estimate
Estimate posterior distribution of Bayesian vector autoregression (VAR) model parameters
Since R2020a
Syntax
Description
returns the Bayesian VAR(p) model
PosteriorMdl
= estimate(PriorMdl
,Y
)PosteriorMdl
that characterizes the joint posterior distributions of the coefficients Λ and innovations covariance matrix Σ. PriorMdl
specifies the joint prior distribution of the parameters and the structure of the VAR model. Y
is the multivariate response data. PriorMdl
and PosteriorMdl
might not be the same object type.
NaN
s in the data indicate missing values, which estimate
removes by using list-wise deletion.
specifies additional options using one or more name-value pair arguments. For example, you can specify presample data to initialize the VAR model by using the PosteriorMdl
= estimate(PriorMdl
,Y
,Name,Value
)'Y0'
name-value pair argument.
[
also returns an estimation summary of the posterior distribution PosteriorMdl
,Summary
]
= estimate(___)Summary
, using any of the input argument combinations in the previous syntaxes.
Examples
Estimate Posterior Distribution
Consider the 3-D VAR(4) model for the US inflation (INFL
), unemployment (UNRATE
), and federal funds (FEDFUNDS
) rates.
For all , is a series of independent 3-D normal innovations with a mean of 0 and covariance . Assume that the joint prior distribution of the VAR model parameters is diffuse.
Load and Preprocess Data
Load the US macroeconomic data set. Compute the inflation rate. Plot all response series.
load Data_USEconModel seriesnames = ["INFL" "UNRATE" "FEDFUNDS"]; DataTimeTable.INFL = 100*[NaN; price2ret(DataTimeTable.CPIAUCSL)]; figure plot(DataTimeTable.Time,DataTimeTable{:,seriesnames}) legend(seriesnames)
Stabilize the unemployment and federal funds rates by applying the first difference to each series.
DataTimeTable.DUNRATE = [NaN; diff(DataTimeTable.UNRATE)];
DataTimeTable.DFEDFUNDS = [NaN; diff(DataTimeTable.FEDFUNDS)];
seriesnames(2:3) = "D" + seriesnames(2:3);
Remove all missing values from the data.
rmDataTimeTable = rmmissing(DataTimeTable);
Create Prior Model
Create a diffuse Bayesian VAR(4) prior model for the three response series. Specify the response variable names.
numseries = numel(seriesnames);
numlags = 4;
PriorMdl = bayesvarm(numseries,numlags,'SeriesNames',seriesnames)
PriorMdl = diffusebvarm with properties: Description: "3-Dimensional VAR(4) Model" NumSeries: 3 P: 4 SeriesNames: ["INFL" "DUNRATE" "DFEDFUNDS"] IncludeConstant: 1 IncludeTrend: 0 NumPredictors: 0 AR: {[3x3 double] [3x3 double] [3x3 double] [3x3 double]} Constant: [3x1 double] Trend: [3x0 double] Beta: [3x0 double] Covariance: [3x3 double]
PriorMdl
is a diffusebvarm
model object.
Estimate Posterior Distribution
Estimate the posterior distribution by passing the prior model and entire data series to estimate
.
PosteriorMdl = estimate(PriorMdl,rmDataTimeTable{:,seriesnames})
Bayesian VAR under diffuse priors Effective Sample Size: 197 Number of equations: 3 Number of estimated Parameters: 39 | Mean Std ------------------------------- Constant(1) | 0.1007 0.0832 Constant(2) | -0.0499 0.0450 Constant(3) | -0.4221 0.1781 AR{1}(1,1) | 0.1241 0.0762 AR{1}(2,1) | -0.0219 0.0413 AR{1}(3,1) | -0.1586 0.1632 AR{1}(1,2) | -0.4809 0.1536 AR{1}(2,2) | 0.4716 0.0831 AR{1}(3,2) | -1.4368 0.3287 AR{1}(1,3) | 0.1005 0.0390 AR{1}(2,3) | 0.0391 0.0211 AR{1}(3,3) | -0.2905 0.0835 AR{2}(1,1) | 0.3236 0.0868 AR{2}(2,1) | 0.0913 0.0469 AR{2}(3,1) | 0.3403 0.1857 AR{2}(1,2) | -0.0503 0.1647 AR{2}(2,2) | 0.2414 0.0891 AR{2}(3,2) | -0.2968 0.3526 AR{2}(1,3) | 0.0450 0.0413 AR{2}(2,3) | 0.0536 0.0223 AR{2}(3,3) | -0.3117 0.0883 AR{3}(1,1) | 0.4272 0.0860 AR{3}(2,1) | -0.0389 0.0465 AR{3}(3,1) | 0.2848 0.1841 AR{3}(1,2) | 0.2738 0.1620 AR{3}(2,2) | 0.0552 0.0876 AR{3}(3,2) | -0.7401 0.3466 AR{3}(1,3) | 0.0523 0.0428 AR{3}(2,3) | 0.0008 0.0232 AR{3}(3,3) | 0.0028 0.0917 AR{4}(1,1) | 0.0167 0.0901 AR{4}(2,1) | 0.0285 0.0488 AR{4}(3,1) | -0.0690 0.1928 AR{4}(1,2) | -0.1830 0.1520 AR{4}(2,2) | -0.1795 0.0822 AR{4}(3,2) | 0.1494 0.3253 AR{4}(1,3) | 0.0067 0.0395 AR{4}(2,3) | 0.0088 0.0214 AR{4}(3,3) | -0.1372 0.0845 Innovations Covariance Matrix | INFL DUNRATE DFEDFUNDS ------------------------------------------- INFL | 0.3028 -0.0217 0.1579 | (0.0321) (0.0124) (0.0499) DUNRATE | -0.0217 0.0887 -0.1435 | (0.0124) (0.0094) (0.0283) DFEDFUNDS | 0.1579 -0.1435 1.3872 | (0.0499) (0.0283) (0.1470)
PosteriorMdl = conjugatebvarm with properties: Description: "3-Dimensional VAR(4) Model" NumSeries: 3 P: 4 SeriesNames: ["INFL" "DUNRATE" "DFEDFUNDS"] IncludeConstant: 1 IncludeTrend: 0 NumPredictors: 0 Mu: [39x1 double] V: [13x13 double] Omega: [3x3 double] DoF: 184 AR: {[3x3 double] [3x3 double] [3x3 double] [3x3 double]} Constant: [3x1 double] Trend: [3x0 double] Beta: [3x0 double] Covariance: [3x3 double]
PosteriorMdl
is a conjugatebvarm
model object; the posterior is analytically tractable. The command line displays the posterior means (Mean
) and standard deviations (Std
) of all coefficients and the innovations covariance matrix. Row AR{
k
}(
i
,
j
)
contains the posterior estimates of , the lag k
AR coefficient of response variable j
in response equation i
. By default, estimate
uses the first four observations as a presample to initialize the model.
