conjugateblm

Bayesian linear regression model with conjugate prior for data likelihood

Description

The Bayesian linear regression model object conjugateblm specifies that the joint prior distribution of the regression coefficients and the disturbance variance, that is, (β, σ²) is the dependent, normal-inverse-gamma conjugate model. The conditional prior distribution of β|σ² is multivariate Gaussian with mean μ and variance σ²V. The prior distribution of σ² is inverse gamma with shape A and scale B.

The data likelihood is $${\displaystyle \prod _{t=1}^T\varphi \left(y_t;x_t\beta ,{\sigma }^2\right)},$$ where ϕ(y_t;x_tβ,σ²) is the Gaussian probability density evaluated at y_t with mean x_tβ and variance σ². The specified priors are conjugate for the likelihood, and the resulting marginal and conditional posterior distributions are analytically tractable. For details on the posterior distribution, see Analytically Tractable Posteriors.

In general, when you create a Bayesian linear regression model object, it specifies the joint prior distribution and characteristics of the linear regression model only. That is, the model object is a template intended for further use. Specifically, to incorporate data into the model for posterior distribution analysis, pass the model object and data to the appropriate object function.

Creation

Syntax

PriorMdl = conjugateblm(NumPredictors)

PriorMdl = conjugateblm(NumPredictors,Name,Value)

Description

PriorMdl = conjugateblm(NumPredictors) creates a Bayesian linear regression model object (PriorMdl) composed of NumPredictors predictors and an intercept, and sets the NumPredictors property. The joint prior distribution of (β, σ²) is the dependent normal-inverse-gamma conjugate model. PriorMdl is a template that defines the prior distributions and the dimensionality of β.

PriorMdl = conjugateblm(NumPredictors,Name,Value) sets properties (except NumPredictors) using name-value pair arguments. Enclose each property name in quotes. For example, conjugateblm(2,'VarNames',["UnemploymentRate"; "CPI"]) specifies the names of the two predictor variables in the model.

Properties

You can set writable property values when you create the model object by using name-value argument syntax, or after you create the model object by using dot notation. For example, to set a more diffuse prior covariance matrix for PriorMdl than the default value, a Bayesian linear regression model containing three model coefficients, enter

PriorMdl.V = 100*eye(3);

NumPredictors — Number of predictor variables

nonnegative integer

Number of predictor variables in the Bayesian multiple linear regression model, specified as a nonnegative integer.

NumPredictors must be the same as the number of columns in your predictor data, which you specify during model estimation or simulation.

When specifying NumPredictors, exclude any intercept term for the value.

After creating a model, if you change the of value NumPredictors using dot notation, then these parameters revert to the default values:

• Variable names (VarNames)

• Prior mean of β (Mu)

• Prior covariance matrix of β (V)

Data Types: double

Intercept — Flag for including regression model intercept

true (default) | false

Flag for including a regression model intercept, specified as a value in this table.

Value	Description
false	Exclude an intercept from the regression model. Therefore, β is a p-dimensional vector, where p is the value of NumPredictors.
true	Include an intercept in the regression model. Therefore, β is a (p + 1)-dimensional vector. This specification causes a T-by-1 vector of ones to be prepended to the predictor data during estimation and simulation.

If you include a column of ones in the predictor data for an intercept term, then set Intercept to false.

Example: 'Intercept',false

Data Types: logical

VarNames — Predictor variable names

string vector | cell vector of character vectors

Predictor variable names for displays, specified as a string vector or cell vector of character vectors. VarNames must contain NumPredictors elements. VarNames(j) is the name of the variable in column j of the predictor data set, which you specify during estimation, simulation, or forecasting.

The default is {'Beta(1)','Beta(2),...,Beta(p)}, where p is the value of NumPredictors.

Example: 'VarNames',["UnemploymentRate"; "CPI"]

Data Types: string | cell | char

Mu — Mean hyperparameter of Gaussian prior on β

zeros(Intercept + NumPredictors,1) (default) | numeric scalar | numeric vector

Mean parameter of the Gaussian prior on β, specified as a numeric scalar or vector.

