Designing a Trading Strategy to Buy and Sell the Stock of Companies Listed on the New York Stock Exchange Based on Classification Learning Algorithms

الموضوعات :

Nasser Heydari ¹ , Majid Zanjirdar ² , Ali Lalbar ³

1 - Department of Finance , Arak Branch, Islamic Azad University, Arak, Iran
2 - Department of Finance , Arak Branch, Islamic Azad University, Arak, Iran
3 - Department of Accounting, Arak Branch, Islamic Azad University, Arak, Iran

تاريخ الإرسال : 10 الثلاثاء , صفر, 1444 تاريخ التأكيد : 21 الخميس , جمادى الأولى, 1444 تاريخ الإصدار : 18 السبت , شوال, 1445

الکلمات المفتاحية: Trading Strategy , Machine Learning, Classification Algorithms,

ملخص المقالة :

This research investigated the development of a stock trading strategy for companies on the New York Stock Exchange (NYSE), a prominent global market. Data was acquired from established libraries and the Yahoo Finance database. The model employed technical analysis indicators and oscillators as input features. Machine learning classification algorithms were used to design trading strategies, and the optimal model was identified based on statistical performance metrics. Accuracy, recall, and F-measure were utilized to evaluate the classification algorithms. Additionally, advanced statistical methods and various software tools were implemented, including Python, Spyder, SPSS, and Excel. The Kruskal-Wallis test was employed to assess the statistical differences between the designed strategies. A sample of 41 actively traded NYSE companies across diverse sectors such as financial services, healthcare, technology, communication services, consumer cyclicals, consumer staples, and energy were chosen using a filter-based approach on June 28th, 2021. The selection criteria included a market capitalization exceeding $200 billion and an average daily trading volume surpassing 1 million shares. Evaluation metrics revealed that the designed random forest trading strategy achieved a good fit with the data and exhibited statistically significant differences from other strategies based on classification learning algorithm.

المصادر:

