INSURANCE VALUATION: A TWO-STEP GENERALISED REGRESSION APPROACH

Abstract Current approaches to fair valuation in insurance often follow a two-step approach, combining quadratic hedging with application of risk measure on the residual liability, obtain cost-of-capital margin. In such approaches, preferences represented by regulatory are not reflected process. We address this issue an alternative procedure, based generalised regression arguments, which leads portfolios that neutral respect measure, as Value-at-Risk or expectile. First, portfolio traded assets aimed at replicating liability is determined local hedging. Second, hedged using objective function. The margin then defined cost capital required hedge liability. case quantile used second step, yearly solvency constraints naturally satisfied; furthermore, minimiser among all satisfy constraints. present neural network algorithm for and liabilities backward iterations scheme. fairly general easily applicable, it only requires simulated paths drivers.