Comparing Single- and Multichannel Distribution

Lucinda trigo-gamarra/Christian growitsch*

C

omparing

S

ingle

-

and

m

ultiChannel

d

iStribution

S

trategieS

in

the

g

erman

l

ife

i

nSuranCe

m

arket

:

a

n

a

nalySiS

of

C

oSt

and

p

rofit

e

ffiCienCy

**

abStraCt

to explain the success of different distribution strategies in the life insurance industry, we analyze the performance of single- and multichannel distribution firms in the ger-man life insurance industry. We estimate cost and profit efficiency for three groups of life insurers: multichannel insurers, direct insurers, and independent agent insurers. We use nonparametric data envelopment analysis to calculate efficiencies for german life insurers for the years 1997-2005. We identify the absence of performance advantages of specialized insurers as the most important reason for the coexistence of different distri-bution systems.

JeL-Classification: g22, L15, L22.

Keywords: dea; distribution Systems; efficiency analysis; insurance markets.

1 introduCtion

Before 1994, German automobile, life, and health insurers were price-regulated, but following the liberalization of the European insurance sector in 1994, German insur-ance markets were deregulated. Deregulation has allowed German insurinsur-ance companies to choose their prices (premium levels) freely, which has, in turn, led to increasing price competition. In addition, insurers are no longer required to obtain authorization for the design of their products from the regulatory agency, resulting in a greater variety of prod-ucts in the market. Both effects have been intensified by the introduction of the Euro-pean Single Market, which has enabled EuroEuro-pean insurance firms to operate throughout the EU under a single license. The German government’s promotion of the private old-age provision has also resulted in the creation of new insurance products.

* Lucinda Trigo-Gamarra, Institute of Economics, University of Rostock, Germany. Christian Growitsch, Univer-sity of Cologne and Martin Luther UniverUniver-sity Halle-Wittenberg, Germany.

These developments were supposed to have a strong impact on the structure of the German life insurance firms’ distribution systems. These systems had been dominated by exclusive, firm-owned agents who were tied to one or several insurance firms. The increased price competition was also expected to increase lower-cost direct distribution channels that did not use agents to distribute their products. These new channels are underpinned by the increasing number of internet users, and particularly by the increasing insurance sales via the internet. The anticipated increased product variety raised the expec-tation that distribution by independent insurance brokers would become more important in the German market, since these agents can compare many insurance products and are free to choose the products they sell and the companies with which they work. Indepen-dent insurance agents and brokers work predominantly on behalf of their customers and deliver better service quality to their customers (see, e.g., Finsinger and Schmid (1993)). Since market liberalization, both direct distribution and distribution via independent agents have gained importance and distribution via exclusive agents has decreased. However, specialized insurance firms that use only direct distribution or independent agents (single-channel insurers) have experienced only a small increase in their market shares. For German life insurance companies, most of the firms that have traditionally distributed their products through dependent or exclusive agents now use a multichannel distribution strategy comprising exclusive agents, independent agents, and brokers, direct distribution, and distribution via bank offices. Approximately 85% of life insurance firms in Germany use a multichannel approach (M-type insurers) that combines at least two channels. These firms are mainly comprised of exclusive and independent agents or insur-ance brokers. In contrast, very few specialized life insurinsur-ance firms in the German market use only a single distribution channel. Direct (D-type) insurers distribute their prod-ucts without the use of salespeople, and independent agency insurers (I-type) distribute exclusively through independent agencies and insurance brokers_{. In absolute terms, the}

number of D-type insurers on the German life insurance market remained relatively stable over the observation period from 1997-2005, with eight direct life insurers in 1997 and nine in 2005. There were ten I-type insurers: one in the market in 1997 and nine in 2005. Thus, the distribution of life insurance products is dominated by M-type firms; specialized single-distribution insurers have only small market shares. Within our sample, premium income by direct insurers was 3.3% percent in 1997 and increased to 4.3% in 2005; among independent agency insurers, the premium income was 4.5% in 1997 and increased to 5% percent in 2005. The remaining premium income is generated by multichannel insurance firms. This development shows the predominance of multichannel distribution compared to insurers that use specialized distribution systems.

In this paper, by comparing the performance of both distribution systems, we are able to analyze the reasons for the development of the specialized and M-type market shares in the German life insurance market. According to previous studies, specialized suppliers should be superior to M-type insurers if they are able to realize either cost advantages (D-type) or 1 There is only one life insurance firm in the German market that distributes its products only through bank

of-fices, and only one insurer that sells its products only through exclusive agents. Due to these very small subpop-ulation sizes, we excluded both insurers from the sample.

quality advantages (I-type). To our knowledge, we are the first to compare different types of single- and M-type insurers, and to add a new facet to the discussion of the coexistence of different distribution channels in insurance markets.

