Constructing Call Warrant Trading Strategies with Deep Learning in Taiwan
碩士 === 國立中山大學 === 財務管理學系研究所 === 106 === This study uses all call warrants from 2003 to 2017 in Taiwan to design trading strategy about Bollinger Bands ,breakouts and moving averages, and then to use two feature selection methods : learning vector quantization plus five warrant features and random fo...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | zh-TW |
Published: |
2018
|
Online Access: | http://ndltd.ncl.edu.tw/handle/h38536 |
id |
ndltd-TW-106NSYS5305048 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-106NSYS53050482019-10-31T05:22:27Z http://ndltd.ncl.edu.tw/handle/h38536 Constructing Call Warrant Trading Strategies with Deep Learning in Taiwan 深度學習在台灣認購權證交易策略之運用 Jung-Chun Mi 糜榮峻 碩士 國立中山大學 財務管理學系研究所 106 This study uses all call warrants from 2003 to 2017 in Taiwan to design trading strategy about Bollinger Bands ,breakouts and moving averages, and then to use two feature selection methods : learning vector quantization plus five warrant features and random forest plus five warrant features to select important features. Then, multi-layer perceptrons in deep learning are used to predict the buy signals, and take the intersection of the above two methods to become third result. Finally, to invest in the best of the three methods and observe it performance. Input variables in this study are selected by feature selection methods. Call warrants price of trading strategies is up or down as output variables, and up is defined as call warrants’ return is greater than zero, down is defined as call warrants’ return is less than zero. The design of data period is from 2003 to 2008 as training data, and 2009 as a testing data;2003 to 2009 as a training data, and 2010 as a testing data and so on. The empirical results show that the performance of feature selection using learning vector quantization with deep learning is the best. As for the results of funding, the annualized rate of return is 44.28%, and the assets grow from 1,000,000 at the beginning of 2009 to about 23,000,000 at the end of 2017, which grow by nearly 22 times. This result show that the deep learning method can be used not only in stock market in Taiwan, but also in call warrant market in Taiwan . Huang Jen-Jsung 黃振聰 2018 學位論文 ; thesis 84 zh-TW |
collection |
NDLTD |
language |
zh-TW |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 國立中山大學 === 財務管理學系研究所 === 106 === This study uses all call warrants from 2003 to 2017 in Taiwan to design trading strategy about Bollinger Bands ,breakouts and moving averages, and then to use two feature selection methods : learning vector quantization plus five warrant features and random forest plus five warrant features to select important features. Then, multi-layer perceptrons in deep learning are used to predict the buy signals, and take the intersection of the above two methods to become third result. Finally, to invest in the best of the three methods and observe it performance.
Input variables in this study are selected by feature selection methods. Call warrants price of trading strategies is up or down as output variables, and up is defined as call warrants’ return is greater than zero, down is defined as call warrants’ return is less than zero. The design of data period is from 2003 to 2008 as training data, and 2009 as a testing data;2003 to 2009 as a training data, and 2010 as a testing data and so on.
The empirical results show that the performance of feature selection using learning vector quantization with deep learning is the best. As for the results of funding, the annualized rate of return is 44.28%, and the assets grow from 1,000,000 at the beginning of 2009 to about 23,000,000 at the end of 2017, which grow by nearly 22 times. This result show that the deep learning method can be used not only in stock market in Taiwan, but also in call warrant market in Taiwan .
|
author2 |
Huang Jen-Jsung |
author_facet |
Huang Jen-Jsung Jung-Chun Mi 糜榮峻 |
author |
Jung-Chun Mi 糜榮峻 |
spellingShingle |
Jung-Chun Mi 糜榮峻 Constructing Call Warrant Trading Strategies with Deep Learning in Taiwan |
author_sort |
Jung-Chun Mi |
title |
Constructing Call Warrant Trading Strategies with Deep Learning in Taiwan |
title_short |
Constructing Call Warrant Trading Strategies with Deep Learning in Taiwan |
title_full |
Constructing Call Warrant Trading Strategies with Deep Learning in Taiwan |
title_fullStr |
Constructing Call Warrant Trading Strategies with Deep Learning in Taiwan |
title_full_unstemmed |
Constructing Call Warrant Trading Strategies with Deep Learning in Taiwan |
title_sort |
constructing call warrant trading strategies with deep learning in taiwan |
publishDate |
2018 |
url |
http://ndltd.ncl.edu.tw/handle/h38536 |
work_keys_str_mv |
AT jungchunmi constructingcallwarranttradingstrategieswithdeeplearningintaiwan AT míróngjùn constructingcallwarranttradingstrategieswithdeeplearningintaiwan AT jungchunmi shēndùxuéxízàitáiwānrèngòuquánzhèngjiāoyìcèlüèzhīyùnyòng AT míróngjùn shēndùxuéxízàitáiwānrèngòuquánzhèngjiāoyìcèlüèzhīyùnyòng |
_version_ |
1719284546794422272 |