Display the posterior means of the AR coefficient matrices by using dot notation.
AR1 = PosteriorMdl.AR{1}
AR1 = 3×3
0.1241 -0.4809 0.1005
-0.0219 0.4716 0.0391
-0.1586 -1.4368 -0.2905
AR2 = PosteriorMdl.AR{2}
AR2 = 3×3
0.3236 -0.0503 0.0450
0.0913 0.2414 0.0536
0.3403 -0.2968 -0.3117
AR3 = PosteriorMdl.AR{3}
AR3 = 3×3
0.4272 0.2738 0.0523
-0.0389 0.0552 0.0008
0.2848 -0.7401 0.0028
AR4 = PosteriorMdl.AR{4}
AR4 = 3×3
0.0167 -0.1830 0.0067
0.0285 -0.1795 0.0088
-0.0690 0.1494 -0.1372
Specify Presample Data
Consider the 3-D VAR(4) model of Estimate Posterior Distribution. In this case, fit the model to the data starting in 1970.
Load and Preprocess Data
Load the US macroeconomic data set. Compute the inflation rate, stabilize the unemployment and federal funds rates, and remove missing values.
load Data_USEconModel seriesnames = ["INFL" "UNRATE" "FEDFUNDS"]; DataTimeTable.INFL = 100*[NaN; price2ret(DataTimeTable.CPIAUCSL)]; DataTimeTable.DUNRATE = [NaN; diff(DataTimeTable.UNRATE)]; DataTimeTable.DFEDFUNDS = [NaN; diff(DataTimeTable.FEDFUNDS)]; seriesnames(2:3) = "D" + seriesnames(2:3); rmDataTimeTable = rmmissing(DataTimeTable);
Create Prior Model
Create a diffuse Bayesian VAR(4) prior model for the three response series. Specify the response variable names.
numseries = numel(seriesnames);
numlags = 4;
PriorMdl = diffusebvarm(numseries,numlags,'SeriesNames',seriesnames);
Partition Time Base for Subsamples
A VAR(4) model requires p = 4 presample observations to initialize the AR component for estimation. Define index sets corresponding to the required presample and estimation samples.
idxpre = rmDataTimeTable.Time < datetime('1970','InputFormat','yyyy'); % Presample indices idxest = ~idxpre; % Estimation sample indices T = sum(idxest)
T = 157
The effective sample size is 157
observations.
Estimate Posterior Distribution
Estimate the posterior distribution. Specify only the required presample observations by using the 'Y0'
name-value pair argument.
Y0 = rmDataTimeTable{find(idxpre,PriorMdl.P,'last'),seriesnames}; PosteriorMdl = estimate(PriorMdl,rmDataTimeTable{idxest,seriesnames},... 'Y0',Y0);
Bayesian VAR under diffuse priors Effective Sample Size: 157 Number of equations: 3 Number of estimated Parameters: 39 | Mean Std ------------------------------- Constant(1) | 0.1431 0.1134 Constant(2) | -0.0132 0.0588 Constant(3) | -0.6864 0.2418 AR{1}(1,1) | 0.1314 0.0869 AR{1}(2,1) | -0.0187 0.0450 AR{1}(3,1) | -0.2009 0.1854 AR{1}(1,2) | -0.5009 0.1834 AR{1}(2,2) | 0.4881 0.0950 AR{1}(3,2) | -1.6913 0.3912 AR{1}(1,3) | 0.1089 0.0446 AR{1}(2,3) | 0.0555 0.0231 AR{1}(3,3) | -0.3588 0.0951 AR{2}(1,1) | 0.2942 0.1012 AR{2}(2,1) | 0.0786 0.0524 AR{2}(3,1) | 0.3767 0.2157 AR{2}(1,2) | 0.0208 0.2042 AR{2}(2,2) | 0.3238 0.1058 AR{2}(3,2) | -0.4530 0.4354 AR{2}(1,3) | 0.0634 0.0487 AR{2}(2,3) | 0.0747 0.0252 AR{2}(3,3) | -0.3594 0.1038 AR{3}(1,1) | 0.4503 0.1002 AR{3}(2,1) | -0.0388 0.0519 AR{3}(3,1) | 0.3580 0.2136 AR{3}(1,2) | 0.3119 0.2008 AR{3}(2,2) | 0.0966 0.1040 AR{3}(3,2) | -0.8212 0.4282 AR{3}(1,3) | 0.0659 0.0502 AR{3}(2,3) | 0.0155 0.0260 AR{3}(3,3) | -0.0269 0.1070 AR{4}(1,1) | -0.0141 0.1046 AR{4}(2,1) | 0.0105 0.0542 AR{4}(3,1) | 0.0263 0.2231 AR{4}(1,2) | -0.2274 0.1875 AR{4}(2,2) | -0.1734 0.0972 AR{4}(3,2) | 0.1328 0.3999 AR{4}(1,3) | 0.0028 0.0456 AR{4}(2,3) | 0.0094 0.0236 AR{4}(3,3) | -0.1487 0.0973 Innovations Covariance Matrix | INFL DUNRATE DFEDFUNDS ------------------------------------------- INFL | 0.3597 -0.0333 0.1987 | (0.0433) (0.0161) (0.0672) DUNRATE | -0.0333 0.0966 -0.1647 | (0.0161) (0.0116) (0.0365) DFEDFUNDS | 0.1987 -0.1647 1.6355 | (0.0672) (0.0365) (0.1969)
Display and Return Estimation Summaries
Consider the 3-D VAR(4) model of Estimate Posterior Distribution.
