Practical Tools for Derivative Instruments based on Nonlinear Time Series Prediction

Derivative instruments are financial engineering tools. Th ey are linked to underlying financial instruments and are often better suited to manage exposure to risk. A number of tools exist for example for the management of interest rate risks. The optimal choice of a sp ecific instrument depends on a variety of factors. One crucial factor is the client’s estimate of inte rest rate developments in comparison to implied forward rates. The goal of the presented project is to develop practical com puter-based tools to assist market participants in assessing rate developments. In the area of int erest rate developments, we will analyze and predict rates using modern nonlinear models, such as multil ayer perceptrons. Unlike other approaches with nonlinear models, this consists of several steps: a mod el-selection step, based on an estimate of mean prediction error, including well known statistical me thods for time series prediction (ARMA); a step to estimate the predictive distribution, and a step to e valuate the predictive distribution in order to establish standard errors and confidence intervals. Predic tive distributions are estimated by less assumptive approaches, i. e., bootstrap methods. Another important aspect is user interaction. We will integ rate assisting tools into existing software at the trade center of the Bankgesellschaft to facilitate it s use. Model selection and estimation is done automatically while parameters depending on the actual ris k management strategy, i. e., the width of confidence intervals or the needed quantile, are chosen by th e user.