金融實(shí)證方法講義.ppt

1、金融實(shí)證方法,上海財(cái)經(jīng)大學(xué)金融學(xué)院 韓其恒 ,參考資料,美本杰明. 格雷厄姆, 戴維. 多德 (2004), 證券分析, 海南出版社 菲利普.A.費(fèi)舍(1999)，怎樣選擇成長(zhǎng)股，海南出版社 約翰. Y. 坎貝爾(2003)，金融市場(chǎng)計(jì)量經(jīng)濟(jì)學(xué)，上海財(cái)經(jīng)大學(xué)出版社 美埃德加.E.彼得斯（1999），資本市場(chǎng)的混沌與秩序，經(jīng)濟(jì)科學(xué)出版社 John Murphy(1994)，Technical Analysis of the Future Markets，地震出版社 Aswath Damodaran(2004), Investment Fables,機(jī)械工業(yè)出版社 Michael E. Edles

2、on(2008), 價(jià)值平均策略(Value Averaging),上海財(cái)經(jīng)大學(xué)出版社 Michael W. Covel(2007), Trend Following-How great traders make millions in up or down markets, FT Press,參考資料,美Alan S. Farley(2007), 高明的趨勢(shì)波段交易師，機(jī)械工業(yè)出版社 John Parker Burg(1975), Maximum Entropy Spectral Analysis, Dissertation of Stanford University Doctor Deg

3、ree 美A.R.奧本海姆, R.W.謝佛, J.R.巴克（2001）， 離散時(shí)間信號(hào)處理，西安交通大學(xué)出版社 美羅伯特哈根(2002),新金融學(xué): 有效市場(chǎng)的反例 Fama, E.F. and K.R. French, 1992. “The cross-section of expected stock returns”, The Journal of Finance Fama, E.F and K.R. French, 1993. “Common Risk Factors in the Returns on Stocks and Bonds”, Journal of Financial E

4、conomics 上海財(cái)經(jīng)大學(xué)金融學(xué)院證券期貨系碩士學(xué)位論文,主要內(nèi)容,市場(chǎng)的有效性：一切金融理論、投資策略的基礎(chǔ) 弱式有效(SHCI)：隨機(jī)性檢驗(yàn)、價(jià)值平均法、趨勢(shì)跟蹤法、技術(shù)指標(biāo)、周期、非線性技術(shù)，選時(shí)(timing)。 半強(qiáng)式有效(A share)：事件研究法；財(cái)務(wù)指標(biāo)；因素模型；動(dòng)量交易策略；選股(stock selection),有效市場(chǎng)假說的類型,弱式有效市場(chǎng)：如果所有關(guān)于過去價(jià)格變化的信息都反映在現(xiàn)行股價(jià)上。 半強(qiáng)式有效市場(chǎng)：假定所有公開可得的信息反映在股票價(jià)格上。 強(qiáng)式有效市場(chǎng)：假定所有信息（尤其包括非公開信息）都反映在股價(jià)上。,有效市場(chǎng)的實(shí)證檢驗(yàn),一、我國(guó)股市的實(shí)證檢驗(yàn)結(jié)果：

5、自從全國(guó)性股票市場(chǎng)建立以來，對(duì)我國(guó)股票市場(chǎng)有效性的討論和檢驗(yàn)從未間斷過，遺憾的是，至今仍未能形成統(tǒng)一的令人信服的結(jié)論。,二、弱式有效檢驗(yàn),背景：弱式有效檢驗(yàn)考察過去價(jià)格的時(shí)間序列是否能用于預(yù)測(cè)未來的股價(jià)。,1、回歸分析,研究：Fama（1965），股票市場(chǎng)價(jià)格行為一文中對(duì)30支股票進(jìn)行了間隔一天的回歸分析。 結(jié)論：過去價(jià)格序列確實(shí)包含一些有關(guān)未來股價(jià)行為的信息，但基于過去數(shù)據(jù)的任何交易方式可能不具價(jià)值，即便最小的交易費(fèi)用也會(huì)淹沒超額報(bào)酬。,2、Autocorrelation Test,Ljung-Box的Q統(tǒng)計(jì)量：是通過計(jì)算序列自相關(guān)系數(shù)平方的加權(quán)平均來檢驗(yàn)序列是否獨(dú)立，是一種傳統(tǒng)直觀的方法。

