Chinese Currency Exchange Rates Forecasting with EMD-Based Neural Network

Table 6

Robust tests for NMSE comparisons with different activation functions.

	1-day	5-day	10-day	20-day	30-day
Panel A: MLP model
MLP(3)	0.0281	0.0877	0.2123	0.3729	0.5579
MLP(5)	0.0245	0.0820	0.2075	0.3729	0.5149
MLP(5,3)	0.0260	0.0859	0.2079	0.3491	0.4450
MLP(6,4)	0.0222	0.0887	0.1996	0.3436	0.4733
Panel B: EMD-MLP model
EMD(-1)-MLP(3)		0.0830	0.2045	0.3717	0.5285
EMD(-1)-MLP(5,3)		0.0862	0.2097	0.3662	0.5282
EMD(-2)-MLP(3)				0.3654	0.5571
EMD(-2)-MLP(5,3)				0.3665	0.4772
EMD(-3)-MLP(3)	0.0417				0.5782
EMD(-3)-MLP(5,3)	0.0393				0.3996
Panel C: EMD-MLP model
EMD(0)-MLP(3)
EMD(0)-MLP(5,3)
EMD(-1)-MLP(3)
EMD(-1)-MLP(5,3)
EMD(-2)-MLP(3)
EMD(-2)-MLP(5,3)

Consider the CNY from January 2, 2006, to December 21, 2015, with a total of 2584 observations. In this table, a hyperbolic tangent function is used as the activation function. This table compares the forecasting performance, in terms of the NMSE, for the MLP, EMD-MLP, and EMD-MLP models. We report the NMSE as percentage for l-day ahead predictions where and 30. The DM test [38] is used to compare the forecast accuracy of EMD-MLP (EMD-MLP) model and the corresponding MLP model. , , and denote statistical significance at 1%, 5%, and 10%, respectively. For each length of prediction, we mark the minimum NMSE as bold.