Load the US macroeconomic data set. Compute the inflation rate, stabilize the unemployment and federal funds rates, and remove missing values.
load Data_USEconModel seriesnames = ["INFL" "UNRATE" "FEDFUNDS"]; DataTimeTable.INFL = 100*[NaN; price2ret(DataTimeTable.CPIAUCSL)]; DataTimeTable.DUNRATE = [NaN; diff(DataTimeTable.UNRATE)]; DataTimeTable.DFEDFUNDS = [NaN; diff(DataTimeTable.FEDFUNDS)]; seriesnames(2:3) = "D" + seriesnames(2:3); rmDataTimeTable = rmmissing(DataTimeTable);
Create a diffuse Bayesian VAR(4) prior model for the three response series. Specify the response variable names.
numseries = numel(seriesnames);
numlags = 4;
PriorMdl = diffusebvarm(numseries,numlags,'SeriesNames',seriesnames);
You can display estimation output in three ways, or turn off the display. Compare the display types.
estimate(PriorMdl,rmDataTimeTable{:,seriesnames}); % 'table', the default
Bayesian VAR under diffuse priors Effective Sample Size: 197 Number of equations: 3 Number of estimated Parameters: 39 | Mean Std ------------------------------- Constant(1) | 0.1007 0.0832 Constant(2) | -0.0499 0.0450 Constant(3) | -0.4221 0.1781 AR{1}(1,1) | 0.1241 0.0762 AR{1}(2,1) | -0.0219 0.0413 AR{1}(3,1) | -0.1586 0.1632 AR{1}(1,2) | -0.4809 0.1536 AR{1}(2,2) | 0.4716 0.0831 AR{1}(3,2) | -1.4368 0.3287 AR{1}(1,3) | 0.1005 0.0390 AR{1}(2,3) | 0.0391 0.0211 AR{1}(3,3) | -0.2905 0.0835 AR{2}(1,1) | 0.3236 0.0868 AR{2}(2,1) | 0.0913 0.0469 AR{2}(3,1) | 0.3403 0.1857 AR{2}(1,2) | -0.0503 0.1647 AR{2}(2,2) | 0.2414 0.0891 AR{2}(3,2) | -0.2968 0.3526 AR{2}(1,3) | 0.0450 0.0413 AR{2}(2,3) | 0.0536 0.0223 AR{2}(3,3) | -0.3117 0.0883 AR{3}(1,1) | 0.4272 0.0860 AR{3}(2,1) | -0.0389 0.0465 AR{3}(3,1) | 0.2848 0.1841 AR{3}(1,2) | 0.2738 0.1620 AR{3}(2,2) | 0.0552 0.0876 AR{3}(3,2) | -0.7401 0.3466 AR{3}(1,3) | 0.0523 0.0428 AR{3}(2,3) | 0.0008 0.0232 AR{3}(3,3) | 0.0028 0.0917 AR{4}(1,1) | 0.0167 0.0901 AR{4}(2,1) | 0.0285 0.0488 AR{4}(3,1) | -0.0690 0.1928 AR{4}(1,2) | -0.1830 0.1520 AR{4}(2,2) | -0.1795 0.0822 AR{4}(3,2) | 0.1494 0.3253 AR{4}(1,3) | 0.0067 0.0395 AR{4}(2,3) | 0.0088 0.0214 AR{4}(3,3) | -0.1372 0.0845 Innovations Covariance Matrix | INFL DUNRATE DFEDFUNDS ------------------------------------------- INFL | 0.3028 -0.0217 0.1579 | (0.0321) (0.0124) (0.0499) DUNRATE | -0.0217 0.0887 -0.1435 | (0.0124) (0.0094) (0.0283) DFEDFUNDS | 0.1579 -0.1435 1.3872 | (0.0499) (0.0283) (0.1470)
estimate(PriorMdl,rmDataTimeTable{:,seriesnames},... 'Display','equation');
Bayesian VAR under diffuse priors Effective Sample Size: 197 Number of equations: 3 Number of estimated Parameters: 39 VAR Equations | INFL(-1) DUNRATE(-1) DFEDFUNDS(-1) INFL(-2) DUNRATE(-2) DFEDFUNDS(-2) INFL(-3) DUNRATE(-3) DFEDFUNDS(-3) INFL(-4) DUNRATE(-4) DFEDFUNDS(-4) Constant ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ INFL | 0.1241 -0.4809 0.1005 0.3236 -0.0503 0.0450 0.4272 0.2738 0.0523 0.0167 -0.1830 0.0067 0.1007 | (0.0762) (0.1536) (0.0390) (0.0868) (0.1647) (0.0413) (0.0860) (0.1620) (0.0428) (0.0901) (0.1520) (0.0395) (0.0832) DUNRATE | -0.0219 0.4716 0.0391 0.0913 0.2414 0.0536 -0.0389 0.0552 0.0008 0.0285 -0.1795 0.0088 -0.0499 | (0.0413) (0.0831) (0.0211) (0.0469) (0.0891) (0.0223) (0.0465) (0.0876) (0.0232) (0.0488) (0.0822) (0.0214) (0.0450) DFEDFUNDS | -0.1586 -1.4368 -0.2905 0.3403 -0.2968 -0.3117 0.2848 -0.7401 0.0028 -0.0690 0.1494 -0.1372 -0.4221 | (0.1632) (0.3287) (0.0835) (0.1857) (0.3526) (0.0883) (0.1841) (0.3466) (0.0917) (0.1928) (0.3253) (0.0845) (0.1781) Innovations Covariance Matrix | INFL DUNRATE DFEDFUNDS ------------------------------------------- INFL | 0.3028 -0.0217 0.1579 | (0.0321) (0.0124) (0.0499) DUNRATE | -0.0217 0.0887 -0.1435 | (0.0124) (0.0094) (0.0283) DFEDFUNDS | 0.1579 -0.1435 1.3872 | (0.0499) (0.0283) (0.1470)
estimate(PriorMdl,rmDataTimeTable{:,seriesnames},... 'Display','matrix');