6、Q統(tǒng)計(jì)量如下式所示：,其中rj是滯后為j的相關(guān)系數(shù)，T是樣本容量，p為滯后階數(shù)。其原假設(shè)為：序列獨(dú)立。,3、游程檢驗(yàn),檢驗(yàn)時(shí)間序列是否獨(dú)立。,游程檢驗(yàn),列出觀測(cè)值 計(jì)算觀測(cè)值中值 將觀測(cè)值中大于中值的記為+，小于中值的記為-，等于中值的記為+。 正號(hào)的個(gè)數(shù)記為n1，負(fù)號(hào)的個(gè)數(shù)記為n2。 按照觀測(cè)值的順序，將正負(fù)號(hào)的改變個(gè)數(shù)加1記為RANS。,當(dāng)n1或n2大于20時(shí)，以下統(tǒng)計(jì)量近似服從標(biāo)準(zhǔn)正態(tài)分布。,當(dāng)Z0值小于1.96時(shí)，在0.05置信水平上不能拒絕觀測(cè)值是獨(dú)立的這一原假設(shè)。,SHCI,順逆檢驗(yàn)(Sequences and reversals),Cowles和Jones，1937年，CJ統(tǒng)計(jì)量

7、,SHCI,鞅理論指出在公平游戲中，無論一個(gè)賭博者多么聰明，都不可能獲得超額盈利。,4、鞅過程的檢驗(yàn),定義：一個(gè)隨機(jī)變量序列Mn：0=n是關(guān)于隨機(jī)變量Xn：0=n的一個(gè)鞅，如果序列Mn滿足以下兩個(gè)條件：,廣義譜分析,SHCI weekly: 12/21/1990-10/16/2008,SHCI,價(jià)值平均策略(Value Averaging)獲得高投資收益的安全簡(jiǎn)便方法,Michael E. Edleson 上海財(cái)經(jīng)大學(xué)出版社，2008,發(fā)明于1988年 Michael E. Edleson, 1988. Value Averaging: A New Approach to Accumulati

8、on. AAII Journal X, no. 7 著名金融史學(xué)家威廉*伯恩斯坦將該書與格雷漢姆的智慧投資者、麥基爾的漫步華爾街相提并論，認(rèn)為：“自1991年首次出版以來價(jià)值平均策略聲譽(yù)鵲起，已逐步成為經(jīng)典作品?！?優(yōu)點(diǎn)：屬于被動(dòng)的公式化策略，簡(jiǎn)單，變化較少。 更加積極的動(dòng)態(tài)公式化策略：根據(jù)紅利、PE、短期利率等動(dòng)態(tài)調(diào)整不同資產(chǎn)類別的比例，如資產(chǎn)配置策略。,三種被動(dòng)投資策略的比較,CS：等量份額，每月購(gòu)買一定量的股份份額。 DCA：幣值成本平均，每月投資相同數(shù)目的金額。 VA：價(jià)值平均，是使投資者持有的股票股票價(jià)值每月固定增長(zhǎng)相同數(shù)目金額的美元。 比較基準(zhǔn)：內(nèi)部收益率（IRR） 養(yǎng)老基金等 M

9、omentum; Buy Contravian,幣值成本平均策略案例,價(jià)值平均策略案例,程序化交易(programing trading),據(jù)相關(guān)統(tǒng)計(jì)，2007年美國(guó)市場(chǎng)的金融交易中，有20%是通過程序化交易完成，2008年這一比例已經(jīng)上升至30%。在全球期貨期權(quán)市場(chǎng)投機(jī)交易中，程序化交易占據(jù)20%的比例，而在套利交易領(lǐng)域，這個(gè)比例高達(dá)80%。 程序化交易以技術(shù)分析為理論基礎(chǔ)。,程序化交易系統(tǒng)的特點(diǎn),順勢(shì)交易 純粹分析性：系統(tǒng)交易方法完全排除任何基本面分析的影響。 客觀性 數(shù)量化 機(jī)械化 資金管理制度化 價(jià)值交易與趨勢(shì)交易：左邊交易與右邊交易,應(yīng)注意的問題,交易的成功在于堅(jiān)持自己的交易系統(tǒng)。

