Selection of data and training strategy.

Published data from Colombo Stock Exchange from 2 nd April 2001 to 23 rd May 2001 was used for ASPI and Milanka Price Index. However it was not possible to obtain daily figures of Call Money interest rate which is very sensitive and instrumental in capital market fluctuations. The set of data used in the process of training the ANN is annexed as table 4.

 

 

 

 

 

 

 

 

ID

DATE

ASPI

MLANKA

TBILLS

PLR

CMR

1

02-04-01

423.10

647.10

18.90

22.03

21.75

2

03-04-01

420.70

643.10

18.90

22.03

21.75

3

04-04-01

419.20

643.40

18.90

22.03

21.75

4

05-04-01

419.80

643.50

18.90

18.90

21.75

5

06-04-01

421.00

641.20

18.90

18.90

21.75

6

09-04-01

420.10

640.40

18.90

18.90

21.75

7

10-04-01

420.60

639.80

18.90

18.90

21.75

8

11-04-01

421.10

641.00

18.90

18.90

21.75

9

16-04-01

424.00

644.10

18.90

18.90

21.75

10

17-04-01

422.90

639.90

18.90

18.90

21.75

11

18-04-01

424.00

648.40

18.90

18.90

21.75

12

19-04-01

419.40

636.80

18.90

18.90

21.75

13

20-04-01

416.50

633.50

18.90

18.90

21.75

14

23-04-01

416.90

633.60

18.90

18.90

21.75

15

24-04-01

417.10

633.90

18.90

18.90

21.75

16

25-04-01

417.10

632.50

18.90

18.90

21.75

17

26-04-01

417.20

632.10

18.90

18.90

21.75

18

27-04-01

416.80

630.60

18.90

18.90

21.75

19

30-04-01

416.40

629.00

18.90

22.03

21.75

20

02-05-01

416.80

629.90

18.80

22.03

21.35

21

03-05-01

415.80

627.70

18.80

22.03

21.35

22

04-05-01

417.50

631.50

18.80

22.03

21.35

23

09-05-01

417.60

632.70

18.80

22.03

21.35

24

10-05-01

417.20

632.70

18.80

22.03

21.35

25

11-05-01

417.70

634.30

18.80

22.15

21.35

26

14-05-01

416.30

630.40

18.79

22.15

21.41

27

15-05-01

416.30

628.10

18.79

22.15

21.41

28

16-05-01

416.20

628.30

18.79

22.15

21.41

29

17-05-01

417.70

632.70

18.79

22.15

21.41

30

18-05-01

415.90

633.20

18.79

21.19

21.41

31

21-05-01

415.90

630.90

18.79

21.19

21.41

32

22-05-01

418.60

636.20

18.79

21.19

21.41

33

23-05-01

416.80

637.50

18.79

21.19

21.41

Table 4

 

 

 

 

 

 

 

 

 

 

Results

Table 5 shows the values obtained for each weight at the end of the training session for the training set.

 

ID

weight

value

1

ww(1)

26.051286198506

2

ww(2)

26.051286198506

3

ww(3)

26.051286198506

4

ww(4)

26.051286198506

5

ww(5)

26.051286198506

6

ww(6)

26.051286198506

7

ww(7)

26.051286198506

8

ww(8)

26.051286198506

9

w(1,1)

3.84240618472063E-02

10

w(1,2)

5.80050884218188E-02

11

w(1,3)

1.7347099091662E-03

12

w(1,4)

2.04490798549169E-03

13

w(1,5)

1.97659059231841E-03

14

w(1,6)

3.84240618472063E-02

15

w(1,7)

5.80050884218188E-02

16

w(1,8)

1.7347099091662E-03

17

w(2,1)

2.04490798549169E-03

18

w(2,2)

1.97659059231841E-03

19

w(2,3)

3.84240618472063E-02

20

w(2,4)

5.80050884218188E-02

21

w(2,5)

1.7347099091662E-03

22

w(2,6)

2.04490798549169E-03

23

w(2,7)

1.97659059231841E-03

24

w(2,8)

3.84240618472063E-02

25

w(3,1)

5.80050884218188E-02

26

w(3,2)

1.7347099091662E-03

27

w(3,3)

2.04490798549169E-03

28

w(3,4)

1.97659059231841E-03

29

w(3,5)

3.84240618472063E-02

30

w(3,6)

5.80050884218188E-02

31

w(3,7)

1.7347099091662E-03

32

w(3,8)

2.04490798549169E-03

33

w(4,1)

1.97659059231841E-03

34

w(4,2)

3.84240618472063E-02

35

w(4,3)

5.80050884218188E-02

36

w(4,4)

1.7347099091662E-03

37

w(4,5)

2.04490798549169E-03

38

w(4,6)

1.97659059231841E-03

39

w(4,7)

3.84240618472063E-02

40

w(4,8)

5.80050884218188E-02

41

w(5,1)

1.7347099091662E-03

42

w(5,2)

2.04490798549169E-03

43

w(5,3)

1.97659059231841E-03

44

w(5,4)

3.84240618472063E-02

45

w(5,5)

5.80050884218188E-02

46

w(5,6)

1.7347099091662E-03

47

w(5,7)

2.04490798549169E-03

48

w(5,8)

1.97659059231841E-03

 

 

Table 5

 

Discussion

Although the ANN gives a satisfactory results following further studies on the net is required to improve the quality of the performance.

  1. Hidden layers were selected arbitrarily when designing and improvement of results may possible by improving the hidden layers which needs further studies.
  2. Only sigmoidal transfer function was tested in the input layer and the hidden layer.

HOME