margin even in the absence of market power. Thus, any analysis of the price–cost margin should account for these factors to the best of its ability. The second concern with viewing MLR regulations as limiting insurer market power is that the MLR is only one component of the price–cost margin; the other component is the share of premiums spent on administrative costs. Therefore, insurers could respond to the MLR regulation by altering this other component of the price–cost margin in ways that leave the price–cost margin on net unchanged. For example, insurers could reduce their efforts regarding utilization management, leading to lower administrative expenses, higher claims payments, and higher MLRs. While some reduction in utilization management may be desirable for improving access to efficient health care (e.g., through lower levels of denials or pre-approvals), this reduction could also lead to increased claims for medical care of low value.

Because the premise behind MLR regulation is tied to fears of excessive market power, it is necessary to explore the presence of market power in this industry in the economic spirit of justifying the regulation. On the other hand, assessments of market power directly based on differences in insurers' price–cost margins would also be misleading because of the reasons stated earlier. One possible approach, however, is to compare the two components of the price–cost margin of firms in markets believed to be more competitive to those that are not, all else equal. This is the approach we take in this paper.

To accomplish our objective of evaluating whether the MLR is a good target measure for regulation, our basic exercise compares components of the price–cost margin, namely the MLR, and the percentage of premiums spent on administrative expenses in markets that are more competitive (those that have more insurers) versus those that are not. We also control for various firm and market characteristics to account for the fact that factors other than market power can generate variation in price–cost margins across insurers. For example, if price–cost margins vary systematically by firm size, business tenure, ownership type, and other characteristics, differences in price–cost margin would not necessarily reflect differences in market power.

No empirical research to date has examined the ways in which health insurance market structure and insurer characteristics are related to insurers' MLRs or administrative costs. In addition to examining MLRs, we also consider the numerator (claims) and denominator (premiums) of the ratio separately. We use data on health insurer annual filing statements from the National Association of Insurance Commissioners (NAIC) for 2001–2009 and other secondary sources. A key empirical challenge is the potential endogeneity of the insurance market structure, which we address using multiple approaches.

We find that when insurers are the only credible insurer (defined as having at least 1000 member-years of enrollment) in their market, they have lower MLRs, controlling for insurer characteristics, healthcare provider market structure, and other market attributes, as well as population-level demographics and health status. Several insurer characteristics, such as presence in the group market, business tenure, and HMO status, are positively associated with MLRs. Moreover, we find no evidence that insurers' administrative expenses, expressed as a percentage of premiums, are related to insurance market structure. Thus, our results are largely consistent with the interpretation that MLRs could serve as target measures of market power in regulating the individual market for health insurance but with notable caveats relating to measurement issues, limited ability to capture product and firm heterogeneity that can influence differentials in price–cost margins, and other potential unintended consequences of the regulation.

Section 2 introduces a conceptual framework for analyzing the MLR in the context of price–cost margins. Section 3 outlines the empirical strategy, including discussion of the data and measures and our econometric approach. Section 4 presents the results, and Section 5 identifies implications for policy and key conclusions.

2 CONCEPTUAL FRAMEWORK

In the context of a differentiated goods industry with similar but not identical products, one can conceptualize the MLR as a component of the price–cost margin. Assuming, for simplicity, that each firm f of F firms sells one product (a framework that could be expanded to multi-product firms without loss of generality), the profits πf of firm f can be written as

display math(1)

where pf is the price of firm f's product, mcf is the marginal cost of production, M is the exogenously determined size of the market, sf (p) is the share of firm f 's product, with p representing a vector of prices in the market, and finally Cf is the fixed cost of production. Under a pure-strategy Bertrand–Nash equilibrium in prices, the price–cost margin of firm f can be written as

display math(2)

where the left-hand side of the equation is the well-known price–cost margin and the right-hand side of the equation is the inverse demand elasticity. One approach to measuring market power would be to estimate the vector of own-price and cross-price demand elasticities. However, such an approach would require detailed data on product-level attributes, prices, and shares as well as an identification strategy that relies on exogenous price variation.

In the context of the individual market for health insurance, such data are not available in any generalizable way. Currently, regulators rely on aggregate data on insurer-level premiums, claims, member-years, and detailed administrative expenses for comprehensive medical insurance in each state and year. Therefore, using information on insurers' financial data to compute the average price–cost margin becomes attractive from a regulatory perspective.

For health insurance products, price is the premium per member-year. The marginal cost (mc) includes expected claims (claims) and administrative expenses (adm) per member-year. The price–cost margin of an insurer, ignoring for now the subscript f, can be written as

display math(3)

The MLR is the ratio of claims to premiums, so we can express (3) as

display math

For a higher MLR to imply a lower price–cost margin, the MLR needs to be monotonically non-decreasing in math formula. As a counter example, after premiums are set for a given plan year, an insurer could reduce its utilization management effort, leading to a decrease in administrative expenses. This effort would result in an increase in claims and thus the MLR, but the decrease in math formula could offset the increase in MLR in the price–cost margin.

If the price–cost margin could directly be associated with market power, the components of the price–cost margin, such as the MLR and math formula, could be used as attractive targets for regulation. However, there are caveats regarding the desirability of measuring market power using the price–cost margin, which we mention in the introduction and echo in the conclusion, such as product heterogeneity and measurement issues.

Insights from oligopoly theory suggest that a reasonable approach might compare price–cost margins in markets that look more competitive to those that do not in order to assess the extent that price–cost margins may relate to market power. Our empirical test of whether the MLR as a regulatory target is closely related to market power thus investigates whether firms in more concentrated markets have lower MLRs.

Unlike pure premium or price regulation, MLR regulation monitors the ratio of claims to premiums. An insurer with an MLR below the minimum threshold could either adjust premiums downward or allow claims to rise (e.g., reducing utilization management). Also, as argued earlier, even with MLR adjustments, a firm could achieve the same price–cost margin by changing math formula. Because the regulation's overall impact depends on an insurer's response regarding claims and premiums, plus other components of the price–cost margin (e.g., administrative expenses), there is value in analyzing not only the ratio itself but also these components.

When analyzing the MLR and price–cost margins, we consider a large set of insurer and market characteristics that would be related to the MLR, its components (premiums and claims), and administrative expenses. These characteristics allow for capturing firm-level heterogeneity and other market characteristics that could influence the MLR and administrative expenses, and thus result in price–cost differentials not directly related to insurer market power. For example, insurer size as well as presence in other health insurance markets could influence premiums, claims, and administrative expenses because of economies of scale and scope. Similarly, smaller and younger firms may experience lower MLRs because of higher administrative expenses associated with establishing their supply chain (e.g., establishing or leasing a provider network) and developing their marketing/distribution channels. Insurers' affiliation with large groups (e.g., Aetna or Cigna), as well as their ownership (non-profit or not) and business types (HMO or not), could also be related to their claims, premiums, and administrative expenses.