The paper is organized as follows: In Section 2, we give a brief overview of earlier studies. In Section 3 we present our hypotheses. In Section 4 we illustrate our methods and modeling approach. In Section 5 we describe the data and the estimation model, and in Section 6 we present the results of our efficiency estimations. Section 7 concludes. 2 literature review

Several empirical studies focus on the coexistence of different distribution systems. However, most of these studies only compare exclusive agency insurers with I-type insurers. Joskow (1973) finds that American insurers working with independent agents incur much higher costs than do insurers using exclusive agents. Cummins and VanDerhei (1979) and Barrese and Nelson (1992) also find that independent agency insurers incur higher underwriting costs. However, none of these three studies compares the (average) profit levels of both systems. Barrese, Doerpinghaus, and Nelson (1995) incorporate a quality dimension into their analysis by using private passenger automobile insurance complaint data as a proxy for service quality. According to their study, American inde-pendent agency insurers in the private passenger automobile insurance line provide higher service quality compared to exclusive agency insurers. Berger, Cummins, and Weiss (1997) analyze a sample of 472 U.S. insurers and conclude that exclusive agency insurers are more cost efficient, but that this performance advantage disappears when they take revenues into account. Brockett et al. (2005) also find that U.S. property-liability I-type insurers are more revenue efficient than is a second group comprised of exclusive-agent and D-type insurers. Klumpes (2004) analyzes a sample of U.K. life insurance firms and estimates cost and profit efficiency levels. In contrast to Berger, Cummins, and Weiss (1997), Klumpes finds that I-type insurers are both less cost efficient and less profit efficient compared to exclusive agency insurers. But in his analysis of the performance of different distribution systems in the U.S. life insurance industry for the period 1988-1995, only Cummins (1999) compares insurers that rely exclusively on the agency system with those that mainly use the agency system. He finds that compared to agent-based insurers, D-type insurers are less cost efficient and revenue efficient, but technically more efficient, and that M-type insurers tend to be more efficient.

3 multiChannel Compared to Single-Channel diStribution SyStemS: hypotheSeS

M-type insurers: The use of multiple channels allows insurance firms to extend their market coverage by addressing different customer groups (Coelho and Easingwood (2004)). More-over, the use of multichannel distribution may be better able to meet the needs of existing customers (Tsay and Agrawal (2004)), because existing customers can purchase the firm’s products through the channel that best suits them. Further, the use of additional channels

may prevent the loss of market share to new rivals that enter the market at low prices via specialized channels. Kumar and Ruan (2006) find that the addition of a direct channel (i.e., an online channel or an exclusive agency channel) may help increase the level of support from existing independent retail channels (i.e., independent agencies or brokers). And customers can switch the distribution channel if their preferences change, an advan-tage that prevents possible cannibalization effects.

However, there are also potential disadvantages to the use of multiple channels. Cost disadvantages can arise because of the high investment costs of establishing an addi-tional distribution channel and of coordinating between the channels (Easingwood and Storey (1996)). Channel cannibalization may also occur: instead of increasing turnover and profits, additional channels simply redirect business from one channel to another.

D-type insurers: Because they save on the costs of establishing a distribution network, constructing their own branches, and so on, D-type insurers can provide their services at lower costs compared to insurance firms that use agents, bank branches, and other third parties to distribute their products. However, complex insurance products are more difficult to sell without the personal advice of an intermediary or staff member at a branch office. Even with less complex products, direct insurers will not acquire customers who prefer a different distribution system. Thus, the growth of direct life insurance firms could be limited. Limited growth in a highly competitive market, combined with high investments for market entry could further prevent a new entrant from realizing possible economies of scale (e.g., lessening of fixed administration and acquisition costs). If direct insurers cannot realize a sufficient firm size, they may not be able to realize the expected cost advantages.

I

-type insurers: Independent agents usually receive high commissions from insurance firms. Such commissions are intended to ensure that the agent places the most profitable busi-ness with that particular firm2_{. However, insurance brokers can compensate for their high}

cost levels by offering a higher level of service quality compared to other distribution chan-nels (see, e.g., Berger, Cummins, and Weiss (1997)). From the insurers’ perspective, using independent agents enables the insurers to write more profitable business (Anderson, Ross, and Weitz (1998)). From the clients’ point of view, the higher quality of service from inde-pendent agents results in a reduction in search costs (Posey and Tennyson (1998)), better information and a better market overview, increased confidence, and better monitoring of the insurer for example, on appropriate coverages, low prices, and financial stability (Regan (1997)). Further, Mayers and Smith (1981) and Barrese and Nelson (1992) also observe that independent agents are better able to deal with insurers when there is a conflict between a policyholder and the insurance company, because they can threaten to move the client to another insurer. Thus, distribution via independent agents is profit-able if this channel can generate the higher revenues that could result from a higher level of quality.

2 We note that in contrast to the U.S. American insurance market for property-liability insurance, German in-dependent insurance agents do not own the client list, i.e., they do not make the decisions on contract renew-al, rather, it is the insurance firm that decides on the renewal of the contract. Nevertheless, independent agents usually receive high commission levels from the insurance firms to ensure that the independent agent places the most profitable clients with them.

From our theoretical considerations, we derive several hypotheses to compare direct insurers with multichannel insurers.

H1: D-type insurers are more cost efficient than M-type insurers.

We compare the second single-channel distribution strategy – I-type insurers – to multi-channel insurers by testing two hypotheses:

H2.1: Compared to M-type insurers, I-type insurers are less cost efficient because of the higher costs of the independent agency system.

H2.2: I-type insurers recoup the disadvantage of M-type insurers in terms of cost efficiency with higher revenues resulting from higher service quality, which lead to similar or higher levels of profit efficiency.

If we find evidence for these hypotheses, then specialized single-distribution strategies should be superior to broader M-type distribution systems. By focusing either on a cost or a quality advantage, D-type and I-type insurers will outperform M-type insurers. In contrast, if we must reject both hypotheses, then the advantages of a M-type distribution system will outweigh its disadvantages. A broad multi-distribution strategy would then be superior to single-channel distribution strategies, and would explain why specialized single-distribution channel insurers have not gained a larger market share.

