PRICING CHALLENGES A CONTINUOUSLY CHANGING MARKET +34 (0) (0)

Transcription

1 PRICING CHALLENGES IN A CONTINUOUSLY CHANGING MARKET Michaël Noack Senior consultant, ADDACTIS Ibérica michael.noack@addactis.com Ming Roest CEO, ADDACTIS Netherlands ming.roest@addactis.com +31 (0) (0) AMSTERDAM - BOGOTA - BRUSSELS - LYON - LUXEMBOURG - MADRID - MUNICH - PARIS - ROME - SINGAPORE - WARSAW

2 Content Introduction Data preparation & pre-analysis Risk models based on historic data Consolidation of models to obtain Risk estimation & commercial premiums (scenarios) Scenario test & impact analysis Behaviour models & price elasticity Re-evaluation of scenarios/new scenarios 16 Latest adjustments with large scale simulation Monitoring/Permanent update Conclusion 18 20

3 INTRODUCTION Business models exchanges have accelerated and customer behaviour is also changing. This is mainly explained by websites when comparing premiums and tariffs on a daily basis. These developments have made it more important than ever for insurers to get their tariffs right. This implies that insurance companies need to be: adequately compensated for the risk they take continue to be competitive avoid anti selection. Actuaries must also make decisions knowing that a single error could potentially lead to significant losses. The intensity of competition between non-life insurance companies is increasing rapidly and margins are tightening. Emergence of price comparison websites and new technologies have led to increased competition on price, making it more difficult for insurers to make (or retain) profitable margins and keep existing customers satisfied. Following all these developments, insurers will need to: operate very rapidly in a continuously changing and competitive market; price their products correctly and therefore use appropriate data, sophisticated and best practice pricing models; have well documented and transparent pricing processes in place, triggered by Solvency II. To overcome all these increasing challenges that insurers are facing, it is important to master a well-organized pricing process that covers all steps and makes it possible to include all involved parties. In this article, the best practice pricing process is described, that takes into account all these points. 2-3

4 DATA PREPARATION & PRE-ANALYSIS In a pricing process, statistical models such as GLM that have the capability to estimate claim costs for different risk profiles need to be fed correctly. To ensure that these models work, it is required to have clean data so that the correct model decisions can be made. Therefore, the first step in a sound pricing process is to ensure that the data are appropriate, adequate and complete. In the preparation phase policy and claims data are extracted. Here, several checks and adjustments are required to ensure that high data quality (no error, no missing data and homogeneous formats) is in place. As a first check it is recommended to conduct basic statistical analyses (1-way, 2-way analysis and correlations analysis). Policy data should contain at least a unique identifier (for example a unique policy number), a start date and an end date. This is required for all versions in order to calculate exposures and to be able to assign claims to the correct policy version. In addition it is important to know what cover there is for each policy, bearing in mind that cover tends to change during the lifetime of a policy (ie. Third Party Liability only, Damage, Comprehensive, Liability + Collision + Theft + etc) Claim data should have: a claim date and the policy identifier (so that it can be merged to the correct policy), a claim description that makes it possible to define which product cover the claim belongs to and that can be used to create homogenous claim segments and finally claims cost; in most situations it will be an amount. In this phase, homogenous 1 claim segments should be created using descriptive information from the claim data set to join claim types with a similar behaviour or the claim amount to split an existing segment in two or to isolate a claim type with fixed cost for example. In addition, decisions should be made on how to treat large claims. There are several methods (Mean excess function, Hill estimate) that help the pricing actuary or specialist to find the most appropriate threshold to cut large claims. 4-5

5 RISK MODELS BASED ON HISTORIC DATA GLM is the market standard for statistical models that gives a prediction on the estimated claim costs for different risk profiles. It is recommended to develop models for frequency and claim costs separately and aggregate these in order to obtain an estimation for the risk (or pure) premium. For some business classes (such as health insurance) the risk premium can be modelled directly. Based on these estimates, a future tariff can be built. What makes the main difference between using GLM in insurance and using it in other fields like biostatistics, clinical studies? First the huge amount of data that makes the law of large number work well and second the practical application. It is not necessary to exactly target the 5% significance level to take a decision. The model phase is a key element of the pricing process here a powerful GLM engine should be combined with the experience of a pricing specialist. This specialist will apply his statistical knowledge together with his business experience to obtain the models that best explain the observed historic date and are predictive at the same time. Several statistical methods are available to find this equilibrium (Chi2 test, AIC). All advanced methods should be used to find the most appropriate model for each segment (all kind of factors, splines, polynomials, interactions, conditions, geospatial smoothing). The main target of this phase is not obtaining a perfect description of what has happened over previous years the main interest is to find out what will happen during the next year(s) when the new tariff will be applied. 6-7

