本文選題：中國股票市場性質(zhì) + 概念漂移�。� 參考：《廣東外語外貿(mào)大學(xué)》2014年碩士論文

【摘要】：自從Fama和French發(fā)表了經(jīng)典的論著，闡述了證券市場有效性的討論，近些年來，該論題一直是學(xué)者關(guān)注的焦點(diǎn)。本文從基本面和技術(shù)面兩個角度檢驗(yàn)中國股票市場的有效性，都否認(rèn)了有效市場假說。尤其是本文構(gòu)建的技術(shù)面分析模型，相比之前的研究，我們的模型在扣除了交易費(fèi)用之后能夠在不同的市場階段內(nèi)攫取穩(wěn)定的超額回報。 在基本面分析部分，使用橫截面回歸模型論證上市公司披露的財(cái)務(wù)數(shù)據(jù)與股票期望回報之間的相關(guān)性，發(fā)現(xiàn)賬面市值比，市值等財(cái)務(wù)因素對股票預(yù)期回報存在較強(qiáng)的解釋力。但是中國市場與美國市場在特定指標(biāo)的解釋力上有些不同。更為有趣的是，本文的得出了與之前美國股票市場的研究類似的結(jié)論，中國股票市場也發(fā)生了時間上的概念漂移現(xiàn)象，即在不同的測試時間段內(nèi)賬面市值比指標(biāo)與預(yù)期回報的相關(guān)性發(fā)生了反轉(zhuǎn)。股票市場中存在“異象”。 在技術(shù)面分析部分，考慮到顯性知識和概念漂移是構(gòu)建金融數(shù)據(jù)模型過程中不可忽視的兩個重要的因素。本文使用eXtended Classifier Systems（XCS）算法來構(gòu)建智能交易模型—eTrend。模型基于當(dāng)前的股票價格行情數(shù)據(jù)特征以及趨勢跟蹤投資規(guī)則進(jìn)行股票買賣決策，在交易過程中，eTrend不斷的自動適應(yīng)股票市場的變化并記憶顯性交易規(guī)則，實(shí)時的提供最優(yōu)的投資決策支持。文章選取上海證券市場的三只指數(shù)作為測試樣例，在12年的測試周期內(nèi)，扣除了交易費(fèi)用的收益序列與大盤的收益序列相比，eTrend能夠在牛/熊市不同的市場階段內(nèi)獲得穩(wěn)定的超額收益，并保持較低的下行風(fēng)險，反映在統(tǒng)計(jì)指標(biāo)上是接近于1.0的索提諾比率。同時從對比的實(shí)驗(yàn)中可以看出，無論從總收益還是收益的穩(wěn)定性方面基于XCS的eTrend模型預(yù)測效果明顯要好于Decision Tree（DT）和Artificial Neural Network （ANN）。 本文主要的貢獻(xiàn)體現(xiàn)在：(1)在學(xué)術(shù)層面，基本面分析部分的結(jié)果較為有趣，在測試樣本上發(fā)現(xiàn)的地理位置，時間序列上的反轉(zhuǎn)異象，使得在金融研究中概念漂移問題的解決顯得尤為重要；相應(yīng)的技術(shù)面分析部分，eTrend自適應(yīng)智能交易模型的提出，對于有效解決股票數(shù)據(jù)的概念漂移問題和顯性知識挖掘的論題提供了一定的幫助。(2)從實(shí)踐層面，為投資者、學(xué)者更好的認(rèn)識股票市場提供了一定的依據(jù)；尤其是eTrend模型，，使用實(shí)踐中的投資規(guī)則與人工智能算法融合進(jìn)行約束學(xué)習(xí)，在長期的測試周期內(nèi)都取得的較好的效果，對于股票市場中的投資策略的研發(fā)者而言極具參考價值。
[Abstract]:Since Fama and French published their classic treatises on the validity of the securities market, this topic has been the focus of scholars' attention in recent years. In this paper, the validity of Chinese stock market is tested in terms of fundamental and technical aspects, and the efficient market hypothesis is denied. In particular, the technical surface analysis model is constructed in this paper. Compared with previous studies, our model can capture stable excess returns in different market stages after deducting transaction costs. In the part of fundamental analysis, the cross-section regression model is used to prove the correlation between the financial data disclosed by listed companies and the expected return of stock. It is found that the financial factors such as book market value ratio, market value and other financial factors have strong explanatory power to the expected return of stocks. But the Chinese market and the U. S. market in the interpretation of specific indicators are somewhat different. What is more interesting is that this paper draws a conclusion similar to the previous research on the American stock market. The Chinese stock market also has a phenomenon of conceptual drift in time. That is, in different test periods, the correlation between book market value ratio and expected return is reversed. There are anomalies in the stock market. In the part of technical surface analysis, it is considered that explicit knowledge and conceptual drift are two important factors which can not be ignored in the process of constructing financial data model. In this paper, we use the extended Classifier Systems (XCSS) algorithm to construct the intelligent trading model-e trend. Based on the characteristics of the current stock price data and the trend tracking investment rules, the model makes stock trading decisions. During the trading process, trend constantly adapts to the changes of the stock market and memorizes the dominant trading rules. Provide optimal investment decision support in real time. In this paper, three indexes of Shanghai stock market are selected as test samples, and the test period is 12 years. After deducting the transaction cost income series compared with the larger market earnings series, trend can obtain stable excess returns at different market stages of the bull / bear market and maintain lower downside risk. Reflected in the statistical indicators is close to 1.0 of the Sodino ratio. At the same time, it can be seen from the comparative experiments that the prediction effect of the eTrend model based on XCS-based model is better than that of decision Tree DTT and Artificial Neural Network in terms of the stability of total income and income. The main contribution of this paper is that at the academic level, the results of the fundamental analysis are more interesting, the geographical location found in the test samples, the inversion anomalies in the time series, Therefore, it is very important to solve the problem of concept drift in financial research. It provides some help to solve the problem of concept drift of stock data and the topic of explicit knowledge mining. It provides some basis for investors and scholars to better understand the stock market from the practical level, especially the eTrend model. Using the combination of investment rules in practice and artificial intelligence algorithm for constraint learning, good results have been achieved in the long test cycle, which is of great reference value to the R & D of investment strategy in the stock market.
【學(xué)位授予單位】：廣東外語外貿(mào)大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：F832.51;F224

【參考文獻(xiàn)】

相關(guān)期刊論文 前6條

1 吳微;張凌;;自適應(yīng)參數(shù)的AOSVR算法及其在股票預(yù)測中應(yīng)用[J];大連理工大學(xué)學(xué)報;2009年04期

2 胡代平,劉豹;多agent股票預(yù)測支持系統(tǒng)的設(shè)計(jì)[J];系統(tǒng)工程;2001年02期

3 陳安斌,蘇俊輔;XCS分類元系統(tǒng)于金融投資決策上之應(yīng)用——以臺灣加權(quán)指數(shù)為例[J];管理學(xué)報;2005年S2期

4 彭麗芳;孟志青;姜華;田密;;基于時間序列的支持向量機(jī)在股票預(yù)測中的應(yīng)用[J];計(jì)算技術(shù)與自動化;2006年03期

5 林倩瑜;馮少榮;張東站;;基于神經(jīng)網(wǎng)絡(luò)和模式匹配的股票預(yù)測研究[J];計(jì)算機(jī)技術(shù)與發(fā)展;2010年05期

6 葉德謙;金大兵;楊櫻;;基于強(qiáng)化學(xué)習(xí)的股票預(yù)測系統(tǒng)的研究與設(shè)計(jì)[J];微計(jì)算機(jī)信息;2006年06期

本文編號：2025501