Bayesian VAR under diffuse priors Effective Sample Size: 197 Number of equations: 3 Number of estimated Parameters: 39 VAR Coefficient Matrix of Lag 1 | INFL(-1) DUNRATE(-1) DFEDFUNDS(-1) -------------------------------------------------- INFL | 0.1241 -0.4809 0.1005 | (0.0762) (0.1536) (0.0390) DUNRATE | -0.0219 0.4716 0.0391 | (0.0413) (0.0831) (0.0211) DFEDFUNDS | -0.1586 -1.4368 -0.2905 | (0.1632) (0.3287) (0.0835) VAR Coefficient Matrix of Lag 2 | INFL(-2) DUNRATE(-2) DFEDFUNDS(-2) -------------------------------------------------- INFL | 0.3236 -0.0503 0.0450 | (0.0868) (0.1647) (0.0413) DUNRATE | 0.0913 0.2414 0.0536 | (0.0469) (0.0891) (0.0223) DFEDFUNDS | 0.3403 -0.2968 -0.3117 | (0.1857) (0.3526) (0.0883) VAR Coefficient Matrix of Lag 3 | INFL(-3) DUNRATE(-3) DFEDFUNDS(-3) -------------------------------------------------- INFL | 0.4272 0.2738 0.0523 | (0.0860) (0.1620) (0.0428) DUNRATE | -0.0389 0.0552 0.0008 | (0.0465) (0.0876) (0.0232) DFEDFUNDS | 0.2848 -0.7401 0.0028 | (0.1841) (0.3466) (0.0917) VAR Coefficient Matrix of Lag 4 | INFL(-4) DUNRATE(-4) DFEDFUNDS(-4) -------------------------------------------------- INFL | 0.0167 -0.1830 0.0067 | (0.0901) (0.1520) (0.0395) DUNRATE | 0.0285 -0.1795 0.0088 | (0.0488) (0.0822) (0.0214) DFEDFUNDS | -0.0690 0.1494 -0.1372 | (0.1928) (0.3253) (0.0845) Constant Term INFL | 0.1007 | (0.0832) DUNRATE | -0.0499 | 0.0450 DFEDFUNDS | -0.4221 | 0.1781 Innovations Covariance Matrix | INFL DUNRATE DFEDFUNDS ------------------------------------------- INFL | 0.3028 -0.0217 0.1579 | (0.0321) (0.0124) (0.0499) DUNRATE | -0.0217 0.0887 -0.1435 | (0.0124) (0.0094) (0.0283) DFEDFUNDS | 0.1579 -0.1435 1.3872 | (0.0499) (0.0283) (0.1470)
Return the estimation summary, which is a structure that contains the same information regardless of display type.
[PosteriorMdl,Summary] = estimate(PriorMdl,rmDataTimeTable{:,seriesnames});
Bayesian VAR under diffuse priors Effective Sample Size: 197 Number of equations: 3 Number of estimated Parameters: 39 | Mean Std ------------------------------- Constant(1) | 0.1007 0.0832 Constant(2) | -0.0499 0.0450 Constant(3) | -0.4221 0.1781 AR{1}(1,1) | 0.1241 0.0762 AR{1}(2,1) | -0.0219 0.0413 AR{1}(3,1) | -0.1586 0.1632 AR{1}(1,2) | -0.4809 0.1536 AR{1}(2,2) | 0.4716 0.0831 AR{1}(3,2) | -1.4368 0.3287 AR{1}(1,3) | 0.1005 0.0390 AR{1}(2,3) | 0.0391 0.0211 AR{1}(3,3) | -0.2905 0.0835 AR{2}(1,1) | 0.3236 0.0868 AR{2}(2,1) | 0.0913 0.0469 AR{2}(3,1) | 0.3403 0.1857 AR{2}(1,2) | -0.0503 0.1647 AR{2}(2,2) | 0.2414 0.0891 AR{2}(3,2) | -0.2968 0.3526 AR{2}(1,3) | 0.0450 0.0413 AR{2}(2,3) | 0.0536 0.0223 AR{2}(3,3) | -0.3117 0.0883 AR{3}(1,1) | 0.4272 0.0860 AR{3}(2,1) | -0.0389 0.0465 AR{3}(3,1) | 0.2848 0.1841 AR{3}(1,2) | 0.2738 0.1620 AR{3}(2,2) | 0.0552 0.0876 AR{3}(3,2) | -0.7401 0.3466 AR{3}(1,3) | 0.0523 0.0428 AR{3}(2,3) | 0.0008 0.0232 AR{3}(3,3) | 0.0028 0.0917 AR{4}(1,1) | 0.0167 0.0901 AR{4}(2,1) | 0.0285 0.0488 AR{4}(3,1) | -0.0690 0.1928 AR{4}(1,2) | -0.1830 0.1520 AR{4}(2,2) | -0.1795 0.0822 AR{4}(3,2) | 0.1494 0.3253 AR{4}(1,3) | 0.0067 0.0395 AR{4}(2,3) | 0.0088 0.0214 AR{4}(3,3) | -0.1372 0.0845 Innovations Covariance Matrix | INFL DUNRATE DFEDFUNDS ------------------------------------------- INFL | 0.3028 -0.0217 0.1579 | (0.0321) (0.0124) (0.0499) DUNRATE | -0.0217 0.0887 -0.1435 | (0.0124) (0.0094) (0.0283) DFEDFUNDS | 0.1579 -0.1435 1.3872 | (0.0499) (0.0283) (0.1470)
Summary
Summary = struct with fields:
Description: "3-Dimensional VAR(4) Model"
NumEstimatedParameters: 39
Table: [39x2 table]
CoeffMap: [39x1 string]
CoeffMean: [39x1 double]
CoeffStd: [39x1 double]
SigmaMean: [3x3 double]
SigmaStd: [3x3 double]
The CoeffMap
field contains a list of the coefficient names. The order of the names corresponds to the order of all coefficient vector inputs and outputs. Display CoeffMap
.
Summary.CoeffMap
ans = 39x1 string
"AR{1}(1,1)"
"AR{1}(1,2)"
"AR{1}(1,3)"
"AR{2}(1,1)"
"AR{2}(1,2)"
"AR{2}(1,3)"
"AR{3}(1,1)"
"AR{3}(1,2)"
"AR{3}(1,3)"
"AR{4}(1,1)"
"AR{4}(1,2)"
"AR{4}(1,3)"
"Constant(1)"
"AR{1}(2,1)"
"AR{1}(2,2)"
"AR{1}(2,3)"
"AR{2}(2,1)"
"AR{2}(2,2)"
"AR{2}(2,3)"
"AR{3}(2,1)"
"AR{3}(2,2)"
"AR{3}(2,3)"
"AR{4}(2,1)"
"AR{4}(2,2)"
"AR{4}(2,3)"
"Constant(2)"
"AR{1}(3,1)"
"AR{1}(3,2)"
"AR{1}(3,3)"
"AR{2}(3,1)"
⋮
Estimate Posterior with Sparse Coefficients
Consider the 3-D VAR(4) model of Estimate Posterior Distribution In this example, create a normal conjugate prior model with a fixed coefficient matrix instead of a diffuse model. The model contains 39 coefficients. For coefficient sparsity in the posterior, apply the Minnesota regularization method during estimation.
