Application of Neural Networks in Economic Forecasting

The complexity of economic processes is reflected in the time series which register their state. Not all the aspects of the economic process can be registered. In order to obtain the useful information from statistical data, it is necessary to apply many labor-consuming and sophisticated procedures. Economic conditions are represented by objective processes, such as industrial production, product prices, export and import, employment and unemployment, job vacancies, etc. on the one hand, and behavioural modes of businessmen and consumers, their assessments and expectations as regards prices, sales, employment, and other economic indexes on the other hand. So we are dealing with objective facts (quantitative data) and subjective facts (qualitative data). Moreover, the analysed processes are all interdependent. Such a situation requires extreme methodological flexibility - a sort of methodological eclecticism. In view of the multidimensional object, the use of merely one method may yield a distorted image. That is why several different methods have been used in this study: the naive no-change method, simple linear regression, auto-regressive (integrated) moving-average model (ARMA and ARIMA), and artificial neuronal networks