Our primary focus is the competitiveness of the individual insurance market as characterized by the counts of credible health insurers, non-credible health insurers, and life insurers selling health insurance in this market. We hypothesize that decreases in insurance market competition could lead to increases in market power and in the ability of insurers to demand higher premiums. However, increased insurance market concentration can also put insurers in stronger negotiating positions with hospitals and physicians, leading to lower reimbursement rates, lower claims costs, and potentially lower premiums. Therefore, it is important to control for provider market attributes, including hospital market concentration and physician supply. In particular, we hypothesize that higher hospital market concentration will diminish insurers' ability to negotiate lower payment rates. Thus, we expect that this factor should lead to higher claims costs (assuming there is not a large response in quantity demanded) and potentially higher premiums. Because MLRs reflect the ratio of claims to premiums, it is not possible to sign the predicted overall effect.

Both insurance market structure and provider market attributes are subject to endogeneity concerns because unobserved market-level demand and cost shocks could influence the entry/exit of both insurers and providers while also influencing premiums, claims, and administrative expenses. We discuss these concerns and our approaches in the next section.

3 EMPIRICAL STRATEGY

3.1 Data sources

Our primary data source is the NAIC Health InfoPro database for years 2001 to 2009. The NAIC is the organization of insurance regulators from the 50 states, the District of Columbia, and the five U.S. territories. These data are used by the insurance industry to determine market share, conduct research, and monitor trends; they have also been used in prior research (Abraham and Karaca-Mandic, 2011). We use a compilation of health insurer filings of Annual Statements (also known as Health Blanks) to the Insurance Department for the individual market. To extract information on their business for each state, we rely on the Exhibit of Premiums, Enrollment and Utilization (also known as the Health State Pages) where insurers report enrollment, premiums, and claims by line of business and state. A major limitation of the data is that the vast majority of insurers operating within California are regulated by the California Department of Managed Health Care and do not file with the NAIC. Second, while most insurers that write comprehensive medical insurance file with the NAIC as accident and health insurers, life insurers do not file Health Blanks, although they account for about 20% of individual market premiums (Abraham and Karaca-Mandic, 2011; Abraham et al., 2013).

We note that the omission of these life insurers is an important limitation in the construction of relevant market-share variables and thus market structure. For example, even if only one health insurer appears in the NAIC data for a given state and year, there may be other types of insurers selling health insurance. To construct measures that capture life insurer presence in the individual market, we also use the NAIC Life/Accident & Health InfoPro database, which includes life insurer filings for aggregate premiums earned from accident and health policies sold to individuals in every state in which that insurer sells its product (these filings are known as Life State Pages).

We supplement the NAIC data with several other sources of relevant state and year level information to capture other market characteristics, including the Area Resource File, the American Hospital Association Annual Survey, the Medical Expenditure Panel Survey—Insurance Component, the Behavioral Risk Factor Surveillance System, the Current Population Survey, the Bureau of Economic Analysis, the National Conference of State Legislatures, the Kaiser Family Foundation State Health Facts, and the U.S. Bureau of the Census.

3.2 Sample selection

The unit of analysis is an insurance company-state-year. We imposed several sample selection criteria. First, we restricted attention to only those companies with active operations and complete information on claims incurred and premiums earned during the calendar year. Second, we removed from consideration any observations corresponding to insurer operations in California, given their incompleteness. Third, we excluded insurers in Washington D.C., as the market definition for insurers and providers may be broader than district political boundaries. Fourth, we followed a two-step process to identify and eliminate erroneous data based on outliers starting from our raw data. We flagged insurer-state-year observations that had both claims per member month (CPMM) and premiums per member month (PPMM) in the bottom 1% or top 1% of the overall distribution. Next, we flagged insurers with MLRs in the top 1% or bottom 1%. These two steps resulted in 76 insurer-state-year observations that we identified as potentially erroneous. In the 2011 U.S. Department of Health and Human Services' interim final rule, insurers with experience of less than 1000 member-years in a state are deemed to have ‘non-credible’ MLRs for the purpose of regulatory enforcement. Sixty-seven of the potentially erroneous observations were ‘non-credible’. We then closely examined the nine remaining potential candidates that seemed to have erroneous reporting issues for a single year of the data, excluding them accordingly. Finally, we analyzed the outcomes for ‘credible’ company-state-year observations for the purposes of regulatory enforcement. Imposing this last restriction led to a final analytic sample of 1335 insurer-year observations.

3.3 Empirical specification

Our principal outcome measures are the MLR, PPMM, CPMM, and administrative expenses expressed as a percentage of premiums. On the NAIC filings, insurers classify general administrative and/or claims adjustment expenses. We consider these two types of administrative expenses separately and measure them as a percentage of premiums, which more closely aligns with the conceptual framework. All dollar amounts are inflated to 2009 dollars.

We estimate multivariate regression models that relate our outcomes to insurer characteristics (Insurer), individual insurance market characteristics (InsMkt), provider market characteristics (ProvMkt), economic characteristics (EconomicChar), the political environment (PoliticalChar), the regulatory environment (RegulatoryChar), population health status (Health), and the demographic composition of the population (PopComp).

display math

γj captures state-fixed effects, λt captures year-fixed effects, and εijt is a random error term. We cluster standard errors at the company level, which is appropriate given that we often observe an insurer multiple times in different years and states. Given that the primary variation of the insurance market structure is at the state level, we also tried clustering standard errors at the state level rather than the insurer level and found that our findings were robust to this alternative specification.

Provider market attributes (Hirschman–Herfindahl Indices (HHIs) for hospitals and physician supply) are potentially endogenous. Unobserved market-level factors can influence provider entry and exit as well as health insurance premiums. We use aggregate state-level measures of the provider market, mitigating this concern under the assumption that entry at the state level is exogenous, although the specific city or area within the state may be selected endogenously. One limitation is that our measures of insurance and provider market structures are at the state level. In general, state geopolitical boundaries may not necessarily reflect economic markets.

Another concern is the appropriate characterization of the insurance market structure. We follow Dafny (2010), characterizing structure by defining variables to capture the number of firms. In general, the number of firms in the market may be endogenous if entry decisions reflect firms' evaluations of the demand and cost conditions of particular markets. Positive demand shocks unobserved to the researcher could both induce entry into the market and increase output prices (i.e., premiums), resulting in a positive spurious correlation between number of firms and prices. For the model in which premiums per member month is the outcome, we hypothesize an inverse relationship between the number of insurers in the market and the outcome. If the number of firms is endogenous, our estimates will be biased toward zero. In contrast, holding demand conditions constant, a market characterized with inefficient incumbent firms could attract new entry by more efficient firms, leading to a decline in output prices and a bias in estimates away from zero. Thus, the overall sign of the bias is ambiguous in the premium model.