10、交易系統(tǒng)有其高峰期和低谷期。交易系統(tǒng)從大類來分可分為趨勢(shì)型和振蕩型。 有沒有既能在振蕩市中賺錢又能在趨勢(shì)中獲利的交易系統(tǒng)？答案是否定的。兩種交易系統(tǒng)只能在相應(yīng)的市場(chǎng)中發(fā)揮作用。統(tǒng)計(jì)學(xué)表明，如果始終堅(jiān)持一種交易系統(tǒng)，就能成功獲利，但這一點(diǎn)大多數(shù)交易者未能做到，他們總是一會(huì)兒用趨勢(shì)型交易系統(tǒng)一會(huì)兒又用振蕩型交易系統(tǒng)，從而導(dǎo)致最終交易結(jié)果的虧損。 不同的交易系統(tǒng)有不同的風(fēng)險(xiǎn)-收益特性。,結(jié)論,總體來看，程序化交易系統(tǒng)的原則是評(píng)估而非預(yù)測(cè)，基本是被動(dòng)等待出現(xiàn)買賣信號(hào)后，再由程序化系統(tǒng)自動(dòng)判斷是否入場(chǎng)或離場(chǎng)，這樣的系統(tǒng)追求的是較低但穩(wěn)定的收益，要做到很高回報(bào)幾乎是不可能的，因?yàn)楦呤找姘殡S的必然是高風(fēng)險(xiǎn)，

11、人力做不到的由人設(shè)計(jì)的系統(tǒng)也不可能做到?？梢赃@樣看：一個(gè)好的程序化交易系統(tǒng)就是一部真正意義上的賺錢機(jī)器。,基于RAT指標(biāo)的證券市場(chǎng)有效性研究,上海財(cái)經(jīng)大學(xué)金融學(xué)院證券期貨系2007級(jí)碩士研究生,RAT,RAT（reverse arrangement test,逆向安排檢驗(yàn)）是統(tǒng)計(jì)學(xué)上用于檢驗(yàn)?zāi)骋粫r(shí)間序列數(shù)據(jù)是否具有趨勢(shì)性的指標(biāo)。 RAT的概念和計(jì)算方法由M.G. Kendall 于1938年在A New Measure of Rank Correlation中首次提出。,RAT指標(biāo)的計(jì)算方法,RAT指標(biāo)具有一個(gè)參數(shù)n，是指數(shù)列的數(shù)字個(gè)數(shù)，可以代表交易日的天數(shù)。,置信區(qū)間的臨界值,交易策略設(shè)定,

12、短期趨勢(shì)預(yù)測(cè)：RAT(10)；長(zhǎng)期趨勢(shì)預(yù)測(cè)：RAT(50)；也可以將短期和長(zhǎng)期的趨勢(shì)值結(jié)合起來使用。 本文的交易策略的設(shè)定為：如果50天的RAT值大于或等于0.19，或者10天的RAT值大于或等于0.467，則買入市場(chǎng)組合；如果50天的RAT值小于或等于-0.192，或者10天的RAT值小于或等于-0.511，則賣出市場(chǎng)組合。 本文的移動(dòng)平均線交易策略的設(shè)定為：如果10日均線向上穿過50日均線0.01時(shí)，買入市場(chǎng)組合；如果10日均線向下穿過50日均線0.01時(shí)，賣出市場(chǎng)組合。 初始日持有現(xiàn)金1，000，000元。 將每次交易的手續(xù)費(fèi)考慮在內(nèi)。,數(shù)據(jù)選取,上證綜指、深成指、上證B股指數(shù)和納斯達(dá)克

13、指數(shù)從1998年1月1日至2009年4月10日。,上證指數(shù)累積資本序列,累積資本比的對(duì)數(shù)值,q1=ln（RAT策略累積資本/基準(zhǔn)策略累積資本） q2=ln（RAT策略累積資本/ MA策略累積資本）,深成指累積資本序列,上海B股指數(shù)的檢驗(yàn),納斯達(dá)克指數(shù)的檢驗(yàn),結(jié)論,上證A股市場(chǎng)，上證B股市場(chǎng)，深圳A股市場(chǎng)和納斯達(dá)克市場(chǎng)均已經(jīng)具有了一定的效率，市場(chǎng)價(jià)格已經(jīng)能夠反映出最近50天歷史價(jià)格中包含趨勢(shì)的信息。其中，上海B股市場(chǎng)的效率從98年到09年有顯著增強(qiáng)的過程。至于市場(chǎng)是否達(dá)到弱勢(shì)有效，還不能夠根據(jù)這一實(shí)證結(jié)果絕對(duì)地推出。由于本文只使用了50階以內(nèi)的RAT策略和MA策略做模擬投資，所以只否定了使用這兩

