Many countries with a competitive health insurance market have government regulations such as open enrollment for a basic benefit package, premium regulation, and risk-equalization. The goal of risk-equalization is to create a level playing field for health insurers and to prevent risk selection. The key element of ex-ante risk-equalization is to find the best prediction of an individual’s healthcare expenses for the new insurance year. After more than 30 years of research these predictions have been substantially improved [1, 2]. This raises the question: assuming we have a ‘perfect’ prediction of everyone’s future medical claims, is then the goal of risk-equalization achieved? The answer is no because in addition to medical claims enrollees also have to pay a loading fee. The loading fee is the excess of the premium above the expected medical claims to be paid by the insurer [3, p. 1237, 4, p. 181]. Depending on the characteristics of the insurance market, the share of the loading fee is about 5–20% of total premium payments. If insurers need to charge a higher loading fee for a high-risk than for a low-risk enrollee, there is still no level playing field and incentives for risk selection remain present.

In this paper, we focus on administrative health insurance costs which is a clearly demarcated cost category and, as we show, is the main cost component of the loading fee. In many countries, insurers have the obligation to report administrative costs annually. Administrative costs contain many different components often depending on the country regulations and type of insurance market [5, 6]. On one hand, some components are suitable for risk-equalization as they represent activities that an insurer has to undertake when providing health insurance. This holds, e.g., for administrative costs related to checking bills, fraud prevention, administrative contacts (phone, visits, email, etc.), costs for handling defaulters, purchasing healthcare, contracting healthcare providers, quality improvement, utilization management and the coordination of care might be lower for an insurer with a low-risk population than one with a high-risk population. On the other hand, not all administrative cost categories are suitable for risk-equalization. Examples are costs related to differences in administrative efficiency, for example due to (dis)economies of scale resulting from group size [4], unintended activities such as creaming, skimping or dumping of enrollees [7], marketing and advertising. These latter aspects should not be equalized because that would distort competition and create an unlevel playing field.

So far, however, the focus of risk-equalization by policymakers and in the literature has been primarily on the equalization of expenses for medical claims only. Often, the loading fee or administrative costs are overlooked. For example, in most papers on risk-equalization the loading fee is not even mentioned at all [1, 8]. A reason may be that in the past decades the efforts to improve the risk-equalization were primarily focused on the low-hanging fruit such as major morbidity indicators. Furthermore, administrative costs are not available at the enrollee level, as is the case with medical claims, making it difficult to show a causal relationship between the medical claims of an enrollee and the corresponding administrative costs.

Many countries with a risk-equalization system do not take administrative costs or the loading fee into account when applying risk-equalization. In this paper, we discuss three of these countries: Australia, the Netherlands and Switzerland, and two countries that do take administrative costs into account, Germany and the US. Germany found in an empirical analysis that around 50% of average administrative costs vary with medical claims and uses this number to risk-equalize insurers [9]. In the US Marketplaces, the rule was 100% until 2017. However, in 2018, the regulator substantially reduced the importance of administrative costs in risk-equalization, taking into account that (a proportion of) administrative costs do not vary with medical claims [10]. We discuss Germany and the US briefly in Sect. 4.2.

This paper is inspired by the experience in Germany [9] and, to the best of our knowledge, is the first paper that discusses to what extent the loading fee should be included in the risk-equalization. First, we study whether administrative costs vary with morbidity, measured by predicted expenses of medical claims. To obtain more causal evidence we test empirically the relationship for insurance markets in Australia, Germany, the Netherlands, Switzerland, and two markets in the US. We gathered several years of data per insurance market which allows us to use a fixed effect panel data regression, and to control for unobservable constant differences across insurers in a market. Although it is difficult to measure effects precisely, we find for most markets a positive correlation between administrative costs and the morbidity of an insurer’s population which enhances the premise of causality. Second, we discuss how administrative costs can be included by policymakers in a risk-equalization system and compare our method with the policy rules implemented by Germany and the US. Third, while this paper focusses on administrative costs, we also briefly discuss whether or not the residual part of the loading fee, such as costs related to profits or risk bearing, should be risk-equalized.

