The Efficiency of UK General Insurance Companies

Transcription

1 THE UNIVERSITY OF NOTTINGHAM Centre for Risk & Insurance Studies The Efficiency of UK General Insurance Companies Professor Stephen Diacon 24 October 2001 CRIS Discussion Paper Series 2001.III

2 The Efficiency of UK General Insurance Companies Professor Stephen Diacon Worshipful Company of Insurers Professor of Insurance Management Nottingham University Business School 24 October 2001 Professor Stephen Diacon Nottingham University Business School Jubilee Campus, Wollaton Road Nottingham NG8 1BB Tel +44 (0) ; Fax +44 (0) ;

3 The Efficiency of UK General Insurance Companies Abstract This paper explores the efficiency of UK specialist and composite insurers transacting general insurance business. The concept of efficiency concerns an insurer s ability to produce a given set of outputs (such as premiums and investment income) via the use of inputs such as administrative and sales staff and financial capital. An insurer is said to be technically efficient if it cannot reduce its resource usage without some corresponding reduction in outputs, given the current state of production technology in the industry. An exploration of the value-based technical efficiency of UK general insurers is undertaken by comparing the relative performance of 431 general insurers licensed in six European countries using data from Standard & Poor s Eurothesys database. The data has been made available on a (roughly) comparable basis as a result of the EU insurance Accounts Directive which only came fully into operation in The study uses the variable returns to scale formulation of the well-known data envelopment analysis to identify the locally-efficient and inefficient insurers within each country. A comparison is then undertaken for all insurers after adjusting for the impact of their local efficiency. The results for 1999 (the latest year of available data) indicate that UK general and composite insurance companies have the potential to be among the most efficient in Europe. On average, after adjusting for local inefficiencies, UK insurers demonstrate an average efficiency score of around 77% - which is substantially higher than the corresponding average figures for France (67%), Germany (70%), Italy (56%), The Netherlands (69%) and Switzerland (66%). However there is also evidence that many UK companies are not currently realising their potential for efficiency improvements in comparison with their European counterparts. 2

4 The Efficiency of UK General Insurance Companies Stephen Diacon 1. Introduction The implementation of the single insurance licence within the European Union in 1994 provided insurers licensed in the EU with the opportunity to transact general insurance business in any EU state, either by starting up a subsidiary or branch, or by direct sale. This EU-wide insurance market has also been extended to countries within the European Economic Area (such as Norway) and the European Free Trade Association (eg Switzerland). Thus companies transacting general insurance business now potentially compete in a European-wide market 1. It is not surprising therefore that there is growing interest and concern about the international competitiveness and efficiency of European insurers. This paper attempts to build on earlier attempts by Rai (1996), Donni & Fecher (1997) and Katrishen & Scordis (1998) to undertake an international comparison of the efficiency of companies transacting general insurance business, and to benchmark UK insurers - both large and small - against their European competitors. International comparisons of efficiency are crucial as efficiency is a relative concept: it is not possible to define some ideal level of efficiency instead, companies have to be compared with those that currently constitute best practice in the market (given the current state of production technology in the industry). The key contribution of this paper is therefore to identify the best practice companies operating in the European general insurance market, and then to benchmark all other insurers against these. It has been widely recognised that inter-firm differences in efficiency can arise due to environmental factors - some of which will be country-specific - and the paper also explores two aspects of such differences. First, the inter- and intra-country differences in average insurer efficiency (termed global and local efficiencies respectively) are examined to determine the extent to which local operating conditions and regulations 3

5 can cause insurers to be disadvantaged in particular local environments. Secondly, a regression analysis is undertaken to explore the differences in efficiency between firms in order to determine the extent to which such differences can be explained by organisation size, structure and risk. This paper is organised as follows. Section 2 provides a brief review of the issues involved in the measurement of value-based technical efficiency and discusses the problems in applying these measures to insurance and other financial services firms. Section 3 then describes the application of Data Envelopment Analysis (DEA) to insurance companies licensed to transact general insurance business in the six largest European countries (namely France, Germany, Italy, The Netherlands, Switzerland and the UK). The results are tabulated and reviewed in Section 4, and the final section provides some discussion and conclusions. 2. A Brief Overview of Efficiency and Measurement The concept of efficiency concerns an insurer s ability to produce a given set of outputs (such as premiums and investment income) via the use of inputs such as administrative and sales staff and financial capital. An insurer is said to be technically efficient if it cannot reduce its resource usage without some corresponding reduction in outputs, given the current state of production technology in the industry. Technical inefficiencies can arise from a variety of sources. Companies which are operating at an inappropriate size (either too large or too small) may display what are termed scale inefficiencies, while others may be utilising their inputs (or producing their outputs) in the wrong proportions ( mix inefficiencies ). All companies have to contend with environmental factors that may damage their ability to operate efficiently. Others may simply be badly managed. A detailed discussion of the potential sources of insurer inefficiency can be found in Cummins & Santomero (1999) and Cummins, Weiss & Zi (1999). 1 A brief review of the European insurance industry can be found in Rassam (1998) 4