4 method

To determine if specialized single-distribution strategies are superior to broader M-type distribution systems, we apply frontier efficiency analysis to estimate cost and profit efficiency in the German life insurance industry. This approach allows for the analysis of multiple input-output technologies. We measure the performance of each firm by comparing it to the efficient frontier of the industry, which comprises the efficient firms in the reference set (e.g., the industry). Thus, we can obtain firm-specific efficiency measures relative to a “best practice” frontier. We estimate firm-specific efficiency by applying nonparametric Data Envelopment Analysis (DEA). When we use DEA, we do not need an a priori specification of the underlying production function, because we can estimate the efficient best practice frontier by solving linear programming models to envelop the observed data as tightly as possible (Charnes, Cooper, and Rhodes (1978)). Doing so requires only that the production possibility set be convex and that the inputs and outputs be disposable.

Taking into account input price information we can determine a firm’s cost efficiency (CE). A firm is fully cost efficient if it is able to produce a given output y0 at minimum cost. We estimate cost efficiency as follows: Using data on K inputs and M outputs for each of the firms (i = 1, ..., N), the ith _{firm uses an input vector xki (k = 1, …, K)} to produce an output vector ym0 (m = 1, …, M). The vector xkirepresents the mone-tary input quantities, i.e., costs. Tone (2002) and Cooper, Seiford, and Tone (2006) label

this approach as “new” cost efficiency. Their focus differs from ours, in that they account for different input prices by considering wki,which is an input price vector for the ithfirm. However, this approach can also be used when input prices are either not available or only partially available if information about costs is present, as in our case. The (“new”) cost efficiency scores contain both technical and allocative inefficiencies, since the firm’s deci-sion about the optimal use of input factors, depending on the given input prices, is already contained in the cost information_.

We solve the following linear programming problem (LP):

Ci* = Min _{λ, x} k

∑

k = 1 K wki xki, (1) s.t.

∑

i = 1 N λi xki ≤ xk,

∑

i = 1 N λi ymi ≥ ym0, λi ≥ 0.

In equation (1), λi is a N*1 vector of constants and represents the weights of a compa-ny’s inputs and outputs to project a cost efficient combination of both. The constraints in equation (1) ensure that the calculated cost-minimizing input-output combination lies within the defined technology set.

We calculate the cost efficiency of the ith _{firm as the ratio of minimum cost to observed} cost, as shown in equation (2). Here, x *ki is the minimum cost vector for theithfirm obtained by the LP in equation (1).

CE =

∑

k = 1 K wki xki* ________

∑

k = 1 K wki xki ; 0 ≤ CE ≤ 1. (2)

The measure of cost efficiency is bounded between zero and one. A value of one represents a fully cost-efficient firm: 1-CE represents the amount by which the firm could reduce its costs and still produce at least the same amount of output. The LP approach calculates CE

under the assumption of constant returns to scale (CRS). To calculate CE under variable 3 We do not derive firm-specific allocative inefficiencies because the difference between them is of only minor

im-portance for the purpose of our study. Recently, Banker, Chang, and Natarajan (2007) derived a consistent esti-mator of aggregate technical and allocative efficiency by using aggregate cost and revenue data. They also show how technical and allocative efficiency can be disentangled in this case.

returns to scale (VRS), we add a convexity constraint

∑

i = 1

N

λi = 1 (Banker, Charnes, and

Cooper (1984)). The CRS assumption says that all firms are operating at optimal scale, i.e., under minimum average costs. Under VRS, firms may exhibit increasing or decreasing returns to scale, so we take into account possible (dis-)economies of scale when calcu-lating cost efficiency. We calculate a firm’s scale efficiency (SE) by dividing the CRS effi-ciency score by the VRS effieffi-ciency score. Doing so makes it possible for us to determine the amount by which a firm’s efficiency could be improved by moving to its optimal scale (see, e.g., Ray (2004)).

If we consider that output quantities are also choice variables, then we can calculate profit efficiency if information about both input and output prices is available. But this is not the case here. However, information about revenues – the product of output quan-tities and prices – is available. Therefore, we can calculate Cooper, Seiford, and Tone’s (2006) “new” profit efficiency. For this calculation, the output is represented by the vector

ymi, where ymi represents the revenues of a firm i. By technically assuming a constant output price of one, we use the revenue information to model the revenue term of the profit optimization process. The input is represented by a vector xki with xki, which repre-sents the monetary input quantities, i.e., the costs. For outputs, we set input prices wki to one. We estimate profit efficiency under the assumption of VRS, because under the assumption of CRS, maximum profit would be zero or undefined (see, e.g., Ray (2004)). The fact that profit efficiency cannot be calculated under the assumption of CRS meets our assumption of imperfect markets. Although technically all firms face the same input and output prices of one, our profit efficiency model allows for the different output prices that can result from product differentiation among firms, because this information is contained in the revenues.

Profit efficiency is the ratio between a firm’s actual profits and the maximum attainable profits, given the input and output prices.

We solve the profit maximization LP as follows:

Pi* = Max _{λ, x} k, ym _{m = 1}

∑

M pmi ymi –

∑

k = 1 K wki xki, (3) s.t.

∑

i = 1 N λi xki ≤ xk,

∑

i = 1 N λi ymi ≥ ym, λi ≥ 0,

∑

i = 1 N λi = 1.