The number of firms in the market may also be correlated with input prices (claims) due to unobservables. Were the market to have favorable cost conditions, entry would be attractive, and input prices would be low. Similarly, if markets with inefficient firms attract new entry by firms with lower costs as described earlier, input prices after entry would decrease. Therefore, for models with claims as the outcome variable, the estimated relationship between the number of insurers and claims is likely attenuated.

We address the endogeneity concern using multiple approaches. First, we incorporate state-fixed effects into the estimation to account for unobservable demand and cost conditions that are time-invariant across markets. We also incorporate year-fixed effects to capture factors that vary over time that are common to all markets. Second, we attempt to directly control for cost and demand shocks by using information on the number of non-credible firms (insurers with experience of less than 1000 member-years in a state-year). The numbers of such small firms in the market can proxy for similar demand and cost conditions that could influence the entry of larger, credible firms. Similarly, we control for the number of life insurers that sell accident and health insurance products. Third, we control for a rich set of state-year level characteristics that reflect factors that could influence market demand and cost. These factors include the economic characteristics, political environment (which could influence actions of state insurance commissioners), regulatory environment (such as presence of state high risk pools), population health status (which could influence claims), and the demographic composition of the population, a rich set of market-level demographic, health, and economic conditions, and include a rich set of company-specific characteristics to account for firm heterogeneity—factors that may also influence insurers' entry and exit decisions. For each approach, we conduct sensitivity tests in which we assess the extent of bias by removing the observed demand and cost related factors from the model that influence the outcomes we study.

An ideal approach for addressing endogeneity requires a source of exogenous variation correlated with insurer entry but uncorrelated with factors affecting premiums and claims—an inherently difficult task. Recent research on the employer-sponsored insurance market utilizes a large national merger, simulating the merger's impact across geographic regions (Dafny et al., 2012). Because this merger was national, the study assumes its effects on each local region to be exogenous.

The simulation of a one-time large national merger on different markets over time requires narrowing the study period around the merger. Nevertheless, we take this approach in one of our sensitivity analyses by exploiting variation in the market structure resulting from Anthem's acquisition of Well Point (announced 10/27/2003; effective 2/10/2004), the largest national merger during this period with $16.4 billion in transaction value (American Medical News, 2004). The combined company, WellPoint, became the largest seller of health insurance in the USA (USA Today, 2004).

3.4 Measures

Annually, health insurers file the NAIC Exhibit of Premiums, Enrollment and Utilization for their business, by line of business and state. The data we use cover fully insured plans for comprehensive (hospital and medical) coverage. For each company-state-year observation, we extracted for the individual market the following: (i) member months of enrollment; (ii) health premiums earned; and (iii) claims incurred for provision of healthcare services. From the Analysis of Operations filings, we extracted the change in contract reserves, which represents the change in financial reserves held by a company for paying the claims it expects to incur under a contract after the valuation. We then computed the MLR as the ratio of incurred claims plus the change in contract reserves to earned premiums at the company-state observation and multiplied by 100 to convert it to a percentage.

Also from the Analysis of Operations filings, we obtained information on the general administrative expenses and claims adjustment expenses of each insurer by year for comprehensive medical insurance sold for the individual and group markets combined. Unfortunately, the NAIC data only permitted identification of these two types of expenses separately for the individual and group markets starting in 2010. Therefore, we used member-years of each insurer reported by state, year, and individual versus group market to apportion administrative expenses to the insurer's individual market for each state-year and expressed them as a percentage of premiums. This approach assumes that per premium administrative costs are the same in the individual and group markets. If, however, they are actually smaller in the group market, we would underestimate the administrative expenses per premium dollar in the individual market. To account for this limitation, our specifications control for the insurer's share in the group market.

Company characteristics (Insurer) include the following: the number of states in which the company sells health insurance in the individual market (operating only in 1 state, in 2–10 states, or in 11 or more states); indicators for operating in the group, supplemental Medicare, Federal Employees' Health Benefits, Medicare, or Medicaid markets; share of the insurer in the group market; a set of 30 indicators to capture an insurer's association with a top 30 group (Blue Cross, UnitedHealth, Aetna, WellPoint, etc.); ownership type; business type (HMO or not); number of years in business; and member-years of individual market enrollment (logged) in a given state-year.

To capture insurance market characteristics (InsMkt), we included indicators for whether there is only one credible insurer (at least 1000 member-years of enrollment) in the state during that year (including the insurer) or whether there are two to four credible insurers (reference category is the presence of five or more credible insurers). We also considered the presence of ‘non-credible’ insurers for characterizing market structure utilizing the categories of zero to one, two to four, and five or more non-credible firms present in the market. To account for life insurers, we extracted the number of companies selling accident and health policies to individuals in each state-year using the Life State Pages. Unfortunately, the data are less precise because life insurers do not file individual market enrollment and premiums associated with comprehensive medical coverage by state. For example, other policies sold to individuals include short-term medical, limited benefit, and disability insurance. We then characterized these insurers into small, medium, large, and very large categories for each state-year based on the quartiles of annual premiums. All other measures and data sources are described in Table 2.

4 RESULTS AND DISCUSSION

4.1 Descriptive analyses

Table 1 summarizes key insurer characteristics for all credible firms operating in the individual market overall and by insurance market structure. Average premiums per member month are $243 (SD $147), while the average claims incurred per member month are $206 (SD $138). Average premiums in the NAIC data are close in value to those in the AHIP Center for Policy and Research Individual Health Insurance 2009 report, which indicates average monthly premiums for single coverage to be $249 for the non-elderly population (AHIP, 2011). The average MLR is 84%, but considerable variation exists (SD 21%). Insurers that are the only credible firm in the market, on average, have higher enrollment and lower MLRs (78.82% relative to 80.32% for those in markets with 2–4 credible insurers and 88.29% for those with 5 or more credible insurers). Average administrative expenses per member month are $26 (SD $20), and average claims adjustment expenses per member month are $10 (SD $7). On average, expenses for general administration and claims adjustment account for 13% and 5% of premiums, respectively.

Table 1. Summary statistics of insurer characteristics, credible insurers only, by insurance market structure: 2001–2009
 All credible insurersInsurers that are the only credible firm in the marketInsurers in markets with 2–4 credible insurersInsurers in markets with 5+ credible insurers
MeanSDMeanSDMeanSDMeanSD
  1. FEHB, Federal Employees' Health Benefits. Credible insurers have at least 1,000 member-years of enrollment in the market. Market represents a state-year.

  2. Data source: National Association of Insurance Commissioners, by permission. The NAIC does not endorse any analysis or conclusions based upon the use of its data.

