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.

148 7 A Kalman filter based conditional multifactor pric<strong>in</strong>g model<br />

follow<strong>in</strong>g form:<br />

Rt = λ1,t ˆβ 1,t−1 + λ2,t ˆβ 2,t−1 + λ3,t ˆβ 3,t−1 + λ4,t ˆβ 4,t−1 + λ5,t ˆβ 5,t−1<br />

λ6,t ˆ β 6,t−1 + νt, for (7.24)<br />

for t = 164, . . . , 683, with Rt and ˆ β 1,t−1, . . . , ˆ β 6,t−1 denot<strong>in</strong>g 18 × 1 vectors <strong>of</strong> excess<br />

sector returns and factor load<strong>in</strong>gs, respectively. The factor load<strong>in</strong>gs used as <strong>in</strong>dependent<br />

variables have been estimated from one <strong>of</strong> the alternative time series regressions (KF,<br />

RLS, RR5 or RR1) <strong>of</strong> excess sector returns on the set <strong>of</strong> explanatory variables given<br />

by (7.18)–(7.20). Estimation <strong>of</strong> (7.24) for each date t <strong>of</strong> the out-<strong>of</strong>-sample period yields<br />

a time series <strong>of</strong> associated risk premia λk,t, k = 1, . . . , 6, each <strong>of</strong> length 520. 24 In<br />

accordance with (7.12)–(7.14) the significance <strong>of</strong> the chosen risk factors is tested by<br />

employ<strong>in</strong>g a t-test to the time series means <strong>of</strong> estimated risk premia.<br />

The cross-sectional regression results are summarized <strong>in</strong> Table 7.8, where the coefficient<br />

averages together with the correspond<strong>in</strong>g Fama-MacBeth t-statistics are reported.<br />

To ascerta<strong>in</strong> whether the proposed Kalman filter based conditional factor load<strong>in</strong>gs are<br />

relatively better <strong>in</strong> expla<strong>in</strong><strong>in</strong>g sector pric<strong>in</strong>g, the cross-sectional regressions have also<br />

been conducted for the betas estimated by recursive least squares and by roll<strong>in</strong>g OLS.<br />

Table 7.8 is divided <strong>in</strong>to four panels: Panel A covers the complete out-<strong>of</strong>-sample period<br />

from 22 February 1995 to 2 February 2005; Panels B to D refer to the three subperiods <strong>of</strong><br />

observed bull market, bear market and market recovery regimes. Overall, there is only<br />

little evidence that the multifactor model is better able to describe the cross-section <strong>of</strong><br />

returns when conditional Kalman filter based factor load<strong>in</strong>gs are used as <strong>in</strong>dependent<br />

variables. Over the entire estimation period, the average ¯ R 2 is only slightly higher than<br />

<strong>in</strong> case <strong>of</strong> the three alternative least squares specifications. The four different <strong>models</strong><br />

expla<strong>in</strong> between 28.9% (KF ) and 28.0% (RR1) <strong>of</strong> the total cross-sectional variation <strong>of</strong><br />

excess sector returns. The only risk premium that is significantly different from zero<br />

over the complete sample is estimated for the Kalman filter based T S factor.<br />

Divid<strong>in</strong>g the sample under consideration <strong>in</strong>to three market-regime dependent subperiods<br />

reveals that both the average ability to expla<strong>in</strong> the cross-sectional variability and<br />

the significance <strong>of</strong> estimated risk premia differ across subperiods. For the bull market<br />

period (Panel B) the risk premium <strong>of</strong> the SIZ factor is significantly positive for all <strong>models</strong>.<br />

The sign on SIZ reflects the outperformance <strong>of</strong> large caps over small caps dur<strong>in</strong>g<br />

the dotcom bubble. Besides, for the KF and RR1 based factor load<strong>in</strong>gs the risk premium<br />

<strong>of</strong> the T S factor is significantly negative at the 10% level. This means that dur<strong>in</strong>g<br />

24 For each cross-sectional regression at a given date t, six parameters have to be estimated<br />

with the dependent variable consist<strong>in</strong>g <strong>of</strong> 18 data po<strong>in</strong>ts only. Similar sett<strong>in</strong>gs with comparable<br />

degrees <strong>of</strong> freedom are commonly employed <strong>in</strong> the literature, see, for example, Chen et al.<br />

(1986) who use 20 portfolios as dependent variables to estimate up to seven parameters <strong>in</strong><br />

the cross-section. Nevertheless, <strong>in</strong> order to check whether the conclusion to be drawn <strong>in</strong> this<br />

subsection is sensitive to the dimension <strong>of</strong> the vector <strong>of</strong> dependent variables, (7.24) has also<br />

been estimated us<strong>in</strong>g 37 sectors <strong>in</strong>stead <strong>of</strong> the 18 supersectors employed throughout this thesis.<br />

As the employment <strong>of</strong> a f<strong>in</strong>er market segmentation leads to the same conclusion with respect<br />

to the practical relevance <strong>of</strong> time-vary<strong>in</strong>g factor load<strong>in</strong>gs, only the case with 18 supersectors<br />

is discussed here; results <strong>of</strong> the Fama-MacBeth procedure us<strong>in</strong>g 37 sectors are available on<br />

request.

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

Saved successfully!

Ooh no, something went wrong!