swz_mardd.m: computes the marginal data density p(Y) for constant-parameter structural VAR models (including over-identified cases
  with linear restrictions).  It may call a few M files that can be downloaded from Tao Zha's Archived Matlab Library ZhaZippedCode in
  http://home.earthlink.net/~tzha02/programCode.html

The standard prior for quarterly data is based on Sims and Zha (98), which improves upon the Minnesota prior by
  incorporating the cointegration and unit root prior.
The algorithm is based on Waggoner and Zha (03) and Chib (95).

A few things that need to change.

The file ftd_cholesky.m gives an example for a two-variable reduced-form (Choleski) VAR.  If you have, say, 4 variables,
  change from

%======== The second equation ===========
Qi(1:1,:,2) = [
   1 0
        ];

  to the following patterns:

%======== The second equation ===========
Qi(1:1,:,2) = [
   1 0 0 0
        ];

%======== The third equation ===========
Qi(1:2,:,3) = [
   1 0 0 0
   0 1 0 0
        ];

%======== The fourth equation ===========
Qi(1:3,:,4) = [
   1 0 0 0
   0 1 0 0
   0 0 1 0
        ];


In the file msstart_setup.m, make sure the data range, the sample range, the lag length, etc.  The following variables
  may have to be changed accordingly.

yrBin=1960;   % beginning of the year
qmBin=1;    % begining of the quarter or month
yrFin=2003;   % final year
qmFin=4;    % final month or quarter
vlist = [1 2];    % 1: U; 2: PCE inflation.
varlist={'U', 'Inflation'};
vlistlog = [ ];   % subset of "vlist.  Variables in log level so that differences are in **monthly** growth, unlike R and U which are in annual percent (divided by 100 already).
vlistper = [1 2];           % subset of "vlist"
%------- A specific sample is considered for estimation -------
yrStart=1960;
qmStart=1;
yrEnd=2003;
qmEnd=4;
%------- Prior, etc. -------
lags = 4        % number of lags


*****************************  REFERENCES ********************
The prior is discussed in
   author = {Christopher A. Sims and Tao Zha},
   title = {Bayesian Methods for Dynamic Multivariate Models},
   journal = ier,
   year = {1998},
   pages = {949--968},
   volume = {39},
   number = {4}

The algorithm is based on
   author = {Daniel F. Waggoner and Tao Zha},
   title = {A Gibbs Sampler for Structural Vector Autoregressions},
   journal = jedc,
   year = {2003},
   volume = {28},
   number = {2},
   pages = {349--366}

and

   author = {Siddhartha Chib},
   title = {Marginal Likelihood from the Gibbs Output},
   journal = jasa,
   year = {1995},
   pages = {1313--1321},
   volume = {90}