btvcvar |

Estimate a Bayesian time-varying coefficients VAR, or BTVCVAR, model.

Syntax

var_name.btvcvar(options) lag_pairs endog_list [@ exog_list]

btvcvar estimates a Bayesian time-varying coefficients VAR. The order of the VAR is specified using a lag pair, followed by a list of series or groups for endogenous variables. Exogenous variables can be included using the @-sign followed by a list of series or groups. A constant is automatically added to the list of exogenous variables; to estimate a specification without a constant, use the noconst option.

Options

Prior hyper-parameters

T0 = int (default = 0) | Set prior sample size . A prior sample is not used if T0 is set to 0. To use a prior sample, T0 must be set to an integer larger than the number of coefficients per equation in a standard VAR version of the model. |

tau0 = num (default = 5.0) | Set prior scaling parameter for the initial state . |

tau1 = num (default = 1.0) | Set prior scaling parameter for the observation covariance . |

nu1 = num (default = 5.0) | Set prior dof parameter for the observation covariance . |

tau2 = num (default = 0.01) | Set prior scaling parameter for the process covariance . |

nu2 = num (default = 5.0) | Set prior dof parameter for the process covariance . |

Display options

usemean | Use posterior mean as the point estimate. The posterior median is used if usemean is not included in the options list. |

showci | Show credibility intervals (bands). |

cilevels = arg (default = "0.95") | Set credibility levels. For multiple levels, enter a space-delimited list of values surrounded by quotation marks, e.g., "0.3 0.5 0.8". |

uselines | Use lines instead of shading for credibility intervals. |

MCMC options

burn = int (default = 5000) | Set burn-in size. |

size = int (default = 5000) | Set posterior sample size. |

thin = int (default = 1) | Set thinning size. A thinning size of indicates that every -th draw after the burn-in period is stored. |

nsub = int | Set the number of subchains. |

seed = int | Set the random seed. EViews will generate a seed if one is not specified. |

rng = arg (default = “kn” or method set via rndseed) | Set random number generator type. Available types are: improved Knuth generator (“kn”), improved Mersenne Twister (“mt”), Knuth’s (1997) lagged Fibonacci generator used in EViews 4 (“kn4”), L’Ecuyer’s (1999) combined multiple recursive generator (“le”), Matsumoto and Nishimura’s (1998) Mersenne Twister used in EViews 4 (“mt4”). |

Other options

smoother = arg (default = "CFA") | Set simulation smoothing method. Available methods are "CFA" (Cholesky factor algorithm), "KFS" (Kalman filter and smoother), and "MMP" (McCausland, Miller, & Pelletier, 2011). |

stable | Use the method of Cogley & Sargent for obtaining stable draws. |

maxattempts = int (default = 100) | Set the maximum number of attempts for the sampler to draw stable VAR coefficients for all dates in the data sample. |

noconst | Do not include a constant in the exogenous regressors list. |

Examples

To declare and estimate a BTVCVAR named MYVAR with endogenous variables DGP and UNEMP, a constant, the first lag, and a prior sample of size 40, run

var myvar.btvcvar(t0=40) 1 1 gdp unemp

in the command window. Running the command

var myvar.btvcvar(t0=40, showci, cilevels="0.3 0.5") 1 1 gdp unemp

will also display shaded 30% and 50% credibility bands. For reproducible results, also set the number of subchains (nsub), the random seed (seed), and the random number generator type (rng):

var myvar.btvcvar(t0=40, nsub=12, seed=342458900, rng=kn, showci, cilevels="0.3 0.5") 1 1 gdp unemp

Command capture will always show the nsub, seed, and rng options.

Cross-references

See
“Bayesian Time-varying Coefficients VAR Models” for details.