| Summary: | Interest rate forecasting is one of the most challenging tasks in modern nance and economics.
Several studies, examining dierent factors and statistical models, have been
employed, however they often failed to beat a simple random walk. This study suggests
the use of nonlinear, nonparametric models, namely Support Vector Machines (SVM) and
Articial Neural Networks (ANN). The methodology employed uses interest rate levels
and spreads to predict the daily changes of UK six-month, one-year, three, ve and tenyear
spot rates, six months forward. Results are promising, providing evidence in support
of the term structures predictive content. Both methods are able to identify the overall
trend of future interest rate movements and perform well in terms of Root Mean Squared
Error and Mean Absolute Error outperforming, for all maturities, the simple random walk
model. Greater accuracy is achieved in terms of predicting the correct direction and by
considering longer-maturity rates. Comparison between the two methods shows that the
accuracy and generalisation performance of SVMs is superior to that of ANNs in almost
every aspect.
ii
|
|---|