6 CONSOLIDATION OF MODELS TO OBTAIN RISK ESTIMATION & COMMERCIAL PREMIUMS (SCENARIOS) In the consolidation phase different models from the previous phase are combined including their adjustments. Pricing actuaries will need to take into account the time horizon and also decide on whether the calculated premium can be used (is it for example a pure risk premium or already a possible commercial premium). The consolidated premium for a cover should take into account the information for all claim segments included in this cover and other cost components that could affect the final premium (uncovered but paid claims, security charges, charges for large claims, charges/ discounts for IBNR). could affect the final premium (uncovered but paid claims, security charges, charges for large claims, charges/ discounts for IBNR). The tariff models that are developed are typically based on historical data from the last 3-5 years, but the calculated tariffs shall be applied for the coming month or years. This is why it is necessary to include adjustments for certain trends (e.g legal changes, new car technologies, demographic changes) and inflation into the calculations. The final estimation should also reflect possible changed market conditions (new competitors, marketing campaigns) and global economic parameters (fuel cost, unemployment rates and improved cars). These parameters should not be defined by the actuaries only, companies should use a well-defined procedure to obtain these parameters that include a definition of who is in charge of each individual parameters from different departments (marketing, legal, financial). In this step it is necessary to obtain the best estimation of future claim costs, but nevertheless several alternative scenarios for possible commercial premiums to go to the market are selected. The last ones in addition to the previously defined parameters should also include a component for internal cost, margins and other components to be included in the commercial premium. The final results from this phase are one (or more) risk premium and several commercial premiums (scenarios). 8-9

7 SCENARIO TEST & IMPACT ANALYSIS In the scenario test and impact analysis phase, the tariff structures of different scenarios will be tested onto the whole portfolio. For all policies (or prospects) estimations of their future claim cost and possible premium scenarios are combined. It is of course also possible to compare premium scenarios with the current premium structure. Any premium change could have a major impact for an insurance company. Therefore, it is essential to understand the potential (financial) impact. In this step impacts of a premium change for several scenarios are studied: the overall impact, the impact on target groups, the minimum and maximum premium changes for an individual risk, checks if the strategic targets are reflected by the scenarios, coordination with marketing activities, etc. Any report that can include all important business indicators by different segments can be created and analysed. One should take notice of the fact that the analyses are static of nature. It is not straightforward to predict the response of the customer/client to the applied premium changes

8 BEHAVIOUR MODELS & PRICE ELASTICITY It is obviously of great importance to understand what the impact of the proposed tariff structure will be on the behaviour of customers. It is therefore recommended to address this customer behaviour phase, which analyses the possible clients response to premium actions. For this analysis, GLM techniques can be used to explain the observed historic customer answer by different risk (behaviour) factors and the offered premium. The objective is to see how, for a fixed profile, the customer response changes with a premium change (price elasticity). It should be realised that this phase is challenging for two reasons: first of all, the data may not always be available and secondly, even if data is available, the models cannot always give an answer. If for example, in the past the same risk profile was always priced identically it is by definition not possible to predict how clients will react on a premium change, as the data is not available. Even if enough premium variation is present to develop the models, there will always be a high correlation between the risk factors and the premium. Special attention is needed to control negative price elasticity. It is more difficult to develop behaviour models than risk models, data are less homogenous and many decisive factors are out of observation

9 In this step (the second test phase) different scenarios are applied to the risk profiles or to the whole portfolio including the customer response. This makes the comparison more realistic and gives a real economic evaluation for different scenarios. Compared to the first static test phase this test phase is of a more dynamic nature, as this test phase includes estimations of retention, conversion, GWP, loss ratio etc. for each alternatives. When such an analysis is carried out, insurance companies get a better understanding of scenarios implications which makes it possible to make more reliable business decisions. RE-EVALUATION OF It is important to be aware that the objective of the pricing process is not just to define a premium structure with base premium and multipliers; it should also include estimations on the behaviour of the main business indicators after the introduction of the new tariff. SCENARIOS/NEW SCENARIOS 14-15

10 LATEST ADJUSTMENTS WITH LARGE SCALE SIMULATION Based on the selected favourite scenario (or the actual tariff) insurance companies can analyse the impact of small premium variations on the principal business indicators (retention, conversion, margin, loss ratio). This is carried out in the simulation phase where previously defined target functions for the entire portfolio for thousands of premium scenarios at once are simulated. This can help the insurance company to get insights into business opportunities or business risks and outliers. This is not possible to get with traditional methods or tests where only a comparison of a limited number of scenarios is made. This simulation phase could also include an optimization method to find out the individual premiums to maximize one business indicator respecting a number of individual and global constraints. For all kinds of simulation, it is essential to act close to the observed experience. Even if the mathematical functions for price elasticity can be calculated for any premium (for example for a premium of in car insurance the result will still be a retention different from zero), the practical use is only valid if there exists historic data in the corresponding premium range

11 Once the new tariff is introduced in the market, actuaries and pricing specialists should permanently control business indicators. The monitoring phase checks if the assumptions and expectations were correct. Any significant deviation should be reported, analyzed and trigger an action plan which was previously defined. Once the new tariff is introduced in the market, actuaries and pricing specialists should permanently control business indicators. The monitoring phase checks if the assumptions and expectations were correct. Any significant deviation should be reported, analyzed and trigger an action plan which was previously defined. MONITORING / PERMANENT UPDATE 18-19

12 CONCLUSION The pricing process has changed over the last years. It is still based on statistical models but it has become a complex business process with an increased number of stakeholders. The exact function and responsibility of all stakeholders needs to be clearly defined. The final result of the pricing process is not just a tariff but a business case with main business indicators estimations in addition to planned actions (contingency plans) in case of (significant) deviations. All this has to be well defined and documented which is in line with Solvency II requirements. All this is only possible with a well-designed application that makes life easier for actuaries and helps them concentrate their efforts on statistical analysis and business decisions instead of data manipulation

13 +33 (0) Amsterdam - Bogota - Brussels - Lyon - Luxembourg - Madrid - Munich - Paris - Rome - Singapore - Warsaw