Detecting Business Cycle Asymmetries Using Artificial Neural Networks and Time Series Models

Abstract

This study examines possible existence of business cycle asymmetries in Canada, France, Japan, UK, and USA real GDP growth rates using neural networks nonlinearity tests and tests based on a number of nonlinear time series models. These tests are constructed using in-sample forecasts from artificial neural networks (ANN) as well as time series models. Our study results based on neural network tests show that there is statistically significant evidence of business cycle asymmetries in these industrialized countries. Similarly, our study results based on a number of time series models also show that business cycle asymmetries do prevail in these countries. So we are not able to evaluate the impact of monetary policy or any other shocks on GDP in these countries based on linear models. Copyright Springer Science + Business Media, Inc. 2005B22, C32, C45, E32,

Predicting water allocations and trading prices to assist water markets

“…These include forecasting financial and economic time series ; stock price predictions (Tsang et al, 2007); mutual fund asset value forecasting (Chiang et al, 1996); and a comparison of artificial neural network and time series models for forecasting commodity prices (Kohzadi et al, 1996;Shahwan and Odening, 2007); electronic commerce systems for selling agricultural products (Wen, 2007); short-term food price forecasting (Haofei et al, 2007); forecasting hog prices (Hamm and Brorsen, 1997); forecasting residential property prices (Wilson et al, 2002); cost functions predictions (Fleissig et al, 2000); business cycle asymmetries and GDP growth rates (Kiani, 2005); and a review of neural network applications in finance (Wong et al, 1997;Wong and Selvi, 1998). Water economics.…”

Testing Forecast Accuracy of Foreign Exchange Rates: Predictions from Feed Forward and Various Recurrent Neural Network Architectures

“…However, over fitting issues associated with neural networks can be eliminated with careful construction of the neural network architecture. Therefore, as recommended in Kiani (2005), and the studies cited therein, we incorporated a small number of biased nodes in our neural network models so as to avoid overfitting issues associated with neural network approximations. Likewise, we are interested in the accurate forecasts from the model rather than parameter estimates because in this study we seek to predict changes in forward foreign exchange rates in British pound, Canadian dollar, and Japanese yen.…”

“…Therefore, in the present research, we employ fewer hidden nodes 2 in our neural networks as was recommended in Kiani (2005) and Kastens et al (1995), since neural networks with fewer hidden nodes have better generalization and less over fitting problems, although there are number of rules of thumbs for selecting biased or hidden nodes for neural networks models. For example, Lippmann (1987) suggested 2n + 1 nodes, Wong (1991) 2n, Tang andFishwick (1993) n, andKang (1991) recommended n/2 hidden nodes but suggested trial and error as the best way for selecting the number of hidden nodes for a neural network model.…”

“…Qi and Wu (2003), who employ an ANN model with monetary fundamentals, find that their model cannot beat the random walk model. Alternatively, Kiani (2005) and Kiani et al (2005) used ANN models with macroeconomic time series and found that they outperformed the linear as well as other nonlinear models they employed. Lisi and Schiavo (1999) perform a detailed comparison of neural network and chaotic models for predicting monthly exchange rates.…”

“…Instead we employ an alternative strategy of selecting a minimum number of hidden nodes due toKiani (2005) andKastens et al (1995) to avoid the overfitting problem in our neural networks. Moreover, rather than employing neuro-coefficient smooth transition autoregression (NCSTAR) due toMedeiros and Veiga (2000), we employ various types of recurrent neural network models as was done due toTenti (1996) and others.…”