Load the US macroeconomic data set. Compute the inflation rate, stabilize the unemployment and federal funds rates, and remove missing values.
load Data_USEconModel seriesnames = ["INFL" "UNRATE" "FEDFUNDS"]; DataTimeTable.INFL = 100*[NaN; price2ret(DataTimeTable.CPIAUCSL)]; DataTimeTable.DUNRATE = [NaN; diff(DataTimeTable.UNRATE)]; DataTimeTable.DFEDFUNDS = [NaN; diff(DataTimeTable.FEDFUNDS)]; seriesnames(2:3) = "D" + seriesnames(2:3); rmDataTimeTable = rmmissing(DataTimeTable);
Create a normal conjugate Bayesian VAR(4) prior model for the three response series. Specify the response variable names, and set the innovations covariance matrix
According to the Minnesota regularization method, specify the following:
Each response is an AR(1) model, on average, with lag 1 coefficient 0.75.
The prior self-lag coefficients have variance 100. This large variance setting allows the data to influence the posterior more than the prior.
The prior cross-lag coefficients have variance 0.01. This small variance setting tightens the cross-lag coefficients to zero during estimation.
Prior coefficient covariances decay with increasing lag at a rate of 10 (that is, lower lags are more important than higher lags).
numseries = numel(seriesnames); numlags = 4; Sigma = [10e-5 0 10e-4; 0 0.1 -0.2; 10e-4 -0.2 1.6]; PriorMdl = bayesvarm(numseries,numlags,'Model','normal','SeriesNames',seriesnames,... 'Center',0.75,'SelfLag',100,'CrossLag',0.01,'Decay',10,... 'Sigma',Sigma);
Estimate the posterior distribution, and display the posterior response equations.
PosteriorMdl = estimate(PriorMdl,rmDataTimeTable{:,seriesnames},'Display','equation');
Bayesian VAR under normal priors and fixed Sigma Effective Sample Size: 197 Number of equations: 3 Number of estimated Parameters: 39 VAR Equations | INFL(-1) DUNRATE(-1) DFEDFUNDS(-1) INFL(-2) DUNRATE(-2) DFEDFUNDS(-2) INFL(-3) DUNRATE(-3) DFEDFUNDS(-3) INFL(-4) DUNRATE(-4) DFEDFUNDS(-4) Constant ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ INFL | 0.1234 -0.4373 0.1050 0.3343 -0.0342 0.0308 0.4441 0.0031 0.0090 0.0083 -0.0003 0.0003 0.0820 | (0.0014) (0.0027) (0.0007) (0.0015) (0.0021) (0.0006) (0.0015) (0.0004) (0.0003) (0.0015) (0.0001) (0.0001) (0.0014) DUNRATE | 0.0521 0.3636 0.0125 0.0012 0.1720 0.0009 0.0000 -0.0741 -0.0000 0.0000 0.0007 -0.0000 -0.0413 | (0.0252) (0.0723) (0.0191) (0.0031) (0.0666) (0.0031) (0.0004) (0.0348) (0.0004) (0.0001) (0.0096) (0.0001) (0.0339) DFEDFUNDS | -0.0105 -0.1394 -0.1368 0.0002 -0.0000 -0.1227 0.0000 -0.0000 0.0085 -0.0000 0.0000 -0.0041 -0.0113 | (0.0749) (0.0948) (0.0713) (0.0031) (0.0031) (0.0633) (0.0004) (0.0004) (0.0344) (0.0001) (0.0001) (0.0097) (0.1176) Innovations Covariance Matrix | INFL DUNRATE DFEDFUNDS ---------------------------------------- INFL | 0.0001 0 0.0010 | (0) (0) (0) DUNRATE | 0 0.1000 -0.2000 | (0) (0) (0) DFEDFUNDS | 0.0010 -0.2000 1.6000 | (0) (0) (0)
Compare the results to a posterior in which you specify no prior regularization.
PriorMdlNoReg = bayesvarm(numseries,numlags,'Model','normal','SeriesNames',seriesnames,... 'Sigma',Sigma); PosteriorMdlNoReg = estimate(PriorMdlNoReg,rmDataTimeTable{:,seriesnames},'Display','equation');
Bayesian VAR under normal priors and fixed Sigma Effective Sample Size: 197 Number of equations: 3 Number of estimated Parameters: 39 VAR Equations | INFL(-1) DUNRATE(-1) DFEDFUNDS(-1) INFL(-2) DUNRATE(-2) DFEDFUNDS(-2) INFL(-3) DUNRATE(-3) DFEDFUNDS(-3) INFL(-4) DUNRATE(-4) DFEDFUNDS(-4) Constant ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ INFL | 0.1242 -0.4794 0.1007 0.3233 -0.0502 0.0450 0.4270 0.2734 0.0523 0.0168 -0.1823 0.0068 0.1010 | (0.0014) (0.0028) (0.0007) (0.0016) (0.0030) (0.0007) (0.0016) (0.0029) (0.0008) (0.0016) (0.0027) (0.0007) (0.0015) DUNRATE | -0.0264 0.3428 0.0089 0.0969 0.1578 0.0292 0.0042 -0.0309 -0.0114 0.0221 -0.1071 0.0072 -0.0873 | (0.0347) (0.0714) (0.0203) (0.0356) (0.0714) (0.0203) (0.0337) (0.0670) (0.0200) (0.0326) (0.0615) (0.0186) (0.0422) DFEDFUNDS | -0.0351 -0.1248 -0.0411 0.0416 -0.0224 -0.1358 0.0014 -0.0302 0.1557 -0.0074 -0.0010 -0.0785 -0.0205 | (0.0787) (0.0949) (0.0696) (0.0631) (0.0689) (0.0663) (0.0533) (0.0567) (0.0630) (0.0470) (0.0493) (0.0608) (0.1347) Innovations Covariance Matrix | INFL DUNRATE DFEDFUNDS ---------------------------------------- INFL | 0.0001 0 0.0010 | (0) (0) (0) DUNRATE | 0 0.1000 -0.2000 | (0) (0) (0) DFEDFUNDS | 0.0010 -0.2000 1.6000 | (0) (0) (0)
The posterior estimates of the Minnesota prior have lower magnitude, in general, compared to the estimates of the default normal conjugate prior model.
