Competitive Insight: How machine learning algorithms outperform traditional pricing methods


Bor Harej, actuary Prime Re Solutions, expert in the field of machine learning and former President of the Slovenian Actuarial Association, presented his latest findings at the traditional Bahnhofskolloquium of the Swiss Actuarial Association in Zurich in early January this year. He examined whether machine learning algorithms could outperform traditional pricing methods when it comes to the calculation of insurance policy premium rates. For this purpose, he compared several algorithms calibrated on synthetic policy and claim data of a total of 1 million motor liability insurance policies, where „true“ expected claims are known.


He demonstrated that in case of two imaginary insurers on the market, where one uses a traditional pricing method, another one can gain important competitive insight with the use of other machine learning algorithms. The use of machine learning algorithms, specifically the use of Light Gradient Boosting and Neural Networks, leads to more accurate calibrations and thus increased the profitability of the investigated motor liability insurance policies in comparison with Generalized Linear Models. Assuming an elastic market, a subtle winner’s curse mechanism sets in and compels the worst risks to buy cheaper policies from the traditional tariffs.


For further information on machine learning or if you would like to conduct a case study with your real data, contact us at solutions@prs-zug.com.


Bor’s slides can be downloaded from the website of the Swiss Actuarial Association.


Find out more about Bor Harej.


#machinelearning #NN #neuralnetworks #insurance #reinsurance #pricing #lightgradientboosting #profitability #insurancepolicy #insurancepolicies #premiumrates #competitiveinsight


396 views