<P ><FONT size=3>我正在对股票收市价格进行随机游走检测,下面是我的测试步骤,不知道正确与否,同时其中有很多不明白的地方,希望各位大侠能解答一二,毕竟国内的对时间序列问题似乎研究得不太多,谢谢了!</FONT></P>
<P ><FONT face="Times New Roman"><FONT size=3> <p></p></FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">1</FONT>平稳测试<FONT face="Times New Roman">(</FONT>对收市价<FONT face="Times New Roman">Closing Price</FONT>而言<FONT face="Times New Roman">)</FONT></FONT></P>
<P ><FONT size=3>Δy<SUB>t</SUB>=βy<SUB>t-1</SUB>ε<SUB><FONT face="Times New Roman">t</FONT></SUB></FONT><FONT face="Times New Roman" size=3> </FONT><v:shapetype><v:stroke joinstyle="miter"></v:stroke><v:formulas><v:f eqn="if lineDrawn pixelLineWidth 0"></v:f><v:f eqn="sum @0 1 0"></v:f><v:f eqn="sum 0 0 @1"></v:f><v:f eqn="prod @2 1 2"></v:f><v:f eqn="prod @3 21600 pixelWidth"></v:f><v:f eqn="prod @3 21600 pixelHeight"></v:f><v:f eqn="sum @0 0 1"></v:f><v:f eqn="prod @6 1 2"></v:f><v:f eqn="prod @7 21600 pixelWidth"></v:f><v:f eqn="sum @8 21600 0"></v:f><v:f eqn="prod @7 21600 pixelHeight"></v:f><v:f eqn="sum @10 21600 0"></v:f></v:formulas><v:path connecttype="rect" gradientshapeok="t" extrusionok="f"></v:path><lock aspectratio="t" v:ext="edit"></lock></v:shapetype><v:shape><v:imagedata><FONT face="Times New Roman" size=3></FONT></v:imagedata></v:shape><FONT size=3>Δ<FONT face="Times New Roman">y<SUB>t</SUB>=lnp</FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">H<SUB>0</SUB>: </FONT>β<FONT face="Times New Roman">=0 </FONT>不平稳<FONT face="Times New Roman"> H<SUB>1</SUB></FONT>:β&lt;<FONT face="Times New Roman">0 </FONT>平稳</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">TS</FONT>=β<FONT face="Times New Roman">/St(</FONT>β<FONT face="Times New Roman">) Logarithms----transform a multiplicative form into an additive</FONT></FONT></P>
<P ><FONT face="Times New Roman"><FONT size=3>1)</FONT> <FONT size=3>PACF</FONT></FONT><FONT size=3>与<FONT face="Times New Roman">ACF(</FONT>图像上看<FONT face="Times New Roman">)</FONT></FONT></P>
<P ><FONT face="Times New Roman" size=3>PACF for AR(1)</FONT></P>
<P ><FONT size=3>在图象上看,滞后一期处有显著性,在其他期处于置信区间,此为序列不稳定的特征之一</FONT></P>
<P ><FONT face="Times New Roman"><FONT size=3>2)</FONT> </FONT><FONT size=3>正式的稳定性测试<FONT face="Times New Roman">(</FONT>用数学方法<FONT face="Times New Roman">)</FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">a Dickey-Fuller </FONT>测试(滞后一期),采用<FONT face="Times New Roman">5</FONT>%置信区间</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">Z(t)=-2.111(</FONT>落受接收域,存在单位根<FONT face="Times New Roman">unit-root, </FONT>因此为不稳定<FONT face="Times New Roman"> )</FONT></FONT></P>
<P ><FONT size=3>用自相关一次,Δy<SUB>t</SUB>=αy<SUB>t-1</SUB>+C有(<FONT face="Times New Roman">H<SUB>0</SUB></FONT>:α=<FONT face="Times New Roman">0</FONT>不稳定<FONT face="Times New Roman"> H<SUB>1</SUB>:</FONT>α&lt;<FONT face="Times New Roman">0</FONT>稳定)</FONT></P>
<P ><FONT size=3>一般而言,此处得出是不稳定的。<FONT face="Times New Roman"> </FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">b </FONT>对Δy<SUB>t</SUB>=αy<SUB>t-1</SUB>+C的残差进行检验</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">c </FONT>对Δy<SUB>t</SUB>=αy<SUB>t-1</SUB>+C检验,发现残差自相关(<FONT face="Times New Roman">LM</FONT>)</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">d </FONT>发现显著性不明显,因此滞后<FONT face="Times New Roman">5</FONT>期(此处不明白,我参考的文章只是简单表述)</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman"> </FONT>用<FONT face="Times New Roman">LM</FONT>检测残差,<FONT face="Times New Roman">nR<SUP>2 </SUP>&gt;X<SUP>2 </SUP> </FONT>分布</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">H0</FONT>:所有系数为零<FONT face="Times New Roman"> </FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">H1</FONT>:至少有一个系数不为零</FONT></P>
<P ><FONT size=3>ε<SUB>t</SUB>=α+β<SUB>1</SUB>ε<SUB>t-1</SUB>+β<SUB>2</SUB>ε<SUB>t-2</SUB>+---+β<SUB>n</SUB>ε<SUB>t-n</SUB>+C</FONT></P>