Number of insurer-year observations1335 147 477 711 
Unique number of insurers222 36 117 144 
Member-years33,94956,93346,07562,72236,94357,50929,43454,836
Premiums earned per member month ($)242.83146.56219.5173.84215.4796.34266.01178.5
Claims incurred per member month ($)206.21138.11172.5368.76174.3689.24234.54166.76
Medical loss ratio (MLR)84.4021.0378.8214.980.3216.8988.2923.69
General administrative expenses per member month ($)25.8519.9022.798.0824.679.9327.2725.68
Claims adjustment expenses per member month ($)9.586.4911.376.169.596.799.216.29
Percentage of premiums spent on general administrative expenses13.0511.3511.185.0913.238.2213.3313.8
Percentage of premiums spent on claims adjustment expenses4.784.155.42.895.094.874.453.8
Other market segments the insurer operates in (1/0)        
Group market0.920.270.980.140.950.230.890.32
Medicare supplement market0.480.500.930.250.480.50.390.49
FEHB market0.460.500.80.40.520.50.350.48
Title 18 Medicare market0.480.500.280.450.480.50.510.5
Title 19 Medicaid market0.240.430.140.340.160.370.310.46
Number of states the insurer operates in2.926.742.56.123.37.322.756.44
Operates in 1 state (1/0)0.670.470.760.430.640.480.680.47
Operates in 2–10 states (1/0)0.280.450.210.410.310.460.280.45
Operates in 11 plus states (1/0)0.050.210.030.180.050.230.040.2
Number of years in business32.4522.4443.2724.530.5521.5231.522.02
In business less than 30 years (1/0)0.610.490.390.490.640.480.640.48
In business at least 30–59 years (1/0)0.150.350.170.380.160.370.130.34
In business 60 or more years (1/0)0.240.430.440.50.20.40.230.42
HMO (1/0)0.470.500.070.260.510.50.520.5
Non-profit (1/0)0.350.480.460.50.250.430.410.49
Affiliated with a group (1/0)0.840.370.70.460.860.340.850.36

Table 2 describes the structure of insurance markets and other key market characteristics. Approximately 11% of observations have only 1 credible insurer, 36% have 2 to 4 credible insurers, and 53% have 5 or more credible insurers. About 27% observations have 0–1 non-credible insurers, 46% have 2–4 non-credible insurers, and 27% have 5 or more.

Table 2. Market characteristics: 2001–2009 (N = 1335)
 MeanSD
  1. Credible insurers have at least 1,000 member-years of enrollment in the market. Life insurers were characterized into small, medium, large, and very large categories for each state-year based on the quartiles of their annual premiums

  2. 1. For the hospital Hirschman–Herfindahl Index (HHI) measure, we first constructed hospital market HHIs for all Metropolitan State Areas (MSAs) based on adjusted admissions, which capture all patient care activity (inpatient and outpatient) at a hospital. Next, we took a population-weighted average of these MSA-level HHIs to aggregate them to the state-year level. Data from the Area Resource File were used to construct measures of primary and specialty physician to population ratios by state and year.

  3. 2. We used several sources to capture economic characteristics of state-year observations. Using the Medical Expenditure Panel Survey—Insurance Component State and Metro Area Summary Tables, we constructed the percentage of total business establishments in the state that are small (<100 workers) and the percentage that are very large (1000 or more workers). We obtained data on unemployment rate from the U.S. Bureau of Labor Statistics; and personal income per-capita from the Bureau of Economic Analysis.

  4. 3. Measures for the political environment were constructed using data from the National Conference of State Legislatures. We constructed an indicator for whether the Democratic Party had a majority in both the Senate and the House. We also included an additional indicator for whether the Governor had Democratic Party affiliation.

  5. 4. Measures for the regulatory environment (state high risk pools) were constructed using the Kaiser Family Foundation State Health Facts. We also considered controls for rate restrictions (e.g., rate bands and community rating) and state-based medical loss ratio regulations. However, for the study period, we found no variation over time within states.

  6. 5. To capture population health attributes, we used the Behavioral Risk Factor Surveillance System and controlled for rates of disease prevalence for asthma, hypertension, diabetes, and obesity. We also defined a set of measures to capture lifestyle choices, such as smoking prevalence, excessive alcohol consumption, and regular physical activity.

  7. 6. Information on population demographic composition measures including total population, age, sex, race, and ethnicity was extracted from the U.S. Census.

Individual insurance market characteristics (state-year)  
1 credible insurer in the market (1/0)0.110.31
2–4 credible insurers in the market (1/0)0.360.48
0–1 non-credible insurer in the market (1/0)0.270.44
2–4 non-credible insurers in the market (1/0)0.460.50
Number of small life insurers55.456.37
Number of medium life insurers56.697.17
Number of large life insurers55.879.63
Number of very large life insurers65.3926.49
Provider market characteristics (state-year)  
Hospital HHI (population-weighted at the MSA)26711515
% of state pop outside MSA9.4410.91
Total office-based primary care physicians in state/100K28.129.62
Total office-based specialist physicians in state/100K68.1418.38
Economic/political/regulatory characteristics (state-year)  
% establishments under 10079.213.52
% establishments over 100013.893.12
Unemployment rate (%)5.591.74
Personal income per capita (in $1000)39.525.33
Democrat control of State and House (1/0)0.320.47
Democratic Party affiliation of the Governor (1/0)0.520.50
Presence of high risk pool (1/0)0.570.49
Health (state-year)  
% adults ever been told have asthma12.821.53
% adults ever been told they have diabetes7.581.40
% adult males 2+ drinks/day; adult females 1+ drink/day5.321.13
% adults reporting current adult smokers20.313.21
% of pop with BMI 30 plus24.193.38
Population composition (state-year)  
% Hispanic10.688.62
% White81.7510.08
% Black11.388.41
% Asian3.684.47
% female50.780.70
% ages 45–65 years25.191.77

4.2 Variation in insurance market structure

Given that our models include state-fixed and year-fixed effects, identification of the insurance market structure effect relies on observing variation in the number of credible insurers within states over time. We identified 32 states with only one credible health insurer (monopoly market) during at least one year in 2001–2009. Among them, eight were a monopoly market throughout the entire study period, 20 experienced several years as a monopoly market followed by entry, and the remaining four states experienced both entry and exit. From sources other than the NAIC data, such as the websites of the Center for Consumer Information & Insurance Oversight and the Office of the Health Insurance Commissioner of several states, we also confirmed the accuracy of the definition of a health insurance monopoly as an insurer with only one credible health insurer market. While these sources may list more than one insurer serving the individual market even for those markets we characterize as monopoly markets, we confirmed that such additional insurers filed as life insurers (online Appendix 1).

We also investigated whether the health insurance market structure was associated with particular trends in key state-year covariates in order to examine whether markets with fewer insurers had differential trends in key state-year level characteristics. We estimated a linear regression model of the number of credible insurers in the state-year market on lagged values of provider market characteristics as well as on lagged values of all the economic, political, regulatory, and population composition characteristics discussed earlier. These characteristics were not statistically significant predictors of the number of credible health insurers, with the exception of logged population.

