In:
Zeitschrift für die gesamte Versicherungswissenschaft, Springer Science and Business Media LLC, Vol. 110, No. 1 ( 2021-02), p. 21-48
Kurzfassung:
The possibility to adapt a premium in German private health insurance depends on the so-called triggering factor. Its value is determined via a linear extrapolation of the loss ratios of the three preceding years. To predict this value early and in a reliable way is of great importance for risk management purposes. We therefore examine the performance of various prediction methods that range from classical time series methods and regression to neural networks and hybrid methods. While regression with ARIMA errors performs best among the classical methods, the overall best method is the combination of time series predictions with a neural network that is trained on deseasonalized and detrended data.
Materialart:
Online-Ressource
ISSN:
0044-2585
,
1865-9748
DOI:
10.1007/s12297-021-00493-1
Sprache:
Deutsch
Verlag:
Springer Science and Business Media LLC
Publikationsdatum:
2021
ZDB Id:
200636-4
ZDB Id:
2425530-0
SSG:
3,2
Permalink