6 Farrell Efficiency Traditionally analysts have analysed the efficiency of organisations (or decisionmaking units DMUs) by focussing on certain simple ratios such as labour productivity (output per unit of labour employed) or capital intensification (ie capital/labour). Similarly the efficiency of insurance firms has often been measured by key ratios such as the expenses and claims ratios, the solvency margin, and the return on invested assets. There are, however a number of problems associated a simplistic multiple-ratio analysis. First, it is generally impossible to identify bestpractice DMUs since it is unlikely that all ratios will point to the same firm/s. Secondly, if the ratios disagree, it may be difficult to decide in advance which ratio should be given most weight in order to compare DMUs. Thirdly, efficiency comparisons should properly be made on the basis of like-for-like so that inefficient firms are identified because, in some way, they are inferior to other similar DMUs. Fourthly, the traditional measures do not readily allow firms to identify the source of any inefficiency. More modern approaches to efficiency and benchmarking try to circumvent the problems associated with traditional methods by using frontier efficiency methodologies. Essentially, the various methods proceed by first identifying best practice frontiers (and the DMUs which lie nearest to these frontiers). The frontier represents the best performance that can be achieved using the currently available production technology. By definition, a DMU which is part of the frontier set uses a minimum amount of inputs in order to achieve any given level of contemporaneous outputs. The efficiency of each DMU can then be measured by comparing it to the frontier firms which are nearest to it. Perhaps the best-known such measure of efficiency is the input-orientated Farrell measure which defines a DMU s technical efficiency as the proportion by which all inputs need to be reduced in order to adopt the most efficient production (Farrell, 1957). The term technical efficiency refers explicitly to the utilisation of units of input to produce units of output without any allowance for input prices (to compute costs) or output prices (to compute revenues). Explicit allowance for input and output prices may enable an analysis of allocative (by exploring cost minimisation) and economic efficiency (by examining revenue maximisation): for further details see Coelli, Rao & Battese (1998, ch 7). 5

7 Some efficiency methodologies also provide recommendations (or projections ) on what specific input-reduction and/or output increases are required in order for any given DMU to become efficient. The widespread practice of using values measures (such as revenue, costs, capital) as proxies for the inputs and outputs of financial services firms raises questions about exactly what type of efficiency is being measured. Technical efficiency strictly requires inputs and outputs to be measured in units, while the optimal allocation of inputs/outputs in response to market prices (allocative efficiency) requires separate data on input and output prices and quantities - in order to compare the actual versus the cost-minimising (revenue maximising) level of inputs (outputs). However, the intangible nature of financial services output often means that no homogeneous unit of output can be identified (sometimes even conceptually) and output prices cannot be quality adjusted. Similarly inputs like capital can only exist in value terms, and the unit cost of capital is difficult to measure in firms that are not publicly quoted. Thus it should it borne in mind that the technical efficiencies measured in this paper are value-based rather than the more traditional units-based measures found in many nonfinancial efficiency studies. The term value-based is used to recognise that the technical efficiency measures are based on monetary values for inputs and outputs (using both cost/revenue flows, and capital stocks), but do not capture optimal choices in response to market prices. The Best Practice Frontier Two main alternative approaches are available to identify the best practice frontier and the firms in the frontier set. The econometric or parametric approach of stochastic frontier analysis (SFA) fits a pre-specified production function (with multiple inputs but usually only one output) and allows DMUs to be off the frontier either because they are inefficient or because of random shocks or measurement errors. Some SFA methods require the asymmetric probability distribution of the inefficiency component to be specified in advance, while others (such as the distribution free method of Berger, 1993) do not. In contrast, non-parametric methods such as data envelopment analysis use linear programming techniques to discover the frontier firms and construct a convex piece- 6

8 wise linear surface or frontier over these firms. In terms of technical efficiency, the most efficient firms are those with the largest value (over all firms) of the ratio of the weighted sum of outputs to the weighted sum of inputs, where the optimal weights for each firm maximises that particular firm s ratio. Non-parametric methods have the advantage of not having to specify the form of the production function or error distribution and can also handle multiple outputs as well as multiple inputs. On the other hand, non-parametric frontiers do not normally have any stochastic component, so that any departure from the frontier must be categorised as inefficiency. A detailed comparison of parametric and non-parametric methods is provided by Coelli, Rao & Battese (1998), while Cummins & Zi (1998) apply the different methods to explore the efficiency of US life insurance firms. Inputs and Outputs Although the choice and inputs and outputs is fundamental to the success of any efficiency analysis, it has proved to be problematic in the case of financial services firms. Particular difficulties can arise in classifying intermediate goods and services, which can have both input and output characteristics. In general, inputs such as land, labour and capital represent the resources that are utilised to produce the firm s output, and the acquisition of these inputs represents a cost to the firm. Outputs, on the other hand, represent those goods or services which the customers of the firm are prepared to purchase, and the sale of these outputs generates revenue. Cooper, Seiford & Tone (2000) comment that ceteris paribus DMUs should generally prefer smaller inputs and larger outputs, and that this relationship should be reflected in the efficiency scores. The management of the DMU should be able to control either inputs and/or outputs in order to improve efficiency. For financial services companies such as banks and insurers, the output is often intangible and therefore difficult to measure (and control). The pragmatic approach is therefore to identify the services provided by such firms and find measurable proxies that are highly correlated with these services. Berger & Humphrey (1997) describe two main ways of conceptualising the flow of services produced by a financial services firm: the production service views financial firms as producing services for 7