If Mu is a vector, then it must have NumPredictors or NumPredictors + 1 elements.

• For NumPredictors elements, conjugateblm sets the prior mean of the NumPredictors predictors only. Predictors correspond to the columns in the predictor data (specified during estimation, simulation, or forecasting). conjugateblm ignores the intercept in the model, that is, conjugateblm specifies the default prior mean to any intercept.

• For NumPredictors + 1 elements, the first element corresponds to the prior mean of the intercept, and all other elements correspond to the predictors.

Example: 'Mu',[1; 0.08; 2]

Data Types: double

V — Conditional covariance matrix hyperparameter of Gaussian prior on β

10000*eye(Intercept + NumPredictors) (default) | symmetric, positive-definite matrix | diag(Inf(Intercept + NumPredictors,1))

Conditional covariance matrix of Gaussian prior on β, specified as a c-by-c symmetric, positive definite matrix. c can be NumPredictors or NumPredictors + 1.

• If c is NumPredictors, then conjugateblm sets the prior covariance matrix to

  $$\left[\begin{array}{cccc}1e5& 0& \cdots & 0\\ 0& & & \\ ⋮& & V& \\ 0& & & \end{array}\right].$$

  conjugateblm attributes the default prior covariances to the intercept, and attributes V to the coefficients of the predictor variables in the data. Rows and columns of V correspond to columns (variables) in the predictor data.

• If c is NumPredictors + 1, then conjugateblm sets the entire prior covariance to V. The first row and column correspond to the intercept. All other rows and columns correspond to the columns in the predictor data.

The default value is a flat prior. For an adaptive prior, specify diag(Inf(Intercept + NumPredictors,1)). Adaptive priors indicate zero precision in order for the prior distribution to have as little influence as possible on the posterior distribution.

V is the prior covariance of β up to a factor of σ².

Example: 'V',diag(Inf(3,1))

Data Types: double

A — Shape hyperparameter of inverse gamma prior on σ²

3 (default) | numeric scalar

Shape hyperparameter of the inverse gamma prior on σ², specified as a numeric scalar.

A must be at least –(Intercept + NumPredictors)/2.

With B held fixed, the inverse gamma distribution becomes taller and more concentrated as A increases. This characteristic weighs the prior model of σ² more heavily than the likelihood during posterior estimation.

For the functional form of the inverse gamma distribution, see Analytically Tractable Posteriors.

Example: 'A',0.1

Data Types: double

B — Scale hyperparameter of inverse gamma prior on σ²

1 (default) | positive scalar | Inf

Scale parameter of inverse gamma prior on σ², specified as a positive scalar or Inf.

With A held fixed, the inverse gamma distribution becomes taller and more concentrated as B increases. This characteristic weighs the prior model of σ² more heavily than the likelihood during posterior estimation.

Example: 'B',5

Data Types: double

Object Functions

estimate	Estimate posterior distribution of Bayesian linear regression model parameters
simulate	Simulate regression coefficients and disturbance variance of Bayesian linear regression model
forecast	Forecast responses of Bayesian linear regression model
plot	Visualize prior and posterior densities of Bayesian linear regression model parameters
summarize	Distribution summary statistics of standard Bayesian linear regression model

Examples

Create Normal-Inverse-Gamma Conjugate Prior Model

Consider the multiple linear regression model that predicts U.S. real gross national product (GNPR) using a linear combination of industrial production index (IPI), total employment (E), and real wages (WR).

$${\text{GNPR}}_t={\beta }_0+{\beta }_1{\text{IPI}}_t+{\beta }_2{\text{E}}_t+{\beta }_3{\text{WR}}_t+{\epsilon }_t.$$

For all $$t$$ time points, $${\epsilon }_t$$ is a series of independent Gaussian disturbances with a mean of 0 and variance $${\sigma }^2$$.

Assume that the prior distributions are:

• $$\beta |{\sigma }^2\sim N_4(M,{\sigma }^{2}V)$$. $$M$$ is a 4-by-1 vector of means, and $$V$$ is a scaled 4-by-4 positive definite covariance matrix.

