Modeling the CNB's Monetary Policy Interest Rate by Artificial Neural Networks

Knowledgeability about interest rates set by a central bank is very important for all participants in an economy. In this paper we have used publicly available data to model how Czech National Bank manipulates its 2W repo rate when conducts its monetary policy. For this purpose, eight indicators are chosen. They are the Consumer Price Index (CPI), GDP growth rate (HDP), the monthly exchange rate EURCZK (KURZ), the monthly growth rate of monetary aggregate M2 (M2), the monthly unemployment rate (NEZAM), the monetary policy interest rate of the European Central Bank (EBC), the two-week Prague Interbank Interest rate PRIBOR14 and Economic Sentiment Indicator (IES). First, they are used as explanatory variables and then as the input signals to two different artificial neural network types with different architecture: the multilayer perceptron (MLP) and radial basis function (RBF) nets with different number of hidden neurons to model 2W repo rate of CNB. As a result, we fi nd that while the RBF network fails to provide stable results superior to the one of the linear model, the MLP network always can deliver better results than the one of the linear model. The best results are achieved with a network with only two hidden neurons. Further, these results are relatively stable with minimum time needed to complete the calculation. The MLP network therefore seems to be a promising tool for modeling the 2W repo rate of CNB.