Analiza zmian kursu euro/dolara: model VAR i perceptron wielowarstwowy

In the financial time series analysis one often recommends the application of the multidimensional methods such as the vector autoregression model – VAR, that was proposed by Sims in 1980. The feedforward artificial neural networks, especially multilayer perceptron – MLP, can be considered as an alternative, for VAR model, tool. In the paper we discuss the results of the euro/dollar time series investigation that is provided employing VAR and MLP models. The efficiency of both methods is evaluated in terms of ex-post errors. The source of the analysed series is REUTERS data base from the 4th of January 1999 till the 5th of December 2003. In our investigation we consider the euro/dollar exchange rate and selected financial instrument time series. The models are constructed for the euro/dollar exchange rate that is transformed into daily logarithmic rate of returns