University of Twente Student Theses


Modeling Lapse Rates: Investigating the Variables That Drive Lapse Rates

Michorius, Cas Z. (2011) Modeling Lapse Rates: Investigating the Variables That Drive Lapse Rates.

[img] PDF
Abstract:In the life insurance industry individually closed contracts are accompanied by risks. This report focuses on one of these risks, more specifically, the risk involving the termination of policies by the policyholders or, as it is called, “lapse” risk The possibility of a lapse can influence the prices of contracts, necessary liquidity of an insurer and the regulatory capital which should be preserved. The possibility of a lapse is reckoned to account for up to 50% of the contract‟s fair value and one of the largest components of the regulatory capital. For these reasons it is of great importance to prognosticate lapse rates accurately. These were the main reasons for conducting this research on behalf of Achmea and for investigating models and explanatory variables. The research question which functioned as the guide line for this research at Achmea is the following: Can the current calculation model for lapse rates be improved1, while staying in compliance with the Solvency II directive, and which variables have a significant2 relation to the lapse rates? The model applied and the explanatory variables analyzed are the result of a literature study. This study provided the Generalized Linear Model [GLM] to be a suitable choice and led to a list of 38 possible explanatory variables of which 9 were tested 3. The GLM was applied to the data of CBA and FBTO corresponding to the years 1996 to 2010 and aggregated per product group. The seven product groups that were analyzed were: mortgages, risk, savings (regular premium), savings (Single premium), unit-linked (Regular premium), unit-linked (Single premium) and whole life & funeral. The aggregation of the data has been done using Data Conversion System and Glean, two products of Sungard, and the data were analyzed using SPSS 17, a product of IBM. The research provided seven models, one for each product group, including variables as “buyer confidence”, “first difference in lapse rates”, “gross domestic product [GDP]”, “inflation”, “reference market rate” and “return on stock market”. Every model provided more accurate predictions than the application of the mean of the data would. It should be noted that, due to lack of data, this comparison has been done on the training set. The performance of the models, when compared with the model provided by regulatory bodies (standard formula), is dependent on the level of expected lapse rates as well as the relative error of the predicted values. The level of the expected lapse rates greatly influences the standard formula, whereas the relative error of the predicted values is one of the great contributors to the prediction interval of the developed model. Additional research showed that the choice for division of the data into several product groups is supported by the huge diversity in lapse rate developments amongst the product groups. Further analysis of the lapse rates with respect to the duration of policies also provided a reason for further research. The analysis indicated that the effect of macro-economic variables on lapse rates is dependent on its duration, indicating that the data per product group can be subdivided or duration can be used as explanatory variable. Based on the research results it is recommended to analyze the possibility of generalizing the results by extending the research to other parts of Achmea. Next to that, it is recommended to investigate the data on a policy level in order to assess the significance of other variables. These additional researches will also increase the statistical strength and accuracy of the inferences that can be made. It is also recommended to clarify the importance of (accurate recording of) lapse rates and to denote a universal definition of a lapse, all to make sure that the lapse data become unpolluted. Finally, it is advised to monitor the models and to examine their performance and sensitivity to new data.1 The performance of the model has been measured in terms of accuracy, on which it has also been compared. 2 The significance of the variables has been tested by statistical measures using a 5% level of significance. 3 Lagged values of these variables have been included as well, which led to a total of 14 analyzed variables.
Item Type:Essay (Master)
Achmea holding
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:85 business administration, organizational science
Programme:Business Administration MSc (60644)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page