Research On Annuity Portfolio Risk Prediction Based On Transfer Learning

Variable annuities are important financial products that contain complex guarantees that are expensive to calculate in terms of risk.Thus,insurers use meta-modeling to value large portfolios of variable annuities to save time.However,the current meta-modeling method still has a low valuation accuracy,and a key problem is that it is very difficult to find a small number of representative annuity contracts.To solve this problem,two novel portfolio risk assessment models are proposed: Transfer Learning(TL)model and Dimensionality-Transfer Learning(DR-TL)model.The TL and dr-tl models proposed in this paper are implemented in Python3.5.Their main ideas are as follows: firstly,z-Score is used to normalize the input and output data sets;Then,the source domain data set is trained to generate the pre-training model,and k-means clustering is used to generate the representative contract in the representation space.Then monte Carlo simulation is used to evaluate the risk of representative contracts.Finally,the pre-training model of representative contract is used to predict the risk of new data.The difference between DR-TL and TL model is that DR-TL can well overcome the sensitive problem of high-dimensional clustering,so as to generate uniform and dispersed representative contracts.The contribution of this paper is that k-means clustering can solve the problem of invalid selection of representative contracts through representation space instead of input space,which lays a good foundation for improving model accuracy.In addition,because the transfer learning avoids the training from scratch,the training time is greatly saved,and the matrix inversion problem of Kriging model is solved.In this article,through the experiment proves that the proposed two models are better than baseline Kriging model performance,especially after DR TL model can be improved in greatly reduce the computational cost at the same time significantly improve the model accuracy,which not only can save insurance company in the aspect of hardware expensive cost,also can make a qualitative leap in risk management implementation has great application value.