All variables are as before. P*

i represents the maximum profit of firm i and pmi

is the vector of output prices for the ith _{firm. We can obtain a measure of profit efficiency} by calculating the ratio of observed profit to maximum (potential) profit.

PE = m = 1

∑

M pmi ymi –

∑

k = 1 K wki xki _______________________

∑

m = 1 M pmi y *mi –

∑

k = 1 K wki x *ki ; – ∞ ≤ PE ≤ 1, (4) where y*

mi is the revenue-maximizing vector of output quantities for the ith firm. Given input and output prices, we calculate x *ki and y *mi by the LP presented in equation (3) so that – ∞ ≤ PE ≤ 1 describes the maximum amount by which the profits of an inef-ficient firm can be increased until it achieves full profit efficiency. A fully profit-efinef-ficient firm shows a profit efficiency of one. Just as profits can be negative, profit efficiency is not bounded by zero at the lower end, but can turn negative (zero) if profits are nega-tive (zero)_.

5 data Set and variableS

5.1 Data Set

We obtain the data used in this study from insurance industry reports and insurers’ income statements for the years 1997-2005 (Hoppenstedt (1999-2007)). Hoppenstedt registers every licensed insurance firm in Germany, so the database also contains informa-tion about firms that do not actively participate in the insurance market. We eliminate firms that had either not delivered any information at all, or that had negative observa-tions for inputs or outputs. We also remove firms that operate in only very specialized product niches, i.e., those that offer products to a very specialized customer base, such as civil servants or doctors, or offer only single, specialized insurance products, such as term life insurance, because such firms are not representative of the industry as a whole. Thus, our data set accounts for approximately 90% of the industry’s total premium income. The German life insurance industry is characterized by great heterogeneity among firms, so we correct for outliers in the sample by applying the outlier correction model suggested

4 To solve this problem, some authors (e.g., Banker and Maindiratta (1988)) suggest that the researcher eliminates firms that exhibit negative profits before calculating efficiency scores. We choose not to remove such firms from the sample, because it is possible for firms to incur short-term losses but still be able to establish themselves in the market in the long run. This argument is especially true for young firms, which incur high initial investments. Our sample contains several firms that entered the market after the liberalization of the German insurance mar-ket in 1994. Since we found only a very few firms that showed only small negative PE scores, and because none of these firms showed negative profit efficiency scores over the whole observation period, the impact on the av-erage profit efficiency scores is very small.

by Wilson (1993)_{. We find that, in each year, the firms detected as outliers were among} the largest in the sample.

However, we note that the results of the efficiency estimations differ only slightly if we do not exclude the detected outliers from the sample and that all of the qualitative results of the study remain unchanged. This stability of results may be explained by the fact that all the excluded firms are M-type insurers. Thus, we exclude only firms that belong to the largest subsample of the data set. The results of the comparison of the different subgroups are not changed significantly by the exclusion of the outliers.

5.2 VariableS

Using DEA requires that we identify the relevant inputs and outputs of an insurance firm. A review of the relevant literature does not show a clear consensus on a single input/ output specification, so in this study we use the value-added approach, which is common in the insurance economics literature (see, e.g., Cummins and Weiss (2000)). According to the value-added approach, the main services provided by life insurers can be categorized in three major groups:risk-bearing/risk-pooling services, “real” financial services related to financial planning and counseling for individuals, and pension and benefit plan adminis-tration for businesses and intermediation services. We approximate the risk-bearing func-tion by using incurred benefits net of reinsurance. Incurred benefits represent payments received by policyholders in the current year. These factors serve as proxies for the risk-bearing/risk-pooling function, because they measure the amount of funds pooled by the insurer and redistributed to the policyholders as benefits. The funds received by insurers that are not needed for benefit payments and expenses are added to policyholder reserves, so additions to reserves is a good proxy for the intermediation function of the insurer (see, e.g., Cummins, Tennyson, and Weiss (1999); Fenn at al. (2008)). Finally, we include

bonuses and rebates in our output measure, because these funds benefit the policyholders_. All three output measures are correlated with real services provided by life insurers, and are summed up into one output. We measure life insurers’ revenue by the sum of premium and investment income (see, e.g., Fenn et al. (2008)). We measure premium income by the sum of gross written premiums, less ceded reinsurance premiums, less the change in the provision for unearned premiums. We classify insurers’ inputs into three principal groups: labor, business services and materials, and capital. In most studies, physical meas-ures for these inputs (e.g., the number of employees) are not available, but there is infor-mation about the costs an insurance firm incurs for using such inputs. Since these inputs are already valued by the corresponding input prices, they represent the product of input quantities and prices. By using the “new” cost/“new” profit efficiency approach (Tone (2002); Cooper, Seiford, and Tone (2006)), we are able to take cost measures into account directly. Most studies derive input quantities by dividing cost values by a uniform price/ 5 Wilson (1993) extends a method developed by Andrews and Pregibon (1978) that allows for the detection of

outliers without basing the analysis on residual analysis. For DEA models, the frontier is not parameterized, and hence diagnostics based on parameter estimation cannot be used.