9 customers, while the intermediation service is to intermediate funds between savers and investors. Cummins & Santomero (1999) discuss the nature of insurance company services, making the usual categorisation risk financing, pooling and transfer, investment, and real services and advice. Financial service firms not only deal with identifiable inputs and outputs, but also in unknown elements of risk. A key part of the success of banks and insurers may be due to their effectiveness in risk management and Berger & Humphrey (1997) report several studies which have found a positive relationship between risk and efficiency. Clearly there would be enormous concern (among customers, regulators etc) if the opposite had been the case, since insurers would then be tempted to increase risk (by reducing capital, for example) in order to improve efficiency. However there are still real conceptual problems in undertaking a joint evaluation of risk management and efficiency (eg see Pastor, 1999) and more work needs to be done in this area. There has been considerable disagreement over the appropriate proxies to use for the output of insurance services. A further, but largely unresolved problem, arises because insurance inputs and outputs rarely take place contemporaneously: for example, in some classes of business, input resources are utilised substantially in advance of the production of insurance services (eg the payment of claims). When it comes to considering insurance company output, the majority of efficiency studies have used premium income as a proxy for the output (of non-investment related) insurance services even though premiums are really a form of revenue, that is price times quantity rather than a count of output units (Yuengert, 1993). The problems associated with using premium income as a proxy for output are discussed in detail in Diacon (1990, ch 10). There are also issues concerning whether premiums should be net or gross or reinsurance (Brown, 2000), and calculated on a cash-flow (written) or accruals (earned) basis. The latter issue can be particularly important in life and pensions business and long-tailed general insurance business where there is a substantial delay between the collection of premiums and the payment of claims. Investment income is often used to proxy for the investment-related services provided by insurers (since again there is no available count of investment units). 8

10 The problems with using premium income to proxy output have led some authors particularly Professor David Cummins to use the value of claims payments instead (eg Cummins & Weiss, 1993; Berger, Cummins & Weiss, 1997; Cummins & Zi, 1998; Cummins, Tennyson & Weiss, 1999; Cummins, Weiss & Zi, 1999; Cummins & Santomero, 1999). However it is difficult to understand why the management of insurance companies would seek to maximise the value of insurance claims (particularly for general insurance), and this therefore violates the principle output characteristic identified by Cooper, Seiford & Tone (2000) that more output should be preferred to less. Indeed some researchers have included insurance claims as an input rather than an output. The time-lag in the payment of claims means that accounting entries for insurance losses accrued (ie claims incurred) involve a substantial element of estimation and year-on-year readjustment. Furthermore, the random nature of insurance loss experience makes claims data unsuitable for use in non-parametric frontier methods. When it comes to the choice of inputs, there is general agreement that labour (administrative, managerial and sales) and capital are the main input resources utilised in the production of insurance. Although it may be possible to undertake a head-count of staff, most studies use total operating and selling costs as a proxy. In the insurance industry, this approximation is a necessity because of the widespread industry practice of outsourcing administrative and sales functions (so that so a simple head-count would seriously underestimate staff inputs). 3. Data, Methodology and Farrell Efficiency Scores Sample Selection and Characteristics Sample data is obtained from the Standard & Poor s Eurothesys 1999 database of specialist general and composite insurance companies licensed in six European countries, namely France, Germany, Italy, The Netherlands, Switzerland and the United Kingdom. The Standard & Poor s database contains information drawn from the annual consolidated report and accounts filed to the respective registrar of companies in each country. The sample therefore contains a mixture of consolidated groups (such as the Allianz Group in Germany Europe s largest company in terms 9

