Anna Górecka Maciej Szmit Exchange Rates Prediction by Arima and Neural Networks Models Article from Artificial Neural Networks Workshop on 47th International Atlantic Economic Conference The newest history of Warsaw Stock Exchange begun in March 1991 when Act on Public Trading in Securities and Trust Fund was adopted. The State Treasury established the Warsaw Stock Exchange (WSE in polish WGPW) joint-stock company in April 1991. At the same time, the Polish Securities Commission, with its chairman appointed by the Prime Minister, was created. Both the structure and the legal regulations of the Polish capital market were patterned after the most modern and efficient system based on French experience and was adopted and implemented with the help of experts from Societe des Bourses Francaises and the French Depository (SICOVAM). At the first period of its history of WSE was using the ? call market? system. It is also known as the ? Single-price auction? (French: par casier or German: Einheitskurs). Its main feature is that a single price per security emerges at each session as a result of the orders submitted. Transaction can be concluded even for one individual security. In their orders, clients of brokerage houses define the quantity and the price of securities. In his order investor can indicate the price limit or "at the market". The validity of an order can not exceed the end of the next month. The price on the equities market can only be higher (upper limitation) or lower (lower limitation) from the previous session's price by a maximum of 10%. For bonds maximum price change is 5 percentage points. The main index of WSE is named Warsaw Stock Index (WSI in polish WIG) is Total return index, which includes dividends and pre-emptive rights (subscription rights). Base value of it was 1000 and the highest value in history was about 20000. Construction of WIG based on the following formula: In our research we used the multi-layer ANN trained by backpropagation algorithm to predict exchange rates on Warsaw Stock Exchange. We compared forecasts obtained by ANN to the ones generated by for Auto-Regressive Integrated Moving Average Process - ARIMA.