東華大學圖書館 |

Research on Trend Issues of Stock Price Time Series based on Fractal Theory.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Research on Trend Issues of Stock Price Time Series based on Fractal Theory./
作者:	Wei, Yi.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2021,
面頁冊數:	110 p.
附註:	Source: Dissertations Abstracts International, Volume: 83-01, Section: B.
Contained By:	Dissertations Abstracts International83-01B.
標題:	Computer science. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28498438
ISBN:	9798516940712

Research on Trend Issues of Stock Price Time Series based on Fractal Theory.
Wei, Yi.

Research on Trend Issues of Stock Price Time Series based on Fractal Theory. - Ann Arbor : ProQuest Dissertations & Theses, 2021 - 110 p.

Source: Dissertations Abstracts International, Volume: 83-01, Section: B.

Thesis (Ph.D.)--State University of New York at Buffalo, 2021.

This item must not be sold to any third party vendors.

The movement of stock price follows a random walk in the short term, so it is extremely difficult to predict. However, the movement of stock price follows a fractal trend in the long term. This argument has become the research basis of this doctoral dissertation. The profits obtained through financial trading are mainly derived from trends, so the study of the trend of stock price time series is the essence of ensuring the profit of stock trading. This doctoral dissertation not only focuses on the trend feature reconstruction, the trend turning points extraction, the trend linear segmentation, and the fractal trend representation, but also focuses on an in-depth analysis of the non-trend drawback of model evaluation results on the stock price prediction. Finally, a fractal trend evaluation method is created that can guide actual securities trading, thus ending the long-standing situation where the classic prediction error based model evaluation method cannot guide actual securities trading. Under the guidance of fractal theory, this doctoral dissertation has made great progress in addressing the trend related issue of the above-mentioned stock price time series.The first research proposed NARX neural network model, the sample data of stock AAPL have been trained in NASDAQ from 2006/01/01 to 2015/01/01 are utilized for training. The results show that if the data are used with the trend information as a training dataset, it will significantly reduce the prediction error and improve the generalization ability of the NARX model, so as to predict the stock trend change at a certain time. In addition, the results of training the nonlinear time series data show that the outputs in probability series form with respect to original price have practical significance. During stock trading, it is sometimes difficult for investors to make decisions, because their price is affected by a variety of factors. Therefore, providing the analytical results in an expected probability form to investors is one of the main goals for researchers. Through the training on nonlinear time series data, the network model has provided the probability sequence corresponding to its price movements, which is important guidance for stock trading. It has successfully estimated the possibility of buying and selling, which provides the necessary theoretical basis on how to determine the stock trading points.The second research has achieved the goal to quickly understand the macroscopic change of NASDAQ stock price data. The provided stock data distribution with effective trading points plays an important role in real-time trading. According to the statistical results, the dual period DMAC approach has a better outcome than both golden-cross and dead-cross in daily and weekly periods. It also has practical significance to determine the trading points with respect to buy and sell, and theoretical significance for the research on expecting stock profits and stop-loss. The market data analysis illustrates that the number of stocks which have the average rate of profit of overall buy points higher than 3% is 373, including GALT, etc.; the number of stocks which have the average rate of profit of overall sell points higher than 3% is 193 including AMDA, etc. In this study, the volatility segmentation approach reduces the uncertainty of stock selection and estimation in a single period DMAC, which helps investors to achieve the desired objectives to improve the stock investment income. It provides a novel application of data-intensive computing on quantitative analysis stock trading. The computation time can be expected to reduce significantly below a second by implementing this approach using Message Passing Interface instead of Matlab by making the code more efficient and scalable.The third research proposed TST system that can be targeted to improve the effectiveness of the piecewise linear representation of the bottom-up segmentation algorithm on stock time series. The training results show that the determined subsequence with proper trend feature is suitable for training when its training error is at 0.196. The accuracy of the predicted trend is satisfactory and can reach between 70%-80%. The TST approach can systematically improve the ability of neural networks to identify the trend feature of the selected stock time series. After the RNN training, only those subsequences generated by designated threshold error corresponding to the trend features with low training error can obtain satisfactory learning performance. In other words, such a stock trend feature generated by the linear representation of the stock time series can be well learned by RNN. This result shows that the training error as feedback from neural networks is an indispensably essential procedure in the proposed TST system.In the fourth research, the trend recognition of stock price subsequence is investigated as the fundamental work for Elliott wave theory when applying to machine learning research and applications, however, at present there are few related research work in this field. As known to all, the stock wave trend recognition is the key task of swing trading of stocks, and its implementation is challenging. This study attempts to use the RNN to classify the stock wave and identify its trend direction. It provides guidance for swing trading as well as theoretical support. The AUC is used on the classification performance of RNN on the wave trend of stock sequence. It can be seen that no matter whether it is the 20 Chinese stocks or the 20 US stocks, for the recognition performance of wave trend in RNN, the average price sequence is better than the unit price sequence. Its advantage is also reflected in the testing wave sequences of Chinese and American stocks. In the trend recognition result of the unit price sequence, the average AUC of Chinese stocks is 0.548, and the average AUC of US stocks is 0.48. In the trend recognition result of the average price sequence, the average AUC of Chinese stocks is 0.611, and the average AUC of US stocks is 0.580. The results show that whether the Chinese stocks or the US stocks, the RNN training performance of the stock average price sequence is better than the stock unit price sequence.In the fifth research, the performance of the current NNSPP is evaluated based on the PE method, but for specific trading in the securities market, this evaluation method has statistical flaws. As known to all, the PE method takes the absolute value of the prediction error, while the movement of the stock time series is directional. Unfortunately, the prediction error fails to characterize the essential fractal characteristics of the financial time series' inherent rise and fall trends. Neural Networks for Stock Price Prediction (NNSPP) have been popular for decades. However, most of its study results remain in the research paper. They cannot play a role in the securities market at all. One of the main reasons leading to this situation is that the Prediction Error (PE) based evaluation results have statistical flaws. Its prediction results cannot represent the most critical financial direction attributes. So it cannot provide investors with convincing, interpretable, and consistent model performance evaluation results for practical applications in the securities market.The sixth research theoretically explained the reason for securities trading's invalidity in the PE evaluation results and conducted an empirical analysis. Pearson's correlation results show that the directional fractal dimension highly correlates with the stock return direction. It has the ability to reflect the direction of the stock price. At the same time, this study shows that PEFT provides a new entry point for the NNSPP research. It also offers a concrete solution to the long-standing problem that the evaluation results of the NNSPP cannot be translated into application-oriented achievements.