4.3 Multivariate analyses

4.3.1 Premiums, claims, and medical loss ratio

We used generalized linear models (GLMs) to estimate our multivariate specification for three principal outcomes: PPMM, CPMM, and MLR. Following Deb, Manning, and Norton (2010), we let the data determine the scale of estimation and the distributional family. The Box-Cox tests suggested a log scale for the PPMM and CPMM, and either a log scale or a square root scale for the MLR as the most appropriate scale of estimation. A modified Park test suggested that we use inverse Gaussian family for PPMM and gamma family for CPMM. Finally for the MLR, the inverse Gaussian family was found most appropriate, using either a log scale or a square root scale (detailed estimates are noted under Table 3).

Table 3. Generalized linear models of premiums, claims, and medical loss ratios
 Premiums per member monthClaims Per Member MonthMedical Loss Ratio
LinkLog Log Log 
FamilyInverse Gaussian Gamma Inverse Gaussian 
 Coeff.SECoeff.SECoeff.SE
  • Credible insurers have at least 1,000 member-years of enrollment in the market. Life insurers were characterized into small, medium, large, and very large categories for each state-year based on the quartiles of their annual premiums.

  • FEHB, Federal Employees' Health Benefits; HHI, Hirschman–Herfindahl Index; MSA, Metropolitan State Area.

  • To determine the appropriate scale of estimation, we used the Box-Cox test. We then used a modified Park test to identify the distributional family. As discussed in Deb, Manning and Norton (2010), Box-Cox test finds the maximum likelihood estimate of λ such that math formula. If math formula, the appropriate functional form is ln(y) =  + ε; if math formula, the appropriate functional form is math formula, and if math formula, the appropriate functional form is y =  + ε. We estimated math formula (SE 0.028) for premiums per member month (PPMM), math formula (SE 0.029) for claims per member month (CPMM), and math formula (SE 0.04) for medical loss ratio (MLR). After specifying the appropriate scale of estimation, the generalized linear model family test determines the relationship between E[y\x] and Var[y\x] by estimating the parameters ∝ and δ for the specification Var[y\x] = ∝ E[y\x]δ. An estimate of δ = 0 suggests Gaussian family, δ = 1 Poisson family, δ = 2  gamma family, and δ = 3 inverse Gaussian family. We estimated δ = 2.83 (SE 0.32) and δ = 2.40 (SE 0.27) for PPMM and CPMM, respectively. For MLR, we estimated δ = 3.37 (SE 0.64) when assuming the scale of estimation to be log scale and δ = 4.78 (SE 0.91) when assuming the square root scale. These findings suggested that we use inverse Gaussian family for PPMM and gamma family for CPMM. Finally, for the MLR, the inverse Gaussian family was the most appropriate, either using the log scale or the square root scale.

  • ***

    p < 0.01,

  • **

    p < 0.05,

  • *

    p < 0.1.

Insurer characteristics      
Insurer's share in the group market0.3940.3070.457***0.1590.093**0.045
Other market segments the insurer operates in (1/0)      
Medicare supplement market0.336**0.1700.266***0.093−0.0090.026
FEHB market−0.0350.071−0.0220.0480.0200.023
Title 18 Medicare market0.0790.1340.0780.050−0.0040.015
Title 19 Medicaid market−0.233***0.090−0.148**0.0670.0160.018
Number of states the insurer operates in      
Operates in 1 state−0.0140.1300.0600.1850.0410.035
Operates in 2–10 states−0.1210.151−0.0210.1700.0370.036
Number of years in business      
In business at least 30–59 years (1/0)−0.0130.0590.0780.0590.0240.020
In business 60 or more years (1/0)−0.0830.1010.122*0.0710.050**0.023
HMO (1/0)0.272**0.1010.325***0.1050.097***0.030
Non-profit (1/0)−0.0900.162−0.152**0.0760.0010.024
HMO * Non-profit0.4340.2690.307***0.112−0.0160.033
Log of member-years (in 10,000)−0.262***0.067−0.183***0.043−0.0130.008
Individual insurance market characteristics (state-year)      
1 credible insurer in the market−0.1280.079−0.151**0.065−0.078**0.034
2–4 credible insurers in the market−0.0520.041−0.0270.037−0.0190.016
0–1 non-credible insurer in the market−0.0390.050−0.066*0.040−0.0260.024
2–4 non-credible insurers in the market−0.0160.039−0.0550.034−0.0140.020
Number of small life insurers0.0000.0120.0050.0030.0020.001
Number of medium life insurers0.0040.0100.008***0.0030.0020.002
Number of large life insurers0.0010.0140.006*0.0040.004*0.002
Number of very large life insurers0.0080.0170.016***0.0050.007**0.003
Provider market characteristics (state-year)      
Hospital HHI (population-weighted at the MSA)−0.4810.464−0.0680.4020.2630.180
% of state pop outside MSA−0.0000.002−0.0010.001−0.0010.001
Total office-based primary care physicians in state/100K0.0160.0110.0060.008−0.0050.004
Total office-based specialist physicians in state/100K−0.0060.013−0.0110.0070.0010.004
Economic/political/regulatory characteristics (state-year)      
% establishments under 1000.0080.0280.0010.009−0.0000.005
% establishments over 10000.0010.0420.0010.0110.0010.006
Unemployment rate−0.033*0.019−0.048**0.021−0.0060.009
Log of personal income per capita (in $1000)−0.3310.865−0.2720.5330.1490.318
Democrat control of State and House (1/0)−0.087*0.048−0.0200.0460.033*0.020
Democratic Party affiliation of the Governor (1/0)0.0400.1100.0240.0350.0070.017
Presence of high risk pool (1/0)0.1520.1080.126**0.060−0.0190.026
Health (state-year)      
% adults ever been told have asthma0.0060.030−0.0110.012−0.0090.006
% adults ever been told they have diabetes0.0200.0310.0220.019−0.0030.008
% adult males 2+ drinks/day; adult females 1+ drink/day0.0000.0290.0110.0180.0080.009
% adults reporting current adult smokers0.0020.0400.0130.0090.0040.005
% of pop with BMI 30 plus−0.033*0.019−0.031***0.012−0.0010.005
Population composition (state-year)      
log of total population−2.303*1.375−2.255*1.3000.0680.444
% Hispanic0.173**0.0860.166**0.0680.043*0.024
% White−0.0120.5120.3570.2600.1600.110
% Black0.1690.4930.493*0.2800.0700.098
% Asian−0.2550.8350.2900.3690.1750.178
% female0.0470.291−0.0860.273−0.1590.116
% ages 45–65 years0.0240.1720.0310.0860.0350.038
Other variables included in all the models      
Group-fixed effects      
State-fixed effects      
Year-fixed effects      

Table 3 presents our results for PPMM, CPMM, and MLR. Most explanatory variables move PPMM and CPMM in the same direction. Insurers that also operate in the Medicare supplemental insurance market have higher PPMM and CPMM. In contrast, insurers that operate in the Medicaid market have lower PPMM and CPMM, which could reflect a more impoverished market environment. Insurers with a larger share in the group market have higher CPMM, which likely reflects product heterogeneity between firms that are dominant players in the group market (and that are thus likely to target consumers with higher demand for more generous policies) versus those that are more focused in the individual market.