6 We tested for the influence of this output measure by leaving it out and re-estimating cost and profit efficiency levels. Our results proved to be robust and did not differ significantly between both models.

wage index over all firms, which leads to the same CE values as our approach (Färe and Grosskopf (1985) and Banker, Chang, and Natarajan (2007)). To measure insurers’ costs, we choose acquisition and administration expenses, which contain expenses for labor and business services and sum up to equal operating expenses, as a proxy for the insurers’ inputs for labor and business services (see, e.g., Berger, Cummins, and Weiss (1997)). Financial capital is also important in considering insurance firms. We use equity capital as an input because insurance is viewed as risky debt (see, e.g., Cummins and Danzon (1997)). We follow the majority of insurance studies by using statutory policyholders’ surplus

(a term that corresponds to the position “Eigenkapital” in German insurance balance sheets) to represent financial equity capital. To measure the cost of equity, we value finan-cial equity capital by the firm-specific price for equity capital. Because of limited data availability and the minor influence of the different approaches on the efficiency results found in other studies, we assume identical prices for equity capital over all firms in a given year. Thus, to approximate the price for equity capital in the industry, we calculate the average return on the book value of equity for the industry in a given year_{. We obtain} equity costs by valuing statutory policyholders’ surplus with the year’s price for equity capital.

Table 1 presents summary statistics for the variables that we use in the analysis as mean values for the whole observation period.

The descriptive statistics show a large dispersion for all the variables between the smallest and largest firms in the sample, as well as among the three groups of insurers. D-type insurers show the smallest average values in terms of operating expenses, outputs, and revenues. . In terms of equity costs, I-type insurers show slightly lower values compared to direct insurers. In general, I-type insurers have higher costs, outputs, and revenues. The largest group of firms is the M-type insurers, which show an output 3.89 times larger than D-type insurers, and an output 3.22 times larger compared to I-type insurers. The differ-ences between these groups are also apparent in terms of costs and revenues.

6 reSultS

Tables 2 and 3 report the average cost efficiency, scale efficiency, and profit efficiency scores of a common cost and a common profit frontier for the three groups of insurers that we analyze. To compare the efficiency scores of the subgroups in the sample, we use the nonparametric Mann-Whitney-U test_{. We start with the comparison of D-type and} M-type insurers before turning to the I-type insurers.

7 Similar approaches can be found in the insurance literature, e.g., Fenn et al. (2008), who use a rate-of-interest variable from long-term government bond rates to proxy the price of capital. Cummins and Rubio-Misas (2006) use the rate of total return of the Spanish Stock Exchange as a proxy for the price of equity capital for every year in their observation period. The return on the book value of equity has been used by Cummins and Weiss (1993), and Cummins and Sommer (1996).

8 As a robustness check, we also applied a Kolmogorov-Smirnov type test to test for differences between the differ-ent groups (Banker (1993)). With few exceptions, the results are iddiffer-entical to the Mann-Whitney-U test results, and all the qualitative results remain the same.

Multichannel insur er D irec t insur ers Independen t agen t insur ers Total m ean (S td . dev .) m in m ax m ean (S td . dev .) m in. m ax. m ean (S td . dev .) m in. m ax. m ean (S td . dev .) m in. m ax. o per ating expenses 110535.61 _(155243.13) 574.00 1229534.54 19519.96 _(22421.85) 420.00 75868.48 60305.65 _(64981.70) 98.00 285024.00 93813.20 (140864.06) 106.72 1229533.54 equit y c osts 11576.26 _(18416.29) 199.25 178698.56 5178.82 (12260.31) 205.53 74515.80 3857.63 _(4762.82) 388.83 28514.04 65844.99 (102159.21) 1268 1101824 o utput 910225.90 (1277991.26) 2878.66 13501076.00 234039.20 _(346108.61) 348.24 1270142.40 282471.89 _(384297.68) 985.49 158009.67 753032.03 (1160998.24) 348.24 13501076.00 Pr emiums (net of r einsur anc e) 641775.05 _(868401.63) 2233.81 8732933.42 204650.16 _(275228.97) 408.87 936858.00 235401.79 _(291156.79) 986.10 1284288.55 540083.90 _(789560.38) 408.87 8732933.42 in vestmen t inc ome 441224.71 _(711351.35) 3468.31 9366708.21 99740.32 (162462.63) 22.55 625372.01 125091.12 _(178460.80) 357.95 689348.88 361954.88 _(640566.11) 22.55 9366708.21 a ll v ar iables ar e e xpr essed in year 2000 thousand eur o units b y defla

ting with the

g er man C onsumer P ric e inde x.