Risk-equalization and administrative costs

Whether the administrative costs in health insurance should be included in the risk-equalization system depends on the goal of risk-equalization. This goal depends on the assumptions made about the health insurance market. In this paper, we assume a competitive health insurance market with regulation such as open enrollment for a basic benefit package and premium regulation in the form of community-rating (by class). The main challenge in a regulated competitive health insurance market is to avoid risk selection. We define risk selection as the actions by consumers and insurers to exploit unpriced risk heterogeneity and to break pooling arrangements [3, 11]. We assume that the goal of risk-equalization is to ensure that each applicant whom the insurer must accept represents an equal insurance risk for the insurer.Footnote 1 This is the case if, and only if, the risk-equalization payment per enrollee provides a full compensation only for all enrollee-related risk characteristics that are not allowed to be used for premium rating. There should be no compensation for insurer-related characteristics such as an insurer’s efficiency, its market power, or the (dis)economies of scale resulting from its number of enrollees because that would distort the competition and create an unlevel playing field [8].

The main question for administrative costs is whether there is unpriced risk heterogeneity related to individual administrative costs (and loading fees). In contrast to medical claims, this question is extremely difficult to answer as administrative costs are not available at the individual level of the enrollee, but only at the insurer level. Moreover, most administrative cost components are strongly aggregated over various costs components making it even harder to attribute certain costs to individuals.

While it has been shown that high-risk enrollees have on average more administrative consumer contacts than low-risk enrollees, and thus are more costly for a health insurer [12], it is unknown to what extent there is unpriced risk heterogeneity related to an aggregate cost measure as total administrative costs. To obtain more insight into this problem, we study in this paper whether administrative insurance costs vary with medical claims at the insurer level. Thus, to what extent are administrative costs indeed higher for insurers with a high-risk population than for insurers with a low-risk population? In the next section, we will study empirically the relationship between administrative costs and an insurer’s population morbidity for six different insurance markets.

Empirical analysis

If there is a causal effect of population morbidity on administrative costs, we expect to observe this relationship in many different insurance markets. Therefore, we selected six insurance markets that should satisfy the following criteria. First, there should exist a system of risk-equalization carried out by a regulator or government, as this allows us to obtain an exogenously determined indicator for population morbidity of each insurer, which in the remaining part of the paper we call the insurer’s risk-score. Second, there should be a sufficient number of insurers in the market to obtain enough cross-sectional variation. For example, we excluded Belgium, Chile, Ireland and Israel because each of these countries has a limited number of insurers. Moreover, there should be multiple years of data available to perform a panel regression with fixed insurer effects. This is important as administrative costs may differ in many unobservable dimensions across insurers, such as differences in efficiency (capital and personnel), profit requirements (for-profit, not-for-profit or social insurer), market power, size, providing other type of insurance activities, different benefit packages, etc. The fixed-effects regression will eliminate constant unobserved differences across insurers. We gathered data for various years in six markets of five countries: Australia, Germany, the Netherlands, Switzerland and two insurance markets in the US, the small group and the individual market (the so-called US Marketplaces). The data are obtained from insurer reports that are publicly available or assembled by the country regulator.Footnote 2 For each insurance market, a description of the market and data is provided in “Appendices 16”. Premium revenues, medical claims and total administrative costs are demarcated cost categories that are available in all annual insurer reports.

To obtain an idea about the different components of the premium, we show in Fig. 1 the mean premium per enrollee per life year for each market in 2019, divided into three components; medical claims, administrative costs and a residual loading fee component.Footnote 3 The sum of this residual component and administrative costs are often denoted as the loading fee [13]. To ease comparability across the countries we converted all amounts into Euros. Figure 1 shows some interesting differences across the six insurance markets. Note that the numbers in the figure can only be broadly compared across markets as markets may differ in various ways, such as in the size of the basic benefit package, cost-sharing arrangements, contracting and efficiency activities by insurers, economies of scope due to activity in other insurance markets, insurance regulations, type and size of insurers in the market, culture, etc. Costs for medical claims and residual loading fees are clearly highest for enrollees in the two US-markets, with mean administrative costs accounting for about 14% of the premium and the loading fee for more than 20%.Footnote 4 These costs are substantially lower for the other four countries; administrative costs (loading fees) range from 3% (3%) of premium payments in the Netherlands to 11% (12%) in Australia. In the Netherlands, we find even slightly negative residual loading fees because some insurers in 2019 used their excessive reserves to lower their annual premiums.