HMOs have higher PPMM and CPMM, likely reflecting more generous benefit designs that may be attractive to individuals with a higher demand for medical services than average. While non-profit status alone does not exhibit an independent association with PPMM, it is negatively associated with CPMM. The coefficient estimate on the interaction of HMO and non-profit status is positive and statistically significant, suggesting that non-profit HMOs have higher CPMM relative to non-profit insurers that are not HMOs. The estimated cross-partial derivative of HMO and non-profit status confirms this conjecture. Finally, firms with larger member-years in the individual market have lower PPMM and CPMM, which may be due to economies of scale. Insurer size is potentially endogenous, and this association may also reflect unobserved factors that are correlated with member-years, premiums, and claims. For example, insurers with more efficient administration could have larger member-years and at the same time lower claims and costs.

Most of the other market characteristics do not exhibit significant relationships with PPMM or CPMM. However, when insurers are the sole credible health insurer in the market, they have lower CPMM on average. This likely reflects their strength to negotiate lower provider payments. Credible insurers in markets with a stronger presence of medium, large, and very large life insurers have higher CPMM. Overall, having a larger number of firms with high bargaining power with providers could imply decreased ability to negotiate lower fees, resulting in higher CPMM.

Relatively few insurer characteristics are associated with the MLR. Exceptions include the firm's market share in the group market for the given state and year, business tenure, and HMO status. Insurers with a higher share in the group market also have higher MLRs, suggesting that they may be leveraging economies of scale and scope. Older firms with business tenure longer than 60 years and HMOs have higher MLRs on average. One possible explanation is that older insurers tend to offer more ‘traditional’ benefit designs, which tend to have higher MLRs relative to high deductible plans (Bernstein, 2010).

The key market characteristic associated with an insurer's MLR involves insurance market structure. Insurers in markets in which they are the only credible insurer have lower MLRs, suggesting that such firms may have higher market power. While the predicted average MLR is 77% for an insurer that is the only credible firm in the insurance market, it is 82% for an insurer with 2–4 other credible firms in the market (the p-value for the difference is 0.04) and 83% for an insurer with 5 or more credible firms in the market (the p-value for the difference from a market with one credible insurer is 0.02). This finding is consistent with an earlier study by Dranove et al. (2003), who investigate competition in HMO markets and find that within HMO types (national vs. local HMOs), each entrant reduces the incumbent's profits at a decreasing rate with a larger number of incumbent firms. Similarly, Abraham et al. (2007) and Brasure et al. (1999) have broadly consistent findings for other healthcare goods and services. These findings are also similar to Breshanan and Reiss (1991), who provide evidence that the entry of second and third firms has the largest competitive impact rather than a gradual effect increasing with the number of entrants.

4.3.2 Investigating administrative expenses

Recall from the conceptual framework that the MLR is only one component of the price–cost margin. A decrease in MLR therefore does not necessarily imply an increase in price–cost margins. We investigate insurers' general administrative expenses and claims adjustment expenses (expressed as a percentage of premiums) by means of the same explanatory variables used in the MLR specification and report the results in Table 4. We estimate our models using GLM with a log scale for both outcomes, an inverse Gaussian family for general administrative expenses, and a gamma family for claims adjustment expenses as suggested by the Box-Cox and modified Park tests. We find no evidence that being the only credible insurer is associated with having higher general administrative expenses and/or claims adjustment expenses. In combination with the MLR results, this suggests that firms that are the only credible insurer in their market have lower MLRs and higher price–cost margins.

Table 4. Generalized linear models of claims adjustment and general administrative expenses per premiums
 Percent of premiums spent on general admin expensesPercent of premiums spent on claims adjustment expenses
Scale of estimationLog Log 
Family of distributionInverse Gaussian Gamma 
 Coeff.SECoeff.SE
  • Credible insurers have at least 1,000 member-years of enrollment in the market. Life insurers were characterized into small, medium, large, and very large categories for each state-year based on the quartiles of their annual premiums.

  • FEHB, Federal Employees' Health Benefits; HHI, Hirschman–Herfindahl Index; MSA, Metropolitan State Area.

  • To determine the appropriate scale of estimation, we used the Box-Cox test. We then used a modified Park test to identify the distributional family as detailed under Table 3. We estimated math formula (SE 0.022) for percent of premiums spent on administrative expenses and math formula (SE 0.021) for percent of premiums spent on claims adjustment expenses. Using the generalized linear model family test, we estimated δ = 2.93 (SE 0.35) and δ = 1.86 (SE 0.28) for the two outcomes, respectively. These findings suggested that we use log scale for both outcomes, inverse Gaussian family for percent of premiums spent on administrative expenses, and gamma scale for percent of premiums spent on claims adjustment expenses.

  • ***

    p < 0.01,

  • **

    p < 0.05,

  • *

    p < 0.1.

Insurer characteristics    
Insurer's share in the group market−0.2450.206−0.540**0.221
Other market segments the insurer operates in (1/0)    
Medicare supplement market−0.237**0.1200.0130.110
FEHB market−0.1470.099−0.159*0.085
Title 18 Medicare market−0.2350.217−0.152*0.079
Title 19 Medicaid market0.0590.0090.0760.089
Number of states the insurer operates in    
Operates in 1 state−0.4120.3210.0230.167
Operates in 2–10 states−0.3730.3640.0900.176
Number of years in business    
In business at least 30–59 years (1/0)−0.3800.2960.0060.112
In business 60 or more years (1/0)−0.591***0.195−0.275*0.143
HMO (1/0)−0.498***0.139−0.419***0.105
Non-profit (1/0)0.1260.2270.339***0.122
HMO * Non-profit−0.548**0.269−0.2760.217
Log of member-years (in 10,000)0.0460.0400.093***0.031
Individual insurance market characteristics (state-year)    
1 credible insurer in the market0.0020.1330.0710.126
2–4 credible insurers in the market0.0050.071−0.0200.076
0–1 non-credible insurer in the market0.1380.1280.1020.101
2–4 non-credible insurers in the market0.0310.1060.197**0.095
Number of small life insurers0.0030.005−0.0010.005
Number of medium life insurers0.0090.0060.0040.006
Number of large life insurers0.017*0.0100.0030.010
Number of very large life insurers0.0160.0120.0160.012
Provider market characteristics (state-year)    
Hospital HHI (population-weighted at the MSA)0.3820.7940.5180.745
% of state pop outside MSA0.0060.0050.0030.004
Total office-based primary care physicians in state/100K−0.0100.019−0.0070.016
Total office-based specialist physicians in state/100K−0.0070.0130.0010.013
Economic/political/regulatory characteristics (state-year)    
% establishments under 100−0.0200.018−0.0160.016
% establishments over 1000−0.0160.020−0.0060.018
Unemployment rate−0.0450.037−0.0120.035
Log of personal income per capita (in $1000)0.8141.446−0.8311.393
Democrat control of State and House (1/0)−0.0520.0710.0170.075
Democratic Party affiliation of the Governor (1/0)0.0550.0560.105*0.057
Presence of high risk pool (1/0)0.2220.193−0.1810.176
Health (state-year)    
% adults ever been told have asthma0.0060.028−0.0200.029
% adults ever been told they have diabetes0.0540.0390.0150.041
% adult males 2+ drinks/day; adult females 1+ drink/day0.0000.032−0.0110.031
% adults reporting current adult smokers0.0020.0200.0120.021
% of pop with BMI 30 plus0.0020.0200.0120.021
Population composition (state-year)    
log of total population−2.3542.854−2.9082.357
% Hispanic0.0720.128−0.0380.146
% White−0.0980.640−0.3170.465
% Black0.0360.544−0.1990.476
% Asian0.1760.833−0.4300.675
% female0.6730.5810.900**0.410
% ages 45–65 years−0.0780.197−0.343*0.194
Other variables included in all the models    
Group-fixed effects    
State-fixed effects    
Year-fixed effects    