Table 2: Comparison of average cost and scale efficiency scores by groups, 1997-2005 Multichannel insur ers D irec t insur ers Independen t agen t insur ers n CE CRS CE VRS SE n CE CRS CE VRS SE n CE CRS CE VRS SE 1997 64 0.3210 _(0.1269) 0.4436 _(0.2213) 0.7736 _(0.2069) 9 0.2383* _(0.3085) 0.4790 _(0.3234) 0.4895* _(0.4190) 10 0.1857* _(0.1289) 0.3475 _(0.2634) 0.6577* _(0.3198) 1998 69 0.3480 _(0.2858) 0.4610 _(0.2270) 0.8011 _(0.1967) 11 0.2185* _(0.2859) 0.4178 _(0.2781) 0.4785* _(0.3974) 12 0.1813* _(0.1247) 0.3136* _(0.2564) 0.6884 _(0.2739) 1999 71 0.2854 _(0.1057) 0.4141 _(0.2190) 0.7692 _(0.2238) 10 0.2504* _(0.2873) 0.4342 _(0.3073) 0.5562* _(0.3579) 12 0.1648* _(0.1009) 0.2906* _(0.2520) 0.7210 _(0.2705) 2000 65 0.4089 _(0.1960) 0.4905 _(0.2384) 0.8416 _(0.1431) 10 0.3021* _(0.2921) 0.4548 _(0.2860) 0.6125* _(0.3309) 12 0.1755* _(0.1499) 0.2810* _(0.2730) 0.7509 _(0.2600) 2001 62 0.3972 _(0.1827) 0.4738 _(0.2216) 0.8603 _(0.1697) 8 0.3807 _(0.3178) 0.5753 _(0.3226) 0.6397 _(0.3799) 11 0.1997* _(0.1380) 0.2467* _(0.1441) 0.7810 _(0.2296) 2002 61 0.4491 _(0.1863) 0.4838 _(0.2018) 0.9288 _(0.1586) 9 0.4012 _(0.2611) 0.5376 _(0.3056) 0.7473* _(0.2990) 10 0.2009* _(0.1594) 0.2232* _(0.1565) 0.8322* _(0.1989) 2003 60 0.4889 _(0.1965) 0.5437 _(0.2140) 0.9071 _(0.1343) 9 0.5104 _(0.2637) 0.6631 _(0.2728) 0.7731 _(0.2725) 11 0.2868* _(0.1739) 0.3255* _(0.1602) 0.8514 _(0.1667) 2004 62 0.4123 _(0.1690) 0.4981 _(0.2166) 0.8477 _(0.1377) 9 0.4380 _(0.2976) 0.5820 _(0.2705) 0.7141 _(0.3371) 10 0.2410* _(0.1100) 0.2860* _(0.1104) 0.8467 _(0.1589) 2005 60 0.4324 _(0.2000) 0.5154 _(0.2443) 0.8607 _(0.1484) 10 0.3642 _(0.2616) 0.5568 _(0.2296) 0.6525 _(0.3601) 10 0.2679* _(0.1196) 0.3525* _(0.1633) 0.7884 _(0.2373) * d iff er en ce s b et w ee n effi ci en cy sc or es a re st at is tic al ly si gn ifi ca nt b et w ee n gr ou ps , a cc or di ng to th e m an n-W hi tn ey -U -t es t. W e te st m ul tic ha nn el in su re rs against dir ec t and independen t agen t insur ers . d etailed t est r esults ar e a vailable fr om the c or responding author on r equest .

Table 3: Comparison of average profit efficiency scores by groups, 1997-2005

Multichannel insurers Direct insurers Independent agent insurers

n PEVRS _{n PE}VRS _{n PE}VRS 1997 64 0.5888 (0.2835) 9 0.5617 (0.3437) 10 0.7099 (0.2521) 1998 69 0.6396 (0.2605) 11 0.4048* (0.3974) 12 0.4643 (0.3887) 1999 71 0.6307 (0.2633) 10 0.4764 (0.3618) 12 0.4681 (0.4082) 2000 65 0.6390 (0.2733) 10 0.5711 (0.3840) 12 0.3896* (0.3603) 2001 62 0.6088 (0.2734) 8 0.6470 (0.4106) 11 0.3341* (0.2684) 2002 61 0.6469 (0.2437) 9 0.6567 (0.3664) 10 0.3664* (0.2975) 2003 60 0.6565 (0.2552) 9 0.7883 (0.2608) 11 0.3494* (0.2516) 2004 62 0.6164 (0.2495) 9 0.6735 (0.3932) 10 0.4243* (0.2198) 2005 60 0.6274 (0.2702) 10 0.6189 (0.3661) 10 0.3897* (0.1925) * differences between efficiency scores are statistically significant between groups, according to the

mann-Whitney-U-test. We test multichannel insurers against direct and independent agent insurers. detailed test results are available from the corresponding author on request.

D-type insurers show lower cost efficiency under the assumption of constant returns to scale (CECRS_{), compared to M-type insurers. The differences between the groups are} significant until the year 2000, after which the analysis of cost efficiency under variable returns to scale (CEVRS_{) shows that the differences in CE between the groups eventually} disappear. By the end of the observation period, D-type insurers even show slightly higher efficiency scores than do M-type insurers. Hence, D-type insurers show much lower scale efficiency levels in most years, indicating that they have not reached their optimal size. This scale inefficiency is underlined by the fact that most D-type insurers operate under increasing returns to scale, i.e., they are not operating at their optimal size.

From our results, we conclude that hypothesis H1 must be rejected, since D-type insurers do not show the expected cost advantage compared to M-type insurers. This result seems to be due to their low scale efficiency, which prevents them from realizing their cost advan-tages. Although D-type insurers are able to recoup some of their cost disadvantages over time, we can assume that compared to M-type insurers, they have not yet reached a

suffi-cient firm size to realize their theoretical cost advantages. Differences in profit efficiency under variable returns to scale (PEVRS_{) between the groups are small and} nonsignifi-cant, with the exception of 1998. In terms of CEVRS_{, the relation between the groups} translates into PE, so there appear to be no systematic differences in the service quality of the groups.