Fig. 1
figure 1

Mean premium paid per person per life year and its major components (to facilitate comparison across countries all numbers are in euros, 2019). The numbers above the bar represent the average premium in the insurance market in 2019 for the basic benefit package. The premiums are in euros using the following exchange rates (April 10, 2021): 1 Euro = 1.56 Australian Dollar = 1.19 US Dollar = 1.10 Swiss Franc. The premium includes subsidized premium payments and risk-adjusted payments insurers receive from (or pay to) the regulator. Medical claims represent payments from insurers to health care providers. Payments of claims by consumers related to cost-sharing arrangements are not included in these payments. Administrative costs are total administrative costs. The residual loading fee is computed by subtracting medical claims and administrative costs from the premium. The percentage numbers in the bars represent the share of average costs for medical claims in the premium. One minus this fraction represents the share of the loading fee (administrative costs plus the residual loading fee) in the premium. All numbers are obtained from annual reports of health insurers or assembled by the country regulator. For the numbers of health insurers included see Table 1. For more information, we refer to “Appendices 16

Table 1 Summary statistics of health insurance market per country

Table 1 provides the summary statistics of the data we use in our panel regressions. Again, we refer to “Appendices 16” for a detailed description of each market. We have five markets that offer basic insurance and one market, Australia, that offers supplementary insurance. The number of insurers and the population size of insurers vary across markets. For all countries, the information is at the level of the insurance carrier, except for the Netherlands where we could obtain data at the holding (several carriers under one roof) level only. This is reflected in Table 1, where Dutch holdings have on average more than one million enrollees. The insurer carriers in the two US markets are smallest in size with on average less than 50,000 enrollees per insurer. Insurer size seems also to be (partly) reflected in the administrative costs as in general, insurers with more enrollees have fewer administrative costs per enrollee due to economies of scale.

To study whether administrative costs vary with expected medical claims we constructed the risk-score for each insurer, using the reported risk-equalization payments that insurers receive from, or have to pay to, the regulator in the market. To make the risk-scores broadly comparable across countries we performed a similar strategy among countries and scaled the payments with average medical claims in a market to obtain average risk-scores around 1 (see notes below Table 1). Note that the rules for applying risk-equalization are determined in all markets ex-ante by the regulator and, therefore, the risk-scores can be considered as exogenous with respect to the medical claims in the new insurance year. However, the constructed risk-scores may contain potential measurement errors.Footnote 5 We discuss this point more extensively in “Appendix 1”, where we also run regressions with an ex-post risk-score variable that we constructed with medical claims in the current year.

Figure 2 shows the correlation in the raw data between mean administrative costs per life year and the risk-score for each insurer in the six markets. Because economies of scale play an important role in administrative costs [13], we divided insurers in each market into three equally sized groups: small (0–33 percentile), medium (34–66 percentile) and large (67 + percentile) insurers, each group represented by a different color in the figure. Five of the six graphs show a positive correlation between administrative costs and an insurer’s risk-score. The insurer’s risk-scores for most markets range between 0.5 and 1.5, only for the individual market in the US we find risk-scores above 1.5 for some insurers. For Germany and Switzerland, and to a lesser extent Australia, the observations are about equally distributed on the x-axis. Also, small, medium, and large insurers are distributed over the whole x-axis which indicates that there is sufficient variation to obtain plausible estimates. For the Netherlands, we have limited variation and find a negative relationship. However, this relationship seems to be driven by economies of scale as population size is strongly positively correlated with the risk-score. This relationship is more clearly visible in the Netherlands, and less so in other markets, because there are only a few Dutch insurers (holdings) in the market and the differences in the market share between the largest insurer (about 27%) and smallest insurer (around 0.5%) is much bigger in the Netherlands than in other countries, which emphasizes the economies of scale effects. Therefore, it is important to control for population size in our estimations. The data in the US small group market are more concentrated around 1 on the x-axis then in the US individual market. Both US markets show substantial variation in administrative costs on the y-axis which suggests that the positive correlations are less clearly present in the data. Note that due to the high administrative costs in the US markets, the scale of the y-axis for the two US markets is a factor five of the other four markets.