11 of general net earned premium) and local subsidiaries (such as AGF Insurance Ltd, Cornhill Insurance plc, Credit & Guarantee Insurance Company plc and EULER Trade Indemnity plc all of which are fully-owned subsidiaries of Allianz operating in the UK). However companies are excluded from the sample if it was obvious that their results were consolidated into another insurer in the same country unless they appeared to be trading as a separate unit. Companies are also excluded if they had non-positive values for total assets, total technical reserves, total capital, total operating expenses, total investment income, general gross written premium or general net earned premium. Some companies were affected by corporate restructuring over the period , and are only included if they had meaningful values for capital and technical reserves in In the 1999 dataset, a total of 431 companies are included as follows: France (64), Germany (121), Italy (27), The Netherlands (31), Switzerland (22) and the United Kingdom (166). Chart 1 about here Chart 1 provides an indication of the relative sizes of sample companies by providing a breakdown by quartile based on the value of total assets (converted to US$ million using year-end exchange rates). The Top Quartile shows the distribution of the 107 largest companies with assets exceeding $2610 million: approximately 29% of these insurers are French, 23% Germany, 10% Italian, 8% Dutch, 10% Swiss, and 19% British. It is apparent from the Chart that a disproportionate number of French and Swiss insurers are in the largest quartile, while British insurers are more heavily represented in the smaller quartiles. Choice of Inputs and Outputs This study uses staff and capital resources as the main inputs, and premiums and investment income as the main outputs. The inputs of sales, administrative and managerial staff are proxied by the insurer s total operating expenses and commissions. Capital inputs are split between shareholders capital and reserves, technical provisions, and debt all measured at the start of the financial year. Premium and investment income are used to represent output as it is felt that these revenue measures are the available best proxies of the services that insurers provide to their customers. No explicit measures of risk have 10

12 been used for either input or output: instead, the resulting efficiency scores are regressed again risk variables using an econometric approach in section 4. Details of the four input and three output variables covering the 431 European insurers worldwide general and long-term business are given in the following boxes: Inputs TOTOPEX total operating expenses, net of reinsurance commissions, from the general and long-term technical accounts and the non-technical account CAPLAG total capital (including shareholders capital, capital and reserves, participating rights capital, special untaxed reserves, minority interests, subordinated liabilities, subordinated debt, and the long-term fund for future appropriations) at start of 1999 ie year-end 1998 TECHLAG total technical reserves for general, and linked and unlinked long-term business at year-end 1998 CREDLAG total borrowings from creditors at year-end 1998 Outputs GNEP LTNEP TOTINV general insurance net earned premiums long-term insurance net earned premiums total investment income from all technical and nontechnical accounts, including realised and unrealised capital gains and losses, net of investment expenses and charges A summary of inputs and outputs is provided in Table 1. All values have been converted into $US million at year-end exchange rates. A simple breakdown of the relative performance of European insurers is illustrated in Chart 2 (based on the figures in Table 1). The chart shows that, in 1999, UK insurers had the highest average return on assets (at 12.6%) and an expense ratio very near to the European average (at 24% of total premiums). Table 1 about here 11

13 Farrell Efficiency Scores Estimates of Farrell efficiency and the projected values for inputs and outputs are obtained by using the input-orientated variable returns to scale (VRS) formulation of data envelopment analysis (DEA) pioneered by Banker, Charnes & Cooper (1983). The variable returns to scale model recognises that firms may not be operating at their optimal scale of production, and produces a frontier which has increasing returns to scale at low input levels and decreasing returns to scale at high input levels. In essence, this means that inefficient firms are only compared to others that are more or less the same size. Chart 2 about here The Farrell efficiency estimates are computed using the two-stage method for exploring efficiency differences across countries illustrated by Berg, Forsund, Hjalmarsson & Suominen (1993), Berger & Humphrey (1997) and Athanassopoulos, Soteriou & Zenios (2000). These are reported in Table 2 for the largest 35 UK companies transacting general insurance business based on general net earned premiums in $million (1999). All efficiency estimates are produced using the DEA- Solver Software of Cooper, Seiford & Tone (2000). Of the thirty-five UK companies in Table 2, only nine have a Global efficiency score of unity, indicating that they are 100% efficient when assessed in the European market context. The Projected Global column shows that (the pre-cngu merger) Norwich Union plc could join this elite group if it could reduce its inputs by a mere 3%. It is interesting to note that several of the largest inefficient companies in 1999 were then involved in takeover and consolidation by much larger groups (eg Guardian Royal Exchange, Eagle Star, London & Edinburgh). Table 2 about here In the first stage, a VRS DEA analysis is conducted at the country level so that insurers are compared only to those operating in the same local environment: this produces the Local Farrell efficiency scores. This procedure picks out those insurers that are inefficient in comparison with their local competitors. Each insurer was then projected to its local efficiency frontier (ie by reducing inputs and increasing outputs so that all insurers appear locally efficient) and a second VRS DEA analysis undertaken of all 431 companies together: this produces the Projected Global efficiency scores. This analysis identifies insurers whose inefficiency is due solely to the existence of more efficient foreign companies. Finally a Global score for each 12

