UBC Theses and Dissertations
Bank asset and liability management Kusy, Martin
The inherent uncertainty of a bank's cash flows, cost of funds and return on investment, along with the increased variability of economic conditions during the past decade, have emphasized the need for greater efficiency in the management of a bank's assets and liabilities. A consequence has been an increased number of studies on how to structure a bank's assets and liabilities so that an "optimal" trade-off exists between risk, return and liquidity. Except for the Bradley and Crane (BC) model, the solution techniques proposed in the literature are computationally tractable only if uncertainty is ignored. Unfortunately, the BC model is not operationally appealing due to severe computational limitations, and a number of undesirable formulation features (such as the restricted feasible region for first period decisions). Given these deficiencies in the literature, the purposes of this dissertation are to develop an asset and liability management model (ALM) that is computationally tractable for large realistic problems and to demonstrate that this model is superior to existing models. The ALM model developed in this dissertation is a stochastic linear program with simple recourse (SLPR). This model incorporates the following essential features of asset and liability management: 1) the stochastic nature of the problem (by utilizing a set of random cash flows (deposits) with a given discrete probability distribution), 2) simultaneous consideration of assets and liabilities, 3) transactions costs, and 4) multi-periodicity. The ALM model was applied to Vancouver City Savings Credit Union's asset and liability management for a five year planning period in order to demonstrate the effort necessary to implement the model. Computational tractability for this large problem was maintained by using Wets' algorithm for solving SLPR. A simulation was run on a real (uncertain) environment to compare the decision making effectiveness of the solutions generated by the SLPR and stochastic dynamic programming (SDP) models. The findings of this dissertation are: 1) the ALM model is superior to an equivalent deterministic model, 2) the solution of the ALM model is sensitive to the asymmetry of the probability distributions of the cash flows, 3) the effort required for the implementation of the ALM model is comparable to that of an equivalent deterministic model, 4) the SLPR formulation is computationally superior to the SDP formulation utilized by Bradley and Crane, and 5) the simulation indicates that the SLPR formulation results in a better initial period decision than the SDP formulation (this is due to the restrictions imposed by the SDP formulation of maintaining feasibility for all possible forecasted economic scenarios for the first period decision).
Item Citations and Data