Market-consistent economic scenarios are at the core of the measurement of the Technical Provisions under Solvency II. To model interest rates in a risk-neutral environment, a calibration procedure is used to match the observable financial instruments in the market. Weights reflecting the sensitivity of the insurer’s liabilities are therefore to be assigned to each market information within the calibration process. This paper presents how neural networks can help deal with the intensive recalibrations required to achieve this objective. Our discussion includes the following sections:
- The weights calibration problem
- Modelling interest rates
- Calibration using neural networks
- Weights design results
Neural network calibration of the DDSVLMM interest rates model, and application to weights calculation
We present a calibration technique for one complex risk neutral model, relying on neural networks and significantly reducing computational time.