Buy and Hold Strategy

Assessing the performance of B&H compared with trading system based on TA

1. Comparison of the Buy and Hold S trategy with Trading System of T echnical R ules Enhanced by ANN and GA Case Study: Tehran Stock Exchange By: K.Dehghan Manshadi Sep 2012

2. Table of Contents Definitions Goals of the research Previous Research Study steps Technical Rules as the trading system parts GA structure used ELMAN Network Testing Hypothesis Approach Key Results 2 4 5 8 10 15 18 20 22

3. Some Definitions Trading System Technical Analysis Trading Policy Using set of tools and techniques in order to make investment decisions Methods and strategies used to forecast future prices based on different factors e.g. past prices, volume, trends ,.. One turning point is a point in time where one price trend change into another one. In general there are 3 main trends: upward, downward, and uniform trends Turning Points The approach that one trader choose in order to do his/her trades to gain from positions he/she gets in the market 2

4. Table of Contents Definitions Goals of the research Previous Research Study steps Technical Rules as the trading system parts GA structure used ELMAN Network Testing Hypothesis Approach Key Results 2 4 5 8 10 15 18 20 22

5. Research Goals Research OBJ. Dependency of Parameter setting to Investors Experience Different Signals from different Trading Rule at the same Time Difficulty of changing different signals from different rules to one trading decision Difficulties for using Technical Analysis Key Issues Technical Rules are based on parameters that if are set properly, will lead to profitable positions in market. The main challenge regarding technical rules are their different mechanism to produce trading signals. This will result in different signals by different rules at the same time. And this will mixed the traders. Building Up the new Intelligent Trading System to omit the Dependency of investments to Investors experience 4

6. Table of Contents Definitions Goals of the research Previous Research Study steps Technical Rules as the trading system parts GA structure used ELMAN Network Testing Hypothesis Approach Key Results 2 4 5 8 10 15 18 20 22

7. Studies Categorization Category one Studies done to develop scientific framework for formulating TA Netcci , Brok , Murphi , Bollinger, Achelis , Osle Category Two Category Three Category Four Studies Categories Focus of the Research's Top Researchers Studies done to investigate the forecasting power of technical rules compered to other forecasting tools Studies done to evaluate the statistical aspects and quality of the rules outputs . Studies done to optimize the TA indicators and rules and developing new trading tools Fama , Blume , James, Chang, Osler,Alexander Scatchell Thomson, Williams, Bollinger 5

8. Previous Research's Alejandro Rodríguez Researcher Year Subejct Key Take Away Using ANN to enhance the TA indices ANN had a remarkable effect on TA indices performance 2011 Xiaowei Lin Using GA to improve the forecasting parameters in TA and enhancing the ESN parameters to reach better forecasted turning point The system based on GA resulted in more profitability compared with B&H strategy 2011 Liu, Chang , et.al Building up an efficient forecasting model in order to producing trading signals CBDWNN had a better performance than other studied models 2009 Baba,Inoue , & Yanjun Establish a system composed of ANN and GA to forecast the TOPIX in future market The composite model had a good performance in forecasting the market Index 2002 Kuo , Chen and Hwang Intelligent system to support decision making based on GA and fuzzy ANN The new system enables quantification of qualitative variables affecting stock price 2001 6

9. Table of Contents Definitions Goals of the research Previous Research Study steps Technical Rules as the trading system parts GA structure used ELMAN Network Testing Hypothesis Approach Key Results 2 4 5 8 10 15 18 20 22

10. Study steps and Trading System Architecture Setting Parameters by GA and Turning Point Diagnoses Network Build up and Training Testing Hypothesis and Assess the performance The society and selected Sample Society: Stocks in Tehran 50 Company Indices Sample: randomly chosen 15 stocks Timeframe: 8 years 2005-2012 Suitable training of the GA parameters for each trading rule to forecast the trading signals Changing different trading signals from different rules to one trading signal with the help of ELMAN network Calculating the portfolio %return by considering uniform weighting across all assets and running Mann Whitney non-parametric Test Trading System Arch. 8

11. Table of Contents Definitions Goals of the research Previous Research Study steps Technical Rules as the trading system parts GA structure used ELMAN Network Testing Hypothesis Approach Key Results 2 4 5 8 10 15 18 20 22

12. Technical Rules – 1 of 4 Golden Cross and Dead Cross Simple MA is a popular technical indicator which calculates the mean price in a specified period in which MA(N ) means long-term MA while MA(n) means short-term MA . Cross section of these two represent a trading point. Approach Figure Parameters   Moving Average Envelope MA envelope forms a channel or zone of commitment around a MA . If price breaks the upper band in downtrend, then it is time to buy ; if it breaks through the lower band in uptrend, then it is time to sell a and n   MA(n) MA(N) 10

13. Technical Rules – 2 of 4 Relative strength Index System RSI ranges from 0 to 100. Generally, if the RSI rises above overbought level (usually 80), it indicates a selling signal; if it falls below oversold level (usually 20), it indicates a buying signal. Approach Figure Parameteres Rate of change Index The divergence of different ROCs can indicate possible reversal of price trend. Generally, when long-term ROC reaches a new high while short-term ROC locates near the equilibrium line (usually with the value of 100 ), the price will possibly fall down; similarly, when long-term ROC reaches a new low while short-term ROC is near the equilibrium line, the price may ascend       11

