02.12.2012 Views

Applications of state space models in finance

Applications of state space models in finance

Applications of state space models in finance

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

38 3 L<strong>in</strong>ear Gaussian <strong>state</strong> <strong>space</strong> <strong>models</strong> and the Kalman filter<br />

3.6.2 Goodness <strong>of</strong> fit<br />

Measures <strong>of</strong> the goodness <strong>of</strong> fit are usually computed on the basis <strong>of</strong> forecast errors.<br />

The basic goodness <strong>of</strong> fit measure for time series <strong>models</strong> is the prediction error variance,<br />

which also serves as an <strong>in</strong>put to calculate the coefficient <strong>of</strong> determ<strong>in</strong>ation. A comparison<br />

between <strong>models</strong> with different numbers <strong>of</strong> parameters is commonly made on the basis<br />

<strong>of</strong> <strong>in</strong>formation criteria.<br />

3.6.2.1 Prediction error variance<br />

In case <strong>of</strong> time-homogeneity, the prediction error variance (p.e.v.) is def<strong>in</strong>ed as<br />

σ 2 = σ 2 ∗ ¯ f, (3.87)<br />

where ¯ f represents the value to which ft converges <strong>in</strong> steady-<strong>state</strong>. When a concentrated<br />

likelihood function is used with σ 2 ∗ be<strong>in</strong>g estimated by (3.61), σ2 can be estimated as<br />

˜σ 2 = ˜σ 2 ∗ ¯ f. (3.88)<br />

The <strong>in</strong>corporation <strong>of</strong> regression effects <strong>in</strong>to a model slightly changes the def<strong>in</strong>ition <strong>of</strong> the<br />

. When deal<strong>in</strong>g with time-vary<strong>in</strong>g regression <strong>models</strong>, the ML estimator<br />

estimator <strong>of</strong> σ2 ∗<br />

<strong>of</strong> σ2 ∗ for stochastic βt is a function <strong>of</strong> the generalized recursive residuals:<br />

s 2 ∗<br />

= (T − d − k)−1<br />

T�<br />

t=d+k+1<br />

˜v †2<br />

t , (3.89)<br />

where k is equal to the dimension <strong>of</strong> the vector <strong>of</strong> explanatory variables. For fixed β t ,<br />

the ML estimator <strong>of</strong> σ 2 ∗ depends on the GLS residuals:<br />

˜σ +2<br />

∗<br />

= (T − d)−1<br />

T�<br />

t=d+1<br />

˜v +2<br />

t , (3.90)<br />

where the sums <strong>of</strong> squares are identical <strong>in</strong> both cases. Tak<strong>in</strong>g (3.89) and (3.90) <strong>in</strong>to<br />

account, the prediction error variance for <strong>models</strong> that conta<strong>in</strong> explanatory variables can<br />

be estimated as<br />

or as<br />

s 2 = s 2 ∗ ¯ f, (3.91)<br />

˜σ +2 = ˜σ +2<br />

∗ ¯ f. (3.92)<br />

Accord<strong>in</strong>g to Harvey (1989, ¢ 7.4.3) the prediction error variance can be approximated<br />

for large samples <strong>in</strong> terms <strong>of</strong> the unstandardized GLS residuals as def<strong>in</strong>ed <strong>in</strong> (3.85):<br />

˜σ 2 =<br />

¯ f<br />

T − d<br />

T�<br />

t=d+1<br />

˜v +2<br />

t<br />

= 1<br />

T − d<br />

T�<br />

t=d+1<br />

v +2<br />

t<br />

ft<br />

¯f ≈ 1<br />

T − d<br />

T�<br />

t=d+1<br />

v +2<br />

t . (3.93)

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

Saved successfully!

Ooh no, something went wrong!