Adjust Posterior Sampling Options
Consider the 3-D VAR(4) model of Estimate Posterior Distribution In this case, assume that the coefficients and innovations covariance matrix are independent (a semiconjugate prior model).
Load the US macroeconomic data set. Compute the inflation rate, stabilize the unemployment and federal funds rates, and remove missing values.
load Data_USEconModel seriesnames = ["INFL" "UNRATE" "FEDFUNDS"]; DataTimeTable.INFL = 100*[NaN; price2ret(DataTimeTable.CPIAUCSL)]; DataTimeTable.DUNRATE = [NaN; diff(DataTimeTable.UNRATE)]; DataTimeTable.DFEDFUNDS = [NaN; diff(DataTimeTable.FEDFUNDS)]; seriesnames(2:3) = "D" + seriesnames(2:3); rmDataTimeTable = rmmissing(DataTimeTable);
Create a semiconjugate Bayesian VAR(4) prior model for the three response series. Specify the response variable names.
numseries = numel(seriesnames); numlags = 4; PriorMdl = bayesvarm(numseries,numlags,'Model','semiconjugate',... 'SeriesNames',seriesnames);
Because the joint posterior of a semiconjugate prior model is analytically intractable, estimate
uses a Gibbs sampler to form the joint posterior by sampling from the tractable full conditionals.
Estimate the posterior distribution. For the Gibbs sampler, specify an effective number of draws of 20,000, a burn-in period of 5000, and a thinning factor of 10.
rng(1) % For reproducibility PosteriorMdl = estimate(PriorMdl,rmDataTimeTable{:,seriesnames},... 'Display','equation','NumDraws',20000,'Burnin',5000,'Thin',10);
Bayesian VAR under semiconjugate priors Effective Sample Size: 197 Number of equations: 3 Number of estimated Parameters: 39 VAR Equations | INFL(-1) DUNRATE(-1) DFEDFUNDS(-1) INFL(-2) DUNRATE(-2) DFEDFUNDS(-2) INFL(-3) DUNRATE(-3) DFEDFUNDS(-3) INFL(-4) DUNRATE(-4) DFEDFUNDS(-4) Constant ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ INFL | 0.2243 -0.0824 0.1365 0.2515 -0.0098 0.0329 0.2888 0.0311 0.0368 0.0458 -0.0206 0.0176 0.1836 | (0.0662) (0.0821) (0.0319) (0.0701) (0.0636) (0.0309) (0.0662) (0.0534) (0.0297) (0.0649) (0.0470) (0.0274) (0.0720) DUNRATE | -0.0262 0.3666 0.0148 0.0929 0.1637 0.0336 0.0016 -0.0147 -0.0089 0.0222 -0.1133 0.0082 -0.0808 | (0.0342) (0.0728) (0.0197) (0.0354) (0.0713) (0.0198) (0.0334) (0.0671) (0.0194) (0.0320) (0.0606) (0.0179) (0.0407) DFEDFUNDS | -0.0251 -0.1285 -0.0527 0.0379 -0.0256 -0.1452 -0.0040 -0.0360 0.1516 -0.0090 0.0008 -0.0823 -0.0193 | (0.0785) (0.0962) (0.0673) (0.0630) (0.0688) (0.0643) (0.0531) (0.0567) (0.0610) (0.0467) (0.0492) (0.0586) (0.1302) Innovations Covariance Matrix | INFL DUNRATE DFEDFUNDS ------------------------------------------- INFL | 0.2984 -0.0219 0.1754 | (0.0305) (0.0121) (0.0499) DUNRATE | -0.0219 0.0890 -0.1496 | (0.0121) (0.0092) (0.0292) DFEDFUNDS | 0.1754 -0.1496 1.4754 | (0.0499) (0.0292) (0.1506)
PosteriorMdl
is an empiricalbvarm
model represented by draws from the full conditionals. After removing the first burn-in period draws and thinning the remaining draws by keeping every 10th draw, estimate
stores the draws in the CoeffDraws
and SigmaDraws
properties.
Estimate VARX Model
Consider the 2-D VARX(1) model for the US real GDP (RGDP
) and investment (GCE
) rates that treats the personal consumption (PCEC
) rate as exogenous:
For all , is a series of independent 2-D normal innovations with a mean of 0 and covariance . Assume the following prior distributions:
, where M is a 4-by-2 matrix of means and is the 4-by-4 among-coefficient scale matrix. Equivalently, .
, where Ω is the 2-by-2 scale matrix and is the degrees of freedom.
Load the US macroeconomic data set. Compute the real GDP, investment, and personal consumption rate series. Remove all missing values from the resulting series.
load Data_USEconModel DataTimeTable.RGDP = DataTimeTable.GDP./DataTimeTable.GDPDEF; seriesnames = ["PCEC"; "RGDP"; "GCE"]; rates = varfun(@price2ret,DataTimeTable,'InputVariables',seriesnames); rates = rmmissing(rates); rates.Properties.VariableNames = seriesnames;
Create a conjugate prior model for the 2-D VARX(1) model parameters.
numseries = 2; numlags = 1; numpredictors = 1; PriorMdl = conjugatebvarm(numseries,numlags,'NumPredictors',numpredictors,... 'SeriesNames',seriesnames(2:end));
Estimate the posterior distribution. Specify the exogenous predictor data.
PosteriorMdl = estimate(PriorMdl,rates{:,2:end},... 'X',rates{:,1},'Display','equation');
Bayesian VAR under conjugate priors Effective Sample Size: 247 Number of equations: 2 Number of estimated Parameters: 8 VAR Equations | RGDP(-1) GCE(-1) Constant X1 ----------------------------------------------- RGDP | 0.0083 -0.0027 0.0078 0.0105 | (0.0625) (0.0606) (0.0043) (0.0625) GCE | 0.0059 0.0477 0.0166 0.0058 | (0.0644) (0.0624) (0.0044) (0.0645) Innovations Covariance Matrix | RGDP GCE --------------------------- RGDP | 0.0040 0.0000 | (0.0004) (0.0003) GCE | 0.0000 0.0043 | (0.0003) (0.0004)
By default, estimate
uses the first p = 1 observations in the specified response data as a presample, and it removes the corresponding observations in the predictor data from the sample.