We explain the limited growth of D-type insurers as resulting from two factors. First, the nature of most life insurance products is complex, so life insurance products can be regarded as counselling-intensive products. Since direct insurers do not provide their customers with personal advice, customers might prefer to rely on multichannel insurers for life insurance products while using direct insurers primarily for the purchase of more standardized products. For example, term life insurance is a more standardized, less complex, insurance service. Further, our data set shows that on average, the share of term life insurance policies in direct insurers’ portfolios is larger than that in M-type insurers’ portfolios. A second reason for the limited growth of D-type insurers may be that M-type insurers are increasingly adopting direct distribution as an additional distribution channel, so customers who are willing to use direct distribution channels do not neces-sarily need to switch to a D-type insurer. This behaviour underscores the importance of reputation in insurance markets: insurance products are credence (trust) goods. Since D-type insurers are mainly young firms founded after the liberalization of the market they have not yet built up a long-term reputation, in contrast to established M-type insurers. Thus, customers might prefer to use additional channels of an established M-type insurer instead of switching to a D-type insurer.

When we compare M-type insurers and I-type insurers over the whole observation period, we find that under both CRS and VRS assumptions, I-type insurers are significantly less cost efficient. According to the theoretical considerations presented in Section 3, this result is to be expected, because distribution via independent agents incurs higher costs. Thus, we cannot reject hypothesis H2.1. Compared to M-type insurers, I-type insurers also show lower scores in cost efficiency, but the differences are much smaller than they are in the case of D-type insurers and are significant for only two years within the obser-vation period.

We find that I-type insurers show lower average profit efficiency scores over the whole observation period, and that from 2000 on, the differences are statistically significant. Thus, we reject H2.2: Compared to M-type insurers, I-type insurers do not show similar or even higher profit efficiency levels, so they are not able to recoup their higher costs with higher revenues. However, this result does not suggest that independent agents would not be able to provide their customers with higher service quality; it states only that the specialized distribution system of I-type insurers is not superior in terms of either costs or of average revenues, to distribution via multiple channels. A possible explanation may also be that differences in service quality are not rewarded. The differences in profit effi-ciency between both groups have increased since the beginning of the observation period. This increase could indicate that specialized I-type insurers have lost part of their client base over time because of the increasing importance to multichannel insurers of distribu-tion by independent agents. Thus, insurance clients who want to make use of the

serv-ices of independent agents are no longer limited to the product range of insurance firms that work exclusively with independent agents; increasingly they have the opportunity to purchase products from M-type insurers.

7 ConCluSionS

Our analysis of the performance of single- and multichannel distribution firms in the German life insurance explains the structure of the industries’ distribution systems. In Germany life insurance sales are dominated by M-type firms and specialized single-distri-bution insurers have only small market shares.

We apply an empirical framework developed by Berger, Cummins, and Weiss (1997) to estimate cost and profit efficiency for three sample groups of German life insurance firms with different distribution systems, M-type, D-type, and I-type insurers. Testing a set of hypotheses, we find that the absence of performance advantages for specialized insurers provides economic evidence for the coexistence of the different distribution systems. Our results show that contrary to theoretical predictions, single-channel insurers do not outper-form M-type insurers in terms of either cost or profit efficiency and, thus, do not repre-sent a superior distribution system: D-type insurers are not able to realize their expected cost advantage over M-type insurers, I-type insurers are unable to take advantage of their hypothesized service superiority This result explains why their market share has remained small despite the increasing importance of direct distribution and the increasing use of independent-agent insurers in the German life insurance market.

Our results also explain the developments in the distribution systems of the German life insurance industry after its liberalization in 1994. The dominance of exclusive agents that prevailed in the German life insurance industry until 1994 has declined in favor of distribution via direct channels and independent agents. However, it is not the special-ized D-type and I-type insurers that have been the primary beneficiaries of this decline in exclusive agency, but the M-type insurers. M-type insurers have benefited by incorpo-rating additional channels into their distribution systems. Thus, we conclude that distri-bution via multiple channels is superior to specialized distridistri-bution systems in the life insurance industry. We suggest that future research should focus on a closer examination of the question of how different insurance products are sold via different channels within a M-type insurance firm.

referenCeS

Anderson, Erin, William T. Ross, and Barton Weitz (1998), Commitment and Its Consequences in the American Agency System of Selling Insurance, Journal of Risk and Insurance 65, 637-669.

Andrews, David F. and Daryl Pregibon (1978), Finding the Outliers that Matter, Journal of the Royal Statistical Society, Ser. B, 40, 85-93.

Banker, Rajiv D. (1993), Maximum Likelihood, Consistency and Data Envelopment Analysis: a Statistical Foun-dation, Management Science 39, 1265-1273.

Banker, Rajiv D., Hsihui Chang, and Ram Natarajan (2007), Estimating DEA Technical and Allocative Inefficiency Using Aggregate Cost or Revenue Data, Journal of Productivity Analysis 27, 115-121.

Banker, Rajiv D., Abraham Charnes, and William W. Cooper (1984), Some Models for Estimating Technical and Scale Inefficiencies in Data Envelopment Analysis, Management Science 30, 1078-1092.

Banker, Rajiv D. and Ajay Maindiratta (1988), Nonparametric Analysis of Technical and Allocative Efficiencies in Production, Econometrica 56, 1315-1332.

Barrese, James, Helen I. Doerpinghaus, and Jack M. Nelson (1995), Do Independent Agent Insurers Provide Superior Service? The Insurance Marketing Puzzle, Journal of Risk and Insurance 62, 297-308.

Barrese, James and Jack M. Nelson (1992), Independent and Exclusive Agency Insurers: a Reexamination of the Cost Differential, Journal of Risk and Insurance 49, 375-397.

