# Econ226_IIIAD - III Linear state-space models A State-space...

This preview shows pages 1–8. Sign up to view the full content.

1 III. Linear state-space models A. State-space representation of a dynamic system Consider following model State equation: r ± 1 8 t ² 1 ³ r ± r F r ± 1 8 t ² r ± 1 v t ² 1 Observation equation: n ± 1 y t ³ n ± k A U k ± 1 x t ² n ± r H U r ± 1 8 t ² n ± 1 w t Observed variables: y t , x t Unobserved variables: 8 t , v t , w t Matrices of parameters: F , A , H v t w t ~ i.i.d. N 0 0 , Q0 0R Q ³ r ± r R ³ n ± n

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
2 Example 1: 8 t ± 1 ² C 1 C 2 C C r " 1 C r 10 C 00 01 C BBC B B C 8 t ± / t ± 1 0 0 B 0 8 j , t ± 1 ² L j " 1 8 1 t for j ² 2,3,. .., r 8 1, t ± 1 ² C 1 8 1 t ± C 2 L 1 8 1 t ± C 3 L 2 8 1 t ± C ± C p L p " 1 8 1 t ±/ t ± 1 C ± L   8 1, t ± 1 ²/ t ± 1 Observation equation: y t ² 6 ± 1 2 1 2 2 C 2 r " 1 8 t y t " 6 ² 2 ± L   8 1 t put together with state equation: C ± L   8 1 t t C ± L  ± y t " 6   ² 2 ± L   / t Conclusion: any ARMA process can be written as a state-space model.
3 Example 2: C t ± state of business cycle D it ± idiosyncratic component for sector i C t , D it unobserved y it ± growth in sector i (observed) 8 t ± ± C t , D 1 t , D 2 t ,..., D nt   U 8 t ² 1 ± F 8 t ² v t ² 1 F ± C C 00 C 0 0 C 1 0 C 0 C 2 C 0 BBC BB 000 C C r Observation equation: y 1 t y 2 t B y nt ± 6 1 6 2 B 6 n ² + 1 10 C 0 + 2 01 C 0 BBBC B + n C 1 8 t

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
4 Purpose of state-space representation: state vector 8 t contains all information about system dynamics and forecasting. 8 t ± 1 ² F 8 t ± v t ± 1 y t ² A U x t ± H U 8 t ± w t E ± y t ± j | 8 t , 8 t " 1 ,..., 8 1 , y t , y t " 1 , ..., y 1 , x t ± j , x t ± j " 1 x 1   ² A U x t ± j ± H U F j 8 t III. Linear state-space models A. State-space representation of a dynamic system B. Kalman filter Purpose of Kalman filter: calculate distribution of 8 t conditional on ( t ² £ y t , y t " 1 y 1 , x t , x t " 1 x 1 ¤ 8 t | ( t ~ N ± § 8 t | t , P t | t
5 8 t ± 1 ² F 8 t ± v t ± 1 y t ² A U x t ± H U 8 t ± w t v t w t ~ i.i.d. N 0 0 , Q0 0R Begin with the prior: 8 0 ~ N ± § 8 0|0 , P 0|0   § 8 0|0 ² prior best guess as to value of 8 0 P 0|0 ² uncertainty about this guess (much uncertainty ² large diagonal elements of P 0|0 ) 8 1 ² F 8 0 ± v 1 8 1 ~ N ± § 8 1|0 , P 1|0   § 8 1|0 ² F § 8 0|0 P 1|0 ² FP 0|0 F U ± Q

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
6 Useful result: suppose that y 1 | x y 2 | x L N 6 1 6 2 , % 11 % 12 % 21 % 22 where 6 i and % ij may depend on x .Th en y 2 | y 1 , x L N ± m ' , M '   m ' ± 6 2 ² % 21 % 11 " 1 ± y 1 " 6 1   M ' ± % 22 " % 21 % 11 " 1 % 12 Here y 1 | x 1 , ( 0 8 1 | x 1 , ( 0 L N 6 1 6 2 , % 11 % 12 % 21 % 22 6 2 ± § 8 1|0 % 22 ± P 1|0 6 1 ± A U x 1 ² H U § 8 1|0 % 11 ± H U P 1|0 H ² R % 21 ± P 1|0 H Hence 8 1 | y 1 , x 1 , ( 0 ± 8 1 | ( 1 L N ± § 8 1|1 , P 1|1   § 8 1|1 ± § 8 1|0 ² P 1|0 H ± H U P 1|0 H ² R   " 1 ³ y 1 " A U x 1 " H U § 8 1|0 P 1|1 ± P 1|0 " P 1|0 H ± H U P 1|0 H ² R   " 1 H U P 1|0
7 Identical calculations: if 8 t | ( t ~ N ± 8 t | t , P t | t   , then 8 t ± 1 | ( t ± 1 ~ N ± § 8 t ± 1| t ± 1 , P t ± 1| t ± 1   P t ± 1| t ² FP t | t F U ± Q P t ± 1| t ± 1 ² P t ± 1| t " P t ± 1| t H ± H U P t ± 1| t H ± R   " 1 H U P t ± 1| t § 8 t ± 1| t ² F § 8 t | t § / t ± 1| t ² y t ± 1 " A U x t ± 1 " H U § 8 t ± 1| t § 8 t ± 1| t ± 1 ² § 8 t ± 1| t ± P t ± 1| t H ± H U P t ± 1| t H ± R

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

## This note was uploaded on 03/02/2012 for the course ECON 226 taught by Professor Jameshamilton during the Winter '09 term at UCSD.

### Page1 / 29

Econ226_IIIAD - III Linear state-space models A State-space...

This preview shows document pages 1 - 8. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online