The posterior means (and standard deviations) of the regression coefficients appear below the X1
column of the estimation summary table.
Input Arguments
PriorMdl
— Prior Bayesian VAR model
conjugatebvarm
model object | semiconjugatebvarm
model object | diffusebvarm
model object | normalbvarm
model object
Prior Bayesian VAR model, specified as a model object in this table.
Model Object | Description |
---|---|
conjugatebvarm | Dependent, matrix-normal-inverse-Wishart conjugate model returned by bayesvarm , conjugatebvarm , or estimate |
semiconjugatebvarm | Independent, normal-inverse-Wishart semiconjugate prior model returned by bayesvarm or semiconjugatebvarm |
diffusebvarm | Diffuse prior model returned by bayesvarm or diffusebvarm |
normalbvarm | Normal conjugate model with a fixed innovations covariance matrix, returned by bayesvarm , normalbvarm , or estimate |
PriorMdl
can also represent a joint posterior model returned by estimate
, either a conjugatebvarm
or normalbvarm
model object. In this case, estimate
updates the joint posterior distribution using the new observations.
For a semiconjugatebvarm
model, estimate
uses a Gibbs sampler to estimate the posterior distribution. To tune the sampler, see Options for Semiconjugate Prior Distributions.
Y
— Observed multivariate response series
numeric matrix
Observed multivariate response series to which estimate
fits the model, specified as a numobs
-by-numseries
numeric matrix.
numobs
is the sample size. numseries
is the number of response variables (PriorMdl.NumSeries
).
Rows correspond to observations, and the last row contains the latest observation. Columns correspond to individual response variables.
Y
represents the continuation of the presample response series in Y0
.
Data Types: double
Name-Value Arguments
Specify optional pairs of arguments as
Name1=Value1,...,NameN=ValueN
, where Name
is
the argument name and Value
is the corresponding value.
Name-value arguments must appear after other arguments, but the order of the
pairs does not matter.
Before R2021a, use commas to separate each name and value, and enclose
Name
in quotes.
Example: 'Y0',Y0,'Display','off'
specifies the presample data Y0
and suppresses the estimation display.
Y0
— Presample response data
numeric matrix
Presample response data to initialize the VAR model for estimation, specified as the comma-separated pair consisting of 'Y0'
and a numpreobs
-by-numseries
numeric matrix. numpreobs
is the number of presample observations.
Rows correspond to presample observations, and the last row contains the latest observation. Y0
must have at least PriorMdl.P
rows. If you supply more rows than necessary, estimate
uses the latest PriorMdl.P
observations only.
Columns must correspond to the response series in Y
.
By default, estimate
uses Y(1:PriorMdl.P,:)
as presample observations, and then estimates the posterior using Y((PriorMdl.P + 1):end,:)
. This action reduces the effective sample size.
Data Types: double
X
— Predictor data
numeric matrix
Predictor data for the exogenous regression component in the model, specified as the comma-separated pair consisting of 'X'
and a numobs
-by-PriorMdl.NumPredictors
numeric matrix.
Rows correspond to observations, and the last row contains the latest observation. estimate
does not use the regression component in the presample period. X
must have at least as many observations as the observations used after the presample period.
In either case, if you supply more rows than necessary, estimate
uses the latest observations only.
Columns correspond to individual predictor variables. All predictor variables are present in the regression component of each response equation.
Data Types: double
Display
— Estimation display style
'table'
(default) | 'off'
| 'equation'
| 'matrix'
Estimation display style printed to the command line, specified as the comma-separated pair consisting of 'Display'
and a value in this table.
Value | Description |
---|---|
'off' | estimate does not print to the command line. |
'table' |
|
'equation' |
|
'matrix' |
|
The estimation information includes the effective sample size, the number of equations in the system, and the number of estimated parameters.
Example: 'Display','matrix'
Data Types: char
| string
NumDraws
— Monte Carlo simulation adjusted sample size
1e5
(default) | positive integer
Monte Carlo simulation adjusted sample size, specified as the comma-separated pair consisting of 'NumDraws'
and a positive integer. estimate
actually draws BurnIn
+ NumDraws*
Thin
samples, but bases the estimates off NumDraws
samples. For details on how estimate
reduces the full Monte Carlo sample, see Algorithms.
Example: 'NumDraws',1e7
Data Types: double
BurnIn
— Number of draws to remove from beginning of Monte Carlo sample
5000
(default) | nonnegative scalar
Number of draws to remove from the beginning of the Monte Carlo sample to reduce transient effects, specified as the comma-separated pair consisting of 'BurnIn'
and a nonnegative scalar. For details on how estimate
reduces the full Monte Carlo sample, see Algorithms.
Tip
To help you specify the appropriate burn-in period size:
Determine the extent of the transient behavior in the sample by setting the
BurnIn
name-value argument to0
.Simulate a few thousand observations by using
simulate
.Create trace plots.
Example: 'BurnIn',0
Data Types: double
Thin
— Adjusted sample size multiplier
1
(default) | positive integer
Adjusted sample size multiplier, specified as the comma-separated pair consisting of 'Thin'
and a positive integer.
The actual Monte Carlo sample size is BurnIn
+ NumDraws
*Thin
. After discarding the burn-in, estimate
discards every Thin
– 1
draws, and then retains the next. For details on how estimate
reduces the full Monte Carlo sample, see Algorithms.
Tip
To reduce potential large serial correlation in the Monte Carlo sample, or to reduce the memory consumption of the draws stored in PosteriorMdl
, specify a large value for Thin
.
Example: 'Thin',5
Data Types: double
Coeff0
— Starting values of VAR model coefficients for Gibbs sampler
numeric column vector
Starting values of the VAR model coefficients for the Gibbs sampler, specified as the comma-separated pair consisting of 'Coeff0'
and a numel(PriorMdl.Mu)
-by-1 numeric column vector.
Elements correspond to the elements of PriorMdl.Mu
(see Mu
).
By default, Coeff0
is the ordinary least-squares (OLS) estimate.
