SP7 17 Assessment of Reserving Results PDF

# 17 Assessment of Reserving Results ## Syllabus Objectives - Reserving result analyses. - Factors to consider in assessing the reasonableness of the results of a reserving exercise. - Typical diagnostics commonly used to assess the reasonableness of the results of a reserving exercise....

# 17 Assessment of Reserving Results ## Syllabus Objectives - Reserving result analyses. - Factors to consider in assessing the reasonableness of the results of a reserving exercise. - Typical diagnostics commonly used to assess the reasonableness of the results of a reserving exercise. - Factors to consider in assessing the reasonableness of changes in results of a reserving exercise over time. - Analysis of experience in the context of a reserving exercise. - How alternative results of reserving exercises can arise and professional issues in resolving them. ## Introduction In previous chapters, we have considered the process of calculating reserves. In this chapter, we consider the reasonableness of the outcome of such reserving calculations. It will be important that the actuary checks that the figures are justifiable and that the methodology and assumptions adopted are appropriate. A range of diagnostic tools and analysis of emerging experience can help with this process. In section 1 of this chapter, we consider why it is important to analyze the results of the reserving exercise. Sections 2 and 3 then look at various diagnostics (measures for interpreting data or results) for claims and the claim development pattern. The analysis of emerging experience is covered in section 4, considering why current experience may be different from that expected. Section 5 describes the underwriting cycle and explains why it should be taken into account in the reserving exercise. Section 6 describes the reserving cycle and the affect this can have on an insurer's reserves. In section 7, we consider why different people carrying out the same reserving exercise may have different results. ## 1 Assessing the Results of a Reserving Exercise ### 1.1 Introduction Reserving is a core activity for actuaries in general insurance work. The area of assessing the results of a reserving exercise is one of the most critical stages in the reserving process. It is here that we can be required to make the key judgements that underpin the analysis carried out. In doing so, we often need to ask ourselves two questions: - Are the results selected reasonable? - Are the new results supported by an analysis of emerging experience? In practice, we may have begun a reserving exercise by looking at the approach and results from the previous review. For example, actuaries will often carry out an "Actual versus Expected" exercise. This involves comparing the actual experience in the period since the last review, to the experience that was expected to have emerged in that time period. The selected assumptions may then be updated to reflect the most recent trends in experience. It is important for us to bear in mind the dangers of anchoring error whereby too much weight is given to the previous methodology and assumptions. Anchoring describes the common human tendency to rely too heavily, or "anchor" on one trait or piece of information, perhaps past claims inflation, when making decisions. Usually once the anchor is set, there is a bias toward that value. Let's take a (non-actuarial) example: A person looking to buy a puppy may automatically search for the same breed as their previous dog, without considering that a different breed may by better suited to their current lifestyle. We need to distinguish carefully between changes in methodology and changes in assumptions, the reasons for the changes and quantifying the impacts of change (particularly if these changes affect the booked reserves). The approach that we follow may satisfy only one of the criteria illustrated by the following scenarios: - We do an analysis where we fit new models to estimate reserves. This may yield results that appear reasonable, but the estimates may change over time (due to the change in model), even where there has been no material change in the underlying data and information available. While there may be circumstances where such a change of basis is appropriate, this can make communication of results more difficult. - We follow an approach that considers changes in results over time in response to data, but we do not monitor whether the results themselves still appear reasonable. We may fail to identify situations where results gradually cease to be reasonable over a period of time. We need to be aware of the risk of anchoring to results that are no longer supportable. ### 1.2 Approaches to Analysing the Results It is therefore important that we are careful when reviewing the results of a reserving exercise to avoid the pitfalls that can arise from either of these scenarios mentioned above. We can achieve this by undertaking two types of analysis: 1. Applying diagnostic tests to check that results are reasonable 2. Carrying out an analysis of the emerging experience. A balance needs to be achieved between the additional understanding gained from applying these analyses versus the additional cost and time