Research Based On Data Mining In Forecasting Securities

[I] IntroductionAs an important feature of the market economy and the stock market, from the birth of the day on tens of millions of investors concerned about the heart. High-risk high-return characteristics of the stock market, so investors are always concerned about the stock market, stock market analysis in an attempt to forecast the development trend of the stock market. More than a century, along with some analysis of the stock market have gradually improved and development, such as: Dow analysis, K chart analysis, histogram analysis, the points chart analysis, moving average method, and morphological analysis France, trend analysis, analysis point of view, and the mysterious series of gold than the spiral calendar segmentation, four space law, with computer technology in the field of securities analysis and application of universal, has introduced a new indicator analysis. However, strictly speaking these methods is the only means of analysis, it can not predict the dynamics of the stock market. In addition, people have tried to use statistical tools such as regression analysis model to predict the stock market. However, the use of traditional techniques for forecasting the stock market and the forecast is one of the most fundamental problems, and that is to be dealt with the huge amount of data. As the stock market by the political, economic and other factors, the internal rules are extremely complex, some of the changes in the law of the cycle may be a year or even several years, the need for large amounts of data can be analyzed, and the traditional Technology Forecast Forecast results are unsatisfactory.[II] ResearchThe present design of data mining technology to analyze the trend of the stock forecast. Goals for the use of data mining of several commonly used methods to build predictive models through the course of the forecast and predict the results of the analysis, data mining algorithms to search for stocks with predictable point. The following details:1.Time series refers to time in accordance with the order in which they will be of value to a variable line up to form the sequence. Time-series model is mainly used to predict the future, which forecasts trends in law. This design will be three times the forecast method to predict the trend of the stock, were a simple moving average forecast law, an exponential smoothing forecasting method, the second exponential smoothing prediction.2. Markov chain model is based on Russian mathematicians Markov named after a kind of dynamic random mathematical model, through its analysis of the random variable of the current movement of these variables to predict the future movement. This article first stock index up according to their degree is divided into four types, using Markov chain related to the nature of the stock index on the run trend of short-term probability forecasting. [III] Time Series Forecast stock methodIn this paper, the time sequence of several commonly used method - a simple moving average forecast law, an exponential smoothing forecasting method, two exponential smoothing prediction method to predict the trend of the stock analysis. This innovative design in a smooth index prediction and two exponential smoothing prediction method of weight to carry out sub-paragraph discussion of values in order to achieve the best forecast. Index below Smooth to a prediction, for example, detailed time-series model for forecasting the trend of the stock.Smooth law forecast an index based onα(1 -α)~i as weighted(0<α<1,i=0,1,2,3,…)Of time series (yt)αweighted average forecast of a method, yt's weight, yt-1's weightα(1 -α), yt-2's weightα(1 -α)~2, ... and so on. Formula:One, yt, said t the first period of the actual value, is the first (t +1) forecast period, respectively, said the first (t-1) period and a period of t exponential smoothing value, said that smoothing factor, 0 <α<1. Standard error of prediction:One, n for the time-series data contained in the original number.Asia and Thailand in Group (600,881), for example, we have to predict in November 2007 of the trend of the Group of Asia and Thailand, source of data for the long wisdom of the large financial securities. (yt) time series, we select from May 8, 2007 to October 31, 2007 closing price, for the six months period, a total of 120 data.Smoothing factorαpredictive value of greater impact, but it has not been a good way to value the election, can only be determined in the light of experience. Analysis of the design in accordance with the number of days and the fluctuations of the broken line to do the following:(1) N data for the highest and the lowest value of the difference is less than 3.5 (million) as data showed the level of development trend, in which N is not less than 20 days.(2) of the remaining cases are rising (declining) trend-based. After calculating, on May 8 to Aug. 28, on October 15 to October 31 showed up (down) trend-based, on August 29 to October 12 was the level of development trends. When the time series data showed the trend of development-level,αin value between 0 to 0.3; if the type of sequence data showed an increase (decrease)-type development trend in value between 0.6 to 1.Forecast formula adopted by the Asia and Thailand in November to seek the trend of the stock of the Group, has been predicted to do with the actual value of the error comparison. The results from it, right in the value of the three stages of value (0.7,0.2,0.8), the most accurate results, the error is: S index a smooth error (1 = 0.7,2 = 0.2,3 = 0.8) = 2.03.By the same token, we can complete the second exponential smoothing method of forecasting the trend of the stock's forecast.It appears from the results, forecast an exponential smoothing method and the second exponential smoothing prediction forecast than a simple moving average is better, but to a large extent depend on the value of the right to choose, at the same time a large amount of computing.[IV] Markov chain method forecast the trend of the stockThis article is a selected group Asia and Thailand on April 30, 2007 to October 31, 2007 closing price (in order to facilitate the calculation, with time-series data to select a different method), for the six months period, a total of 121 closing price data Sample data, to observe the Group of Asia and Thailand on April 30, 2007 to October 31, 2007 closing price, or a maximum of 2.98 (million), the highest decrease of 2.79 (million), due to the stock during this period no serious Change , So the forecast in the course of this article ignored the rise of the (stop) the special board, in accordance with the Group of Asia and Thailand's stock index up every day and the actual situation (or) the high rate of the daily closing price on the same day by the state Is defined as down strong, or weak, weak inflation, rose four strong state: daily or 1 (million) contains more than 1 (million) is defined as down strong; day, a decrease of 1 (million) included the following 0 (yuan) Is defined as weak or; per day or 1 (million) included the following 1 yuan for the weak rise (X3); per day or 1 (million) or more is defined as up-keung (X4). P said the state transition matrix, Pij said that the state i to j state appears probability, on October 31 for a state or strong, so the day of the state vector u (120) = (1,0,0,0), according to our formula Be on November 1 Group Asia and Thailand's index in the near future. u (121) =u (120)·P=(0.4019 0.0909 0.3636 0.1429)So on November 1 will be weak up. By the same token to be in November predicted the trend of the stock. Markov chain that may arise due to non-transfer, so this paper, the Markov chain increase in the number of state. It was found that with the increasing number of states, predictable number of days also increased the forecast accuracy has been greatly improved. In this paper, the Markov chain is divided into 600 kinds of state will be able to achieve an accurate forecast of the trend of the stock, calculated (Method By the same token, there Description not),SMarkov chain error=0.35.[V] Summary and OutlookIn this paper, the use of data mining methods in time-series methods, Markov chain model shares the same trend was forecast.The two types of data mining method of forecasting, the paper concluded: time-series forecasting methods of the trend of the stock of the most simple, but the error is too large. Markov chain model better suited to the trend of the stock on the forecast, the smaller the amount calculated to improve the forecasting method is more effective.Due to limited time to study, the design is still a lot to the development of space. Construction Markov chain model, if the data are too large changes in the state means that the number should increase, will increase a lot of invisible computing, can be more flexible in order to increase mobile matrix forecast is correct. In the design model, the design takes into account only the daily closing price, should be opened, the highest price, the lowest price, volume, volume, taking into account the forecast model. In addition, there are special circumstances of the shares (shares were phenomena occur) can not make the right prediction.