Abstract
In recent years, with the rapid development of computer computing power and
artificial intelligence, quantitative investment began to emerge in China's financial
markets. By means of the trend of Big Data, machine learning gradually play an
advantage in the field of quantitative investment. The application of machine learning
technology in quantitative investment field is mainly studied in this paper.
Academic research shows that sentiment of investor can affect the future trend of the
stock market. In order to quantify the investor's mood reasonably, the Python software
is used to write the web crawler program firstly, obtaining 2 million title text data in
the Oriental wealth net “stock”. After the word processing and the statistic of the word
frequency, the initial key word list is formed. Then related search word of the
Baidu-index is used to expand the number of keywords to 69, and the investor
sentiment index is computed through Elastic Net and principal component analysis.
Meanwhile, using a total data of 1500 trading days of the Shanghai Composite Index
from March 24, 2011 to May 24, 2017, the stock market timing trading model is built
based on support vector machine (SVM), after the feature engineering processing.
Then the optimal parameter is learnt by Sliding Window method, so as to predict the
trend of the stock market and the operating signal of trading day in the future. Finally,
the investor sentiment index based on text mining is substituted into SVM as a feature
of the data, discussing if it can improve the predicting performance of the model.
Two conclusions are draw in this paper: First, after the processing of original data by
the feature engineering, the stock market timing trading model based on SVM both
has high prediction accuracy for the rising market, the falling market and the volatile
market. Specially, it captures more of the trading information in volatile market.
Compared to the “buy and hold” strategy, it has a higher annualized rate of return.
Second, the investor sentiment index based on text mining can improve the predicting
performance of the model.
Key words: Text Mining; Investor Sentiment Index; SVM; Timing Trade
评论0