involved. In this chapter, we will look at both of these types of analysis. We should consider undertaking an overview of the whole exercise for reasonableness and to ensure that there are no gaps or overlaps in any analysis performed. Finally, we must make a judgement, taking into account the analysis and our experience, when making a final selection of results. Judgement will form a key part of any actuary's assessment of results. We can think of our experience as providing an implicit benchmark for expected results. We will have a view based on our own experience, for example, on how frequently a particular event may occur, or what we would expect a reasonable range for a particular diagnostic to be. What would be reasonable will vary with the type of business written, the target market and over time. However, we should be careful as our experience may either be limited, or not typical. For example, one particular actuary may have experience of personal lines business but not commercial lines, or London Market business. Where this is the case professional conduct standards require that help from a more experienced actuary is sought. Peer review can help here, where another actuary can review and challenge our methodology and assumptions. ## 2 Diagnostics ### 2.1 An Introduction A diagnostic is a measure used to assist with interpretation of data or results and to help us test and verify the underlying methodology and assumptions. It can indicate that experience is inconsistent with the assumptions. Some diagnostics are a test of results (eg IBNR divided by premium). Others test the data (eg paid claims divided by incurred claims). Others test both (eg ultimate loss ratios). In other words, diagnostics are often used to provide a high-level "reasonableness" check of results, or to indicate which areas of the portfolio should be examined more closely. **Question** Suggest what we can learn from examining: (i) IBNR divided by premium (ii) ultimate loss ratios. **Solution** (i) IBNR divided by premium can highlight errors or inconsistencies in the reserving process or a change in premium rating strength (more on this later in the chapter). (ii) Ultimate loss ratios can indicate where there have been changes in the stringency of claims underwriting, an improvement or worsening of claims experience or a change to the rating basis. ### 2.2 Interpreting the Diagnostics We should ensure that we understand the reasons for changes in diagnostics over time, and any unusual features highlighted by them. **Question** Suggest what we should do if the diagnostics highlight unusual features. **Solution** We should: - investigate the reason for the unusual feature - understand the implications for the reserving process - take appropriate action, eg change methodology or assumptions if necessary. While triangles are typically a key input into reserving methods, they are a key data item for diagnostics to validate reserving selections. This assists to understand how the forecasts of future performance compare to the historical performance. We may have to consider how these diagnostics vary over differing data groupings (that is accident, underwriting, calendar, reporting and development period). For example, we would expect that for any given development period, IBNR divided by premium for employers' liability business would be higher than for household property business. Diagnostics relating to development periods are covered in the next section. Changes in diagnostics over time, or unusually high or low figures, may result from unexpected emerging experience that is considered to be a one-off. For example, there may have been a change to terms and conditions which would preclude this experience from happening again. In this case, we may feel that the methodology and assumptions continue to be appropriate. In other cases, we may identify limitations in the reserving methodology and assumptions and decide to make changes. It is necessary to consider materiality when addressing features revealed in diagnostics. Where an analysis of the diagnostics does highlight unusual features in experience, an actuary may decide not to update their methodology or assumptions if the resulting change in reserves is immaterial relative to the size of the company's total reserves. This decision will also depend on the purpose of the reserving exercise. When we interpret diagnostics, it is critical to understand the underlying reasons for the behaviour shown. It is important to consider whether the diagnostics fall in an expected range. If they do not, we should consider whether this indicates a concern with the results or identifies an underlying feature of the experience that has not previously been taken into account. This is analogous to the feedback stage of the actuarial control cycle, ie notice any deviation, understand the reasons for the differences and then feedback by making appropriate adjustments. Where possible, we should exclude special features such as large losses to avoid providing a distorted picture, but also consider the appropriateness of any reserve for the loss. When examining the diagnostics, exceptional items should be excluded, although of course it will be important that the end reserve