|
| Reference | Data source and method | Findings | Remark |
|
| [6] | Overnight varieties of SHIBOR; back propagation | Mean square error (MSE) of BPNN and WNN were 0.0028 0.0017 | Cuckoo optimisation of WNN prediction model improved MSE to 0.0012 |
| Neural network (BPNN), |
| Wavelet neural network (WNN), Cukoo-WNN |
|
| [2] | 12 months US LIBOR data; genetic algorithm | Variance between prediction and real value was not more than 0.015 | GA + BRNN improved prediction compared to only GA or BRNN |
| Based bidirectional |
| Recurrent neural |
| Network (GA + BRNN) |
|
| [7] | US LIBOR history; | Convergence rate of 1.7 achieved | Faster convergence compared with finite
Difference method (FDM) |
| Partial differential |
| Equation (PDE) in one and two dimensions using |
| Radial basis functions (RBF) |
|
| [8] | Overnight SHIBOR daily data (Jan. 2007 to Dec. 2015); BPNN, WNN, and cuckoo search-wavelet neural network (CS-WNN). | BPNN prediction model gave a mean absolute error (MAE) of 0.0388 and MAE of WNN was 0.0281 | Cuckoo search algorithm for parameter optimisation of the wavelet neural network improved the prediction accuracy of the model |
|
| [9] | Overnight shibor daily data from January 4, 2007 to December 31, 2015; | BPNN gave MSE value of 0.0028 while the WNN gave MSE of 0.0017 | MSE of combined BPNN and WNN improved |
| BPNN |
|
| [10] | Daily american stock index standard and Poor’s 500 (SP), US LIBOR; Elman-Jordan neural network | Mean error, %: | Deutsche mark (DM) futures showed intermediate prediction. |
| S&P500 futures, DM futures, ED future of 0.217, 0.279, 0.201 |
|
