23.01.2014 Views

Estimation in Financial Models - RiskLab

Estimation in Financial Models - RiskLab

Estimation in Financial Models - RiskLab

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

tion<br />

F (x) =<br />

Z x<br />

,1<br />

f(y)dy:<br />

The latter statement can be made precise for <strong>in</strong>stance through the Borel-<br />

Cantelli Lemma and its various renements (see [68]) .<br />

In order to construct a density estimator f n (x) start<strong>in</strong>g from the empirical<br />

distribution function F n (x) <strong>in</strong> (4.1), we have to smooth F n . One possibility<br />

is by us<strong>in</strong>g so-called kernel density estimators<br />

nX x , Xk<br />

f n (x) = 1<br />

nh n<br />

k=1<br />

where the kernel V :IR ,! IR + satises<br />

Z 1<br />

,1<br />

V<br />

V (x)dx =1;<br />

and where the so-called bandwidth sequence (h n )issuch that<br />

h n<br />

h n ,! 0; nh n ,!1:<br />

This class of estimators is referred to as the Parzen-Rosenblatt estimators.<br />

Under certa<strong>in</strong> restrictions on f(x) the sequence f n (x) converges <strong>in</strong> some<br />

probabilistic sense to f(x). See [38], x4.4, and [69] for a detailed discussion.<br />

Ibragimov and Has'm<strong>in</strong>skii [38], x7, present some <strong>in</strong>terest<strong>in</strong>g approaches to<br />

and examples of nonparametric estimators of a signal S(t), belong<strong>in</strong>g to a<br />

certa<strong>in</strong> functional space. Two models for signals are considered. A rst one<br />

has the form<br />

dX(t) =S(t)dt + "db(t); (4.2)<br />

with 0 t 1, (b(t)) a Wiener process, " small (typically " # 0) and X(t) is<br />

an observed signal on [0; 1]. A second model is of the form<br />

<br />

;<br />

dX(t) =S(t)dt + db(t); (4.3)<br />

with 0 t n, S(t) a one-periodic function and X(t) is observed over n<br />

time periods.<br />

As a basic example consider estimation of the functional F on S(t)<br />

F (S) =<br />

Z 1<br />

0<br />

f(t)S(t)dt;<br />

with S; f 2 L 2 (0; 1). The functional F has to be estimated based on observations<br />

of (4.2), respectively (4.3). The estimator<br />

^F 1 =<br />

Z 1<br />

0<br />

f(t)dX(t)<br />

57

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!