does include an allowance for such items. It is important to consider the granularity (that is, what level to subdivide the data) at which the diagnostics are reviewed and assumptions made. We may do this by reviewing the results and diagnostics at a higher level. The remainder of this section discusses how various diagnostics can be used within a reserving analysis. ### 2.3 Loss Ratios We should review changes in loss ratios (some or all of paid, outstanding, IBNR, incurred and ultimate as the situation dictates). These may highlight some of the following: - Changes in premium rating strength (see further discussion later in this chapter on the underwriting cycle). Adjusting underlying premiums for premium rate movements may remove this effect (a rate index can be used for this purpose) but beware of rating increases imposed to reflect increased risk. - Sources of uncertainty (eg exceptionally large open claims in a particular cohort). - Inconsistencies in the model assumptions (eg if the progression of IBNR loss ratios is not monotonically decreasing with age of the claims cohort). A monotonically decreasing function is a non-increasing function. - Errors in the reserving process (eg misapplied Bornhuetter-Fergusson model). The premiums on policies may not be known until some time after the end of the period of exposure. For example, retrospective experience rating may be operating, or there may be an adjustment for exposure at the end of the year (eg on employers' liability to reflect the number of workers covered over the year). In the case where premium estimates are unavailable to derive ultimate loss ratios, we could consider the ultimate claims as a percentage of another measure of exposure (for example, policy or claim count, insured turnover and so on). We can compare loss ratios to premium and claims indices or available benchmarks and decide whether they look sensible in comparison. Other things being equal, an increasing premium rate index should lead to a lower loss ratio. By reviewing the paid and incurred loss ratios, we can see at an early stage how the experience to date has turned out for each origin year. For example, where loss ratios are unexpectedly high, we can then investigate further whether this is due to a unique claim or type of claim, or is an early indicator that claims experience is going to be materially worse than expected. Triangulations of these ratios can be helpful to spot trends and sense check assumptions (eg initial expected loss ratios) for reasonableness. Considering IBNR as a percentage of case estimates is useful in those situations where a complete paid or incurred claims development history is not available. This will be useful for long-tailed classes, or immature portfolios where the earliest development year is yet to be fully run-off. ### 2.4 Paid to Incurred and/or Case Estimates to Incurred Ratios This diagnostic can indicate the strength of case estimates. An increasing ratio trend over time, when we are reviewing a triangle of cumulative paid claims divided by cumulative incurred claims, may have a number of possible explanations, such as: - Case estimate strength has been reduced If the ratio of paid claims to incurred claims has increased, this may indicate that the strength of case estimates has reduced. However, if the ratio of case estimates to incurred claims has increased, this may indicate that the strength of case estimates has increased. Care needs to be taken that the definition of incurred claims (ie what has been included) has not changed. - An underlying change in business - An acceleration in the claims settlement pattern - A slow-down in the rate at which outstanding claims are established - A distorting large loss settlement. **Question** A general insurer writes public liability insurance and reserves using the basic chain ladder method. On examining the relationship between paid and incurred claims it is found that the ratio is decreasing over time. Suggest what impact this will have on claims development and hence the reserves calculated. **Solution** Actual claims development is slower than that suggested by the use of the basic chain ladder. If no adjustment is made to the basic chain ladder method, then the reserves calculated will be an underestimate of the amount required. If the reserves are also discounted, then the payments will be assumed to be made earlier than will eventually be the case, and will therefore be affected to a lesser extent by discounting. This will offset to some extent the inaccuracy in the basic chain ladder model. A revised model should take account of both of these factors. ### 2.5 Average Outstanding Case Estimate A review of this triangle can highlight changes in the strength of case reserves. ### 2.6 Ratio of IBNR to Case Estimates For more mature cohorts, this diagnostic is helpful, particularly where IBNR is expected to be mainly in respect of IBNER, rather than "pure" IBNR claims. Such a diagnostic gives a feel for the outstanding claims and the uncertainty relating to them. ### 2.7 Survival Ratios Survival ratios show how long a reserve or IBNR estimate will last (before all outstanding claims are paid) if current paid or incurred claims development continues at a given