14、種方法獲得超額收益的可能性，但并不排除使用其他可能的策略獲得超額收益的可能性。,Trend FollowingHow great traders make millions in up or down markets,Michael W. Covel, FT Press, 2007 ,The father of Trend Following,Richard Donchian did not start his Trend Following fund until age 65. He traded into his 90s. He is known as the father of Tren

15、d Following.,一個(gè)完整的交易系統(tǒng),市場(chǎng)-買賣什么 頭寸規(guī)模-買賣多少 入市-何時(shí)買賣 止損-何時(shí)退出虧損的頭寸 離市-何時(shí)退出贏利的頭寸 策略-如何買賣,海龜交易什么,芝加哥期貨交易所（CBOT）： 30年期美國(guó)長(zhǎng)期國(guó)債（Treasury Bond）；10年期美國(guó)中期國(guó)庫(kù)券（Treasury Note）；紐約咖啡可可與原糖交易所（NYCSC）；咖啡； 可可； 原糖； 棉花 芝加哥商品交易所（CME）： 瑞士法郎；德國(guó)馬克；英鎊；法國(guó)法郎； 日?qǐng)A；加拿大元；標(biāo)準(zhǔn)普爾500股票指數(shù)；歐洲美元；90天美國(guó)短期國(guó)庫(kù)券（Treasury Bill） 紐約商品期貨交易所（COMEX）：黃金；白

16、銀；銅 紐約商業(yè)期貨交易所（NYMEX）：原油；燃油；無鉛汽油,波動(dòng)性,N就是TR（True Range，實(shí)際范圍）的20日指數(shù)移動(dòng)平均，現(xiàn)在更普遍地稱之為ATR。,頭寸規(guī)模與資金管理,頭寸規(guī)模算法的原理：如果一個(gè)市場(chǎng)的合約價(jià)值波動(dòng)性較強(qiáng)，那么這個(gè)市場(chǎng)中的合約持有量就少一些；相反，如果一個(gè)市場(chǎng)的合約價(jià)值波動(dòng)性較弱，那么這個(gè)市場(chǎng)中的合約持有量就可以大一些。 絕對(duì)波動(dòng)幅度=N每一點(diǎn)數(shù)所代表的美元 頭寸規(guī)模單位=帳戶的1%/市場(chǎng)的絕對(duì)波動(dòng)幅度 目的是讓一個(gè)N相當(dāng)于帳戶凈值的1%。 請(qǐng)注意，如果你沒有足夠的資金，這種分散化是很難實(shí)現(xiàn)的。 逐步建倉(cāng) 止損,Donchians channel breako

17、ut system (通道突破系統(tǒng)),The higher band: the highest high of the past n days. The lower band: the lowest low of the past n days. Assumption: a crossover of these psychologically important lines is the result of a change in the markets opinion, and thus a continuation of the initial movement could possibl

18、y expected.,入市,系統(tǒng)一-以20日突破為基礎(chǔ)的偏短線系統(tǒng)；系統(tǒng)二-以50日突破為基礎(chǔ)的較簡(jiǎn)單的長(zhǎng)線系統(tǒng) 我們完全可以按照自己的意愿自行決定將凈值配置在何種系統(tǒng)上。我們中的一些人選用系統(tǒng)二交易所有的凈值，一些人分別用凈值的50%選擇系統(tǒng)一，50%選擇系統(tǒng)二，而其他人則選擇了不同的組合。,退出,系統(tǒng)1采取10日突破退出法則：對(duì)多頭頭寸來說，在價(jià)格跌破過去10日最低點(diǎn)時(shí)退出；對(duì)空頭頭寸來說，在價(jià)格超過過去10日最高點(diǎn)時(shí)退出。 系統(tǒng)2采取20日突破退出法則：對(duì)多頭頭寸來說，在價(jià)格跌破過去20日最低點(diǎn)時(shí)退出；對(duì)空頭頭寸來說，在價(jià)格超過過去20日最高點(diǎn)時(shí)退出。,1998 Cocoa Break

19、 out Trades,Donchian Trend Results: January 1996 to May 2006 (average compounded rate of 43.7 percent),Way of the Turtle海龜交易法則,美柯蒂斯.費(fèi)思 2007，中信出版社,Markets,The markets in the testing portfolio include the Australian dollar, the British pound, corn, cocoa, the Canadian dollar, crude oil, cotton, the eu

20、ro, the eurodollar, feeder cattle, gold, copper, heating oil, unleaded gas, the Japanese yen, coffee, cattle, hogs, the Mexican peso, natural gas, soybeans, sugar, the Swiss franc, silver, Treasury notes, Treasury bonds, and wheat. These markets were selected from the liquid (high-tradingvolume) U.S