Tip
Create
Coeff0
by vertically stacking the transpose of all initial coefficients in the following order (skip coefficients not in the model):All coefficient matrices ordered by lag
Constant vector
Linear time trend vector
Exogenous regression coefficient matrix
Specify the vectorized result
Coeff0(:)
.A good practice is to run
estimate
multiple times using different parameter starting values. Verify that the solutions from each run converge to similar values.
Data Types: double
Sigma0
— Starting values of innovations covariance matrix for Gibbs sampler
numeric positive definite matrix
Starting values of the innovations covariance matrix for the Gibbs sampler, specified as the comma-separated pair consisting of 'Sigma0'
and a numeric positive definite matrix. Rows and columns correspond to response equations.
By default, Sigma0
is the OLS residual mean squared error.
Tip
A good practice is to run estimate
multiple times using different parameter starting values. Verify that the solutions from each run converge to similar values.
Data Types: double
Output Arguments
PosteriorMdl
— Posterior Bayesian VAR model
conjugatebvarm
model object | normalbvarm
model object | empiricalbvarm
model object
Posterior Bayesian VAR model, returned as a model object in the table.
Model Object | PriorMdl | Posterior Form |
---|---|---|
conjugatebvarm
| conjugatebvarm or diffusebvarm | Analytically tractable |
normalbvarm
| normalbvarm | Analytically tractable |
empiricalbvarm | semiconjugatebvarm | Analytically intractable |
Summary
— Summary of Bayesian estimators
structure array
Summary of Bayesian estimators, returned as a structure array containing the fields in this table.
Field | Description | Data Type |
---|---|---|
Description | Model description | String scalar |
NumEstimatedParameters | Number of estimated coefficients | Numeric scalar |
Table | Table of coefficient posterior means and standard deviations; each row corresponds to a coefficient, and each column corresponds to an estimate type | Table |
CoeffMap | Coefficient names | String vector |
CoeffMean | Coefficient posterior means | Numeric vector; rows correspond to CoeffMap |
CoeffStd | Coefficient posterior standard deviations | Numeric vector; rows correspond to CoeffMap |
SigmaMean | Innovations covariance posterior mean matrix | Numeric matrix; rows and columns correspond to response equations |
SigmaStd | Innovations covariance posterior standard deviation matrix | Numeric matrix; rows and columns correspond to response equations |
Alternatively, pass PosteriorMdl
to summarize
to obtain a summary of Bayesian estimators.
More About
Bayesian Vector Autoregression (VAR) Model
A Bayesian VAR model treats all coefficients and the innovations covariance matrix as random variables in the m-dimensional, stationary VARX(p) model. The model has one of the three forms described in this table.
Model | Equation |
---|---|
Reduced-form VAR(p) in difference-equation notation |
|
Multivariate regression |
|
Matrix regression |
|
For each time t = 1,...,T:
yt is the m-dimensional observed response vector, where m =
numseries
.Φ1,…,Φp are the m-by-m AR coefficient matrices of lags 1 through p, where p =
numlags
.c is the m-by-1 vector of model constants if
IncludeConstant
istrue
.δ is the m-by-1 vector of linear time trend coefficients if
IncludeTrend
istrue
.Β is the m-by-r matrix of regression coefficients of the r-by-1 vector of observed exogenous predictors xt, where r =
NumPredictors
. All predictor variables appear in each equation.which is a 1-by-(mp + r + 2) vector, and Zt is the m-by-m(mp + r + 2) block diagonal matrix
where 0z is a 1-by-(mp + r + 2) vector of zeros.
, which is an (mp + r + 2)-by-m random matrix of the coefficients, and the m(mp + r + 2)-by-1 vector λ = vec(Λ).
εt is an m-by-1 vector of random, serially uncorrelated, multivariate normal innovations with the zero vector for the mean and the m-by-m matrix Σ for the covariance. This assumption implies that the data likelihood is
where f is the m-dimensional multivariate normal density with mean ztΛ and covariance Σ, evaluated at yt.
Before considering the data, you impose a joint prior distribution assumption on (Λ,Σ), which is governed by the distribution π(Λ,Σ). In a Bayesian analysis, the distribution of the parameters is updated with information about the parameters obtained from the data likelihood. The result is the joint posterior distribution π(Λ,Σ|Y,X,Y0), where:
Y is a T-by-m matrix containing the entire response series {yt}, t = 1,…,T.
X is a T-by-m matrix containing the entire exogenous series {xt}, t = 1,…,T.
Y0 is a p-by-m matrix of presample data used to initialize the VAR model for estimation.
Tips
Monte Carlo simulation is subject to variation. If
estimate
uses Monte Carlo simulation, then estimates and inferences might vary when you callestimate
multiple times under seemingly equivalent conditions. To reproduce estimation results, set a random number seed by usingrng
before callingestimate
.
Algorithms
Whenever the prior distribution
PriorMdl
and the data likelihood yield an analytically tractable posterior distribution,estimate
evaluates the closed-form solutions to Bayes estimators. Otherwise,estimate
uses the Gibbs sampler to estimate the posterior.This figure illustrates how
estimate
reduces the Monte Carlo sample using the values ofNumDraws
,Thin
, andBurnIn
. Rectangles represent successive draws from the distribution.estimate
removes the white rectangles from the Monte Carlo sample. The remainingNumDraws
black rectangles compose the Monte Carlo sample.
Version History
Introduced in R2020a
See Also
Objects
Functions
Commande MATLAB
Vous avez cliqué sur un lien qui correspond à cette commande MATLAB :
Pour exécuter la commande, saisissez-la dans la fenêtre de commande de MATLAB. Les navigateurs web ne supportent pas les commandes MATLAB.
Select a Web Site
Choose a web site to get translated content where available and see local events and offers. Based on your location, we recommend that you select: .
You can also select a web site from the following list:
How to Get Best Site Performance
Select the China site (in Chinese or English) for best site performance. Other MathWorks country sites are not optimized for visits from your location.
Americas
- América Latina (Español)
- Canada (English)
- United States (English)
Europe
- Belgium (English)
- Denmark (English)
- Deutschland (Deutsch)
- España (Español)
- Finland (English)
- France (Français)
- Ireland (English)
- Italia (Italiano)
- Luxembourg (English)
- Netherlands (English)
- Norway (English)
- Österreich (Deutsch)
- Portugal (English)
- Sweden (English)
- Switzerland
- United Kingdom (English)