rate. We may consider the ratios on a number of different averages (say one year, three years and five years) paid or incurred claims to avoid years of particularly high or low claims activity. We can compare these survival ratios with other similar portfolios or market benchmarks. In some cases, these may be used to set the reserves for particularly uncertain classes of business. For example, a ratio often used to estimate asbestos loss reserve adequacy is the three-year survival ratio, which is the ratio of loss reserves to the three-year paid loss average. Assuming that average paid losses remain constant, with no additional reserving, the survival ratio indicates how many years the reserves should last. ### 2.8 Claim Frequency and Average Cost per Claim Diagnostics These diagnostics are useful where claim count information is available. Changes in frequency and severity may highlight many of the features described above for other diagnostics. They may also identify inflationary trends in claims costs and trends in claim frequency per unit of exposure. Where policies have a maximum claim value, it may be useful to compare average claim size with this value. **Question** A general insurer writes employers' liability business. On investigation it is found that claim frequency per unit of exposure is increasing. Suggest possible reasons for this. **Solution** Reasons include: - Change in policy terms and/or conditions - Emergence of a new type of claim - Increasingly litigious society - Weaker initial and/or claims underwriting. A triangulation of the number of settled claims divided by the number of reported claims can indicate the stability of the claims settlement process. Similar diagnostics can be considered on the number of nil claims as a percentage of reported claims to identify changes to the definition of a reported claim. For classes of business where open nil claim counts (that is, number of claims where no payment has yet been made) are recorded, monitoring these may provide an early warning of an anticipated increase in claims costs or problems arising in processing claims. ### 2.9 Incremental Development Triangles These diagnostics are used to consider the level of IBNR or reserves for a cohort compared to movements prior cohorts have seen to develop to ultimate. While not a precise technique, this can give a good sense check on the selections where the outcome is particularly judgemental or when considered at a less granular level than the reserves were calculated at. ### 2.10 Reinsurance to Gross Ratios We can apply all the above diagnostics gross and net of reinsurance. In addition, it is valuable to consider how gross, reinsurance and net estimates interact and the ratios between them. If a reinsurance programme remains unchanged for various origin years, and claims experience remains relatively constant, we would expect such ratios to remain fairly uniform. However, changes in these ratios are common and we should understand these when we consider the reasonableness of the reinsurance or net assumptions. While this is often more straightforward when considering proportional reinsurance, it can be useful to consider the ratios for non-proportional business as well; but note that the impact of a fixed limit and deductible is likely to reduce with inflation. **Question** A general insurer writes domestic household business in the US. The ratio of gross to net of reinsurance claims has been declining over the last 10 years. Suggest the possible actions that may need to be taken. **Solution** Lower reinsurance recoveries are being made. The insurer needs to consider whether this is acceptable. A deliberate decision may have been made to self-insure to a greater extent, in which case no action is required. If this is not the case then the insurer should review the reinsurance programme, both the types and extent of cover, in order to achieve the desired reinsurance recovery as a proportion of gross claims. Changes in these ratios can be a result of: - Changes in the amount of business being retained or ceded by the insurer - Changes in the mix of non-proportional and proportional reinsurance cover - Changing policy terms, such as deductibles, limits and reinstatements - Changes in the underlying gross experience - Inconsistencies in the treatment of gross and net claims estimates. Where proportional reinsurance is in place, we can review the selected net to gross ratios to ensure these at least allow for the proportional element. For example, if a 30% quota-share is in place, we would expect at least 30% of premium, paid and incurred to be ceded to reinsurers. Splitting reinsurance by type can also be helpful to sense check the recoveries against each type of reinsurance. Alternatively, we can use net of reinsurance to gross of reinsurance ratios. ## 3 Development Pattern Diagnostics ### 3.1 Introduction One of the most important sets of diagnostics for a reserving exercise are those comparing assumed future development patterns with past development patterns. Clearly, this can only be carried out where there is relevant prior history. We can carry out such a review by considering an array showing data split by origin cohort and development period. A common approach