Fig. 2
figure 2

Insurer’s administrative costs (per person per life year) and the risk-score. To obtain comparability across countries the range of the x-axis is the same for all countries. Only for the US Marketplace (individual market), we extended this range to 1.6, and still 14 observations fall outside this (extended) range. In Switzerland, all observations of one insurer are excluded from the figure because they fall outside the range. A linear trend is plotted in the figure with a 95% confidence interval (95% CI)

Table 2 shows the fixed effects estimates for each market of the following panel regression:

$${\text{adm}}_{it} = \beta_{0} + \beta_{1} {\text{riskscore}}_{it} + \beta_{2} {\text{popsize}}_{it} + \beta_{3} \left( {\frac{1}{{ {\text{popsize}}_{it} }}} \right) + \alpha_{i} + \gamma_{t} + \varepsilon_{it} ,$$

where subscript \(i\) represents the insurer and \(t\) the insurance year. The dependent variable is \({\text{adm}}_{it}\), mean administrative costs per life year and \({\text{riskscore}}_{it}\) is the insurer’s risk-score. \(\alpha_{i}\) denotes insurer fixed effects and \(\gamma_{t} { }\) year fixed effects. Although the insurer fixed effects control for the average market size of an insurer, we also included an insurer’s population size, \({\text{popsize}}_{it} ,\) and the inverse of population size to capture possible non-linear time varying aspects. \(\varepsilon_{it}\) represents the error term. \(\beta_{1}\) is our coefficient of interest as it measures the effect of an increase in the risk score on the administrative costs. Note that we cluster standard errors at the insurer’s level.

Table 2 Fixed effect estimates of \(\beta_{1}\): the effect of an insurer’s risk-score on administrative costs

We find a positive coefficient for the risk-score for all insurance markets. For four markets, we find a statistically significant positive effect, suggesting that insurers with a high-risk population have higher administrative costs. The negative correlation in the raw data for the Netherlands (see Fig. 2) becomes positive after controlling for market share, but the effect is statistically insignificant.Footnote 6 The effects are substantial considering that the risk-scores in most markets range between 0.5 and 1.5. This implies that the estimates in Table 2 can be roughly interpreted as the difference in administrative costs between an insurer with the highest and the lowest risk population. For example, in Germany the difference in risk-scores between the lowest and highest insurer is about 0.8 amounting to a difference in administrative costs of 0.8*86.0 = 69 euros per person per life year.

The results in Table 2 are for our most preferred specification. In “Appendix 1”, we test the robustness of our results and run our panel regressions also with an ex-post risk-score, constructed by dividing an insurer’s average expenses of medical claims per life year by the market average. This ex-post risk-score introduces endogeneity problems as it is determined in the same year as administrative costs. Figure 3 in “Appendix 1” suggest that the correlation between the ex-post risk-score and administrative costs are about equal for Germany, the Netherlands and Switzerland than in Fig. 1, but the positive correlation becomes larger for Australia and the US. The fixed effect estimates that follow from the regressions with the ex-post risk-score are somewhat smaller and statistically insignificant for Australia and the US, which suggest that it is difficult to measure the effects precisely for these countries. The estimates remain positive and statistically significant for Germany and Switzerland.

We conclude that there is evidence for a positive causal effect of an insurer’s population morbidity on its administrative costs. The estimated effect is relatively stable for Germany and Switzerland. However, it is unclear how large the size of the effect is for the US, Australia and the Netherlands as the estimated effects are surrounded with more uncertainty. The likely reason for the US is that there is extensive variation in administrative costs across insurers and the ex-ante risk-score may contain more potential measurement errors than in other countries (see “Appendix 1”). In Australia and the Netherlands, the annual number of insurers in the market is relatively small which complicates measuring the effects precisely.

