Does Surplus Participation Reflect Market Discipline? An Analysis of the German Life Insurance Market

The academic evidence on risk sensitivity of customer demand comes to ambiguous results and supports the skepticism: Zanjani (2002) and Epermanis and Harrington (2006) find evidence for risk sensitivity of customer demand in the U.S. Eling and Schmit (2010) is the only study to analyze the risk sensitivity of customer demand outside the U.S. They find only limited evidence for market discipline and conclude that regulators need to enforce market transparency, if they want market discipline to be a strong element of Solvency II.

The German ordinance on minimum participation for customers in life insurance (MindZV) sets out strict formal requirements. Surplus participation needs to be at least equal to the sum of (1) the maximum of 90% of the investment result and the guaranteed interest rate (specified at the beginning of the contract), (2) 75% of the risk result, and (3) 50% of the cost/other result. If the values for (2) or (3) are negative, it is set to 0, i.e., cross-subsidization among different surplus sources is not allowed.

The earned surplus is calculated as gross premiums earned + investment income − claims and insurance benefits incurred − acquisition and administrative expenses + reinsurance result − other expenses − income taxes. Policyholder participation is measured as net changes in reserves for insurance and investment contracts. The corresponding figures can be directly obtained from the annual profit and loss statement of German life insurers. The market average is obtained by aggregating the values across all companies.

Other possible performance indicators include the so-called Finsinger rating for German life insurers or the net interest return. The Finsinger rating is also easily publicly available and takes the entire business operations into consideration but strongly depends on the underlying rating methodology making it a less objective measure (for details see Section 3.4). The net interest return only takes into account the investment result of a company. Although this is definitely one of the main performance drivers, it neglects other surplus sources completely. Moreover, both measures can also be influenced by management decisions.

There are significant differences in market discipline between banking and insurance. In banking, there is a great deal of market discipline in stock and bond markets because the equity and debt of most large banks is traded on capital markets (see, e.g., Avery et al. 1988; Morgan and Stiroh 2001; Sironi 2003; Distinguin et al. 2006). Much market discipline can thus be observed with traded debt, such as yields on subordinated debt. The insurance sector, however, is fundamentally different as many insurers are mutuals, not stock companies. Furthermore, many of the insurers that are organized as stock companies are not traded on the stock market, thus making effects on stock prices difficult to observe. There is also hardly any traded debt in the insurance industry since the reserves of the policyholders are in general the major part of the insurers’ liabilities. For example, in Germany it is prohibited to have debt other than the reserves for policyholders and only very restrictive exceptions are allowed from this general rule (e.g., with hybrid instruments). We thus cannot observe market discipline on insurance capital markets for either equity or debt as we do in banking. For this reason existing studies on market discipline can only focus on customers and not on investors. See Eling (2010) for more details.

Cottin et al. (2007) define the surplus participation rate spread as the difference of the company specific surplus participation rate and the average surplus participation rate in the market for each year.

A joint linear relationship of independent and response variable is rather unlikely. Having, however, only observations for ten years, the reliability of regression models for single companies is limited (see Cottin et al. 2007, p. 345).

Due to the complex, interest rate volatility smoothing surplus distribution mechanisms in life insurance, the surplus participation rate follows longer term trends with a certain time gap. If the positive gap between market interest rate and surplus participation rate during periods of increasing interest rates exceeds surrender cost, lapse rates are likely to increase.

This guarantee is regulated and currently set to be 2.25% for new contracts, while older tariff generations receive a higher guarantee of up to 4.0%. In order to ensure the long-term fulfillment of insurance contracts this guarantee rate is defined conservatively.

If the empirical analysis does not find a significant relationship, we are not able to determine whether the first or the second aspect does not hold.

Our data set covers an unbalanced panel of 66 to 70 insurance companies (depending on the considered response variable) over a time period of eleven years. The setup of a wide but short data set is typical for panel data. In this case heterogeneity across units is often the central analysis focus (see Greene 2003). Accordingly, we take into account only firm effects but no time effects. Besides the data design, the use of relative modeling approaches (see Sections 3.2 and 3.3) and multicollinearity issues with the considered control variables support the non-consideration of time effects.

We apply the Nerlove method for estimating the variance components. This method assures positive estimates of the variance components (see Baltagi 1995).