is to consider cumulative (paid or incurred) claims development as a proportion of the estimated ultimate claims for a given origin cohort to determine areas to revisit and understand why cohorts are out of line (rather than simply adjusting the ultimate to fit the diagnostic). We can then compare differing origin years readily. **Example** A check for calendar-year distortions could involve examining whether there are diagonals in our run-off triangle with a bias towards "high" or "low" development factors. The median development factor is obtained (M) and then for each development period we can determine whether each development factor is the median (M) or is high (H) or low (L) relative to the median. | Settlement Delay in Years (Development Year) | 0 | 1 | 2 | 3 | 4 | |---|---|---|---|---|---| | 2019 | L | H | H | M | H | | 2020| L | M | L | H | | 2021| L | H | H | | 2022| L | L | | 2023 | H | For each diagonal of the run-off the numbers of H's and L's are counted. In the absence of any calendar-year effects the numbers of H's and L's should be about the same. If they are not, then this may suggest some adjustments need to be made to the development factors. Some actuaries find this is most easily performed graphically, but we should be careful not to rely entirely on graphs as this may mask smaller trends that are difficult to see on a graph. These graphs can mask the more detailed information shown in the triangle above, and do not give an indication of the materiality of each origin year. However, they can give a useful indication of whether the changes in development patterns need to be analysed more fully. ### 3.2 Changes in the Development Pattern Such an analysis can show features of the data and the volatility of the development pattern. It can also show differences in development of each year and evidence of changing development patterns over time where this exists. **Question** You have a run-off triangle of paid claims split by accident year. Suggest possible reasons for: (i) a row of figures that is unusually high (ii) a column of figures that is unusually high (iii) a diagonal of figures that is unusually low. **Solution** (i) This represents unusually heavy claims in a certain accident year. Possible reasons could be heavy flooding or an unusually bad winter resulting in a large number of claims. (ii) A column represents the payments made a certain number of years after the year of the accident. A high figure may imply that we settle many claims in the early years (property damage claims) and then there is a fall off in payments before liability claims are paid. (iii) A low diagonal of figures represents a fall-off in payments being made by the insurer in a given calendar year. This could be due to implementation of a new system, loss of staff or a postal strike. We should consider how recent cohorts develop compared to older cohorts to identify any changes in the trends over time to ensure the development pattern selected does not use history that clearly develops in a different way to recent history. We would normally expect origin years to be more developed (where development has a positive tail) as they get older, unless there are particular reasons for this not to be the case. Such reasons may include: - External influences such as inflation, catastrophes or changes in the underlying nature of the risks - Internal influences such as changes in underwriting, claim handling and settlement and recording procedures or reinsurance arrangements - Changes in the type of business attracted within a class and types of claim emerging - Random fluctuations or large claims (or even lack of expected large claims) in a portfolio - If policy limits have been exhausted on some or all of the policies in a portfolio, that would limit the scope for further deterioration of the incurred claims position. We should seek to understand the reasons for features in the development pattern for each individual origin year and ensure that the underlying assumptions are appropriate in light of this. ### 3.3 Stability of Development Pattern Where the development pattern is volatile, we should consider whether the estimates are reliable considering this. Where development patterns are too volatile or unstable, alternative approaches or benchmarks may be necessary. As the scope for adverse development (that is, an increase in aggregate claims) often exceeds that for favourable development (that is, a reduction in aggregate claims), the distribution of final outcomes will often be positively skewed. We should consider whether we have allowed for a sufficient tail beyond the claims development experience to date. **Question** Give examples of factors that would lead to favourable development and those leading to adverse development. **Solution** Favourable development could be a result of faster claims handling or a propensity towards small/simpler claims with shorter reporting and settlement delays. Adverse development could be a result of a change in experience leading to more complex/large claims with longer reporting and settlement delays. An accumulation of claims due to one event (eg a hurricane) could also cause a slowing down of settlement. If the claims team take longer to deal with