To minimize the difficulties in calculating claims inflation from data, it is recommended that one should use large data sets of claims from many different policyholders. This would minimize the numerical instabilities as far as applicable. The empirical index approach is not recommended for a small data set because it can show erratic results. It is recommended to use robust statistics, however, it also has undesirable side effects such as its focus on the middle distribution.

When the history of estimates is needed, for example, it is essential in the field to record the reporting delays of non-zero claims, the quantification date is used in the system. This date provides factors used in the computation of compensation.

The exposure measure is proportional to the risk of a policyholder. To Increase its objectivity and accuracy, different factors relevant to risk are considered. More factors are introduced in the calculation of exposure over time as experience is accumulated.

These criteria may be used as a starting point in the measure of exposure. The most important of the three is the proportion of the measure to the expected losses. When the measure is objective and easy to obtain, it can be evaluated reliably.

The three components of a Catastrophe Model are the hazard module, vulnerability module, and the financial module. The hazard module is concerned with the catastrophic event’s characteristics such as its strength, location, and frequency. The vulnerability module provides vulnerability curves. These vulnerability curves show the degree of damage such as the magnitude of an earthquake. The financial module translates physical damage to financial loss. Its main outputs are the event loss tables (ELTs) that give a detailed list of events and related losses.

Each risk in the individual risk model can only have a loss once which theoretically justifies the use of the binomial/ multinomial distribution. However, In a collective risk model, it is better to use a distribution with a variance/ mean ratio that is equal to that of a Poisson process because more than one event is possible for a given risk, unlike the individual risk model. In the Poisson process, the interarrival time is distributed exponentially. This process is memoryless because the occurrence of a loss does not change the probability that another will occur in a given interval.

To arrive at the average point of occurrence of the renewal policy, the full incurred amount of all claims is needed to be revalued by using the appropriate inflation index. Using the full incurred amount is essential to making an informed guess as to the value of a claim at a point in time.

Simple statistics are used to summarize data for each policy year. The information needed for data summarization of each policy year consists of total losses, Standard Deviation, and Number of Claims. These pieces of information also have the role of creating data that allows readers to observe the trends and behaviors of the risks. There is also a list of top losses that allow the readers to analyze the risks.

Before the raw set of data is transformed into a set of results, it undergoes the three-step approach in data preparation – data cleansing, data transformation, and data summarization. This is done to filter the raw data and analyze its quality whether or not there are errors or anomalies. Another purpose of data preparation is to revise confidential information for data privacy purposes such as names, amounts, etc.

When all claims are above a certain level of deductible, it negatively pushes up the claims which were previously not part of the data set. As a result, the claims will not be comparable and there is no way of deducing inflation from statistics that are based on the empirical severity distribution above a fixed threshold.

Risk-sharing aims to spread the risk among those involved. When the principal insurer passes some of the risks to another insurance company, another party becomes involved called the reinsurer. In other words, the direct insurer can be likened to the policyholder who also purchases insurance. Therefore, the relationship of the policyholder with the direct insurer and the direct insurer with the reinsurer is parallel.

When a direct insurer enters into a reinsurance agreement, the reinsurer protects the direct insurer from having the sole responsibility for the tails of the distributions of large claims. Thus, the effects of a reinsurance agreement are the reduction in the variability of the claims which is paid out by the direct insurer and the reductions of the mean amount paid out by the direct insurer on claims. Besides. The direct insurer will have a reduction in the probability of paying a large amount on a particular claim.

Under proportional reinsurance, there is an agreement on the proportion in each claim that will be covered by the direct insurer. Both direct insurer and reinsurer have unlimited liability in this reinsurance agreement. Thus, this agreement is not favorable to the direct insurer.

The Burn Cost Analysis is one of the most used pricing techniques because of its simplicity despite its many limitations. With this pricing technique, the results of the frequency/ severity modeling are better understood due to the sense-check that it provides. Thus, this technique achieves many useful results without resorting to Stochastic Modelling.