Berger, Allen N., John D. Cummins, and Mary A. Weiss (1997), The Coexistence of Multiple Distribution Systems for Financial Services: the Case of Property-Liability Insurance, Journal of Business 70, 515-546.

Brockett, Patrick L., William W. Cooper, Linda L. Golden, John J. Rousseau, and Yuying Wang (2005), Financial Intermediary versus Production Efficiency of Marketing Distribution Systems and Organizational Structure of Insurance Companies, Journal of Risk and Insurance 72, 393-412.

Charnes, Abraham, William W. Cooper, and Edwardo Rhodes (1978), Measuring the Efficiency of Decision-making Units, European Journal of Operational Research 2, 429-444.

Coelho, Filipe J. and Chris Easingwood (2004), Multiple Channels in Services: Pro, Cons and Issues, The Service Industries Journal 24, 1-29.

Cooper, William W., Lawrence M. Seiford, and Kaoru Tone (2006), Introduction to Data Envelopment Analysis and Its Uses, New York et al.: Springer.

Cummins, John D. (1999), Efficiency in the U.S. Life Insurance Industry: Are Insurers Minimizing Costs and Maxi-mizing Revenues?, in John D. Cummins and Anthony M. Santomero (eds.), Changes in the Life Insurance Industry: Efficiency, Technology, and Risk Management, Boston et al.: Kluwer.

Cummins, John D. and Patricia M. Danzon (1997), Price Shocks and Capital Flows in Liability Insurance, Journal of Financial Intermediation 6, 3-38.

Cummins, John D. and Maria Rubio-Misas (2006), Deregulation, Consolidation, and Efficiency: Evidence from the Spanish Insurance Industry, Journal of Money, Credit, and Banking 38, 323-355.

Cummins, John D. and David W. Sommer (1996), Capital and Risk in Property-Liability Insurance Markets, Journal of Banking and Finance 20, 1069-1092.

Cummins, John D., Sharon Tennyson, and Mary A. Weiss (1999), Consolidation and Efficiency in the U.S. Life Insurance Industry, Journal of Banking and Finance 23, 325-357.

Cummins, John D. and Jack VanDerhei (1979), A Note on the Relative Efficiency of Property-Liability Insurance Distribution Systems, Bell Journal of Economics and Management Science 10, 708-719.

Cummins, John D. and Mary A. Weiss (1993), Measuring Cost Efficiency in the Property-Liability Insurance Indus-try, Journal of Banking and Finance 17, 463-481.

Cummins, John D. and Mary A. Weiss (2000), Analyzing Firm Performance in the Insurance Industry Using Frontier Efficiency and Productivity Methods, in Georges Dionne (ed.), Handbook of Insurance, Boston et al.: Kluwer, 767-829.

Easingwood, Christopher and Christopher Storey (1996), The Value of Multi-Channel Distribution Systems in the Financial Services Sector, The Services Industries Journal 16, 223-241.

Färe, Rolf and Shawna Grosskopf (1985), A Nonparametric Cost Approach to Scale Efficiency, The Scandinavian Journal of Economics 87, 594-604.

Fenn, Paul, Dev Vencappa, Stephen Diacon, Chris O’Brien, and Paul Klumpes (2008), Market Structure and the Efficiency of European Insurance Companies: a Stochastic Frontier Approach, Journal of Banking and Finance

Finsinger, Jörg and Frank A. Schmidt (1993), Gebundener versus ungebundener Vertrieb, sbr 45, 216-226. Hoppenstedt (1999-2007), Versicherungsjahrbuch 1999-2007, Darmstadt: Hoppenstedt.

Joskow, Paul L. (1973), Cartels, Competition and Regulation in the Property-Liability Insurance Industry, Bell Journal of Economics and Management Science 4, 375-427.

Klumpes, Paul (2004), Performance Benchmarking in Financial Services: evidence from the UK Life Insurance In-dustry, Journal of Business 77, 257-273.

Kumar, Nanda and Ranran Ruan (2006), On Manufacturers Complementing the Traditional Retail Channel with a Direct Online Channel, Quantitative Marketing and Economics 4, 289-323.

Mayers, David and Clifford W. Smith, Jr. (1981), Contractual Provisions, Organizational Structure, and Conflict Control in Insurance Markets, Journal of Business 54, 407-434.

Posey, Lisa L. and Sharon Tennyson (1998), The Coexistence of Distribution Systems under Price Search: Theory and some Evidence from Insurance, Journal of Economic Behavior and Organization 35, 95-115.

Ray, Subhash C. (2004), Data Envelopment Analysis – Theory and Techniques for Economics and Operational Research, Cambridge: Cambridge University Press.

Regan, Laureen (1997), Vertical Integration in the Property-Liability Insurance Industry: a Transaction Cost Approach, Journal of Risk and Insurance 64, 41-62.

Tone, Kaoru (2002), A Strange Case of the Cost and Allocative Efficiencies in DEA, Journal of the Operational Re-search Society 53, 1225-1231.

Tsay, Andy A. and Narendra Agrawal (2004), Modeling Conflict and Coordination in Multi-Channel Distribution Systems, in David Simchi-Levi, David Wu, and Max Shen (eds.), Handbook of Quantitative Supply Chain Analysis, New York et al.: Springer.

Wilson, Paul W. (1993), Detecting Outliers in Deterministic Nonparametric Frontier Models with Multiple Ouputs,