claims this will also adversely affect development. ### 3.4 Comparison Between Classes The speed of development patterns (how quickly claims develop to their ultimate position) will vary by class. Examples of where we would normally expect to see slower development patterns are: - Liability classes compared to property damage classes due to delays caused by disease development, determination of liability through courts, etc - Reinsurance classes compared to the equivalent direct classes due to the delay in being notified of reinsurance recoveries and settlement of them - Classes with policies attaching at higher layers compared to those with policies attaching at lower layers as it tends to take longer for larger claims to develop up to the (higher) attachment point. Larger claims are also likely to be more complex, so it may take longer for the extent of the claim to be fully understood - Business written on a risk attaching basis compared with a claims made basis. This is because late reported claims will fall into a subsequent policy year. This was discussed earlier in the course. We should consider whether the relative speeds of the suggested development patterns for the different classes are appropriate. ### 3.5 Claim Development versus Premium Development Where there is evidence of a changing premium development pattern, we should consider the reasons for this, and whether this should affect the claim development pattern. In some circumstances, it may also be useful to compare earned premium patterns with incurred claim development patterns as we would always expect to earn premium faster than claims are incurred. An earned percentage for an underwriting year lower than the incurred percentage developed would be an indication of either the development pattern being too fast or an error in the earned premium percentage. In theory, the earned premium would be expected to develop to reflect the potential for claims over the period of exposure. In practice however, the earned premium development pattern is often estimated using formulaic techniques and such a comparison may be of limited use. ### 3.6 Comparison to Benchmarks There are a variety of sources of development patterns that could be used as benchmarks. These include: - Industry and market sources - Other classes of business that are closely related, eg household contents as a proxy for household buildings - Similar portfolios within the actuary's experience. ### 3.7 Residuals of Fitted Link Ratios The analysis of run-off patterns often places great reliance on estimating individual link ratios from those observed in the data. Useful insights may be gained from looking at the size and pattern of the residuals on the individually fitted link ratios. The residuals of the link ratios could be calculated by taking the selected link ratio for a particular origin year and development period, and subtracting the actual link ratio for that development period. Widely-spread or skewly-distributed residual patterns may indicate that the estimate of the link ratio is highly uncertain or perhaps even inappropriate. Patterns within the residuals can also help identify distortions in the underlying data. For example, if one diagonal of development factors exhibits consistently high residuals, then this would indicate a calendar year effect, eg an internal exercise to speed up claim settlement. ## 4 Analysis of Emerging Experience ### 4.1 Why Analyse Emerging Experience? It is essential to monitor emerging experience to implement the actuarial control cycle effectively. The environment in which a general insurer operates is constantly changing. Monitoring the effect of past actions can help in revising a firm's strategy for risk management and in reassessing the risks that it faces. **Question** (i) Explain why an insurer might have to revise decisions after an analysis of experience. (ii) A general insurer has analysed the experience of its commercial buildings portfolio. List the key stakeholders to whom deviations in experience from expectations should be communicated. **Solution** (i) Why revise decisions The analysis may have revealed that the original decisions were not suitable because: - they were not good enough, ie wrong - new information has become available - the circumstances have changed. (ii) Key Stakeholders - Senior management - Pricing team - Reserving team - Claims department - Underwriters - Reinsurers - Brokers Where we have estimates and assumptions from the previous review, we are required to understand why these have changed at the current review. The key changes and their reasons for change need to be communicated to the stakeholders so they can understand and provide feedback if there are additional issues that may not have been considered. For this purpose, it is useful to break down the movement since the previous review in the estimated ultimate claims into its component parts, namely: - How emerging experience compares to that expected in the previous model (this will require the actuarial model used to have this predictive capability) - Changes in methodology - Changes in assumptions. Such a reconciliation of the figures can aid in understanding how the changes in experience have arisen and which factors