How can administrative costs be included in the risk-equalization system?

When applying risk-equalization most regulators (or governments) use individual annual medical claims, often from the total population, and regress these claims on individual consumer characteristics [2, 8]. However, administrative costs are not available for individual consumers and are only available at the insurer level. Thus, a different approach is needed.

We first show a methodology for risk-equalizing administrative costs and then discuss the policy rules adopted in Germany and the US.Footnote 7 Our methodology uses the estimated effects of an insurer’s risk-score on the average administrative costs per enrollee (see Table 2).

As in the previous section we denote the average risk-score for an insurer \(i\) with \({\text{riskscore}}_{i}\), where the average is taken over the predicted risk-scores of all enrollees of insurer \(i\). The predicted risk-score of an enrollee is based on the predicted medical claims of an enrollee, based on for example demographics and diagnoses of an enrollee. We scale the average predicted risk-score of an enrollee to 1. Thus, an insurer with a \({\text{riskscore}}_{i}\) > ( <) 1 will have higher (lower) predicted medical claims per enrollee in the new insurance year than an insurer with \({\text{riskscore}}_{i}\) = 1.

Most countries risk-equalize only medical claims, i.e., the regulator predicts \(\widehat{{{\text{MC}}}}\), the predicted average medical claim per enrollee in the market. \(\widehat{{{\text{MC}}}}*{\text{riskscore}}_{i}\) then reflects the predicted amount of medical claims for the average enrollee of insurer \(i\).Footnote 8 Note that in these countries the regulator does not predict the loading fee or administrative costs and thus both components do not vary with medical claims in risk-equalization.

Formula for risk-equalizing administrative costs

A method to risk-equalize administrative costs that vary with the insurer’s risk-score, is to use the regression results in Table 2. From Eq. (1) follows that \(\widehat{{{\text{adm}}}}_{it} = \hat{\delta }_{it} + \hat{\beta }_{1} {\text{riskscore}}_{it}\), where \(\hat{\delta }_{it}\) represents the part that is independent of the risk-score and, thus, does not vary with medical claims.Footnote 9 We define \(\widehat{{{\text{ADM}}}}_{t}\) as the mean predicted administrative costs per life year, with the average taken over all insurers. Since the average annual risk-score of all insurers is constructed to be 1, \(\widehat{{{\text{ADM}}}}_{t} = \hat{\delta }_{t} + \hat{\beta }_{1} ,\) with \(\hat{\delta }_{t}\) the average of \(\hat{\delta }_{it}\) taken over all insurers. Because \(\hat{\delta }_{it}\), the first component of \(\widehat{{{\text{adm}}}}_{it}\), contains insurer-specific aspects (i.e., the insurer fixed effect and its population size) that the regulator (most likely) does not want to equalize, the predicted administrative costs of insurer \(i\) that are acceptable to be equalized under a zero-sum equalization can be specified as \(\widehat{{{\text{adm}}}}_{it} = \hat{\delta }_{t} + \hat{\beta }_{1} {\text{riskscore}}_{i}\). The equalization payment for insurer \(i\) then equals \(n_{i}\) (\(\widehat{{{\text{adm}}}}\)it\(\widehat{{{\text{ADM}}_{{\text{t}}} }}\)) = \(n_{i} \hat{\beta }_{1} ({\text{riskscore}}_{i} - 1),\) with \({\text{n}}_{i}\) equal to the number of enrollees of insurer \(i\). For example, an insurer with a risk-score 0.7 has to pay an equalization payment of 0.3*\(\hat{\beta }_{1}\) per enrollee, while an insurer with a risk-score of 1.3 receives 0.3*\(\hat{\beta }_{1}\) per enrollee. In the six insurance markets 0.6*\(\hat{\beta }_{1}\) ranges between about 40 euro and 300 euro (see Table 2), which implies that risk-equalization of administrative costs that vary with the insurer’s risk-score, is nontrivial. Because an insurer’s risk-score may change after enrollees have switched insurers during the open enrollment period, it should be calculated after the open enrollment period. This equalization of administrative costs can be done in addition to the equalization of medical claims, or it can be combined with it.

