Expert Advisor (EA) Evaluation System Using Web-based ELECTRE Method in Foreign Exchange (Forex) Market

Transcription

1 The 2 nd International Conference on Energy, Environent and Inforation Syste (ICENIS 2017) 15 th 16 th August 2017, Universitas Diponegoro, Searang, Indonesia Expert Advisor (EA) Evaluation Syste Using Web-based ELECTRE Method in Foreign Exchange (Forex) Market Satibi Satibi 13,*, Catur Edi Widodo 2, and Farikhin Farikhin 2 1 Master of Inforation Systes, School of Postgraduate Studies, Diponegoro University, Searang - Indonesia 2 Physics Departent of Science and Matheatics Faculty, Diponegoro University, Searang - Indonesia 3 Departent of Inforatics Manageent, AMIK Selatpanjang, Meranti Island - Indonesia Abstract. This research ais to optiize forex trading profit autoatically using EA but its still keep considering accuracy and drawdown levels. The evaluation syste will classify EA perforance based on trading arket sessions (Sydney, Tokyo, London and New York) to deterine the right EA to be used in certain arket sessions. This evaluation syste is a web-based ELECTRE ethods that interact in real-tie with EA through web service and are able to present real-tie charts perforance dashboard using web socket protocol counications. Web applications are prograed using NodeJs. In the testing period, all EAs had been siulated 24 hours in all arket sessions for three onths, the best EA is valued by its profit, accuracy and drawdown criterias that calculated using web-based ELECTRE ethod. The ideas of this research is to copare the best EA on testing period with collaboration perforances of each best classified EA by arket sessions. This research uses three onths historical data of EUR/USD as testing period and other 3 onths as validation period. As a result, perforance of collaboration four best EA classified by arket sessions can increase profits percentage consistently in testing and validation periods and keep securing accuracy and drawdown levels. Keywords: Perforance; Evaluation syste; Forex; Expert Advisor; DSS; ELECTRE. 1 Introduction In deterining the investent strategy in the foreign exchange arket, decision akers have iportant roles and responsibilities to deterine risks and profit targets. Decision akers are also required to be able to ake three iportant decisions, naely: 1. Prices will ove up or down; 2. Deterine buying or selling decisions; 3. How long to hold the position that has been opened [1]. Due to the iportance of decision-aking role in forex trading arket, it is necessary to have a syste to record and evaluate the history of transactions that have been done before in real-tie so that it can help to provide perforance reports fro previous transactions. Nowadays it is the era of coputer and inforation technology that has been developing quite rapidly. For sure, this evolution affecting the way of trading transactions ethod in the forex arket to continues to grow. Utilizing current technology allows Traders to perfor real-tie analysis and trading over the internet. Forex trading can also be done autoatically based on EA (Expert Advisor) without huan intervention, with autoatically analytics and transactions using EA in forex transaction negotiation is to avoid huan intervention that is usually influenced by eotions, anxiety, stress and other things. Therefore, trading by using EA becoes a solution to eliinate the eleents of huan intervention that are volatile. Even studies in 2010 concluded 45% of negotiation syste in the forex arket was conducted using EA [2]. Although the percentage of the use of EA in forex trading is very large, but analysis process and transaction history evaluation usually had been done only liited to transaction history analysis that found in Metatrader software, which its not optial when used as EA perforance evaluation data. With the integration of EA supported by calculations based on real-tie web-based ELECTRE ethod will be ore optial in providing the best choice of several EA alternatives based on trading arket sessions. In addition to being able to perfor analysis and transaction in the forex arket autoatically, the EA also has the ability to interact with the web service, so it can be integrated with algoriths that are applied to web based applications that provide supporting inforation for the EA itself. Forex trading research on the forex arket has been done a lot before, including forex forecasting using ANN ethods [3], ulti-agency financial decision support syste with Fuzzy logic [4] and Web-based decision support systes with ANN for forecasting and forex trading [5]. However, this research focuses on evaluating the perforance of ten open source EA that available on the official MQL4 (MetaQuotes Language 4) counity website, although this evaluation syste can still be used on all types of EAs either paid or created by traders who aster the MQL4 prograing language, by * Corresponding author: satibi@bata.obi