21、. markets.,Money Management,We employed a figure that was half as aggressive. Instead of equating 1 ATR to 1 percent of our trading equity, we equated it to 0.5 percent.,Test Dates,The testing was performed using data from January 1996 through June 2006 for all the systems.,The Systems,ATR Channel B

22、reakout The higher band: a 350-day moving average+7 ATR The lower band: a 350-day moving average-3 ATR A long trade is entered on the open if the previous days close exceeded the top of the channel; a short trade is entered if the previous days close fell below the bottom of the channel. Trades are

23、exited when the close crosses back through the moving average price.,Bollinger Breakout,The higher band: a 350-day moving average+ 2.5 sdv The lower band: a 350-day moving average- 2.5 sdv Trades are exited when the close crosses back through the moving average price.,Donchian Trend,It uses a 20-day

24、 breakout for entry and a 10-day breakout for exits. Trend filter: 350-day/25-day. If the 25-day moving average is above the 350-day average, only longs may be taken; if the 25-day moving average is below the 350-day average, only shorts may be taken. Stoploss: a 2-ATR stop just as the original Turt

25、le system did.,Donchian Trend with Time Exit,Donchian Trend with Time Exit system, uses a time-based exit instead of a breakout exit. It exits the trade after 80 days and does not use any stops whatsoever.,Dual Moving Average,Slower MA: 350-day Fast MA: 100-day Unlike the other systems, this system

26、is always in the market, either long or short.,Triple Moving Average,Fast MA: 150-day Slower MA: 250-day Trend filter: 350-day 如果兩者都高于350日均線，只能做多，如果兩者都低于350日均線，只能做空。 Unlike the Dual Moving Average system, this system is not always in the market.,Historical System Performance Comparison,Note: Channel

27、 Breakout is abbreviated as CBO, Drawdown as DD, the MAR ratio divides the annual return by the largest drawdown, and CAGR the compound annual growth rate. Keep it simple: Simple systems hold up better over time than do more complex ones.,濾波法則（Filter rules）,背景：根據(jù)過去價(jià)格序列可以設(shè)計(jì)出無窮多個(gè)交易法則。其中流行的最著名的一類，就是濾波法

28、則。Alexander (1961, 1964) pioneered the technique of the filter rule. 百分之y濾波：如果價(jià)格至少上升y%，那么買入證券并持有它，直至其價(jià)格從前一高位至少下跌y%，這時(shí)賣出股票并作空頭；保持賣空頭寸直至價(jià)格從前一低位上漲y%，這時(shí)補(bǔ)倉(cāng)并買進(jìn)股票。如果股價(jià)上漲或下跌低于y%，那么就不作交易。,Filter Rules and Stock-Market Trading,Eugene F. Fama, Marshall E. Blume The Journal of Business , 1966,Comparison of Rate

29、s of Return, Before Comminsions,The significance of technical trading-rule profits in the foreign exchange market: a bootstrap approach,RICHARD M. LEVICH Journal of International Money and Finance, 1993,Abstract,In this paper, we present new evidence on the profitability and statistical significance

30、 of technical trading rules in the foreign exchange market. Our results suggest that simple technical trading rules have very often led to profits that are highly unusual.,Data,Futures prices for five currencies: British pound (BP), Canadian dollar (CD), German mark (DM), Japanese yen (JY) and Swiss

31、 franc (SF) January 1, 1976 to December 31, 1990, approximately 3800 daily observations. A single time series is assembled by bringing together quotations on successive near-term contracts. Samuelson (1976) has proved that near futures contracts show more variability than (sufficiently far) distant

32、ones,Our analysis of futures price changes reveals that there is no significant difference between volatility for far maturities (80 T 110) and near maturities (20 T 50).,Return,Under the hypothesis of no currency risk premium, TR should not be significantly different from zero.,Sample statistics of

33、 daily returns,Autocorrelation functions of daily returns,Simulation procedure,bootstrap (解靴帶; 自舉；自助法) approach, which assumes nothing about the distribution generating function.,Profitability of filter rules, per cent per annum,moving average rules, per cent per annum,Transaction costs,The likely c

34、ost of transacting in the currency futures market is about 2.5 basis points (0.025 per cent) per transaction for a large institution. A more conservative estimate would be roughly 4.0 basis points. At 65 trades per year, a speculator would have his trading profits reduced by 1.62 per cent per year o

