干净的tvpvar,干净的英文

Time-varying parameter VAR model using TVP-VAR package

Jouchi Nakajima

jouchi.nakajima@gmail.com

September 1, 2013

1 Introduction

TVP-VAR (Time-Varying Parameter Vector AutoRegression) is the package for a Bayesian

analysis of the time-varying parameter VAR models. It implements the Markov chain Monte

Carlo (MCMC) algorithm to generate sample from the posterior distribution of the TVP-VAR

models. The TVP-VAR class is written in Ox (Doornik (2006)), and can be used by creating

at the top of the file you write. Alternatively, you can use the package by putting the TVPVAR

folder into the ox\packages folder and adding

#include <packages/TVPVAR/TVPVAR.ox>

at the top of your file.

1.2 Files in TVP-VAR

Files in TVP-VAR package are

• TVPVAR.ox – the main source file of TVPVAR class;

• tvpvar ox.pdf – this document;

included sample programs:

1

y

+ · · · + B

y

+ e

, e

∼ N (0, Ω

for t = s + 1, . . . , n, where y

are (k × k)

matrices of time-varying coefficients, and Ω

is a (k × k) time-varying covariance matrix. A

recursive identification is assumed by the decomposition Ω

= A

Σ

Σ

of the free lower-triangular elements of A

; and h

= (h

, . . . , h

) where h

time-varying parameters follow the random walk process:

= β

+ u

,

ε

O O

a

∼ N (µ

, Σ

). See Nakajima (2011) for the details.

The TVP-VAR package supports several types of the model specification. The following

• Code MCMC(N) to implement the MCMC algorithm with N iterations.

2