The second component of \(\widehat{{{\text{ADM}}}}_{t} ,\) i.e., \(\hat{\beta }_{1}\), depends on the risk-scores and therefore represents the part of \(\widehat{{{\text{ADM}}}}_{t}\) that is risk-equalized. Thus, the percentage of administrative costs that is used for risk-equalization equals \(\frac{{\hat{\beta }_{1} }}{{\widehat{{{\text{ADM}}}}}}_{t} {*}100{\text{\% }}\). Using the mean administrative costs over all relevant years per insurance market from Table 1 and the estimates \(\hat{\beta }_{1}\) from Table 2 yields the following back-of-the-envelope percentages for Australia: 187%, Germany: 59%, the Netherlands: 69%, Switzerland: 120%, US (individual market): 61%, US (group market): 57%.Footnote 10

Germany and the US marketplaces

Germany is, as far as we know, the only country that uses an empirical prediction of administrative costs [9, 14]. Drösler and co-authors find a positive effect of the risk-score on administrative costs which resulted in the policy rule that 50% of the administrative costs should be risk-equalized. The 50% rule is in line with our results where we find 59% (see above). Thus, while Germany includes administrative costs, they do not include the residual loading fee in risk-equalization.

Before 2018, in the US marketplace the risk-equalization payments for insurer \(i\) were based on the rule \(\overline{P}*{\text{riskscore}}_{i}\), where \(\overline{P}\) is the average premium in the market. So, the US applied the rule that the loading fee (administrative costs and residual loading fee) should be equalized for 100%. As of 2018, the US substantially reduced the importance of the loading fee in the risk-equalization. In the formula for risk-equalization, the average premium in the market is reduced by a fixed rate of 14% with the argument that this reduction reflects (the proportion of) administrative costs that do not vary with medical claims [10].Footnote 11 Thus, the new rule implies that risk-equalization will be based on 0.86 \(*\overline{P}*{\text{riskscore}}_{i}\). Our estimation results for the US are too imprecise to judge whether the new rule should be preferred over the old rule. The estimations of the risk-score indicate that the risk-equalization should be based on 0.94 \(*\overline{P}*{\text{riskscore}}_{i}\) which implies lower equalization payments in the US Marketplaces for administrative costs than the old rule, but higher equalization payments than then new rule. The estimation results for the ex-post risk-score in “Appendix 1” implies that risk-equalization should be based on 0.85 \(*\overline{P}*{\text{riskscore}}_{i}\) which is in line with the new rule.Footnote 12


The main message of our analyses is that administrative costs should be taken into account when applying risk-equalization. In this paper, we show how this can be done by relating the mean administrative costs per enrollee to an insurer’s risk-score. However, other routes are possible as well. The most obvious approach is that the regulator requires insurers to decompose their administrative costs into several components. For each of these components the regulator can determine to what extent they are eligible for risk-equalization. For example, one component could be the administrative costs for enrollees that cannot pay their deductible or out-of-pocket premiums and end up in payment arrears. In case of an insurance mandate where insurers cannot terminate coverage, insurers have often to undertake costly activities to collect this money. The probability of ending up in payment arrears is likely to be unevenly distributed across individuals.

Our approach, i.e., considering the mean administrative costs per enrollee and relate this to an insurer’s risk-score, should be seen as a practical approximation.Footnote 13 This approach will only work if there is enough variation in various dimensions in the data to credibly control for all types of insurance aspects. If the number of insurers is relatively small or variables as risk-scores, population size and other administrative activities are unevenly distributed, it is difficult to obtain credible estimates. Moreover, if the risk-score follows from imperfect risk-equalization of medical claims, then relating the risk-score to administrative costs will result in biased estimates (see also “Appendix 1”). Thus, applying a simple rule, like the US and Germany do, seems to be a practical solution to a complicated problem. It is transparent and to be preferred over applying no rule at all. The exact percentage can be calculated as we do in this paper. However, our computations can be improved if there is better and more precise information available about components of (expected) administrative costs, risk-scores, etc.