35、r 2.60 per cent per year if we take our more conservative measure. For the most trading rules we consider, the volume of trading is considerably smaller, and transaction costs do not significantly affect profits.,Statistics on the 10000 simulated samples,Stability test,To measure the stability of th

36、ese results over time, we split the sample period into three, five-year sub-periods and repeated our analysis: 1976 -1980; 1981-1985; 1986 -1990 On average, there is some deterioration over time in the profitability of these rules, but the overall decline is small.,Conclusion,These results strongly

37、suggest that the actual exchange rate series contained significant departures from serial independence that allowed technical trading rules to be profitable., but the nature of that dependency remains unclear.,Simple technical trading rules and the stochastic properties of stock returns,William Broc

38、k, Josef Lakonishock, and Blake Lebaron The Journal of Finance, 1992,Data,Dow Jones Industrial Average (DJIA) Full sample: 1/1/1897-12/31/1986 Subperiod: 1/1/1897-7/30/1914 (closing of the stock exchange during the World War One) 1/1/1915-12/31/1938 (include turbulent times of depression) 1/1/1939-6

39、/30/1962 (include the World War Two and ends in the data CRSP begins its daily price series) 7/1/1962-12/31/1986 (extensively researched),Summary statistics for daily and 10-day returns,Technical trading rules,Two of the simplest and most widely used technical rules are investigated: moving average

40、and trading range break-out (resistance and support levels). Buy (sell): short-period moving average rises above (or falls below) the long-period moving average rises.,VMA(Variable-Length Moving average)可變長(zhǎng)度移動(dòng)平均規(guī)則,當(dāng)S-天移動(dòng)平均線的價(jià)格超過（低于）L-天移動(dòng)平均線的價(jià)格至少B%時(shí)。 1-50，1-150，1-200，5-150和5-200；以0和1作為預(yù)先指定的百分比。,Stand

41、ard test for VMA,FMA(Fixed-Length Moving average)固定長(zhǎng)度移動(dòng)平均規(guī)則,當(dāng)短期移動(dòng)平均線從下面（上面）超過（低于）長(zhǎng)期移動(dòng)平均線時(shí)。當(dāng)信號(hào)發(fā)出后，F(xiàn)MA規(guī)則發(fā)出的信號(hào)持續(xù)一個(gè)固定天數(shù)。 1-50，1-150，1-200，5-150，和2-200；以10天作為預(yù)先指定的固定天數(shù)。,Standard test for FMA,TRB(trading range break-out,交易區(qū)間突破,支撐阻力分析方法),A buy signal is generated when the price penetrates the resistance le

42、vel. The resistance level is defined as the local maximum. A sell signal is generated when the price penetrates the support level which is the local minimum price. In accordance with the moving average strategy, Maximum (or minimum) prices were determined based on the past 50, 150, and 200 days. In

43、addition, the rule is implemented with and without a one percent band.,Standard test for TRB,Bootstrap Methodology,A random walk with a drift,Simulation tests for 500 replications,Bootstrap Methodology,AR(1),Bootstrap Methodology,GARCH-M: Garch in mean model,Bootstrap Methodology,EGARCH-M: exponeati

44、al Garch,Conclusion,The recent studies on predictibility of equity returns from past returns suggest that the conclusion reached by many earlier studies that found technical analysis is useless might have been premature. Overall our results provide strong support for the technical strategies that we

45、 explored. The returns obtained from buy (sell) signals are not likely to be generated by the four popular null models (random walk with a drift, AR(1), Garch-M, and E-Garch).,二、半強(qiáng)式有效檢驗(yàn),背景：弱式有效檢驗(yàn)僅注重股票過去價(jià)格的信息，半強(qiáng)式有效檢驗(yàn)涉及所有公開所得信息，當(dāng)然包括股票價(jià)格；如果市場(chǎng)是半強(qiáng)式有效，那么利好消息已經(jīng)反映在股價(jià)上，在披露信息后，無超額報(bào)酬可掙。,事件研究法,背景：半強(qiáng)式有效市場(chǎng)的檢驗(yàn)可以采用