<P ><FONT size=3>(这里不明白,对残差到底要滞后多少期<FONT face="Times New Roman">?</FONT>)<p></p></FONT></P>
<P ><FONT size=3>如果<FONT face="Times New Roman">H1</FONT>成立,残差为自相关,为单位根,与<FONT face="Times New Roman">Dickey</FONT>-<FONT face="Times New Roman">Fuller</FONT>测试一致</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">e </FONT>用增加滞后变量的办法来清除序列自相关(为什么我们要用多加几次滞后的方法来清除序列残值的自相关<FONT face="Times New Roman">?</FONT>)<p></p></FONT></P>
<P ><FONT size=3>我参考的文章原文是增加了<FONT face="Times New Roman">5</FONT>期这里<FONT face="Times New Roman">(</FONT>为什么要加五期滞后来消除残差的自相关<FONT face="Times New Roman">?</FONT>到底要增加多少期,有什么标准和规则吗?<FONT face="Times New Roman">)<p></p></FONT></FONT></P>
<P ><FONT size=3>y<SUB>t</SUB>=αy<SUB>t-1</SUB>+β<SUB>1</SUB>Δy<SUB>t-1</SUB>+β<SUB>2</SUB>Δy<SUB>t-2</SUB>+β<SUB>3</SUB>Δy<SUB>t-3</SUB>+β<SUB>4</SUB>Δy<SUB>t-4</SUB>+β<SUB>5</SUB>Δy<SUB>t-5</SUB>---</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">f Augmented Dickey-Fuller </FONT>测试</FONT></P>
<P ><FONT size=3>检测<FONT face="Times New Roman">5</FONT>期的自相关</FONT></P>
<P ><FONT size=3>如果<FONT face="Times New Roman">H0</FONT>成立,则存在单位根</FONT></P>
<P ><FONT size=3>然后进行<FONT face="Times New Roman">LM</FONT>检测,发现不相关了,随机游走成立</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">(</FONT>到这里以后,我们应该就结束了,现在已经证明了是随机过程了,为何下面对回报<FONT face="Times New Roman">return,</FONT>考察以后,发现<FONT face="Times New Roman">return</FONT>不是随机的,那不是前后矛盾了<FONT face="Times New Roman">?)<p></p></FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman"> <p></p></FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman"> <p></p></FONT></FONT></P>
<P ><FONT face="Times New Roman"><FONT size=3>3)</FONT> </FONT><FONT size=3>差分检测,如果有单位根,对其进行差,使其稳定</FONT></P>
<P ><FONT size=3>然后对每期回报(<FONT face="Times New Roman">return</FONT>)进行检测:<FONT face="Times New Roman"> </FONT></FONT></P>
<P ><FONT face="Times New Roman"><FONT size=3> R<SUB>t</SUB>=(y<SUB>t</SUB>-y<SUB>t-1</SUB>)/y<SUB>t</SUB>=lny<SUB>t</SUB>-lny<SUB>t-1</SUB></FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">a </FONT>对<FONT face="Times New Roman">R<SUB>t </SUB></FONT>进行<FONT face="Times New Roman">Dickey-Fuller</FONT>检测,如果发现其落入拒绝域,则稳定</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">b PACF </FONT>观测残差(<FONT face="Times New Roman">error</FONT>),然后决定滞后多少期</FONT></P>
<P ><FONT size=3>Δ<FONT face="Times New Roman">y<SUB>t</SUB>=</FONT>αy<SUB>t-1</SUB>+ε</FONT></P>
<P ><FONT size=3>通过图形,看看第几期显著,然后对其进行<FONT face="Times New Roman">LM</FONT>检测,发现<FONT face="Times New Roman">H1</FONT>成立,即至少有一个系数比为了<FONT face="Times New Roman">0</FONT>,因此存在自相关。</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">C </FONT>用<FONT face="Times New Roman">Augmented Dickey</FONT>-<FONT face="Times New Roman">Fuller</FONT>消除自相关</FONT></P>
<P ><FONT size=3>此时发现已经不为单位根了,即已经平稳了,不必继续差分</FONT></P>
<P ><FONT size=3>接着进行<FONT face="Times New Roman">LM</FONT>检测,也为不相关,平稳,不存在单位根</FONT></P>
<P ><FONT face="Times New Roman"><FONT size=3> <p></p></FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">4</FONT>)到这里,我们发现<FONT face="Times New Roman">return</FONT>是稳定的,从而建模</FONT></P>
<P ><FONT size=3>建模:<FONT face="Times New Roman">ARIMA </FONT>(分别滞后<FONT face="Times New Roman">1 2 3 4 5</FONT>期)<FONT face="Times New Roman">(</FONT>我建立自相关模型到底要滞后多少期,用什么标准确定要滞后多少期<FONT face="Times New Roman">?)<p></p></FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">a </FONT>零均值化</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">b </FONT>标准分布</FONT></P>
<P ><FONT size=3><FONT face="Times New Roman">(</FONT>为何要用正态分布来确定是否稳定,如何模拟图像来证明有预测性<FONT face="Times New Roman">?)<p></p></FONT></FONT></P>
<P ><FONT size=3><FONT face="Times New Roman"> <p></p></FONT></FONT></P>(如何用AIC BIC考察定阶,具体如何使用AIC BIC?)