4.4 Sensitivity analyses

We conducted a series of analyses to check the sensitivity of our main result—that insurers have significantly lower MLRs on average when they are the only credible insurer in a market. We describe each analysis. Table 5 reports a summary of our sensitivity analyses for the MLR specification.

Table 5. Summary of sensitivity analyses from generalized linear models of medical loss ratio (MLR; log scale, inverse Gaussian family)
 Mean95% CIMean95% CIMean95% CIDifference in estimates (p-values)
 Insurers that are the only credible firm in the marketInsurers in markets with 2–4 credible insurersInsurers in markets with 5+ credible insurersB − AC − A
Predicted estimates of MLR (%)(A)(B)(C)  
  1. The sensitivity test for excluding firms with erratic reporting pattern had a sample size of 1230 observations. The sensitivity test that excluded states with major enrollment changes had a sample size of 997.

Baseline (from Table 3)77(73.4–81.0)82(80.2–83.2)83(81.5–85.1)0.040.02
Cluster standard errors at the state level77(73.5–80.6)82(80.3–83.2)83(81.7–84.8)0.020.01
Exclude firms with erratic reporting pattern75(71.6–79.2)80(78.8–81.7)83(81–84.8)0.020.005
Exclude states with major enrollment changes during the study period76(72.1–79.6)82(80.7–83.8)86(83.6–87.9)0.003<0.001
Exclude controls for non-credible insurers in the market77(73.4–81.2)82(80.2–83.4)83(81.4–84.9)0.040.03
Exclude controls for non-credible insurers and life insurers in the market77(73.4–81.3)82(80.5–83.6)83(81.2–84.7)0.040.04
Exclude controls for non-credible insurers and life insurers in the market, demographic/population/health characteristics of the market76(71.4–81.4)82(80.2–83.2)84(81.8–85.4)0.0770.03
Exclude state and year-fixed effects76(71–80.4)81(78.3–82.9)85(82.3–87.1)0.060.004
Interact year-fixed effects with insurer characteristics77(74–80.7)82(80.6–83.4)83(81.3–84.7)0.0080.02
Interact year-fixed effects with indicators of 5 largest groups78(74.8–80.8)82(80.2–83)83(81.5–84.6)0.020.01
Interact state-fixed effects with indicators of 5 largest groups78(74–81.2)82(80.3–83.2)83(81.5–84.6)0.050.03

4.4.1 Insurers' reporting behavior over time

First, we directly examined insurer reporting patterns, identifying insurers that exhibited inconsistent reporting behavior (i.e., an insurer reports in a given state in year t, not in t + 1, and reports again in t + 2). There were 17 such cases. Re-estimating our models excluding these observations led to very similar adjusted MLR results.

Next, based on analyses reported in Abraham and Karaca-Mandic (2011), we excluded states that experienced major enrollment changes between 2001 and 2009: AL, CT, FL, IA, ID, IL, IN, MI, OH, SC, TX, and VA. Although this exclusion reduced our sample size to 997 credible company-state-year observations, coefficient estimates were qualitatively and quantitatively similar to the baseline.

Finally, we examined the composition of reporting insurers over time using bivariate analyses. The only systematic difference detected was increased reporting among insurers that operate in 11 or more states (p = 0.004).

4.4.2 Sensitivity to excluding observed market characteristics

Unobserved demand and cost conditions may affect both insurance market structure and MLR outcome, leading to potential endogeneity bias. Our approach relied on including state-fixed and year-fixed effects, counts of non-credible firms and life insurers, and a rich set of economic, political, regulatory, health, and population characteristics of the market in the models. While this is not a formal test of endogeneity, it lets us examine insurance demand based on observed characteristics and serves as a guide for assessing the extent of unobservable characteristics affecting demand (Altonji et al., 2005).

Table 5 presents a specification that excludes counts of non-credible firms and a specification that excludes the counts of non-credible firms and life insurers. Our results remain robust. An additional specification excludes the counts of non-credible insurers and life insurers, as well as the economic/political/regulatory/population/health characteristics of the market. A final model excludes state-fixed and year-fixed effects. The predicted values of the MLR remain robust across insurance market structure. In the last two specifications only, we see an attenuation in the statistical significance of the difference in the predicted MLR of insurers that are the only credible firm relative to those in markets with 2–4 credible firms.

4.4.3 Differential insurer response over time and by State

The year-fixed effects control for unobserved factors that may vary over time and are common to all firms. Yet, it is possible that different types of insurers (e.g., for-profit versus non-profit) may respond differently. To examine this possibility, we conducted three checks. First, we estimated the model interacting year-fixed effects with insurer characteristics (member-years, non-profit status, HMO status, and business tenure older than 60 years). Second, we estimated a version of the model also including interactions of year-fixed effects with indicators for each of the largest five insurer groups. There are 222 unique insurers in the data set and 67 unique insurer groups, although there are on average 6 insurers per state-year; we did this alternative estimation only for the five largest groups. Third, we estimated a model with interactions of state-fixed effects with indicators for each of the largest five insurer groups. Our results were robust in all three checks.

4.4.4 Exploiting exogenous variation in market structure from a national merger

We systematically searched for national mergers during the 2001–2009 period that could affect the individual market.2 We employed several criteria in the selection of which national merger(s) to use. First, we wanted the merger to be national or broadly regional in scope rather than specific to a narrow geographic region. This is important to having the merger be exogenous for any particular state. Second, we wanted the merger to take place relatively early in our study period so as to have sufficient post-merger data to evaluate its impact. Third, we searched for mergers with large transaction values in order to ensure an impact on the individual market. Anthem's acquisition of Well Point, announced 10/27/2003 and effective 2/10/2004, was the largest merger during this period with $16.4 billion in transaction value through which the combined company, WellPoint, became the largest health insurer in the USA (Appleby, 2004).