We are sceptical about following a similar strategy for the residual loading fee, i.e., correlating the loading fee with an insurer’s risk-score.Footnote 14 A first reason is that it is a priori not clear whether the residual loading fee contains many categories that are suitable for risk-equalization. It is doubtful whether a regulator should want to equalize for components of the residual loading fee such as profits related to market power, profit windfalls or shortfalls (that may be correlated across insurers), or solvency requirements that insurers have to meet. However, the cost of risk-bearing could be a potential candidate for risk-equalization as the expected variation in medical claims is often larger for high-risk than for low-risk individuals. All else equal, risk averse insurers might want to charge a higher risk-premium for high-risk individuals. A second reason is that endogeneity problems might also play a role. For example, consider a case of perfect competition (where insurers have to charge the same premium for the basic benefit package) and imperfect risk-equalization of medical claims, then insurers who are undercompensated will face lower profits. A third reason is that several components are not clearly demarcated, also because they often contain transfers from previous years, making it difficult to precisely measure the costs of single components or the total residual loading fee in a year.

Finally, there might be potential disadvantages of including administrative costs in risk-equalization as it might incentivize cost inflation. For example, insurers might see possibilities to game the risk-equalization system by increasing their administrative activities or by shifting administrative costs between several cost components, such as shifting administrative costs related to supplementary insurance to basic insurance. These potential disadvantages are likely to be relevant in concentrated markets where insurers have relatively large market shares, such as for example in the Netherlands and Australia, as the marginal returns from increasing administrative costs increases with the market share. To prevent gaming activities, policymakers should clearly define the various administrative cost categories so that they can be properly monitored. An extreme option to prevent cost inflation is putting a constraint on the size of administrative costs, as is done in the US. In the Affordable Care Act insurers must remit a rebate if the loading fee is larger than 20% of the premium. Another option is to define the total amount of administrative costs on a percentage of medical costs only.


Many countries with a risk-equalization system in health insurance risk-equalize only medical claims of enrollees. We argue in this paper that components of the loading fee should be considered for risk-equalization as well, with the loading fee defined as the excess of the premium above the expected medical claims to be paid by the insurer. The reason is that enrollee characteristics, such as being a high or low-risk enrollee, likely have a causal impact on the loading fee. We show for six insurance markets in five countries that an insurer’s administrative costs, which we show is the major component of the loading fee, is positively correlated with an insurer’s risk-score as an indicator for the morbidity of the insurer’s population. We show how in practice administrative costs can be included in risk-equalization in a simple way and we show that this results in additional equalization payments nontrivial in size. We discuss the examples of Germany and the US marketplaces. The policy rule in Germany that 50% of the administrative costs should be risk-equalized, is consistent with our empirical findings. For the US, our empirical results are too imprecise to judge the current US-rule and more research is needed.

As far as we know, our paper is the first to address this important issue. There are many channels for future research. A first channel would be to obtain better knowledge of why administrative costs may differ across enrollees. For example, Douven and Kauer [12] show that high-risk enrollees cause more consumer contacts then low-risk enrollees. But other interesting administrative differences could occur for handling defaulters, purchasing healthcare, contracting healthcare providers, utilization management, etc. A second channel is to improve our regressions for individual markets by obtaining better data about administrative cost components and risk-scores. Policymakers could require insurers to decompose administrative costs into different components, components that are suitable and not suitable for risk-equalization. Also, the measurement of an insurer’s risk-score could be more precisely computed by taking specific characteristics of insurance markets better into account. A third channel to consider is that administrative costs may depend on characteristics not necessarily related to health risks. For example, insurers may face higher administrative costs for empowered consumers or consumers who have a larger chance to end up in payment arrears.

We are more skeptical about risk-equalizing the residual loading fee, i.e., the difference between the loading fee and administrative costs. The residual loading fee is not a clear demarcated cost category and contains a lot of components that may not be suitable for risk-equalization. More research is needed whether some specific cost components of the loading fee are suitable for risk-equalization.