46、事件研究法，事件如公司配股；拆股信息的頒布；會(huì)計(jì)領(lǐng)域中的事件盈利分紅信息的頒布；送轉(zhuǎn)股；基金經(jīng)理的變更；融資決策對(duì)標(biāo)的股票價(jià)格(或企業(yè)價(jià)值) 的影響；公司控制權(quán)交易；會(huì)計(jì)領(lǐng)域中的事件等，這些信息被視為對(duì)股價(jià)具有主要影響。,事件研究各時(shí)間窗,市場(chǎng)模型（Market Model）,市場(chǎng)指數(shù)調(diào)整模型Market-adjusted return Model,Brown 和Warner在1980年提出的，他假定股票在某個(gè)時(shí)期的期望收益為市場(chǎng)收益率。,經(jīng)濟(jì)模型,較典型的經(jīng)濟(jì)模型包括資本資產(chǎn)定價(jià)模型(CMPA) 與套利定價(jià)模型(APT) 。 Fama 與French (1993) 、Carhart(1997

47、) 基于套利定價(jià)模型思想提出的三因素模型四因素模型。,一種證券的事件窗,市場(chǎng)模型（估計(jì)窗口：T0-T1）,非正常收益（AR）（檢驗(yàn)窗口：T2-T3）,當(dāng)估計(jì)窗口的長(zhǎng)度L1很大時(shí)，非正常收益率AR之間的相關(guān)性將消失，AR將相互獨(dú)立。,累積非正常收益（CAR）（檢驗(yàn)窗口：T2=23 =T3）,當(dāng)估計(jì)窗口的長(zhǎng)度L1很大（如L130）時(shí),N種證券的事件窗,假設(shè)：N個(gè)證券的事件窗在公布日不發(fā)生重疊，協(xié)方差為零。,平均非正常收益（檢驗(yàn)窗口：T2-T3）,當(dāng)估計(jì)窗口的長(zhǎng)度L1很大時(shí),累積非正常收益（CAR）（檢驗(yàn)窗口：T2=23 =T3）,對(duì)美國(guó)股市的實(shí)證研究,股份分割：理論界與務(wù)實(shí)界對(duì)股票分割的經(jīng)濟(jì)價(jià)值及

48、其對(duì)股價(jià)之影響尚未達(dá)成一致見解。大多數(shù)理論家認(rèn)為，股票分割沒有任何經(jīng)濟(jì)價(jià)值。,The Adjustment of Stock Prices to New Information,Eugene Fama; Lawrence Fisher; Michael C. Jensen; Richard Roll International Economic Review, 1969,法馬、費(fèi)雪、Jinsen和Roll(FFJR)（1969年2月）檢驗(yàn)了股份分割對(duì)股票超額報(bào)酬的影響。對(duì)每種股票作月回歸：,其中是超額報(bào)酬，然后計(jì)算了股份分割前30個(gè)月、后30個(gè)月的累積超額收益。,一般來說，只有贏利公司當(dāng)其股價(jià)

49、達(dá)到很高價(jià)位時(shí)才分割股份，因此，分割前30個(gè)月股價(jià)上升是對(duì)收入增加的反映。這樣，股份分割沒有經(jīng)濟(jì)價(jià)值，但可視為未來收入的一種信號(hào)。 若投資者一旦得到股份分割信息便買進(jìn)股票，發(fā)現(xiàn)不能掙到超額報(bào)酬。這一結(jié)果印證了半強(qiáng)式市場(chǎng)有效性。,結(jié)論,THE PERFORMANCE OF STOCKS: PROFESSIONAL VERSUS DARTBOARD PICKS,Youguo Liang, Sanjay Ramchander and Jandhyala L. Sharma Journal Of Financial And Strategic Decisions, Spring 1995,DATA,I

50、nvestment Dartboard Column (ID) was created in October 1988 by John R. Dorfman and is henceforth being published in the first half of every month in the Wall Street Journal. , In this game a quartet of well-regarded investment experts pick their favorite stock to either buy or sell, while four amate

51、urs hurl darts at a list of New York Stock Exchange (NYSE) or Over-the- Counter (OTC) stocks and record the names of the stocks pierced by pure chance. , The pros whose selections do best are invited back in the next month for another round of a similar game. We examined this column on a monthly bas

52、is from October 1988 through June 1991,., A total of 120 recommendations were made by the pros during the sample period, of which 114 were buy recommendations.,METHODOLOGY,Event day 0: the date of publication in the WSJ. Estimation window: -375, -126 250 trading days Market model Rit = ai + biRmt +

53、eit Checking window: 5, 21, 42, 84 and 125 trading days. Moreover, as Pound and Zeckhauser posit, investment professionals have long maintained that their strategies are not supposed to “outsmart” the market over the 30-40 day period typically employed in the event study literature, but rather on a