Similar to Dafny et al. (2012), we used the Anthem–WellPoint merger as providing exogenous variation in insurance market structure. In particular, we conjecture that state-year markets for which WellPoint had a larger market share would attract fewer other health insurers the following year. We estimated our models using the two-stage residual inclusion method (Terza et al., 2008). In our study, the potentially endogenous variable is the market structure: number of credible insurers in a state-year observation. We categorized this measure into three categories: ‘one firm’, ‘2–4 firms’, and ‘5 or more firms’. For each state ‘s’ and year ‘t’, we calculated the WellPoint market share and used it to predict the number of credible firms (1, 2–4, 5, or more) in state ‘s’, year ‘t + 1’ (using ordered logit specification), controlling for other state-year covariates used in all models and discussed earlier. The coefficient estimate on the WellPoint share was negative and statistically significant (p = 0.03), indicating that markets with a strong presence of WellPoint had a lower number of firms.

Next, we predicted the probability of each outcome corresponding to insurance market structure and the residual corresponding to that prediction. Finally, we estimated our specifications for PPMM, CPMM, MLR, Percent of premiums spent on general administrative expenses, and Percent of premiums spent on claims adjustment expenses, including the residuals from the first stage. The model was estimated for a narrower time frame around the merger (2002–2007; 841 observations). As shown in Table 6, the magnitude of the coefficient estimate on one credible insurer was statistically significant only in the MLR specification, suggesting that MLR was lower in markets with one credible firm relative to those with 2–4 credible firms (76.08% vs. 82.58%, p < 0.001). The predicted MLR in markets with 5 or more credible firms was also higher relative to those with one credible firm (81.93%, p = 0.04). These are very similar estimates to our baseline findings.

Table 6. Generalized linear models of premiums, claims, and medical loss ratios using a two-stage residual inclusion model, 2002–2007
 Premiums per member monthClaims per member monthMedical loss ratioPercent of premiums spent on general admin expensesPercent of premiums spent on claims adjustment expenses
Linklog log log log log 
FamilyInverse GaussianGammaInverse GaussianInverse GaussianGamma
 Coeff.SECoeff.SECoeff.SECoeff.SECoeff.SE
  • Credible insurers have at least 1,000 member-years of enrollment in the market.

  • Model includes all other variables in Table 3.

  • ***

    p < 0.01,

  • **

    p < 0.05,

  • *

    p < 0.1.

Individual insurance market characteristics (state-year)          
1 credible insurer in the market−0.0450.087−0.15*0.09−0.074**0.0360.0590.20−0.130.17
2–4 credible insurers in the market−0.0450.06−0.0460.050.0080.0360.110.11−0.140.09

4.4.5 Alternative measure of insurance market structure

An alternative measure of insurance market concentration is insurer HHI. Theoretically, HHI would be a suitable measure to model Cournot competition among homogenous firms, which may not fit the context of the individual insurance market, with firms having non-symmetric cost structures and differentiated products. By using HHI to characterize insurance market structure, our findings are robust (online Appendix 2).

5 POLICY IMPLICATIONS AND CONCLUDING REMARKS

Our key finding is that insurers operating in a market in which they are the only credible insurer have lower MLRs, on average, than those operating in markets with 2–4 credible firms or 5 or more credible firms. Insurers reporting lower MLRs could have lower price–cost margins, but this pattern may also be consistent with higher general administrative or claims adjustment costs. Our analyses of these costs exhibit no relationship to insurance market structure, suggesting that lower MLRs resulting from being the only credible insurer in the market are associated with increased price–cost margins rather than higher administrative expenses.

While the price–cost margin is an attractive target for regulation intended to address market power concerns, several caveats apply as we have mentioned briefly elsewhere. First, although we control for a large set of firm characteristics, we are not able to fully capture product heterogeneity and quality differentiation. Such unobserved heterogeneity could lead to justifiably higher price–cost margins unrelated to market power—for example, if firms offer superior products. A second caveat is that firms may have high price–cost margins as a way to recoup high fixed costs, but this too may be unrelated to market power. There are also measurement issues that complicate the use of the price–cost margin as an accurate measure of market power. The MLR is a measure based on average claims costs per premium, but the true price–cost margin requires knowledge of marginal costs, which are not easily observed. Furthermore, the MLR reflects accounting costs rather than economic costs.

Although it is difficult to anticipate the full range of insurer reactions to the MLR regulations, it is possible that outcomes will also involve strategic as well as unintended responses. For example, despite standardized accounting practices, insurers may vary in terms of how they define and allocate specific expenses. The ACA allows for insurers' expenses for certain quality improvement activities to be classified as clinical benefits and counted similarly to medical claims. This may give insurers incentive to react strategically to the MLR regulations by documenting some of their expenses as ‘qualified’ expenses that may count toward clinical benefits. Similarly, insurers may also have incentives to re-label some administrative costs as quality improvement expenses. Although HHS has responded by providing detailed guidance on reporting requirements or quality improvement expenses, there is likely still some room for reclassification of costs. The ACA also only applied MLR regulations to insurers with under 1000 life-years, and it is unknown if smaller insurers near the threshold may let enrollment fall below this threshold to avoid regulation. An examination of the state small-group health insurance reforms of the 1990s suggested that small firms grew in order to bypass reforms (Kapur et al., 2011).

The MLR regulations may result in reduced cost containment efforts similar to those documented in the broader regulatory economics literature on the use of incentive regulation when industries are natural monopolies. First proposed in a seminal paper by Averch and Johnson and later extended and clarified by others, the basic premise is that regulations on firms' rates of return (profit to capital ratio) may distort input decisions by encouraging the use of more capital to labor than is efficient (Averch and Johnson, 1962; Baumol and Klevorick, 1970). In the MLR context, insurers may have less incentive to pursue efficient utilization management or to negotiate as hard on provider reimbursement if given the chance to do so after premiums are set. It will be important for regulators to understand the specific behavioral responses of insurers to MLR regulation and to assess how these actions intentionally or unintentionally affect consumer well-being.

CONFLICT OF INTEREST

Authors do not have any conflicts of interest to disclose

ACKNOWLEDGEMENTS

We received valuable feedback from seminar participants at the University of Minnesota and conference participants at APPAM, the Midwestern Health Economics Conference, AcademyHealth, IHEA, NBER Health Care Summer Institute, Rice University, University of California, Berkeley, and University of Colorado, Denver. We especially thank Amitabh Chandra, Roger Feldman, Jonathan Gruber, Mark Showalter, James Rebitzer, Vivian Ho, and Richard Boylan for insightful comments.

  1. 1

    Authors' estimates from the 2010 Current Population Survey.

  2. 2

    We thank Leemore Dafny for sharing with us information on mergers collected by her and her co-authors.

Ancillary