are most significant, eg which individual assumptions have had the greatest influence. Different areas of the experience can be examined. For the incurred Bornhuetter-Ferguson method, for example, these could be the: - Difference between the actual and expected movement in incurred - Change in selected development pattern - Change in selected prior loss ratio - Change in selected premium estimate - Changes in other assumptions, for example inflation, if the inflation-adjusted Bornhuetter-Ferguson method is being used. Where we identify a consistent deterioration or improvement in one of these diagnostics, we should consider whether this is adequately reflected in the assumptions. This is particularly the case where emerging experience in prior years might suggest that the prior loss ratios should be increased. Where we have used the incurred average cost per claim method, we should break down the total movement into that relating to claim frequency and that relating to claim severity. We can estimate the expected movement in incurred claims by considering the previous development pattern and the previous estimated IBNR. For short-tail classes, we would expect a larger movement in incurred as a proportion of the previous estimated IBNR (sometimes known as the "IBNR burn rate") than for longer-tailed classes. It is helpful to understand the reasons for the business developing differently to what was expected (though this is not always possible). We can then consider whether the assumptions and methods remain appropriate considering this. For example, the selected method may inadequately capture certain aspects of the emerging experience. For example, if inflation is no longer stable then there may be a need to move to a method which makes an explicit allowance for future inflation as opposed to using a method which assumes past inflation will continue. It may be when we consider the reasons for the experience, this invalidates our prior assumptions, for example adverse experience may be caused by the shape of the development pattern at early stages underestimating the movements. In this case, we may choose to speed up the pattern despite the adverse experience. We should watch out for any trends emerging, and consider these in the reserving process. Where previous reviews have been carried out, there is a danger of anchoring. Where the previous basis has been used as a starting point, we may not adjust the assumptions quickly enough considering unexpected developments. However, there is also a danger that we may overreact in these situations. We should avoid changing assumptions too frequently in reaction to a short period of experience (eg one bad quarter should not necessarily mean the reserves are immediately strengthened materially). We must find a balance between the two tendencies. It is also useful to review periodically how accurate previous estimates and methods have been. We can learn from this what methods are working well in different situations. ## 5 The Underwriting Cycle ### 5.1 What is the Underwriting Cycle? The underwriting cycle is the process occurring over a period of years when premium rates for a given class of business oscillate between a high level at the top of the cycle (a "hard market") where the business written is typically profitable, and a low level at the bottom of the cycle (a "soft market") where the business written is typically unprofitable. **Hard Market** More companies enter market Business becomes profitable Profits are squeezed **Soft Market** Companies exit the market This is described in detail in Chapter 9. The length of the underwriting cycle varies by class of business and territory. For example, in some UK personal lines classes, it is around seven years. It will be dependent on, for example: - Macro-economic effects, eg people pay less for insurance and claim more when economic conditions are poor - Investment conditions (if it is expected that good returns can be made on invested premiums then the insurer may be prepared to offer softer premium rates) - Major industry losses, eg natural disasters or terrorism. ### 5.2 Impact of the Underwriting Cycle on the Assessment of Reserves The underwriting cycle can have an influence on claims development. For example, in a hard market, individuals who perceive themselves as low-risk may choose to self-insure rather than pay high premiums, resulting in anti-selection against the insurer. When assessing the reasonableness of the results of a reserving exercise, we should consider whether we have allowed appropriately for the underwriting cycle. One way to allow for the underwriting cycle in reserving exercises is to use a rate index when deriving the initial expected loss ratios for use in credibility-type methods. We should be careful, however, when selecting appropriate rate indices because: - Rate indices are typically only available for renewal business and therefore may not adequately allow for any differences between new and renewed business. - Rate indices can sometimes be constructed largely based on highly subjective information (such as the underwriters' views rather than hard data). - Rate indices will allow for changes in the attachment and limits on the policy. However these