The revaluation factor in the burning cost analysis, in general, reflects company-specific variables such as product price changes. Therefore, this factor is dependent on the status of the company since the burning cost analysis is mostly based on experience.

The burning cost analysis only relies on historical information which may impair the views on future risks. Since this pricing technique is not model-based, there is no attempt to find regularities in the data and predict beyond its range of experience. It is not possible to estimate the full aggregate loss distribution and predict the richness of possible outcomes.

The modelling process starts with an analyst’s experience and prior knowledge. With an analyst’s professional opinion, models that are better suited for a specific client or data set are chosen. As a result, experience and prior knowledge save time on relying solely on the available data. Its input increases the efficiency and effectiveness of the model due to the real regularities present in the data.

All models are affected by the prediction error to some extent. Selecting the best model is one of the most essential activities in actuarial modelling for general insurance. One method used is selecting from a number of statistical distributions that are the most suitable for the data. The goal of modeling selection is to use the model that has the most predictive power.

Statistical learning is also known as “machine learning”. An important note to keep in mind about statistical learning is that increasing the complexity of models does not increase its accuracy in making predictions. A model that is closer to reality may sometimes be better at making predictions and decisions.

The development of claims count triangles is the most common method for the estimation of the number of IBNR claims. Chain ladder or other similar triangle development techniques are used for projecting the claim count. First, the claims count in a triangle is organized in a cumulative format. Only then can the chain ladder method be applied to project the number of claims to ultimate.

Assumptions of stationary and independent increments are not easy to justify in practice. The number of claims used in the Poisson distribution does not reflect the stability in practice. The Poisson rate is only estimated based on a small number of years, hence the parameter uncertainty.

The correct order is individual loss data, exposure data, portfolio/market information, then cover data. In frequency and severity modeling, exposure data and portfolio information are used for several of its steps.

The Poisson process is only adequate when systemic effects play a negligible role, therefore, elucidating the impression that the use of the Poisson Process can only be justified when the Poisson rate is genuinely constant. However, there are several factors that affect the constancy of the Poisson rate such as the volatility of the loss experience is likely to be underestimated when systemic effects are common and Poisson distribution is assumed. With the presence of seasonal volatility, the loss process can only be modeled roughly and a sophisticated approach is necessary.

Process Uncertainty happens when an element exhibits random fluctuations driven by the loss generation process. It is intrinsic to the stochastic phenomena which does not have a stable pattern or order. Its effect can be calculated by producing models that show the number of losses or its total amount.

Exposure rating is unnecessary in personal lines insurance. Loss is experienced normally which makes the use of an exposure rating inappropriate. On the other hand, exposure ratings are useful when the experience is barely sufficient and diversified.

Extended warranties allow the insured to prolong the warranty of a purchased product in exchange for a premium. It may be provided by retailers, independent insurers, or manufacturers. The failure rate is the most important entity of extended warranty risk.

The exposure information is obtained from the number of sales per day or month, then subdivided the type of product, territory, and other relevant factors. Claims data are information such as Claim ID, purchase date, repair date, type of defect, etc.. This information is needed to determine the failure rate which has the goal to predict the number of failures over the timeframe of the extended warranty.

The base failure rate the failure rate during the mature phase. Adjustments are made to project the failure rate during the extended warranty period. It is calculated by subdividing all sales in a specific number of months.

Standard actuarial techniques do not work for catastrophes. The client’s experience in catastrophes is near-zero credibility. Instead of the client’s experience, they use a database of historical events for a country or the whole world and scientific understandings of phenomena such as hurricanes and earthquakes.

When the client experience is barely sufficient, the portfolio data provides factors to consider in pricing. Portfolio data in this context is the non-client data owned by the insurer or intermediary. With the portfolio data and the client’s data combined, a better estimate of the expected losses will be calculated.