Translated Abstract
With the rapid development of the financial market，more and more reaserches of stock have come to our eyes. In recent years, the prediction of stock price turning points which are acknowledged as effective trade signals based on the machine learning has been a hot research direction. Based on the Gaussian process classification and regression model, the paper studied the prediction problem of stock price turning points. The main contents are as follows:
1. Firstly, study the turning points of stock history price equence. Compare and analyze two turning points abstract methods, one based on the time window and the other based on the piecewise linear representation(PLR). From the experimental comparision, the method based on PLR has a better performance on abstracting the turnning points and the result can react the price reversal case more accurately.
2. Secondly, focus on the prediction problem of stock price turning points, and propose an integrating piecewise linear representation and Gaussian process classification for stock turning points prediction algorithm. PLR is for the turnning points of stock history close price data, which are the history trade points. And GPC is for modeling the relationship between the stock close price and the characters which can influence the stock price in order to make a decision whether it’s the turning point. With PLR-GPC, we can predict the turning points of stock price. For the reason that the output of GPC is the probability that the current point is the turning point, and the probability responses the reliability and risk that the current point is regarded as the trade signal, in investment, different probabilities can be chosen for different investment risk preferences. Based on the probabilistic prediction, the paper discusses the relationship between the selection of probability threshold value and the investment risk preference in detail.
3. Thirdly, for the prediction of long stock trend, propose a stock price trend prediction algothrim by integration stock tangent theory and Gaussian process regression, which is an experiment to combine the machine learning method and the frequently-used stock tangent theory in stock investment. With the method of GPR, we can model and stuty support and resistance point sequence, and predict the value of the next support and resistance point. Because it’s difficult to predict the investment decision time directly, when the stock price is going to break, we can use the probability character of the GPR output, predicting the price probability interval of the next support and resistance point, and it’s the decision time when the real price comes into the price interval.
4. To verify the feasibility of the two prediction algorithms proposed in this paper, we carry out experiments on the real stock data respectively. The results of the experiments show that the algothrims can improve the return on investment, and can be used in actual stock trade.
Corresponding authors email
