本文選題：算法交易 + 趨勢跟蹤交易��； 參考：《中南財(cái)經(jīng)政法大學(xué)》2017年碩士論文

【摘要】：算法交易通過預(yù)先設(shè)定的計(jì)算機(jī)程序?qū)ふ沂袌錾系慕灰讬C(jī)會(huì)并做出交易決定,然后高效、低成本地實(shí)現(xiàn)交易訂單的執(zhí)行和成交。目前,算法交易已經(jīng)成為資本市場的主流。復(fù)雜事件處理引擎能夠?qū)崟r(shí)處理瞬息萬變的市場信息數(shù)據(jù),捕獲稍縱即逝的算法交易時(shí)機(jī),有良好的應(yīng)用前景。然而,復(fù)雜事件處理在算法交易的應(yīng)用還處在新興階段,只是通過比較前后價(jià)格的大小發(fā)現(xiàn)趨勢。這樣的閾值過于敏感,匹配到的模式多且持續(xù)時(shí)間短,有較強(qiáng)的偶然性,并不能準(zhǔn)確判斷趨勢的出現(xiàn)。CUSUM(累積和)控制方法能夠過濾過于微小的波動(dòng)得到一段時(shí)期的整體趨勢,對(duì)微小的偏移較為敏感,而且適用于流處理。針對(duì)使用復(fù)雜事件處理進(jìn)行趨勢跟蹤時(shí)趨勢的界定閥值過于敏感的問題,在CUSUM控制方法的基礎(chǔ)上本文提出了基于CUSUM的趨勢判斷算法,以及根據(jù)趨勢判斷進(jìn)行交易的兩個(gè)基本的交易策略:極值點(diǎn)交易策略和趨勢追隨交易策略。設(shè)計(jì)了CUSUM趨勢跟蹤的復(fù)雜事件處理實(shí)現(xiàn),包括事件處理網(wǎng)絡(luò)設(shè)計(jì)、事件定義以及事件模式定義。最后,通過實(shí)證和性能測試證明基于CUSUM的趨勢跟蹤復(fù)雜事件處理實(shí)現(xiàn)是可行且高效的。為了驗(yàn)證所提出方法是否能夠獲利,本文選取華泰柏瑞滬深300ETF(510300)在2015年1月5日至12月31日的收盤價(jià)作為行情數(shù)據(jù)進(jìn)行模擬交易,并使用了開源的復(fù)雜事件處理引擎Esper來實(shí)現(xiàn)所提出的CUSUM趨勢跟蹤交易策略。實(shí)驗(yàn)表明:1、CUSUM趨勢判斷方法比前后價(jià)格大小比較的趨勢判斷方法產(chǎn)生更少的趨勢信號(hào),說明CUSUM趨勢判斷方式能夠通過閾值的控制過濾微小的波動(dòng),從而有效降低交易次數(shù)。2、在收益方面,CUSUM趨勢判斷方法配合趨勢追隨交易策略能夠獲得更高的收益。這種交易方法的特點(diǎn)是在平穩(wěn)細(xì)微震蕩時(shí)期較少進(jìn)行交易;在大幅波動(dòng)時(shí)期會(huì)進(jìn)行持續(xù)時(shí)間較為短暫的做多操作,但一般操作結(jié)果為虧損;在持續(xù)上升時(shí)期會(huì)進(jìn)行持續(xù)時(shí)間較長的做多操作,此時(shí)一般會(huì)有豐厚的獲利。3、在性能方面,系統(tǒng)在穩(wěn)定階段有96%的事件能夠在10微秒以內(nèi)處理完成,事件平均延遲約為4.5微秒,而且在系統(tǒng)承受范圍內(nèi)規(guī)則數(shù)的增加并不影響事件的平均處理時(shí)間。另外系統(tǒng)在平穩(wěn)運(yùn)行階段占用較少的CPU資源,內(nèi)存使用隨運(yùn)行時(shí)間增加而增加,但最終能夠保持平穩(wěn)不再持續(xù)增加。
[Abstract]:The algorithm deals with the pre set computer program to find the trading opportunities in the market and make the transaction decision. Then, the transaction order is implemented and sold at low level. At present, the algorithm transaction has become the mainstream of the capital market. The complex event processing engine can deal with the changing market information data and capture the fast changing market information. However, the application of complex event processing in the algorithm transaction is still in the emerging phase, only by comparing the size of the price to find the trend. The threshold is too sensitive, the matching pattern is more and the duration is short, there is a strong chance, and the trend is not accurate to judge the trend. The emergence of.CUSUM (accumulation and) control methods can filter too small fluctuations to get a period of overall trend, sensitive to small offset, and suitable for flow processing. The problem that the threshold threshold is too sensitive to trend tracking using complex event processing, based on the CUSUM control method, is proposed in this paper. The trend judgment algorithm based on CUSUM, and two basic trading strategies based on trend judgment: extreme point trading strategy and trend following transaction strategy, design the implementation of complex event processing for CUSUM trend tracking, including event processing network design, event definition and event pattern definition. Finally, through empirical and sexual characteristics In order to verify whether the proposed method is profitable, this paper selects the closing price of huatberi Shanghai and Shenzhen 300ETF (510300) from January 5, 2015 to December 31st as the market data, and uses an open source complex event processing method to verify whether the proposed method is profitable or not. Esper to implement the proposed CUSUM trend tracking transaction strategy. The experiment shows: 1, the trend judgment method of CUSUM trend judgment method produces less trend signal than the trend judgment method compared with the front and back price. It shows that the CUSUM trend judgment method can filter the small wave motion through the control of the threshold, thus effectively reducing the transaction number.2 and in the income aspect. The CUSUM trend judgment method combined with the trend following trading strategy can gain higher returns. This transaction method is characterized by less trading during a smooth and subtle period; a short duration of operation in a period of large volatility, but the general operation results in a loss; it will continue in a sustained period of rise. With a long and long operation, there will be a generous profit.3 at this time. In terms of performance, 96% of the events in the stable phase can be processed within 10 microseconds. The average delay of the event is about 4.5 microseconds, and the increase in the number of rules within the range of the system does not affect the average processing time of the event. In addition, the system is running smoothly. The stage occupies less CPU resources, and the memory usage increases with the increase of running time, but eventually it can keep stable and no longer continue to increase.
【學(xué)位授予單位】：中南財(cái)經(jīng)政法大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2017
【分類號(hào)】：F832.51

【參考文獻(xiàn)】

相關(guān)期刊論文 前10條

1 姚海博;茹少峰;張文明;;基于動(dòng)態(tài)交易量預(yù)測的VWAP算法交易賣出策略[J];運(yùn)籌與管理;2015年02期

2 孟由;欒鐘治;謝明;錢德沛;;一種基于算子的可擴(kuò)展復(fù)雜事件處理模型[J];軟件學(xué)報(bào);2014年11期

3 燕汝貞;李平;曾勇;;一種面向高頻交易的算法交易策略[J];管理科學(xué)學(xué)報(bào);2014年03期

4 王宇超;李心丹;劉海飛;;算法交易的市場影響研究[J];管理科學(xué)學(xué)報(bào);2014年01期

5 曹科寧;王永恒;李仁發(fā);王鳳娟;;面向物聯(lián)網(wǎng)的分布式上下文敏感復(fù)雜事件處理方法[J];計(jì)算機(jī)研究與發(fā)展;2013年06期

6 程蘇s，

本文編號(hào)：1993854