Without exception, the empirical research literature investigating the structure of health insurance markets has investigated only the HMO market (Paul and Chollet, 1997). In part because this literature fails to consider the other sectors of the market, and in part because it fails to consider the types of regulation of interest in this paper, it offers no real precedent for this study.

In the next several sections, we propose several models of insurer behavior in regulated markets. In general, these models are consistent with conventional competitive models of supply and also a growing recent literature that links more restrictive state regulation in the small-group market to declines in coverage in that market.⁴ Finally, we extend the logic of the small-group insurer responses to the individual market, estimating essentially the same models for the individual market. We estimate both sets of models by state and by year, using fixed-effects multiple regression analysis.

A. Insurer Response to Regulation

The types of regulation that states have imposed in the small group market, and to a lesser extent in the individual market, reward greater insurer size, especially if insurers already are operating with increasing economies of scale.⁵ In general, one would expect that regulation which reduces insurers’ ability to deny risk or to rate risk appropriately would enhance rewards to scale by allowing insurers to defray the implicit subsidy to high-cost groups and individuals over a larger base of insured lives. Moreover, it is unlikely that regulation favoring market entry by high-risk groups and individuals (guaranteed issue or renewal, restrictions on preexisting condition exclusions and limits on rate variation for age, health or other factors) expands the overall size of the market (Nichols, 2000). Instead, it is likely that the number of covered lives might decline, given the price elasticity of demand for coverage in these markets (Marquis and Long, 1995), and indeed that is the sense of the emerging literature linking health insurance coverage (usually among workers in small groups) and insurance market regulation. In markets with increasing economies of scale and no net growth in demand for coverage, insurance regulation is likely to increase price; and competition among insurers may force merger or exit to recapture lower average cost and to increase market share.

Number of insurers. We hypothesize that the number of insurers writing coverage will decline in states with regulation that would require insurers to accept risk that they would otherwise deny or to rate risk in ways that subsidize high-cost insureds. This decline would occur either by insurers exiting the market (ceding market share), or by insurers merging with their competitors to increase the surviving insurer’s market share and gain economies of scale.

To test this hypothesis, we estimate a fixed effects model of the following general forms:

(1) NUMINS_st =f (X_st, REG_st)

(2) NUMINS_st =f (X_st, REG_st, BCBS_st)

(3) NUMINS_st =f (X_st, REGDUM_st, BCBS_st)

where NUMINS is the number of insurers in state s and year t, X_st is a vector of control variables, REG_st is a vector of continuous and categorical regulation variables, and REGDUM_st is a vector of regulation variables expressed as categorical variables only (precise variable definitions are provided in Table 8). We tested regulatory variables as both continuous and categorical, the latter to test for a “shock effect” of regulation unrelated to the stringency of the requirement. Also, note that the latter two models control for the Blues’ market share as an exogenous variable; in these specifications we entertain the hypothesis that the Blues’ market share in many states is an artifact of their history and unique position in state regulation, not principally a result of short-term market dynamics.

Market share. We hypothesize that very large insurers, with constant economies of scale, are more likely to thrive in highly regulated markets than small insurers with increasing economies of scale, and that insurer types with characteristically different cost economies systematically gain or lose market share in more highly regulated states. Because different insurer types may have different scale economies,⁶ we first estimate the relationship between states’ regulatory variables and market share for each of the major types of insurers. To reflect the very nonexclusive rules that BCBS plans in some states have adopted to qualify as HMOs in state law, we include BCBS HMOs as BCBS insurers. Specifically, we estimate the following general forms, again as fixed-effect models and again introducing BCBS market share in some specifications as an independent (exogenous) variable:

(4) BCBS_st =f (X_st, REG_st)

(5) HMO_st =f (X_st, REG_st)

(6) HMO_st =f (X_st, REG_st, BCBS_st)

(7) COMM_st =f (X_st, REG_st)

(8) COMM_st =f (X_st, REG_st, BCBS_st)

Market concentration. A significant net exit of insurers, all else being equal, will increase market concentration, reducing competition among insurers. We estimate the impact of state regulation on two conventional measures of market concentration: (a) a Herfindahl index;⁷ and (b) recognizing the problems of the Herfindahl index in describing highly skewed markets, the share of the market held by the largest (arbitrarily, the largest five) competitors in the state.

(9) HERF =f (X_it, REG_it)

(10) TOPFIVE = f (X_it, REG_it)

The price of insurance. Because health insurance products are nonstandard, it is very difficult to measure price in the usual sense. Risk-averse consumers are willing to pay a price in excess of the actuarially fair price (that is, the expected cost of their health care estimated over a class of similar risks), and the price of insurance is measured as the loading that consumers are willing to pay in excess of the actuarially fair price. We express the loading on the insurance policy as its inverse: the insurer’s medical loss ratio (i.e., medical claims paid per premiums earned). In more competitive markets, it is presumed that the price of health insurance is lower – and insurers’ loss ratios are higher.