The omission of potentially significant variables might introduce an estimation bias. Therefore, we run the OLS and RE regression models including an indicator variable for legal form (being 1 if the considered company is a mutual and 0 otherwise). The corresponding results indicate that this variable is significant only in three of 22 model specifications considered. Moreover, the inclusion of this variable does not materially change the coefficient estimates and significance levels for all other variables. We thus conclude that no severe bias is introduced by omitting the corresponding variable. Detailed results are available upon request. Additionally, Harrington and Niehaus (2002) find evidence that capital ratios are on average higher for mutuals than those for stock companies. Thus, the solvency indicator considered in our analyses might partially incorporate information on the legal form.

Regarding new business, one needs to distinguish between actual new business (meaning new contracts) and other inflows from increases in sum insured of existing business. As we are analyzing new business effects, we only take into account the first kind of business.

The annual premium equivalent combines single and regular premium business taking into account the differences in method of payment. It is calculated as sum of regular premiums and 10% of single premiums assuming an average policy duration of ten years for single premium business (see Hardwick and Adams 2002).

The differentiation between single and regular premiums is not possible for lapses as lapse volume is only measured in terms of regular premiums.

The impact of the surplus participation might be more pronounced for endowment and annuity business containing a dominant savings component.

All relative modeling approaches for the surplus participation rate yield consistent results and similar conclusions. Detailed results are available upon request.

If raw data is used for the analysis, all variables are significant at the 1% level. This can be explained by the presence of outliers forcing the significance of the variables. The conclusions are the same when using preprocessed data but the results are more differentiated, i.e., not all variables being always significant.

These and all other threshold values are the result of a trade-off between deleting the most extreme growth rates but keeping the data sample as large as possible. The number of deleted company years is not very sensitive to variations of the threshold. The corresponding analyses are available upon request.

Correlation tests have been performed to check for collinearity between Finsinger rating and solvency, but the values are close to zero. The Finsinger rating is not primarily a measure of financial strength. Instead it focuses on profitability and sustainability of contract performance. Hence, it represents more a risk-return indicator than a pure risk indicator. See Cummins and Doherty (2002) and Pottier and Sommer (2002) for a detailed analysis of solvency in the insurance industry.

The results for sum insured are very similar to those when total number of contracts is considered. For both new business and lapses an OLS model is preferred and the surplus participation rate has a significant positive and negative impact, respectively. The significance levels of the control variables change only slightly. Therefore, we do not present the corresponding results for sum insured in the following. Detailed results are available upon request.

Additionally, we consider the time horizon 2003–08 to account for potential distortions due to the shift of the surplus participation rate following the economic crisis of 2001 to 2003. The overall picture remains unchanged. While there is a significant, positive relationship between new business and participation rates in four out of six model specifications, we find a significant, negative relationship only in two out of five cases for lapses. Due to the reduced level in the surplus participation rate and fixed surrender charges it becomes less profitable to lapse existing contracts. Therefore, the relationship between lapses and participation rates becomes less pronounced. Another factor might be the limited time horizon of six years for the analysis.

Firm effects are identified for total business only when new business premium growth in terms of annual premium equivalent is considered. In this case single premium business is taken into account. Single premiums incur predominantly for savings products, i.e., endowment and annuity business. As we have seen that firm effects are present for those products, the presence of firm effects for growth in annual premium equivalent is not surprising.

As robustness test we analyzed unweighted new business premiums (i.e., regular premiums + 100% of single premiums). The corresponding results are identical, except for minor changes in coefficient estimates and significance levels, but all variables are still significant.

Statements regarding the significance of control variables require the normality assumption for the regression residuals to hold. Although one might assume the residuals to be normally distributed for a large data set as in our case, we explicitly checked this assumption using Q-Q-plots. This test indicates that the normality assumption holds for the residuals of the considered regression models.

This does not hold for the OLS model when absolute market share is considered as response variable which is studied by Cottin et al. (2007). They report the strongest relationship between new business and surplus participation rate for the OLS model with absolute market share and absolute surplus participation rate spread. Using this model specification we find a significant relationship for the fixed firm effects model suggested by the specification tests, but we do not find any significant relationship when the OLS is considered. Moreover, we do not find any significant relationship between changes in market share and the surplus participation rate spread. These observations are in line with the findings of Cottin et al. (2007).

Contrary to Cottin et al. (2007) we even find a significant relationship between changes in market share or absolute market share and surplus participation rate spread. Note that we consider endowment/annuity business jointly, while Cottin et al. (2007) consider them separately.

As Cottin et al. (2007) do not explicitly state the results of their analysis for number of contracts, we are not able to compare our results with their findings in this case.