Efficient SCR Estimation Using Machine Learning Algorithms

...accurately approximating the non-linear effects of complex options and guarantees on the market value of liabilities. This article investigates the application of Artificial Neural Networks (‘ANN’) as an alternative for
polynomial based curve-fitting. The accuracy of the SCR calculation is significantly improved upon with Neural Networks, even when using a dataset as small as 10 data points.

Introduction

Solvency II requires insurers to calculate the SCR. In an internal model setting, calculating the SCR requires many simulations, i.e. stochastic revaluations of the entire balance sheet, which are very time consuming. Reducing the amount of simulations can be achieved through the application of a proxy model. Part of the complexity in calculating the SCR in a proxy model arises from the complex nonlinear relations between movements in risk factors (e.g. credit spread risk) and corresponding movements in the market value of liabilities.
Traditional approximation methods lack the flexibility to capture the non-linear effects of changes in risk factors on the market value of certain liability portfolios. This might lead to the capital requirements being under- or overestimated.

In practice, a number of different approaches to estimate the SCR are used. A polynomial based curve-fitting approach is the most commonly applied technique. This research focuses on the application of ANNs as an alternative for this polynomial based approach. An advantage of ANNs over polynomials is that they are theoretically able to capture any relationship that is in the data. The challenge is to make the ANN provide accurate and stable predictions on a small dataset, as these models typically have a large number of parameters.

To determine whether Neural Networks are an improvement over polynomials, the market risk SCR of a synthetic balance sheet is estimated. In particular, the non-linear relationship between changes in the volatility adjustment and the market value of liabilities is examined. The asset side of this balance sheet consists of bonds with various ratings and durations. The liabilities of this balance sheet are represented by (simple) deterministic future pension payments. This balance sheet is sensitive to credit spread risk (corporate financial, non-financial and sovereign) and interest rate risk.

Read the full article under Download.