14 insurer was obtained by multiplying the Local and Projected Global scores (this method of computing the global score is preferred as it allows for local efficiency improvements via the removal of slacks). To illustrate the interpretation of these results, consider the case of the Independent Insurance Group. The Local score of 0.72 indicates that, in comparison with its nearest UK efficient peer companies (Direct Line and Lloyds TSB General), the Independent needs to reduce its inputs by around 28% to be an efficient producer of its current levels of output. However, even if such a reorganisation were to be implemented, the Projected Global score of 0.57 shows that Independent would still need a further 43% reduction in order to be efficient in comparison with its new European peers (which turn out to be Stichting Ziekenfonds VGZ of The Netherlands and Norwich Winterthur Holdings of the UK). The Global efficiency score of 0.41 makes Independent one of the least efficient of the companies reported in Table 2. The following box provides details of those companies that appear most often as peers or benchmark companies (ie they are 100% efficient on the global efficiency frontier, and are used frequently by the DEA analysis as comparators for inefficient companies): Companies Appearing Most Frequently as Benchmarks Country of Registration France Germany The Netherlands UK Company Name Antarius (parent BNP Paribas) Pensions-Sicherungs-Verein VVaG Stichting Ziekenfonds VGZ Zorgverzekeraar VGZ Groep Cedar Insurance Chemists Defence Association Cooperative Insurance Society Electrical Contractors Insurance Legal & General Group Liverpool Victoria Friendly Society Old Mutual plc Pharmacy Mutual Summaries of the average Farrell efficiency scores for all UK companies transacting general insurance business in 1999 are provided in Table 3, broken down by quartiles based on total assets. A consistent picture emerges from all three types of efficiency 13

15 score: the biggest average scores seem to be obtained by companies in the top and bottom quartiles in terms of assets. On the other hand, the companies with assets in the middle quartiles (ie between $57.3m - $946.4m) seem to be less efficient. Table 3 about here 4. An Analysis of Efficiency Differences National Differences Once the efficiency scores of European general and composite insurers have been evaluated, it seems a natural next step to compare efficiency across countries. The first stage in such a comparison is to examine average scores for insurers by country, and these are shown in Table 4. Table 4 about here The results in Table 4 show substantial efficiency differences among the six countries. This observation is confirmed by Kruskal-Wallis and Jonckheer-Terpstra statistics which test whether the six country-based samples are drawn from the same overall population: both tests rejected the null hypothesis for all three efficiency scores at the 0.1% level. Looking first to the Projected Global scores (which project each insurer to the national efficiency frontier before making an international comparison), it can be seen that UK insurers have the highest average score (of 0.77). In other words, if UK insurers operated at their most efficient level, they would be the most efficient in Europe. In actuality, UK insurers have a low level of average local efficiency with the consequence that their average global efficiency score (0.52) is the lowest amongst the six countries. A more detailed picture of the sources of such international efficiency differences will emerge when inter-company differences are explored in more detail. Table 5 about here Table 5 illustrates the average reductions in inputs and the expansion in outputs necessary to project national insurers to their local and then global efficiency frontiers. The projected values take account of both the Farrell efficiency scores and the input and output slacks as described in Cooper, Seiford & Tone (2000, ch 4). These slacks arise because it may be possible to reduce a frontier firm s inputs (or 14

16 increase outputs) without having any impact on the Farrell efficiency. Cooper, Seiford & Tone (2000) refer to such non-zero slacks as mix-inefficiencies. Table 5 shows projected values as a percentage of original values for the four inputs and three output variables. The lower (higher) the input (output) percentage the greater the source of inefficiency on average. Thus, the figure under the Expenses Local column of shows that UK insurers are required, on average, to reduce total operating expenses by almost 6% relative to their local frontier. The value of in the Expenses Global column indicates a reduction of almost 20% relative to the global frontier. The Global projection percentages of Table 5 enable an international comparison of the source of inefficiency among general insurers. The following box cites those countries whose insurers, on average, are particularly inefficient in the utilisation of inputs or the production of outputs. Source of Inefficiency Country Excessive operating expenses Germany, France, Netherlands Over-capitalised Italy, Germany Excessive technical reserves Switzerland, Germany (ie poor claims experience) Over-borrowing Netherlands, Germany GNEP too low Netherlands LTNEP too low - Insufficient investment income Germany, Switzerland The use of total capital as an input in an insurer s production function (which is a commonplace choice in other efficiency studies cited above) raises some interesting questions about the trade-off between efficiency and capital adequacy. On the one hand, there is clearly a potential conflict between efficiency and solvency if efficient insurers are characterised as ones with low levels of capital. However well-informed policyholders may be prepared to pay higher premiums to financially-secure insurers, so it is possible that efficiency and solvency may be positively related. In essence, this 15

