1. NEURAL NETWORKS FOR TECHNICAL ANALYSIS: A STUDY ON KLCI 授課教師：楊婉秀 報告人：李宗霖

2. Outline Introduction Forecasting the Stock Market Neural Network and its Usage in the Stock Market A Case Study on the Forecasting of the KLCI Discussion Conclusion and Future Research 2

3. Introduction 3

4. It is generally very difficult to forecast the movements of stock markets. Artificial neural network is a well-tested method for financial analysis on the stock market. 4

5. Introduction Using neural networks in equity market applications include: Forecasting the value of a stock index Recognition of patterns in trading charts Rating of corporate bonds Estimation of the market price of options Indication of trading signals of selling and buying 5

6. Introduction Feed-forward back propagation Without the use of extensive market data or knowledge useful prediction can be made and significant paper profit can be achieved. 6

7. Forecasting the Stock Market 7

8. There are three schools of thought in terms of the ability to profit from the equity market. Random Walk Hypothesis & Efficient Market Hypothesis fundamental analysis Technical analysis 8

9. Forecasting the Stock Market Use a variety of techniques to obtain multiple signals. Neural networks are often trained by both technical and fundamental indicators to produce trading signals. 9

10. Forecasting the Stock Market For fundamental methods: retail sales gold prices industrial production indices foreign currency exchange rates For technical methods: delayed time series data technical indicators 10

11. Neural Network and its Usage in the Stock Market 11

12. 3.1 Neural networks Neural networks 12

13. 3.2 Time series forecasting with neural networks 13

14. 3.2 Time series forecasting with neural networks Three major steps in the neural network based forecasting proposed in this research: Preprocessing Architecture Postprocessing 14

15. 3.3 Measurements of neural network training Normalized Mean Squared Error (NMSE) Signs Gradients 15

16. We argue that NMSE may not be the case for trading in the context of time series analysis. 3.3 Measurements of neural network training 16

17. 3.3 Measurements of neural network training 17

18. A Case Study on the Forecasting of the KLCI 18

19. A Case Study on the Forecasting of the KLCI Kuala Lumpur Composite Index (KLCI) Neural networks are trained to approximate the market values. To find the hidden relationship between technical indicators and future KLCI. 19

20. Data choice and pre-processing 20

21. Data choice and pre-processing The major types of indicators moving average (MA) momentum (M) Relative Strength Index (RSI) stochastics (%K) moving average of stochastics (%D) 21

22. Data choice and pre-processing inputs I t−1 I t MA 5 MA 10 MA 50 RSI M %K %D output I t+1 22

23. Data choice and pre-processing Normalization 23

24. Nonlinear analysis of the KLCI data 24

25. Nonlinear analysis of the KLCI data 25

26. Nonlinear analysis of the KLCI data 26

27. Neural network model building Historical data are divided into three parts Training sets (2/3) Validation sets (2/15) Testing sets (3/15) 27

28. Neural network model building A trade-off between convergence and generalization. Number of hidden nodes 28

29. Neural network model building 29

30. Neural network model building Primary sensitive analysis is conducted for input variables. Low influence factors: M 20, M 50, %K, %D High influence factors: I t, RSI, M, MA 5 Chosen for input: I t, MA 5, MA 10, RSI, M, I t −1 30

31. Neural network model building 31

32. Neural network model building 32

33. Paper profits using neural network predictions 33

34. Paper profits using neural network predictions 34

35. Paper profits using neural network predictions In a real situation, this might not be possible as some indexed stocks may not be traded at all on some days. The transaction cost of a big fund trading, which will affect the market prices was not taken into consideration in the calculation of the “paper profit”. 35

36. Benchmark return comparison Benchmark 1: Benchmark 2: 36

37. Benchmark return comparison Benchmark 3: 37

38. Comparison with ARIMA Autoregressive Integrated Moving Average (ARIMA) Model Introduced by George Box and Gwilym Jenkins in 1976. Provided a systematic procedure for the analysis of time series that was sufficiently general to handle virtually all empirically observed time series data patterns. ARIMA(p, d, q) 38

39. 39

40. Discussion 40

41. Discussion A very small NMSE does not necessarily imply good generalization. Better testing results are demonstrated in the period near the end of the training sets. We have no data to test the “best” model. There are two approaches for using the forecasting result. best-so-far approach committee approach 41

42. Discussion Measures NMSE Sign Gradient Four challenges inputs and outputs types of neural networks and the activation functions neural network architecture evaluate the quality of trained neural networks for forecasting 42

43. Conclusion and Future Research 43

44. Conclusion and Future Research It shows that useful prediction could be made for KLCI without the use of extensive market data or knowledge. It shows how a 26% annual return could be achieved by using the proposed model. It highlights the following problems associated with neural network based time series forecasting: the hit rate is a function of the time frame chosen for the testing sets; generalizability of the model over time to other period is weak; there should be some recency trade offs. 44

45. Conclusion and Future Research A mixture of technical and fundamental factors as inputs over different time periods should be considered. Sensitivity analysis should be conducted which can provide pointers to the refinement of neural network models. The characteristics of emerging markets such as KLCI should be further researched to facilitate better modeling of the market using neural networks. 45

46. Thanks for Your Listening 46

47. Forecasting the Stock Market The research done here would be considered a violation of the above two hypotheses above for short-term trading advantages in, Kuala Lumpur Stock Exchange (KLSE for short hereafter) 47

48. Forecasting the Stock Market There are three schools of thought in terms of the ability to profit from the equity market. The first school believes that no investor can achieve above average trading advantages based on the historical and present information. Random Walk Hypothesis Efficient Market Hypothesis Taylor provides compelling evidence to reject the random walk hypothesis and thus offers encouragement for research into better market prediction. 48

49. Forecasting the Stock Market The second school's view is the so-called fundamental analysis. It looks in depth at the financial conditions and operating results of a specific company and the underlying behavior of its common stock. In 1995 US$1.2 trillion of foreign exchange swapped hands on a typical day [10]. The number is roughly 50 times the value of the world trade in goods and services which should be the real fundamental factor. 49

50. Forecasting the Stock Market Technical analysis belongs to the third school of thought. It attempts to use past stock price and volume information to predict future price movements. These five series are open price, close price, highest price, lowest price and trading volume. Most of the techniques used by technical analysts have not been shown to be statistically valid and many lack a rational explanation for their use 50

51. Normalized Mean Squared Error (NMSE) Signs Gradients 3.3 Measurements of neural network training 51

52. 3.3 Measurements of neural network training Normalized Mean Squared Error (NMSE) 52

53. Signs 3.3 Measurements of neural network training 53

54. Gradients 3.3 Measurements of neural network training 54

55. Data choice and pre-processing 55

56. Data choice and pre-processing 56

57. Data choice and pre-processing 57