Open AccessForecasting financial variables using artificial neural networks - dynamic factor model
Author(s) -
Ali Babikir,
Henry Mwambi
Publication year - 2017
Publication title -
journal of economic and financial sciences
Language(s) - English
Resource type - Journals
eISSN - 2312-2803
pISSN - 1995-7076
DOI - 10.4102/jef.v10i1.7
Subject(s) - design for manufacturability , autoregressive model , artificial neural network , dynamic factor , econometrics , computer science , mean squared error , series (stratigraphy) , statistics , artificial intelligence , economics , mathematics , engineering , mechanical engineering , paleontology , biology
In this paper we introduce a new model that uses the dynamic factor model (DFM) framework combined with artificial neural network (ANN) analysis, which accommodates a large cross-section of financial and macroeconomic time series for forecasting. In our new ANN-DF model we use the factor model to extract factors from ANNs in sample forecasts for each single series of the dataset, which contains 228 monthly series. These factors are then used as explanatory variables in order to produce more accurate forecasts. We apply this new model to forecast three South African variables, namely, Rate on three-month trade financing, Lending rate and Short-term interest rate in the period 1992:1 to 2011:12. The model comparison results, based on the root mean square errors of three, six and twelve months ahead out-of-sample forecasts over the period 2007:1 to 2011:12 indicate that, in all of the cases, the ANN-DFM and the DFM statistically outperform the autoregressive (AR) models. In the majority of cases the ANN-DFM outperforms the DFM. The results indicate the usefulness of the factors in forecasting performance. The RMSE results are confirmed by the test of equality of forecast accuracy proposed by Diebold-Mariano.