17 is an empirical issue which can be resolved when inter-company differences in efficiency are explored. On the other hand, questions must be raised about those insurers whose projected levels of capital are so low that they may fall below the European Union minimum solvency margin. In such cases, one must either doubt the future viability of the company or can only suppose that the reported figures for capital are known (by policyholders and regulators) to be understated. In actual fact, 68 out of 431 sampled insurers appeared to have global projected capital figures below the EU minimum (which was computed on the approximate basis of 16% of GNEP + 4% of LT technical reserves). This group of 68 is broken down as follows: 13 French (ie 20.3% of French insurers), 32 German (26.4%), 10 Italian (37.0%), 5 Netherlands (16.1%), 0 Swiss, and 8 UK (4.8%). Since EU insurers still have some discretion in the reporting of balance sheet items (eg in the valuation of investments), it is likely that these figures reflect national differences in the computing capital, with under-reporting being more common in France, Germany and Italy. Inter-Company Differences An exploration of inter-company differences in efficiency necessitates a two-stage analysis whereby efficiency scores from the first-stage DEA process are regressed against environmental variables. By definition, these environmental variables are not decision variables which would otherwise figure in the firm s choice of the nature or level of inputs and/or outputs (as these should have already been included in the DEA analysis). Second-stage regressions commonly utilise a Tobit model for censored data (rather than ordinary least squares) in order to allow for the restricted (0,1] range of Farrell efficiency scores. However, some caution is needed in applying this two-stage approach: Problems can arise if the environmental variables are co-determined or even highly correlated with the DEA inputs or outputs (eg see Coelli, Rao & Battese, 1998, ch 7), as this can lead to biased estimates in the regression analysis. Similar problems can arise in stochastic frontier analysis. Since VRS DEA efficiency scores are naturally dependent on the size of the DMU, there will be an inevitable correlation between any second-stage size variable and DEA inputs/outputs. 16

18 A regression analysis based on Farrell efficiency scores ignores any inefficiencies arising from slacks in the DEA analysis. Coelli, Rao & Battese (1998) suggest that a possible solution would be to analyse the input and output projections (which take into account both Farrell efficiency and slacks) instead. It is not clear that the Tobit model is entirely appropriate for analysing efficiency scores, since the model assumes that the scores are censored to fall in the (0,1] range. In fact, input-orientated Farrell scores can never be outside the unit interval. However this is still a largely unexplored issue. Accepting these cautions, the next step is to undertake a Tobit regression on efficiency scores using explanatory environmental variables. These variables represent aspects of the insurer s environment that may influence efficiency, but which are outside the immediate control of company management. There is clearly the potential for disagreement on exactly what kind of variables can meet these criteria. A list of selected variables and descriptive statistics is provided in Table 6. Table 6 about here Many two-stage studies of company efficiency choose company size as a key environmental variable on the basis that management cannot easily use size as a decision-variable. Since the Pearson correlation coefficient between inputs, outputs and total assets is extremely high (ranging from to 0.990), it is preferable to use the natural logarithm of total assets in $ million (SIZE) as a measure of company size - and this reduces the correlation with inputs and outputs to between and The clear non-linear relationship between size and efficiency apparent in Table 3 for UK insurers suggests that SIZE 2 should also be included. Many of the other environmental variables are included to pick up differences between companies in terms of risk (eg the gearing ratio GEAR, concentration in the asset portfolio HERFASST, liquidity LIQUID, total profitability PROFIT, reliance on reinsurance REINS, and solvency SOLV). However the international differences in the calculation of total capital reported above suggest that the impact of the SOLV variable may vary across countries, and this is therefore included as a series of interaction variables with the six country dummies. Table 7 about here 17

19 The results of the Tobit regression on the Global and Projected Global efficiency scores are presented in Table 7. The sample includes all 431 European insurance companies transacting general insurance business, and country dummy variables are included to pick up any country effects. Tobit regressions were also undertaken of the percentage difference between projected and actual inputs/outputs, but the results are entirely consistent with those outlined in Table 7 and so are not reported here. The following main results emerge: 1. Efficiency scores are strongly associated with insurer size, with clear evidence of a U-shaped relationship ie both small and large insurers appear to have higher efficiency scores. 2. The positive coefficient for HERFASST shows that an increase in the concentration of an insurer s assets (eg in particular classes of investments) is associated with higher levels of efficiency. 3. A higher proportion of premiums in life and pensions insurance business is associated with higher efficiency. This is very likely to be due to: (i) the improved opportunities to earn investment income on such business and (ii) the fact that in force long-term premiums generate comparatively few operating expenses. 4. MUTUAL companies have a higher level of efficiency than stock insurers. This confirms the (more sophisticated) results obtained in the study of US general insurers by Cummins, Weiss & Zi (1999) which finds that mutuals have some comparative advantages over stock companies. 5. The impact of the solvency variable on efficiency does indeed depend on the national environment. For insurers licensed in The Netherlands and the UK, an increase in the solvency ratio (SOLV) is associated with higher efficiency scores, whereas the effect is either negligible or negative for the other countries. Thus the customers of British and Dutch insurers seem to reward their highly solvent insurers with more premium income. It can only be speculated why the customers of French, German and Italian insurers do not do the same. 6. The relationship between efficiency and risk is rather mixed. There seems to be some evidence that efficiency and solvency are positively related in some countries but not in others. The positive coefficients for HERFASST suggest 18

