Estimation in Financial Models - RiskLab
Estimation in Financial Models - RiskLab
Estimation in Financial Models - RiskLab
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
we assume that it follows an AR(1) E-ARCH process:<br />
h i<br />
ln [Y kh ] = ln Y (k,1)h + hh kh 2 + kh " kh ;<br />
i h h i<br />
= ln khi 2 + , ln <br />
2<br />
kh h + C12 " kh<br />
ln h 2 (k+1)h<br />
+ h j" kh j,(2h=) 1=2i ; (2.12)<br />
where [(C 22 , C 2 12)=(1 , 2=)] 1=2 and " kh iid, N(0;h).<br />
If ln( 2 0) is normally distributed, then the ln( 2 t ) process <strong>in</strong> (2.11) is Gaussian.<br />
Otherwise, if > 0, a Gaussian stationary limit distribution for ln( 2 t ) exists.<br />
Thus, the conditional variance <strong>in</strong> cont<strong>in</strong>uous time is lognormal, and as <strong>in</strong><br />
the GARCH(1,1) example above, from this <strong>in</strong>formation Nelson [54] <strong>in</strong>fers<br />
the distribution of the <strong>in</strong>novation process <strong>in</strong> the discrete time model (2.12).<br />
He shows that <strong>in</strong> the discrete time model (with short time <strong>in</strong>tervals) the<br />
distribution of the <strong>in</strong>novations is approximately a normal-lognormal mixture.<br />
Hence, we derived the approximate distributions of GARCH(1,1) and AR(1)<br />
E-ARCH models for small sampl<strong>in</strong>g <strong>in</strong>tervals by us<strong>in</strong>g the distributional results<br />
available for the diusion limit.<br />
15