2 inserting a line of MQL4 code created in this evaluation syste in order to interact with the web service and utilizing historical data of previous relevant currency oveents to siulate the EAs. 2 Method 2.1 Data acquisition Historical currency oveent data that used to siulate trading charts oveent in MetaTrader software is downloaded fro Histdata.co, while the EAs trading transaction data acquired in real-tie on the web-based ELECTRE application through web service which is also integrated with web socket for Displaying EA perforance graphs in real-tie on the dashboard. These acquired transaction data saved in database, this database will be classified based on trading arket sessions to get the best perforing EA in each trading arket session. Classifications process of arket session is recognized by trading open position that done by real-tie EA transaction. The flow of data acquisition process as shown in Figure 1. Fig. 1. Data acquisition process of EA transactions 2.2 ELECTRE ethod The ELECTRE ethod is one of the ulticriteria decision analysis ethods which is an acrony of Eliination Et Choix Traduisant la Realité or in English eans Eliination and Choice Expressing Reality, first introduced in 1965 by the SEMA consulting fir in Europe [6]. The odeling with the ELECTRE ethod includes six stages, ie : 1. Noralization of atrix Noralization process will produce R atrix, that calculated by : r ij = x ij x 2 ij i=1 2. To weight the noralized atrix, ultiply that R atrix values by the weight of each criterion using following equation: (1) 3. Deterining concordance and discordance index sets. Concordance index calculated by : C kl = {j v kj v ij }; untuk j = 1,2,, n (3) Meanwhile, following equation is to calculate discordance index : D kl = {j v kj < v ij }; untuk j = 1,2,, n (4) 4. Concordance and discordance atrix Concordance atrix obtained by : c kl = j Ckl W j (5) Below equation is used to obtain discordance atrix: d kl = ax{ v kj v lj } j Dkl ax{ v kj v lj } j (6) v ij = w i r ij. (2)

3 5. Doinance concordance and discordance atrix To deterine doinance concordance atrix : c = k=1 l=1 c kl ( 1) Eleents of F atrix specified by : f kl = { 1, jika c kl c 0, jika c kl < c To deterine doinance discordance atrix : d = k=1 l=1 d kl ( 1) Eleents of G atrix specified by : g kl = { 1, jika d kl d 0, jika d kl < d (7) (8) (9) (10) 6. Deterining the aggregate doinance atrix Aggregation of the doinance atrix indicates the sequence of partial preferences of the available alternatives, to obtain this aggregate doinance atrix, following equation is used : e kl = f kl x g kl (11) Fro that calculation will produce the E atrix which gives sequence of options fro each available alternative, if e kl = 1 then alternative A k is a better alternative than A l. So, the rows in the E atrix that having less e kl = 1 can be eliinated, as well as other rows calculation can be eliinated until reaining first row as selected best EA alternative. 3 Results and discussion Required data acquisition process in this research is EAs transaction data fro ten siulated EA and percentage of profit, accuracy and drawdown fro each EA perforance. a. Transactions data acquisition process Data acquisition process in web-based ELECTRE application is done in real-tie when trading siulation is executed on MetaTrader software, EA will send trading transaction data to web service such as alt_id, ticket_id, open_tie, close_tie, type, lots, sybol, open_price, close_price and profit fields. b. Meanwhile, the value data that consisting of profit, accuracy and drawdown were obtained by calculating the percentage of that three criterias based on transaction history data, while capital=5000, %Profit = ((profit + loss) / capital) x 100, %Accuracy = ( profit positions / all positions) x 100, Drawdown = (axiu floating / capital) x 100. This calculation will be obtained value as shown in Table 1 and Table 2. Table 1. Value of each criteria without arket classification Alt Cri Alt Cri Value Step ID ID ID ID Value Step Table 2. Value of each criteria by arket sessions Alt Cri ID ID Sydney Tokyo London Newyork Step

4 c. Deterining the weight of each criteria, the criteria weight in this case obtained fro online questionnaire results, the questionnaire gives criteria weight for profit = 3.83%, accuracy = 2.39% and drawdown 3.27%. d. Noralize the atrix by specifying value for each atrix index of all alternatives per criterion, value of profit calculation exaple : x 1 = ( ) + ( ) + + ( ) = To get profit value fro first and second EAs : r 11 = x 11 x 1 = = r 21 = x 11 x 1 = = r = e. The sae way is used to deterine the value of noralization on accuracy and drawdown criterias value. f. Noralized atrix weighting is obtained by ultiplying noralized atrix values by the weight of each criterion, eg. : v 11 = x 3.83 = , v 21 = x 3.83 = , g. Deterining concordance and discordance index sets by coparing an alternative with another alternative based on the sae criterion, if selected value is greater then that index is used as a concordance index, and if the index value is saller it will be used as a discordance index, eg. : C 12 = {1,3}, C 13 = {1,3} h. Calculates concordance and discordance atrices. Concordance atrix is obtained by suing weights based on concordance index, eg. : C 12 = = While discordance atrix is obtained by getting value of difference between two criterion values based on discordance index then taken the highest value fro it, this value will divided by the highest value of difference fro all indexes coparison.. i. Doinance atrix of concordance and discordance is obtained by deterining the threshold value first, then deterining the doinance concordance atrix by coparing concordance and discordance atrix eleents with the threshold value of each atrix, if the value of each eleent is greater than the threshold value will give a value of 1 And gives a value of 0 if it is saller than the threshold value. j. Aggregate doinance atrix is obtained by ultiplying the doinance atrix eleent of concordance and doinance discordance atrix eleents. The alternative with e kl = 1 will be eliinated while alternative with the ost of e kl = 1 will be selected as the best alternative as shown in Figure 2. k. These calculation steps are also used to get the best EAs in the best EA selection process based on arket sessions as shown in Figure 3. l. In order to get optial results, if there are ore than one EAs with sae perforances appeared in the sae arket session, only one EAs will be selected as best EA on that arket session based on highest criteria weight priority. 3.1 Results Evaluation and validation EAs perforance evaluation and validation of ELECTRE ethod calculation had been tested using different test tie data, october to deceber (2016) as testing period and january to arch (2017) as validation period with different transaction data aounts, so that the results obtained as in Figure 4 and 5.