20 that less asset concentration decreases efficiency. There seems to be little relationship between liquidity and efficiency, and profitability has a significantly positive influence in the projected global model only. Strangest of all, the coefficients on the reinsurance variable change sign between the global and projected global models. Although the overall picture is a bit mixed, there is certainly no evidence that risk and efficiency are positively related. 7. Finally, the country dummies confirm the international differences in average efficiency that were identified in Table 4: French and German insurers seem to have higher levels of Global efficiency on average than the UK and other countries. In other words, the current production technologies employed in French and German insurers appear to be superior to that currently employed elsewhere in Europe. On the other hand, The Netherlands and Switzerland have significantly lower average levels of Projected Global efficiency than the UK while those of France and Germany are much the same, and Italy is lower but not strongly significant. In other words, if all insurers were to improve their production technologies to the best locally-available level, then UK, French and German insurers would on a par, and would be superior to the Dutch, Swiss and Italians. This poor performance of UK insurers (relative to the French and Germans) is due to their local inefficiency rather than the superiority of their competitors. Residuals from the Tobit model of Global efficiency are reported for UK companies in Table 2 under the column Residual Global. These residuals represent company efficiency after the effects of the environmental variables have been removed, and perhaps provide the best indication of the inherent ability of company management to turn inputs into outputs. 5. Discussion and Conclusions This paper explores the efficiency of UK specialist and composite insurers transacting general insurance business. The concept of efficiency concerns an insurer s ability to 19

21 produce a given set of outputs (such as premiums and investment income) via the use of inputs such as administrative and sales staff and financial capital. The study uses value measures of insurance company inputs and outputs to undertake an exploration of the technical efficiency of UK general insurers. This is achieved by comparing the relative performance of 431 general insurers licensed in six European countries using data from Standard & Poor s Eurothesys database. Three main issues are explored. First, the value-based technical efficiency measures are produced for the leading general insurers in the six European countries, and these are split between local efficiencies (where firms are assessed relative to their national competitors) and global efficiencies (where firms are compared to all competing European insurers). The effect of this international comparison can be illustrated by looking at the top 35 UK insurers by general premium income: in terms of local efficiency, 21 (ie 60%) are classed as fully efficient; however this number falls to nine (26%) when their global efficiency is considered. An obvious implication is that, when undertaking benchmark comparisons (such as those proposed in the Business Excellence Model), UK insurers need to compare themselves against their European competitors in order to gain a more meaningful insight into their true international competitiveness. Secondly, the study attempts to compare national differences in efficiency. The analysis indicates that real variations in average levels of efficiency do exist across the European insurance industry. These variations can arise for a number of different reasons (particularly the differences in size) but there seems to be some evidence that French and German insurers are, on average, more efficient than their European competitors even after environmental factors have been controlled for. However, although this comparison has been made possible by the recent attempts to introduce some standardisation into European insurance accounts, it is vulnerable to national variations in the accounting practices (particularly in the valuation of assets and liabilities). Finally, inter-insurer differences in efficiency are explored using a Tobit regression model. Although there are a number of difficulties with this (widely used) technique, the results indicate a U-shaped relationship between size and efficiency, and that 20

22 mutual insurers (often small niche-market players) are more efficient than their proprietary counterparts. The relationship between efficiency and solvency seems to vary by country, with a strong positive link in The Netherlands and the UK, but none elsewhere. The interaction between efficiency and risk is clearly problematic, and more work needs to be done to understand this issue: none-the-less there is little strong evidence that risk and efficiency are positively related (which is a cause for some relief). 21