[1] Cortes, C., Vapnik, V., Support-Vector Networks, Machine Learning, 1995; 20:273–297. Doi: 10.1007/BF00994018.
[2] Breiman, L., Random Forests, Machine Learning, 2001; 45:5-32. Doi: 10.1023/A: 1010933404324.
[3] Zhang, Q., Yang, L., Zhou, F., Attention Enhanced Long Short-term Memory Network with Multi-Source Heterogeneous Information Fusion: An Application to BGI Genomics, Information Sciences, 2020; 553: 305-330, Doi: 10.1016/j.ins.2020.10.023.
[4] Siti, H., Mahmoud, G., N And A Noryati., Conceptual Paper of the Trading Strategy: Dogs of the Dow Theory, SSRN Electronic Journal, 2014; Doi:10.2139/ssrn.2697334.
[5] Perwej, Y., Perwej, A., Prediction of the Bombay Stock Exchange (BSE) Market Returns Using Artificial Neural Network and Genetic Algorithm, Journal of Intelligent Learning Systems and Applications, 2012; 40(2): 108–119, Doi:10.4236/jilsa.2012.42010.
[6] Dattatray, P.Gandhmal, K., Systematic Analysis and Review of Stock Market Prediction Techniques, Computer Science, 2019; 34. Doi:10.1016/j.cosrev.2019.08.001.
[7] Benjamin, R., A Tour of Reinforcement Learning: The View from Continuous Control, Mathematics Optimization and Control, 2018; 1. Doi:10.48550/arXiv.1806.09460.
[8] kofi, Nt I., Felix Adekoya, D., Asubam Weyori, B., Eflcient Stock-Market Prediction Using Ensemble Support Vector Machine, Open Computer Science, 2020; 10: 153–163. Doi: 10.1515/comp-2020-0199.
[9] Cervelló-Royo, R., Guijarro, F., Forecasting Stock Market Trend: A Comparison of Machine Learning Algorithms, Finance, Markets and Valuation, 2020; 6:37-49. Doi: 10.46503/NLUF8557.
[10] Dosdoğru, A T., Boru, A., Göçken, M., Özçalici, M., Göçken, T., Assessment of Hybrid Artificial Neural Networks and Metaheuristics for Stock Market Forecasting, Computer Science, 2018; 24(1): 63–78. Doi:10.13140/rg.2.1.1954.1368.
[11] Zhang, X., Pan, Z., Hu, G., Tang, S., Zhou, C., Stock Market Prediction on High-Frequency Data Using Generative Adversarial Nets, Mathematical Problems in Engineering, 2018; 1: 1–11. Doi:10.1155/2018/4907423.
[12] Zhang, X., Zhang, Y., Wang, S., Yao, Y., Fang, B., Yu, P., Improving Stock Market Prediction Via Heterogeneous Information Fusion, Computer Science. 2017; 143: 236–247. Doi:10.1016/j.knosys.2017.12.025.
[13] Saif, S., Jamshidi Navid, B., Ghanbari, M., Ismailpour, M., Predicting The Trend of Iran's Stock Market Using Elliott Wave Profile and Relative Strength Index, Financial Research, 2021; 23(1): 134-157. Doi: 10.22059/frj.2020.310664.1007072 (In persion).
[14] Dayi, A., Obadadi ,AM., Kivan, B., Application of Web Mining in Predicting the Price Direction of Chemical Products Group in the Stock Exchange, Iran Information and Communication Technology Quarterly, 2018; 40:19-48.Doi:20.1001.1.27170414.1398.11.39.2.8. (In persion).
[15] Alotaibi, S., Ensemble Technique with Optimal Feature Selection for Saudi Stock Market Prediction: A Novel Hybrid Red Deer-Grey Algorithm, Institute of Electrical and Electronics Engineers Access, IEEE Access, 2021; 9: P. 64929 – 64944. Doi: 10.1109/ACCESS.2021.3073507.
[16] Alizadeh, H., Zanjirdar, M., Haji, G,The Ability of Elliott Waves Theory to Predict the Information Content of Accounting Profit, Advances in Mathematical Finance & Applications, 2022; 7.Doi:10.22034/amfa.2022.1950621.1685.
[17] Haddadian, H., Haskuee, M., Zomorodain, G., An Algorithmic Trading System Based on Machine Learning in Tehran Stock Exchange, Advances in Mathematical Finance & Applications, 2021; 6(3):653-669, Doi:10.22034/amfa.2020.1894049.1380.
[18] Tavakoli, M., Doosti, H., Forecasting The Tehran Stock Market by Machine ‎Learning Methods Using a New Loss Function, Advances in Mathematical Finance & Applications, 2021; 6(2): 194-205.Doi:10.22034/amfa.2020.1896273.1399.

نص كامل:



Adv. Math. Fin. App., 2024, 9(3), P.1128-1139
	Advances in Mathematical Finance & Applications www.amfa.iau-arak.ac.ir Print ISSN: 2538-5569 Online ISSN: 2645-4610 Doi: 10.22034/amfa.2022.1967149.1796

 

 Original Research

Designing a Trading Strategy to Buy and Sell the Stock of Companies Listed on the New York Stock Exchange Based on Classification Learning Algorithms

 

Nasser Heydaria, Majid Zanjirdara ,*, Ali Lalbarb

aDepartment of Finance, Arak Branch, Islamic Azad University, Arak, Iran bDepartment of Accounting, Arak Branch, Islamic Azad University, Arak, Iran

 

