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

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

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

34 3 L<strong>in</strong>ear Gaussian <strong>state</strong> <strong>space</strong> <strong>models</strong> and the Kalman filter<br />

Follow<strong>in</strong>g Chow (1984) a general expression for the basic univariate time-vary<strong>in</strong>g<br />

coefficient regression model with homoskedastic error terms can be formulated as<br />

yt = ztξ t + ɛt, ɛt ∼ N(0, h), (3.73)<br />

where zt denotes a 1 × m row vector with m = k fixed explanatory variables, <strong>of</strong> which<br />

unity might be the first element. The column vector ξ t conta<strong>in</strong>s the regression coefficients<br />

whose behavior over time is represented by<br />

ξ t+1 = T ξ t + Rη t, η t ∼ N(0, Q), (3.74)<br />

where the system matrices T and R are assumed to be time-homogeneous. To allow for a<br />

mean<strong>in</strong>gful economic <strong>in</strong>terpretation, the <strong>state</strong> equation given <strong>in</strong> (3.74) can be rearranged<br />

by <strong>in</strong>troduc<strong>in</strong>g the mean <strong>state</strong> vector ¯ ξ and by replac<strong>in</strong>g ξ t by ξ t − ¯ ξ. This leads to<br />

a representation <strong>in</strong> which ¯ ξ can be <strong>in</strong>terpreted as the mean coefficient over the entire<br />

sample. The matrix T is referred to as the speed parameter, which measures how fast<br />

the time-vary<strong>in</strong>g <strong>state</strong> vector returns to its mean:<br />

ξ t+1 − ¯ ξ = T (ξ t − ¯ ξ) + Rη t. (3.75)<br />

The literature has not arrived yet at a consensus on how to <strong>in</strong>troduce time-variation<br />

<strong>in</strong>to the coefficients <strong>of</strong> explanatory variables. In the follow<strong>in</strong>g, it will be demonstrated<br />

how different processes for the time-vary<strong>in</strong>g <strong>state</strong> vector can be derived depend<strong>in</strong>g on<br />

the chosen value for T .<br />

3.5.2.1 The random coefficient model<br />

Sett<strong>in</strong>g T to an m × m zero matrix yields a model that describes the time path <strong>of</strong> a<br />

chang<strong>in</strong>g ξ t as random coefficients (RC):<br />

ξ t+1 = ¯ ξ + Rη t. (3.76)<br />

The random coefficient model, orig<strong>in</strong>ally <strong>in</strong>troduced by Hildreth and Houck (1968) <strong>in</strong> a<br />

cross-sectional context, implies a long-run average coefficient around which the current<br />

estimate fluctuates randomly. The parameters to be estimated are ¯ ξ and the variances<br />

<strong>of</strong> the error terms. As it is not possible to dist<strong>in</strong>guish between a randomly behav<strong>in</strong>g<br />

<strong>in</strong>tercept and the observation disturbances, any <strong>in</strong>tercept term has to be <strong>in</strong>cluded as<br />

be<strong>in</strong>g fixed <strong>in</strong> the observation equation. This reduces the <strong>state</strong> vector’s dimension by<br />

one (cf. Wells 1996, ¢ 5.3).<br />

Overall, the practical relevance <strong>of</strong> this model is limited: as the stochastic properties<br />

<strong>of</strong> the underly<strong>in</strong>g process are only reflected <strong>in</strong> the <strong>state</strong> disturbances, the average <strong>state</strong><br />

always represents the best forecast. For a review <strong>of</strong> the random coefficient model, see,<br />

for example, Nicholls and Pagan (1985).<br />

3.5.2.2 The random walk model<br />

By sett<strong>in</strong>g the transition matrix to an m-dimensional identity matrix, one can derive<br />

an important model<strong>in</strong>g class accord<strong>in</strong>g to which the behavior <strong>of</strong> the chang<strong>in</strong>g regression

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

Saved successfully!

Ooh no, something went wrong!