Outputs effective sample sizes (ESSs) of the draws of the parameters in a BTVCVAR model.
Generates three matrix objects (one per parameter block) containing ESSs. The ESS for the draws of a parameter is equal to the relative numerical efficiency (RNE) times the posterior sample size. Values near the posterior sample size are desirable. ESSs much smaller than the posterior sample size indicate slow mixing. An NA is returned if sample autocorrelation does not taper off sufficiently quickly (i.e., if sample autocorrelation does not fall below 0.05 in the first 100 lags).
The ESS is the number of IID draws that is needed to achieve the same level of numerical precision as the obtained set of MCMC draws. It measures the extent of efficiency loss for using MCMC instead of vanilla Monte Carlo for sample generation.
The ESS, inefficiency factor, and relative numerical efficiency (RNE) convey the same information and only differ in interpretation.
Syntax
var_name.makeess c_ess_name s_ess_name q_ess_name
Names for the output matrices can be passed in as arguments. ESS matrix names are entered in the order: coefficients (C), observation covariance (S), then process covariance (Q). Untitled matrices are returned if no names are given.
Example
var var1.btvcvar 1 1 gdp inflation
var1.makeess
The first line declares and estimates a BTVCVAR with two variables. Estimation entails posterior simulation, as the posterior distribution is not known analytically. The second line computes and outputs inefficiency factors of the draws obtained.
Cross-references