14. Technical Rules – 3 of 4 Stochastic System In the up-trend, it tries to measure when the closing price would get close to the lowest price in the given period; in the down-trend, it means when the closing price would get close to the highest price in the given period. The crossover of %K and %D lines may indicate meaningful reversal in price trend. Approach Figure Parameters   C:close price at now LL :lowest price in the period HH :highest price in the priod       12

15. Technical Rules – 4 of 4 Hammer and Hanging man Indicates price reversal in the future Approach Figure Parameters Dark Cloud Cover     Indicates price reversal in the future ndC :next day close price pdO :previous day open price Piercing Line Indicates price reversal in the future Engulfing Pattern Indicates price reversal in the future       13

16. Table of Contents Definitions Goals of the research Previous Research Study steps Technical Rules as the trading system parts GA structure used ELMAN Network Testing Hypothesis Approach Key Results 2 4 5 8 10 15 18 20 22

17. GA Structure – Fitness Function Genetic Structure- Buy Position Assume the sequence of expected trading points:   For each trading point “Ti” we are looking for trading signal “ Sj ” suggested by different technical rules   If Ti is a buy position, then there are three states for fitness function: A) If Sj is a buy signal, then the fitness is:   B) If Sj is a sell signal then we should have punishment for wrong identification Also consider the neighborhood criteri :   C) If no trading operation is suggested by system, then a punishment for missing the trading opportunity must be consider. In this state the fitness function is as bellow:   The overall fitness function of the trading sequence S={S1,S2,…, Sm } is:   If the Ti is an expected selling position then the fitness function will be build in a similar way. 15 1 2 3 4 5 6 7 8 9

18. GA Steps GA Structure – Key Steps Considering following chromosome structure for each feasible solution Creating a random society as chromosomes with above structure Calculating the fitness function for each chromosome In order to generating the next generation, some current chromosomes are selected as parents F (Position) = 2- sp + 2 *( sp -1) * (pos-1) / (n-1) With the following equations each pare of parents reproduce new spring: Offspring 1 = Parent 1 * (rand1) + Parent 2 * (1-rand1) Offspring 2 = Parent 1 * ( rand 2 ) + Parent 2 * ( 1-rand 2 ) Next step is to produce new generation. Next generation is composed of the best current springs and new springs. Parameters and Specifications of the used GA: Population: 50 Gen: 300 GGAP: 0.8 Parent selection approach: Roulette wheel selection New spring creation approach: Recombination Mutation probability: 0.1 Policy to create new generation: keeping 10% of the best current springs+ keeping 10% of the worst current springs+ the random springs of the old and new generation P1 p2 P3 …… Pn 16 1 3 2 5 4 6 7

19. Table of Contents Definitions Goals of the research Previous Research Study steps Technical Rules as the trading system parts GA structure used ELMAN Network Testing Hypothesis Approach Key Results 2 4 5 8 10 15 18 20 22

20. ELMAN Network Network Architecture Network Specification Recurrent Network with two layer The recurrent specification of the network enable detecting time varying trends – high approximating power The main difference of ELMAN with other 2layers networks is to have a recurrent relationship in layer one – delay in this layer keep the past values in the network to use them in future. 18

21. Table of Contents Definitions Goals of the research Previous Research Study steps Technical Rules as the trading system parts GA structure used ELMAN Network Testing Hypothesis Approach Key Results 2 4 5 8 10 15 18 20 22

22. Testing Hypothesis Approach Implications To what extend we can rely on historic data? How much data is suitable to train the network? It’s a rule of thumb that using more data to train the Network don’t result in better performance all the time Price time series nonstationary and changing behavior Challenges with the Network Rolling Window Approach If the time series behavior trough the time is nonstattionary , it means some characteristics of the series such as noise as well as the forecasting parameters change trough the time . Therefore using a static model lead to weak forecast. P-Value of the Mackinnon statistic in dickey-Fuller test for most of the stocks is remarkable(very big) and the unit root hypothesis is rejected that admit the nonstationary of the price time series in our sample 20

23. Table of Contents Definitions Goals of the research Previous Research Study steps Technical Rules as the trading system parts GA structure used ELMAN Network Testing Hypothesis Approach Key Results 2 4 5 8 10 15 18 20 22

24. The system performance in diagnosing turning points Signals %Frequency Correct Signals 31% Zero Signals 61% Wrong Signals 8% Implication The developed trading system have a good performance in diagnosing trading points 22

25. Comparison between B&H Strategy and the developed Trading System performance Implication Both Strategy performance are remarkable. The trading system in all window had positive performance 23 Buy and Hold Trading System

26. Testing Hypothesis Implication Statistically there is no significant difference between the returns in B&H strategy and the intelligent Trading System Non-Parametric Test Parametric Test No significant difference between performance of the two strategy 5 %= α 24

27. Conclusions Suggestions for future studies Key Results TA like the buy and hold strategy possess the potential for profitability in Iran Market Both Active and Passive Strategies can be profitable in Iran Stock Market Artificial Intelligence can help improve the performance of technical Analysis rules The variance of returns in B&H strategy is more than suggested trading system Good performance of the technical analysis can approve the weak efficiency of the market. Comparison of the trading system based on technical rules with other trading strategies such as momentum and reverse. In this study the weights of different assets assumed equal. Rebalancing the portfolio trough the time can be good option to enhance the trading system performance. Using more technical rules to build the system Using other artificial intelligence techniques to set the technical parameters Considering other factors like volume of the trades in trading system to moderate the sensitivity of the system to price changes. 25