5 Fig. 2. Ten EAs perforance in testing period Fig. 3. Ten EAs perforance by arkets classification

6 Fig. 4. EAs perforance coparison Fig. 5. EAs perforance validation 4 Conclusion The decision-aking syste with web-based ELECTRE ethod to deterine the best collaboration EAs to be used based on its best arket session perforance, EAs perforance are calculated by its profit, accuracy and drawdown criterias and the use of cobination four EAs were best on its arket session can increase profit consistently while aintaining accuracy and drawdown levels, this can be proven in the testing and validation processes with the profit increase level of 0.63 (9.20%) in the testing period and 9.03 (219.71%) in the validation period, and keep securing accuracy and drawdown levels when copared to other EAs.

7 References 1. T. Balabanov, I. Zankinski, N. Dobrinkova, Tie Series Prediction by Artificial Neural Networks and Differential Evolution in Distributed Environent. Large-Scale Scientific Coputing (LSSC) 2011, Vol 7116, (2012). 2. R.M.C. Pinto, J.C.M. Silva, Strategic Methods for Autoated Trading in Forex th International Conference on Intelligent Systes Design and Applications (ISDA), (2012). 3. P. Czekalski, M. Niezabitowski, R. Styblinski, ANN for FOREX Forecasting and Trading, th International Conference on Control Systes and Science, (2015). 4. J. Korczak, M. Hernes, M. Bac, Fuzzy Logic in the Multi-agent Financial Decision Support Syste Federated Conference on Coputer Science and Inforation Systes (FedCSIS), (2015). 5. K.K. Lai, L. Yu, S. Wang, A Neural Network and Web-Based Decision Support Syste for Forex Forecasting and Trading. Data Mining and Knowledge Manageent, Vol 3327, (2004). 6. D. Siić, J. Gajić, V. Ilin, V. Svirčević, S. Siić, A Hybrid Bio-inspired ELECTRE Approach for Decision Making in Purchasing Agricultural Equipent. Hybrid Artificial Intelligent Systes, Vol 9648, (2016). 7. R.P. Barbosa, O. Belo, Autonoous Forex Trading Agents. Advances in Data Mining, Medical Applications, E-Coerce, Marketing, and Theoretical Aspects, Vol 5077, (2008). 8. R.F.B. de Brito, A.L.I. Oliveira, Sliding Windowbased Analysis of Multiple Foreign Exchange Trading Syste by Using Soft Coputing Techniques International Joint Conference on Neural Networks (IJCNN), (2014). 9. M.H. Eng, Y. Li, Q.G. Wang, T.H. Lee, Forecast Forex With ANN Using Fundaental Data International Conference on Inforation Manageent, Innovation Manageent and Industrial Engineering, Vol 1, (2008). 10. Y.L. Yong, Y. Lee, D. Nga, An Investigation into the Recurring Patterns of Forex Tie Series Data IEEE International Syposiu on Robotics and Intelligent Sensors (IRIS), (2015). 11. V.H.T. Kiet, R. Wojciechowski, Straight Through Processing for Corporate Foreign Exchange Trading Lessons Learned fro a Web Service Based Ipleentation IEEE International Conference on e-technology, e-coerce and e-service, (2005). 12. Z. Liu, D. Xiao, An Autoated Trading Syste with Multi-indicator Fusion Based on D-S Evidence Theory in Forex Market Sixth International Conference on Fuzzy Systes and Knowledge Discovery, Vol 3, (2009).