ISBN: 9798516940712Subjects--Topical Terms:

523869
Computer science.
Subjects--Index Terms:

Trend issues

Research on Trend Issues of Stock Price Time Series based on Fractal Theory.
LDR:09589nmm a2200421 4500 001 2345949
005 20220613064822.5
008 241004s2021 ||||||||||||||||| ||eng d
020 $a 9798516940712
035 $a (MiAaPQ)AAI28498438
035 $a AAI28498438
040 $a MiAaPQ $c MiAaPQ
100 1 $a Wei, Yi. $3 871211
245 1 0 $a Research on Trend Issues of Stock Price Time Series based on Fractal Theory.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2021
300 $a 110 p.
500 $a Source: Dissertations Abstracts International, Volume: 83-01, Section: B.
500 $a Advisor: Chaudhary, Vipin.
502 $a Thesis (Ph.D.)--State University of New York at Buffalo, 2021.
506 $a This item must not be sold to any third party vendors.
520 $a The movement of stock price follows a random walk in the short term, so it is extremely difficult to predict. However, the movement of stock price follows a fractal trend in the long term. This argument has become the research basis of this doctoral dissertation. The profits obtained through financial trading are mainly derived from trends, so the study of the trend of stock price time series is the essence of ensuring the profit of stock trading. This doctoral dissertation not only focuses on the trend feature reconstruction, the trend turning points extraction, the trend linear segmentation, and the fractal trend representation, but also focuses on an in-depth analysis of the non-trend drawback of model evaluation results on the stock price prediction. Finally, a fractal trend evaluation method is created that can guide actual securities trading, thus ending the long-standing situation where the classic prediction error based model evaluation method cannot guide actual securities trading. Under the guidance of fractal theory, this doctoral dissertation has made great progress in addressing the trend related issue of the above-mentioned stock price time series.The first research proposed NARX neural network model, the sample data of stock AAPL have been trained in NASDAQ from 2006/01/01 to 2015/01/01 are utilized for training. The results show that if the data are used with the trend information as a training dataset, it will significantly reduce the prediction error and improve the generalization ability of the NARX model, so as to predict the stock trend change at a certain time. In addition, the results of training the nonlinear time series data show that the outputs in probability series form with respect to original price have practical significance. During stock trading, it is sometimes difficult for investors to make decisions, because their price is affected by a variety of factors. Therefore, providing the analytical results in an expected probability form to investors is one of the main goals for researchers. Through the training on nonlinear time series data, the network model has provided the probability sequence corresponding to its price movements, which is important guidance for stock trading. It has successfully estimated the possibility of buying and selling, which provides the necessary theoretical basis on how to determine the stock trading points.The second research has achieved the goal to quickly understand the macroscopic change of NASDAQ stock price data. The provided stock data distribution with effective trading points plays an important role in real-time trading. According to the statistical results, the dual period DMAC approach has a better outcome than both golden-cross and dead-cross in daily and weekly periods. It also has practical significance to determine the trading points with respect to buy and sell, and theoretical significance for the research on expecting stock profits and stop-loss. The market data analysis illustrates that the number of stocks which have the average rate of profit of overall buy points higher than 3% is 373, including GALT, etc.; the number of stocks which have the average rate of profit of overall sell points higher than 3% is 193 including AMDA, etc. In this study, the volatility segmentation approach reduces the uncertainty of stock selection and estimation in a single period DMAC, which helps investors to achieve the desired objectives to improve the stock investment income. It provides a novel application of data-intensive computing on quantitative analysis stock trading. The computation time can be expected to reduce significantly below a second by implementing this approach using Message Passing Interface instead of Matlab by making the code more efficient and scalable.The third research proposed TST system that can be targeted to improve the effectiveness of the piecewise linear representation of the bottom-up segmentation algorithm on stock time series. The training results show that the determined subsequence with proper