indices are often based on ILF curves, which may not fully reflect changes to claim profiles. ILF curves are discussed in Subject SP8. - Rate indices often fail to remove the underlying trends in the claims experience fully, leaving a residual loss ratio trend for the reserving actuary to explain. - Rate indices often struggle to allow for changes to terms and conditions adequately. Where there are more material rate changes, the method used to aggregate across all policies can give materially different results (eg simple or volume weighted mean, mean or median). - The selected rate index may be skewed by a small number of policies with above or below average rate change. When using a rate index, it is important to understand the inflationary allowance that is included within this and whether inflation needs to be explicitly allowed for beyond the allowance within the rate index. Changes to the types of claims covered and any claim frequency trends should also be considered and allowed for. It is also important to consider whether pricing already reflects inflation the claims are likely to be subject to or if its is a less precise match. For example, reflecting an increase in the value of a total loss may not reflect the inflation in the repair costs of the same item. Changes to commission rates over time should also be considered if the loss ratios being used are net of commission, as is common in the London Market. As the market moves between "soft" and "hard" conditions, there tend to be more changes to the underlying cover. For example, as the market hardens, terms and conditions tend to tighten with deductibles being added or policies attaching at a higher level. As the market softens, these trends will typically reverse. This can make reserving challenging as the underlying cover will be changing, and rates indices often fail to fully capture the impact of the changes. **Question** Suggest what impact the following will have on claims development patterns: (i) looser terms and conditions (ii) lower deductible. **Solution** (i) Looser terms and conditions - leads to claims arising which have different reporting and/or settlement patterns. It may also lead to more protracted claims due to litigation. (ii) Lower deductible - leads to a shorter claim development pattern. ## 6 The Reserving Cycle Recent studies have also suggested the existence of a "reserving cycle" which is highly correlated with the underwriting cycle. This appears to show that in a soft market, incurred claims development patterns are slower to develop (or longer-tailed) than in a hard market so that an unadjusted projection can underestimate ultimate claims in a soft market (and, equivalently, overestimate them in a hard market, when insurers can afford it). Potential reasons for this phenomenon include: - The effect of weakened terms and conditions - An increasing tendency to dispute claims - A possibly less conservative approach to case reserving when results are worse. The evidence of a reserving cycle is more noticeable for business which is already thought to be long tailed. The initial expected loss ratio can be chosen to take account of changes in the reserving cycle as well as changes in the underwriting cycle. This over-estimation or under-estimation of booked reserves is a decision to be made by the Board. It should not impact the actuary's best estimate, although they may wish to indicate a range of "best estimates" within which they believe the Board's decision should lie. The general insurer is likely to wish to flatten the reserving cycle: - So that reserves are more accurate. This reduces the likelihood of insufficient reserves being set up in past years, which will have a detrimental impact on the ongoing business - So that the profitability of the business can be more readily understood. Appropriate decisions can then be made as to whether to continue, contract or expand a class. ## 7 Comparison to Other Estimates Chapter 14 explains the various reserving bases that can lead to different results and explains the reasons for these. Here, we consider the case where two parties independently investigate the same data over the same period and estimate different results and different diagnostics. Benchmarks based on market data or peers are a useful comparison against the various diagnostics discussed previously. Care needs to be taken when comparing against market data since the policies offered may have different terms and conditions affecting the expected claims. Alternative estimates may come from another actuary or from management (including claims and underwriting teams). It is important to compare such results and diagnostics to identify areas of differences which may put into question the methodology and assumptions used. We should understand areas of difference and communicate these to the stakeholders. Possible reasons for differences between two estimates are as follows: - Data used. - The two estimates may have used different data sets. One dataset may be better in quality or have additional data or benchmark information. An extreme example of this would be in the situation of an acquisition where the buyer of the company will have limited access to data from which to estimate reserves.