54、longer-term approach. The prediction errors: PEit= Rit-(ai + biRmt),*Significant at the 0.05 level. *Significant at the 0.01 level.,Performance Of Pros Recommendation And Dart Picks Over Short Intervals,Performance Of Pros Recommendation And Dart Picks Over Longer Intervals,Conclusion: This leads us

55、 to believe that the publicity effect, is potent only in the short-term which then lends support to a moral hazard problem encountered by investment professionals. In other words, the effect of the recommendation in the long- run is transitory.,Special Information and Insider Trading,Jaffe, J., Jour

56、nal of Business, 1974,Cummulative returns following insider trading,上面的曲線表示內(nèi)部人士?jī)糍I入最多的股票，下面的曲線表示內(nèi)部人士?jī)糍u出最多的股票，顯然，內(nèi)部人士?jī)糍I入的股票具有較高的收益。,Conclusion,總體而言，跟隨內(nèi)部人士的交易是有價(jià)值的，但是這個(gè)價(jià)值不是特別明顯。 策略：關(guān)注小公司高層的股票交易，但不要對(duì)此持抱有過大的期望，因?yàn)檫@只是公共信息。,推薦游戲,分析家推薦 買入，持有，賣出 強(qiáng)買，弱買 盈余預(yù)測(cè)中的羊群效應(yīng)：賣方分析師的羊群行為,一般的，買入評(píng)級(jí)的股票要大大超過賣出評(píng)級(jí)，在2001年，7：1，在199

57、0年，25：1,一開始，投資者對(duì)買入評(píng)級(jí)有反應(yīng)，買入評(píng)級(jí)的影響在6個(gè)月后基本消失。 賣出評(píng)級(jí)對(duì)股價(jià)的負(fù)面影響大大超過買入評(píng)級(jí)的正面影響。,Conflicts of Interests and the Credibility of Underwriter AnalystsRecommendation,Michaely, R. and K.L. Womack, Review of Financial Studies, 1999,很多分析師同樣是市場(chǎng)的參與者，要注意區(qū)分。,Conclusion,We conclude that the recommendations by underwriters

58、analysts show significant evidence of bias. We show also that the market does not recognize the full extent of this bias. The results suggest a potential conflict of interests inherent in the different functions that investment bankers perform.,中國(guó)證券分析師推薦價(jià)值研究,上海財(cái)經(jīng)大學(xué)金融學(xué)院碩士論文,數(shù)據(jù),2005年5月31日起至2007年3月31日止

59、wind資訊系統(tǒng)記錄的全部股票推薦，期間，收錄了來自32家研究機(jī)構(gòu)653名分析師共計(jì)4567個(gè)推薦評(píng)級(jí)樣本，涉及1035家上市公司，其中滬市617家，深市418家。 有效樣本2922個(gè)：剔除，如次新股、ST股票、重復(fù)推薦等。,事件研究法,分別研究分析師推薦的短期效應(yīng)和長(zhǎng)期投資價(jià)值。 估計(jì)窗口：推薦日前67天至前176天共110個(gè)交易日。 短期檢驗(yàn)窗口：推薦日及前后5天共計(jì)11天作為事件期。 長(zhǎng)期檢驗(yàn)窗口：推薦日起后推最長(zhǎng)6個(gè)月作為事件期。,全部樣本推薦日起六個(gè)月內(nèi)的ACAR(平均累積異常收益),平均而言，6個(gè)月能夠獲得超過大盤將近2%的超額收益，整體而言，分析師具有一定的選股擇時(shí)能力，其推薦具有一定的投資價(jià)值。,大盤股、中盤股和小盤股自推薦日起六個(gè)月內(nèi)的ACAR,平均來看，中等市值股票六個(gè)月能獲得相對(duì)大盤約7%的超額收益。分析師推薦的小盤股長(zhǎng)期的平均異常收益顯著為負(fù)，即長(zhǎng)期來看其表現(xiàn)不如市場(chǎng)指數(shù)。,明星分析師和非明星分析師,自2003年開始，我國(guó)著名財(cái)經(jīng)雜志新財(cái)富借鑒國(guó)際慣例，每年由機(jī)構(gòu)投資者投票評(píng)選出當(dāng)年各個(gè)行業(yè)的“最佳分析師“。當(dāng)選“最佳分析師”，意味著該分析師推薦的價(jià)值得到了買方機(jī)構(gòu)投資者的高度認(rèn)可，同時(shí)，也會(huì)給當(dāng)選分析師帶來薪酬和行業(yè)地位的大幅提高，目前該頭銜已