Chinese Currency Exchange Rates Forecasting with EMD-Based Neural Network
Table 10
Performance of trading strategy based on the best EMD-MLP.
SD%
Return with transaction cost (%)
0.0
0.1
0.2
0.3
Panel A:
1-day
EMD(-1)-MLP(3)
812
1.46
13.59
−11.41
−36.41
−61.41
5-day
EMD(-1)-MLP(3)
822
1.53
7.12
2.12
−2.88
−7.88
10-day
EMD(-1)-MLP(5,3)
830
1.70
6.19
3.69
1.19
−1.31
20-day
EMD(-1)-MLP(5,3)
823
1.76
4.98
3.73
2.48
1.23
30-day
EMD(-2)-MLP(5,3)
823
1.73
4.59
3.76
2.92
2.09
Panel B:
5-day
EMD(-1)-MLP(3)
32
3.65
36.78
31.78
26.78
21.78
10-day
EMD(-1)-MLP(5,3)
112
2.80
19.18
16.68
14.18
11.68
20-day
EMD(-1)-MLP(5,3)
220
2.03
12.15
10.90
9.65
8.40
30-day
EMD(-2)-MLP(5,3)
364
1.75
8.30
7.47
6.64
5.80
Panel C:
5-day
EMD(-1)-MLP(3)
5
4.93
85.02
80.02
75.02
70.02
10-day
EMD(-1)-MLP(5,3)
17
4.72
40.20
37.70
35.20
32.70
20-day
EMD(-1)-MLP(5,3)
71
2.29
18.42
17.17
15.92
14.67
30-day
EMD(-2)-MLP(5,3)
131
1.72
13.08
12.24
11.41
10.58
In terms of NMSE and , we select the best models to forecasting l-day ahead predictions where and 30. The third column shows the sample size in each model and is denoted as . By the trading strategy rule as (11) with , we generate l-day returns. The fourth column reports the standard deviation of the annualized returns without transaction cost. The last four columns represent the averages of annualized returns with different transaction costs.