Article Info Article history: Received  2022-09-06 Accepted  2022-12-15   Keywords: Trading Strategy Machine Learning Classification Algorithms		Abstract
		This research investigated the development of a stock trading strategy for companies on the New York Stock Exchange (NYSE), a prominent global market. Data was acquired from established libraries and the Yahoo Finance database. The model employed technical analysis indicators and oscillators as input features. Machine learning classification algorithms were used to design trading strategies, and the optimal model was identified based on statistical performance metrics. Accuracy, recall, and F-measure were utilized to evaluate the classification algorithms. Additionally, advanced statistical methods and various software tools were implemented, including Python, Spyder, SPSS, and Excel. The Kruskal-Wallis test was employed to assess the statistical differences between the designed strategies. A sample of 41 actively traded NYSE companies across diverse sectors such as financial services, healthcare, technology, communication services, consumer cyclicals, consumer staples, and energy were chosen using a filter-based approach on June 28th, 2021. The selection criteria included a market capitalization exceeding $200 billion and an average daily trading volume surpassing 1 million shares. Evaluation metrics revealed that the designed random forest trading strategy achieved a good fit with the data and exhibited statistically significant differences from other strategies based on classification learning algorithm.

 

1 Introduction

     This passage examines algorithmic trading strategies within the context of the New York Stock Exchange (NYSE), the largest stock exchange in the US by market capitalization (approximately $26 trillion as of May 2021). The NYSE hosts over 3,500 prominent and actively traded companies. The research focuses on the growing prevalence of algorithmic trading, a technique utilizing computer programs to execute trades. This method has seen a significant rise in the US market over the past century, with algorithmic trades exceeding 70% of total volume by 2009-2010 (compared to 15% in 2003). Given the crucial role of capital markets for various stakeholders (investment firms, portfolio managers, etc.), researchers have explored the design of optimal trading algorithms. Existing studies suggest the potential of various machine learning approaches: Zhang identified Bayesian neural networks as promising for accurate predictions [3]. Sooud Al-Tabi advocated for the superior performance of hybrid models [15]. Isaac Kufi compared models, highlighting the suitability of support vector machines (SVMs) and the lower error rates observed with neural networks compared to SVMs and decision trees [8]. Corolla Roya and Guijarro reported superior accuracy using the random forest algorithm [9]. Zhang et al. conversely found SVMs to be the most accurate [12]. These conflicting findings regarding the best learning algorithms motivated this study to evaluate trading strategies based on various machine learning models. The research incorporates technical analysis indicators and oscillators as model inputs, aiming to design a trading strategy for NYSE-listed companies using classification algorithms.

§       The criticality of such strategies stems from several factors:

§       The significance of the capital market for investors.

§       The ever-increasing volume of information, exceeding the capacity for traditional analysis methods.

§       The need for enhanced decision-making speed and accuracy to maximize profits and minimize losses.

§       The ability to compare the performance of different classification strategies.

The NYSE market capitalization was approximately $25 trillion in June 2021. The anticipated benefits of this research extend to various stakeholders within the financial landscape, including investors, capital providers, investment firms, holdings, consulting firms, capital market service companies, portfolio management companies, brokerage firms, investment funds (stock, mixed, fixed income), market management funds, financial information processing companies, and digital currency exchanges.

2 Theoretical Fundamentals and Research Background

Some concepts in technical analysis are taken from Charles Dow regarding market theories based on which all information affects prices. Dow sometimes found that a company's share price on the stock market fell after good news came because the news was not as good as expected. These rules are still true for many traders and investors, especially those who use technical analysis tools extensively. The Dow Theory was introduced by Michael O'Higgins and John Downes, known as the Dow Dog Theory, as one of the most well-known investment strategies in the United States to outperform the market [4]. Machine learning theory is a new branch of artificial intelligence, which starts by identifying the learning domain and solves problems by testing and applying the obtained results [5]. Many machine learning algorithms have been developed and used to predict the stock market [6]. Machine learning in the modern sense was created by psychologist Frank Rosenblatt from Cornell University based on the human nervous system. Based on this idea, a group built a machine between 1957 and 1960 to recognize letters of the alphabet and called it perceptron as the first modern example of an artificial neural network close to animal and human models. Novikoff added a convergence condition to the perceptron model, which limited the learning algorithm to several steps for better understanding. The significant development of machine learning during the last two decades has led to divergence in this theory so that the learning of Bristow et al and reinforcement learning can be mentioned [7]. Linear classification methods try to separate the data by constructing a hypersurface as a linear equation. As one of the linear classification methods, the SVM (Support Vector Machine) classification method finds the best hypersurface that separates the data of two classes with the maximum distance. The support vector machine algorithm was first proposed by Vladimir, Vapnik, and Alexey. Bernhard Bowser, Isabel Goyan, Vladimir, and Penik proposed a solution to create a non-linear classification using the Kernel function [1].

