Table 1 Review of ANN articles on financial time series prediction
From: A chemical-reaction-optimization-based neuro-fuzzy hybrid network for stock closing price prediction
Citation | Methodology | Year | Application area/Dataset | Functionalities/Results |
|---|---|---|---|---|
(Yu et al., 2009) | Neural network metamodeling | 2009 | Financial time series/S&P 500, Euro/USD | Better performance over RW, ARIMA, ES, BPNN |
(Turchenko et al., 2011) | MLP with back propagation learning | 2011 | Short term prediction of stock prices of Fiat company, | 91% of prediction result < 5% error and 33% results have < 15 error |
(Aminian et al., 2006) | Artificial neural networks | 2006 | Forecasting economic data/U.S. Real Gross Domestic Production and Industrial Production | Neural networks significantly outperform linear regression due to nonlinearities inherent in the data sets |
(Mostafa, 2010) | MLP | 2010 | Kuwait Stock Exchange data 2001–2003 | MLP performed better to generalized regression neural networks |
(Zhuo et al., 2007) | Improved back propagation neural network | 2007 | Railway passenger traffic volume from 1980 to 1998 | Better to standard back propagation neural network |
(Blinova, 2007) | 28-time-lagged feed-forward ANN | 2007 | Intraregional and interregional passenger traffic for 2006–2010 of Russian air transport network | ANN models developed adequately described the passenger traffic demand for the next two or 3 year |
(Mostafa, 2004) | ARIMA and ANN | 2004 | Forecasting the Suez canal traffic flow | The models gave useful insight into the behavior of maritime traffic flows |
(Darbellay & Slama, 2000) | ANN and ARIMA | 2000 | Forecasting the short-term evolution of the Czech electric load | ANN are superior |
(Rahman et al., 2017) | Feed-forward NN and Elman NN with different learning methods | 2017 | Temporal and spatial atmospheric pollution index of Sterlitamak city | Quite effective prediction with 83% accuracy |
(Calderon & Cheh, 2002) | Review and limitations of ANN based models | 2002 | Audit and risk assessment | Scope and limitations of ANN models are explored |
(Ecer, 2013) | MLP, SVM, RBFNN | 2013 | Turkish banking failures/ 34 Turkish commercial banks, 17 of which failed the periods 1994–2001 and contains 36 ratios available for those types of banks | MLP performed better with 2.94% error. |
(Zhong & Enke, 2017) | ANN with PCA, fuzzy robust PCA, kernel-based PCA | 2017 | Forecasting daily direction of S&P 500 Index ETF return based on 60 financial and economic features | ANN + PCA gives better result than other approaches |
(Zhang, 2003) | ARIMA+ANN | 2003 | Bench mark time series datasets | The hybrid model had superior performance |
(Enke & Thawornwong, 2005) | Multilayer feed-forward NN and Generalized regression NN, Probabilistic NN | 2005 | Monthly data (March 1976–December 1999), total 286 periods from S&P 500 | Effectiveness of NN models were established for level estimation and classification. Trading strategies guided by NN models were able to generate higher profits. |
(Ture & Kurt, 2006) | ANN | 2006 | Future expectations index, CDI interest tax rate, Selic interest tax rate/Formal employment, Brent oil price, Domestic market automobile sales, Consumer confidence index, Investors participation | Percentage of change in direction (POCID) is 93.62% for test set and 87.50% for validation set. |
(Niaki & Hoseinzade, 2013) | ANN | 2013 | S&P 500 index 365 trading days | POCID of linear regression is 51.75. ANN had better POCID than LR |
(Alalaya et al., 2018) | ANN + Fuzzy | 2018 | Amman stock exchange of financial and banking sector from 1/2010 to 12/2016 with record data 265 | Fuzzy-neural models are found prominent in terms of MSE, MAD |