This paper presents a multistage stochastic programming model for refinancing mortgages with non-contractual maturity under liquidity restrictions in the market. An extension to the management of other products such as savings accounts is straightforward. The evolution of interest rates is modelled by principal components for short-term and a two-factor mean reversion model with long rate and spread for long-term planning. Barycentric approximation provides tight lower and upper bounds for the original problem with relative discretization errors in the order of one per cent.
The management of non-maturing account positions in a bank's balance like savings and sight deposits as well as certain types of variable-rate mortgages is complicated by the embedded options that its clients may exercise. In addition to the usual interest rate risk, uncertainty in the timing and amount of cash flows must be taken into account when investment or refinancing strategies are determined. This paper introduces a multistage stochastic programming model where the stochastic evolution of interest rates and volume under management is described by stochastic processes in discrete time. Scenarios are generated by means of barycentric approximation which is particularly useful to deal with the observed correlations between interest rates and volume. Practical experience from the application at a major Swiss bank is reported where the model has been employed since the mid-90s.
The risk management of non-maturing account positions in a bank's balance like savings deposits or certain types of loans is complicated by the embedded options that clients may exercise. In addition to the usual interest rate risk, there is also uncertainty in the timing and amount of future cash flows. Since the corresponding volume risk cannot directly be hedged, the account must be replicated by a portfolio of instruments with explicit maturities. This paper introduces a multistage stochastic programming model that determines an optimal replicating portfolio from scenarios for future outcomes of the relevant risk factors: Market rates, client rates and volume of the non-maturing account. The weights for the allocation of new tranches are frequently adjusted to latest observations of the latter. A case study based on data of a real deposit position demonstrates that the resulting dynamic portfolio provides a significantly higher margin at lower risk compared to a static benchmark.
The incorporation of single-factor interest rate models within the stochastic programming methodology is investigated and applied to multiperiod investment. Barycentric approximation is used for discretizing the stochastic factors and for generating scenario trees which take the various term structure movements into account. It is shown that employing the Vasicek model for the instantaneous rate process preserves convexity of the stochastic multistage program and, hence, guarantees information on the accuracy of the approximate investment strategies. To the contrary, the convexity of the program cannot be assessed if the square root process due to Cox-Ingersoll-Ross is used for the instantaneous rate. In this case, the approximate investment policies and their associated interest surplus may be accepted as estimates. Numerical results for 8-period and 6-period investment problems are discussed.