Table 1: Summary of Research Background

Researcher	Year	Learning Method	Model Evaluation Indicator	Data	Conclusion
Alizadeh et al. [16]	2022	Random forest	Accuracy	Elliott waves	We can predict the price trend based on EW and the RF.
Hadian et al. [17]	2021	NN, FL, and GA	Returns	RSI, PDI, stochastic, CP, and MA	The return on investment with the active method based on research algorithms is higher than buying and holding method.
Tavakoli and Dousti [18]	2021	NN, SVM, and GA	MSE and RMSE	Price	NN have high accuracy with optimization through GA
Seif et al. [13]	2021	DT, SV and SB	Accuracy, correctness, recall and F scores	RSI and Elliott waves	The SVM predicted the stock market index with about 98%, the DT with about 92%, and the simple MA with about 57% accuracy.
Daei et al. [14]	2020	SVM w   ith linear kernel	Accuracy, correctness, and F score	SP Published news	The SVM with a LK could increase the prediction power to 83% on average for the price of chemical products in the stock market and with non-LK to 85%.
Isaac Kufi [8]	2020	SVM, RF, DT, GA, and multilayer NN	The area under the curve, RRMSE, MAE, SD, accuracy, and correctness.	O &C P, L & H P, SMA, EMA, MACD, RSI, VI, and stochastic	The accuracy of the SVM was 94%, and the accuracy of this model was higher than other models.
Corolla Roya and Guijarro [9]	2020	DL, RF, GRA, and GLM	Accuracy index	L & H Aroon, RAR, BB, Chaikin fluctuations, near-C F, CC, DMI, EMA, MFI, MACD, VI, RSI, SMI, PI	The RF had a better performance with an average accuracy of 80% and could predict the coming ten days of the market.
Zhang et al. [3]	2020	SVM, DT, GB, LR	Accuracy, R, and F, and RRMSE, MSE, MAE, and AE were used to evaluate the performance of CA.	Daily trading data (H, L, O, C) and the previous day's stock price, online news, fluctuations and technical indicators such as MA, VI, FS	The optimal F value for the support vector algorithm was equal to 67%, which was more accurate than logistic regression in price prediction
Dosdogru et al. [10]	2018	NN and meta-heuristics hybrid model	Relative root mean square error (RRMSE), mean relative error, and mean square error	Density/Distribution Index, Chaikin Oscillator, MACD, NVI, RSI, BB, PCI, VCI, Momentum	Various NN models were used to improve the accuracy of stock market predictions, and the combination of meta-heuristics and NN models had valuable advantages.
Zhang et al. [11]	2018	From short-term memory and convolutional NN, Arch, Fungi, and ARIMA and AI model of NN and SV	Relative root means square error (RRMSE) and prediction accuracy	Rising, falling, O&C P, fluctuations, RSI, BB, VI, SM, EMA, stochastic, SMA, MACD	The proposed approach of this study can effectively improve the prediction accuracy of stock price direction and reduce the prediction error
Zhang et al. [12]	2017	SVM and confirmatory factor analysis	Accuracy and correlation matrix	CP, industry index, CT, financial ratios and book value	The accuracy rate of the SVM in this research was above 50%

 