Insurance regulation complicates several aspects of this conventional model of competitive markets. Most insurance regulation is intended to increase pooling by insurers, forcing them to develop more heterogeneous classes of risk and to cross-subsidize higher-risk members of the class. With more heterogenous risk classes, higher loss ratios may correlate not only with greater competition (measured by the number of insurers and/or market concentration)⁸ but also with the entry of higher-risk groups or individuals into the health plan.

We hypothesize that average loss ratios are higher in states with insurance regulations that attempt to force insurers to develop more heterogeneous risk classes, both accepting greater risk and limiting rate variation. The model controls for the number of insurers in the market and for market concentration (as measures of competition) as well as the usual vector of control variables defined in Table 10. We estimated the model with alternative measures of market concentration as follow:

(11) LRATIO_st =f (X_st, REG_st, NUMINS_st, HERF_st)

(12) LRATIO_st =f (X_st, REG_st, NUMINS_st, TOP5_st)

B. Measures of regulation

In preparing a regulation database from published sources, we discovered that recognizes sources of information about insurance regulation (Institute for Health Policy Solutions, 1999; Health Policy Tracking Service, 1996-1998; and Blue Cross and Blue Shield Association, 1996-99) occasionally disagree. With funding from the Robert Wood Johnson Foundation and the National Association of Insurance Commissioners, we launched a 50-state survey to reconcile these differences.

The survey entailed calling each state department of insurance to obtain the name of the correct person to respond to questions about regulation, sending to each state department of insurance a customized 7-page survey instrument restating where available sources had agreed about their regulation in the small-group and individual health insurance markets. We asked each state to confirm, correct and complete the available information, especially clarifying dates of implementation versus enactment. We then followed up with each respondent in an hour-long telephone interview to review their responses.

The response rate to this effort was excellent (48 states and the District of Columbia responded and were interviewed). In a number of cases we discovered that, while all published sources agreed, they were wrong (for example, all published sources reflected the date of enactment of a specific provision, not implementation). In these cases, we obtained and reviewed the state’s insurance statute for confirmation. For two states, where we were unable to obtain either a survey response or, with reasonable effort, the state’s relevant insurance statute, we used published information to complete their regulatory profile.

From this information, we developed measures of regulation, using categorical variables where necessary (for example, for guaranteed issue or renewal), but developing continuous measures of regulation wherever possible (specifically, for age, health and composite rate bands; and for limits on preexisting condition exclusions). We measure rate regulation as the inverse of the ratio of the maximum allowable rate to the minimum allowable rate in regulation. Thus, we rate regulation values for each state in each year that varies between 1 (the rate factor is prohibited, in effect allowing only 1:1 rating on that factor) and, in the limiting case, zero (1/(:1). We measured limits on preexisting condition exclusions as the maximum number of months that insurers may exclude coverage for a preexisting condition.⁹

By developing continuous measures for these variables, we are able to avoid the problems of multicollinearity that have forced other researchers to bundle heterogenous reform measures and search for differences related to the presence or absence of regulation, without regard to differences the variety and restrictiveness of the states’ regulation. Instead, our analysis examines the impacts of specific regulatory measures (e.g., health rate bands separately from age rate bands), and also accounts for stricter limits on pricing and preexisting condition exclusions in some states compared to others. Descriptive statistics for all the market structure, regulation and control variables are provided in Tables 9 and 10.

4 - See, for example: Sloan and Conover, 1998; Jensen and Morissey, 1999; Zuckerman and Rajan, 1999; and Simon, 1999a and 1999b.

5 - At least one study (Grace and Timme, 1992) suggests that most of the accident and health insurance industry (including major medical insurers as well as other accident and health insurers) experiences significant increasing economies of scale, with only the very largest insurers experiencing constant returns to scale. It is reasonable to expect that their result also holds for the subset of the industry that writes major medical coverage. Similarly, other studies (Blair and Vogel, 1978; Clement, 1995; and Feldman, Wholey and Christianson, 1996) suggest increasing economies of scale among small HMOs.

6 - Wholey et al. (1995) estimated that HMOs experience increasing economies of scale until they enroll about 100,000 lives, beyond which additional economies of scale are insignificant.

7 - The Herfindahl index is defined as , where m is the market share of each insurer. The Herfindahl index takes on a maximum value of 1 (monopoly) and approaches zero as market share is distributed among more competitors.

8 - Satterthwaite (1979) raised an intriguing alternative to the conventional economic sense of competitive markets as price-reducing, which arguably could apply to health insurance markets. Satterthwaite hypothesized that in markets where consumers place subjective value on the service being sold and are able to discern the value of the product accurately only after experiencing it for a period of time, more sellers in a market may equate to greater monopoly power for each seller and higher prices not greater competition and lower prices as the conventional economic theory of monopolistic competition would predict. We use this logic to justify application of two-tailed significance tests in our estimation.

9 - We also tested the sum of limits on look-back periods and waiting periods, with no difference in our empirical results. In states with no limit on the waiting period for coverage of preexisting conditions we coded the waiting period as 72 months — one year longer than the maximum waiting period that we observed in any state.