23 References Athanassopoulos A, Soteriou A & Zenios S (2000) Disentangling Within- and Between- Country Efficiency Differences of Bank Branches, in Harker & Zenios (2000, ch 10) Berg S, Forsund F, Hjalmarsson L & Suominen M (1993) Banking Efficiency in the Nordic Countries, Journal of Banking & Finance, 17, Berger A (1993) Distribution-Free Estimates of Efficiency in the US Banking Industry and Tests of the Standard Distributional Assumptions, Journal of Productivity Analysis, 4, Berger A, Cummins J & Weiss M (1997) The Coexistence of Multiple Distribution Systems for Financial Services: The Case of Property-Liability Insurance, Journal of Business, 70, 4, Berger A & Humphrey D (1997) Efficiency of Financial Institutions: International Survey and Directions for Future Research, European Journal of Operations Research, 98, Banker, R, Charnes A & Cooper W (1984) Some Models for Estimating Technical and Scale Inefficiencies in Data Envelopment Analysis, Management Science, 30, Brown M (2000) An Investigation of the Relative Efficiency of UK General Insurers, Paper presented at the 22 nd UK Insurance Economists Conference, University of Nottingham, March Coelli T, Rao D & Battese, G (1998) An Introduction to Efficiency and Productivity Analysis, Kluwer Academic Publishers, Boston Cooper W, Seiford L & Tone K (2000) Data Envelopment Analysis: A Comprehensive Text with Models, Applications, References and DEA-Solver Software, Kluwer Academic Publishers, Boston Cummins J & Santomero A (1999) Changes in the Life Insurance Industry: Efficiency, Technology and Risk Management, Kluwer Academic Publishers, Boston Cummins J & Weiss M (1993) Measuring Cost Efficiency in the Property-Liability Insurance Industry, Journal of Banking & Finance, 17, Cummins J, Tennyson S & Weiss M (1999) Consolidation and Efficiency in the US Life Insurance Industry, Journal of Banking & Finance, 23, Cummins J, Weiss M & Zi H (1999) Organizational Form and Efficiency: The Coexistence of Stock and Mutual Property-Liability Insurers, Management Science, 45, 9, Cummins J & Zi H (1998) Comparison of Frontier Efficiency Methods: An Application to the US Life Insurance Industry, Journal of Productivity Analysis, 10, Diacon S (1990) A Guide to Insurance Management, Macmillan Press, London Donni O & Fecher F (1997) Efficiency and Productivity of the Insurance Industry in the OECD Countries, Geneva Papers on Risk and Insurance: Issues and Practice, 85, Farrell M (1957) The Measurement of Productive Efficiency, Journal of the Royal Statistical Society, Series A, 120, Harker P & Zenios S (2000) Performance of Financial Institutions: Efficiency, Innovation, Regulation, Cambridge University Press, Cambridge Katrischen F & Scordis N (1998) Economies of Scale in Services: A Study of Multinational Insurers, Journal of International Business Studies, 29, 2, Pastor, J (1999) Efficiency and Risk Management in Spanish Banking: A method to Decompose Risk, Applied Financial Economics, 9,

24 Rai A (1996) Cost Efficiency of International Insurance Firms, Journal of Financial Services Research, 10, Rassam C (1998) Insurance in Europe: A Report on the Major European Insurance Markets, International Business Intelligence Reports, The Stationery Office, London Yuengert A (1993) The Measurement of Efficiency in Life Insurance: Estimates of a Mixed Normal- Gamma Error Model, Journal of Banking & Finance, 17,

25 Chart 1: European General Insurers by Asset Quartiles, % 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Top 25% 2nd 25% 3rd 25% 4th 25% UK Switzerland Netherlands Italy Germany France Asset Quartile 24

26 Table 1: Input and Output Variables European Insurance Companies Transacting General Business, 1999, $m TOTOPEX CAPLAG TECHLAG CREDLAG GNEP LTNEP TOTINV France Mean N=64 Std. Deviation Minimum Maximum Germany Mean N=121 Std. Deviation Minimum Maximum Italy Mean N=27 Std. Deviation Minimum Maximum Netherlands Mean N=31 Std. Deviation Minimum Maximum Switzerland Mean N=22 Std. Deviation Minimum Maximum UK Mean N=166 Std. Deviation Minimum Maximum Total Mean N=431 Std. Deviation Minimum Maximum Key TOTOPEX CAPLAG TECHLAG CREDLAG GNEP LTNEP TOTINV Total Operating Expenses Total Capital at start of year Total Technical Reserves at start of year Total Borrowing from Creditors at start of year General Net Earned Premium Income Long-term Net Earned Premium Income Total Investment Income 25

27 Chart 2: Performance of European General Insurers, 1999 % Expense Ratio Return on Assets 0 France Germany Italy Netherlands Switzerland UK Total 26

28 Table 2: Farrell Efficiency Scores for UK General Insurers, largest insurers by General Net Earned Premium, $million Company GNEP $m Global Projected Global Local Residual Global* CGU plc Royal & Sun Alliance Insurance Group plc Norwich Union plc Guardian Royal Exchange plc Sun Life & Provincial Holdings plc Eagle Star Insurance Company Ltd Cornhill Insurance plc London & Edinburgh Insurance Company Ltd Direct Line Insurance plc Financial Insurance Company Ltd National Farmers Union Mutual Insurance Soc Independent Insurance Group plc Lloyds TSB General Insurance Ltd Co-operative Insurance Society Ltd National Insurance & Guarantee Corporation CNA International Reinsurance Company Ltd Prudential Corporation plc GAN Insurance Co Ltd Old Mutual plc Churchill Insurance Legal & General Group plc Fortis Insurance Ltd Iron Trades Insurance Group Liverpool Victoria Friendly Society Ltd Ecclesiastical Insurance Group plc QBE International Insurance Ltd CGU Underwriting Ltd Terra Nova Insurance Company Ltd Swiss Reinsurance Company UK Ltd London General Holdings Ltd St Paul Reinsurance Company Ltd SVB Holdings plc UK Insurance Ltd Prime Health Ltd Zurich Specialities London Ltd Note The Residual Global scores are the residuals from the Tobit regression on Global scores reported in Table 7 ie after adjustment for environmental factors. The larger the score, the greater the residual efficiency. 27