The Random Forest algorithm is an easy-to-use machine learning algorithm, providing appropriate results without adjusting its meta-parameters in the form of a random forest based on a group of decision trees. The forest is constructed using the trees and bagging method, and the random forest builds multiple decision trees and merges them to produce more accurate and stable predictions. This method, developed by Berryman, combines the bagging sampling approach of Berryman and the random selection of features independently developed by Hu and Amit and Jeman. Random forest is a machine learning algorithm based on decision trees and bagging, and decision trees are algorithms that help predict and classify based on a series of roles [35]. Prediction and classification rules are provided by nodes and branches, and the final prediction and classification are displayed in the framework of tree leaves [2]. The K-Nearest Neighbors method is a case learning method, which is one of the simplest machine learning algorithms and was first described by Cover and Hart. In this algorithm, the K-value and the nearest neighbor are two critical issues. Euclidean distance, Mahalanobis distance, and the cosine of the Manhattan angle are the most important functions used to measure the nearest neighbor distance [3]. Researchers have used learning methods, as well as various data and model evaluation indicators to design a trading strategy (Table 1).

3 Proposed Methodology

      Research question: Does the trading strategy designed based on classification learning algorithms for buying and selling stocks of companies listed on the New York Stock Exchange fit with the data ?This applied, classified, and post-e   vent research was conducted on active companies accepted in financial services, health care, technology, communication services, consumer cycle, consumer support and energy in the New York Stock Exchange based on the classification made in the Yahoo Finance information base. In the next step, companies with a market value greater than 200 trillion dollars were determined, and finally, companies with an average transaction volume of more than 1 million were selected using the filter writing method on 2021/06/28.

Table 2: Population and Sample

No.	Industry	Number of companies			Number of samples
		Accepted in the stock exchange	With a market capitalization greater than $200 trillion	With an average quarterly transaction volume above 1 million
1	financial services	1606	23	6	6
2	Health Cares	493	6	6	6
3	Technology	467	10	9	9
4	Communication services	187	8	8	8
5	Consumer cycle	401	6	5	5
6	Consumer support	161	5	5	5
7	Energy	218	2	2	2
Total		3533	61	41	41

 

Indicators and oscillators are the input data of the trading strategy and are calculated in Table 3. Theoretical foundations and research literature were collected by reading resources, publications, domestic and foreign books, and information bases. Iranian Research Institute for Information Science and Technology (IranDoc), scientific and research journals approved by the Ministry of Science, Research, and Technology, the database of the Academic Jihad Scientific Information Center, Noor specialized journals (Noormags), and Civilica were used to check domestic resources. In addition, google scholar, ResearchGate, the database of specialized journals, google books, and Educational Resources Information Center (ERIC) were applied for international articles. The data of opening and closing price and highest and lowest price in daily timeframe were extracted through programming in Python software and yahoo finance for the sample stocks of the research. The indicators and oscillators were calculated using available libraries and programming as input data. Some of the essential software libraries in this research are Numpy, Pandas, Seaborn, Pyfolio, and Matplotlib. This study aimed to design a trading strategy to buy and sell the stock of companies listed on the New York stock exchange based on classification learning algorithms. First, the indicators and oscillators such as simple moving average, exponential moving average, converging and diverging moving average, relative strength index, Bollinger bands, price channel, high and low Aroon, momentum oscillator, directional average index, momentum, and density and distribution index were calculated based on the lowest, highest, beginning and ending price of the company's stock. Then, the calculated indicators were used as input data in support vector machine, random forest, and nearest neighbor algorithms with learning and testing rates of 60 and 40%. In the next step, the Accuracy, Recall and F-Measure indicators were used to check the classification algorithms to choose a trading strategy. The evaluation indicators of the model were imported to the SPSS software after processing in the Excel environment. Finally, Kolmogorov-Smirnov and Shapiro-Wilk tests were used to assess the distribution of research data, and the crosstab test, Chi-Square tests, and Kruskal-Wallis’s test were utilized to examine different strategies.

 

Table 3: Methods of Measuring Indicators and Oscillators in Technical Analysis

---