trend feature is suitable for training when its training error is at 0.196. The accuracy of the predicted trend is satisfactory and can reach between 70%-80%. The TST approach can systematically improve the ability of neural networks to identify the trend feature of the selected stock time series. After the RNN training, only those subsequences generated by designated threshold error corresponding to the trend features with low training error can obtain satisfactory learning performance. In other words, such a stock trend feature generated by the linear representation of the stock time series can be well learned by RNN. This result shows that the training error as feedback from neural networks is an indispensably essential procedure in the proposed TST system.In the fourth research, the trend recognition of stock price subsequence is investigated as the fundamental work for Elliott wave theory when applying to machine learning research and applications, however, at present there are few related research work in this field. As known to all, the stock wave trend recognition is the key task of swing trading of stocks, and its implementation is challenging. This study attempts to use the RNN to classify the stock wave and identify its trend direction. It provides guidance for swing trading as well as theoretical support. The AUC is used on the classification performance of RNN on the wave trend of stock sequence. It can be seen that no matter whether it is the 20 Chinese stocks or the 20 US stocks, for the recognition performance of wave trend in RNN, the average price sequence is better than the unit price sequence. Its advantage is also reflected in the testing wave sequences of Chinese and American stocks. In the trend recognition result of the unit price sequence, the average AUC of Chinese stocks is 0.548, and the average AUC of US stocks is 0.48. In the trend recognition result of the average price sequence, the average AUC of Chinese stocks is 0.611, and the average AUC of US stocks is 0.580. The results show that whether the Chinese stocks or the US stocks, the RNN training performance of the stock average price sequence is better than the stock unit price sequence.In the fifth research, the performance of the current NNSPP is evaluated based on the PE method, but for specific trading in the securities market, this evaluation method has statistical flaws. As known to all, the PE method takes the absolute value of the prediction error, while the movement of the stock time series is directional. Unfortunately, the prediction error fails to characterize the essential fractal characteristics of the financial time series' inherent rise and fall trends. Neural Networks for Stock Price Prediction (NNSPP) have been popular for decades. However, most of its study results remain in the research paper. They cannot play a role in the securities market at all. One of the main reasons leading to this situation is that the Prediction Error (PE) based evaluation results have statistical flaws. Its prediction results cannot represent the most critical financial direction attributes. So it cannot provide investors with convincing, interpretable, and consistent model performance evaluation results for practical applications in the securities market.The sixth research theoretically explained the reason for securities trading's invalidity in the PE evaluation results and conducted an empirical analysis. Pearson's correlation results show that the directional fractal dimension highly correlates with the stock return direction. It has the ability to reflect the direction of the stock price. At the same time, this study shows that PEFT provides a new entry point for the NNSPP research. It also offers a concrete solution to the long-standing problem that the evaluation results of the NNSPP cannot be translated into application-oriented achievements.
520 $a The proposed PEFT method in this doctoral dissertation study has achieved the research goals of simultaneously outputting the two evaluation indicators of model prediction direction and prediction accuracy. It fulfills the professional needs of securities trading, and eventually fills the long-standing gap in NNSPP performance evaluation research.
590 $a School code: 0656.
650 4 $a Computer science. $3 523869
650 4 $a Computer engineering. $3 621879
650 4 $a Finance. $3 542899
650 4 $a Information science. $3 554358
653 $a Trend issues
653 $a Stock price
653 $a Fractal theory
653 $a Trend feature reconstruction
653 $a Stock trading
653 $a NASDAQ
653 $a Prediction Error
690 $a 0984
690 $a 0464
690 $a 0723
690 $a 0508
710 2 $a State University of New York at Buffalo. $b Computer Science and Engineering. $3 1035503
773 0 $t Dissertations Abstracts International $g 83-01B.
790 $a 0656
791 $a Ph.D.
792 $a 2021
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28498438