• $${\sigma }^2\sim IG(A,B)$$. $$A$$ and $$B$$ are the shape and scale, respectively, of an inverse gamma distribution.

These assumptions and the data likelihood imply a normal-inverse-gamma conjugate model.

Create a normal-inverse-gamma conjugate prior model for the linear regression parameters. Specify the number of predictors p.

p = 3;
Mdl = bayeslm(p,'ModelType','conjugate')

Mdl = 
  conjugateblm with properties:

    NumPredictors: 3
        Intercept: 1
         VarNames: {4x1 cell}
               Mu: [4x1 double]
                V: [4x4 double]
                A: 3
                B: 1

 
           |  Mean     Std            CI95         Positive       Distribution     
-----------------------------------------------------------------------------------
 Intercept |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 Beta(1)   |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 Beta(2)   |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 Beta(3)   |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 Sigma2    | 0.5000   0.5000    [ 0.138,  1.616]     1.000   IG(3.00,    1)        
 

Mdl is a conjugateblm Bayesian linear regression model object representing the prior distribution of the regression coefficients and disturbance variance. At the command window, bayeslm displays a summary of the prior distributions.

You can set writable property values of created models using dot notation. Set the regression coefficient names to the corresponding variable names.

Mdl.VarNames = ["IPI" "E" "WR"]

Mdl = 
  conjugateblm with properties:

    NumPredictors: 3
        Intercept: 1
         VarNames: {4x1 cell}
               Mu: [4x1 double]
                V: [4x4 double]
                A: 3
                B: 1

 
           |  Mean     Std            CI95         Positive       Distribution     
-----------------------------------------------------------------------------------
 Intercept |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 IPI       |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 E         |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 WR        |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 Sigma2    | 0.5000   0.5000    [ 0.138,  1.616]     1.000   IG(3.00,    1)        
 

Estimate Marginal Posterior Distributions

Consider the linear regression model in Create Normal-Inverse-Gamma Conjugate Prior Model.

Create a normal-inverse-gamma conjugate prior model for the linear regression parameters. Specify the number of predictors p and the names of the regression coefficients.

p = 3;
PriorMdl = bayeslm(p,'ModelType','conjugate','VarNames',["IPI" "E" "WR"]);

Load the Nelson-Plosser data set. Create variables for the response and predictor series.

load Data_NelsonPlosser
X = DataTable{:,PriorMdl.VarNames(2:end)};
y = DataTable{:,'GNPR'};

Estimate the marginal posterior distributions of $$\beta$$ and $${\sigma }^2$$.

PosteriorMdl = estimate(PriorMdl,X,y);

Method: Analytic posterior distributions
Number of observations: 62
Number of predictors:   4
Log marginal likelihood: -259.348
 
           |   Mean      Std          CI95        Positive       Distribution      
-----------------------------------------------------------------------------------
 Intercept | -24.2494  8.7821  [-41.514, -6.985]    0.003   t (-24.25, 8.65^2, 68) 
 IPI       |   4.3913  0.1414   [ 4.113,  4.669]    1.000   t (4.39, 0.14^2, 68)   
 E         |   0.0011  0.0003   [ 0.000,  0.002]    1.000   t (0.00, 0.00^2, 68)   
 WR        |   2.4683  0.3490   [ 1.782,  3.154]    1.000   t (2.47, 0.34^2, 68)   
 Sigma2    |  44.1347  7.8020   [31.427, 61.855]    1.000   IG(34.00, 0.00069)     
 

PosteriorMdl is a conjugateblm model object storing the joint marginal posterior distribution of $$\beta$$ and $${\sigma }^2$$ given the data. estimate displays a summary of the marginal posterior distributions to the command window. Rows of the summary correspond to regression coefficients and the disturbance variance, and columns to characteristics of the posterior distribution. The characteristics include:

• CI95, which contains the 95% Bayesian equitailed credible intervals for the parameters. For example, the posterior probability that the regression coefficient of WR is in [1.782, 3.154] is 0.95.

• Positive, which contains the posterior probability that the parameter is greater than 0. For example, the probability that the intercept is greater than 0 is 0.003.

• Distribution, which contains descriptions of the posterior distributions of the parameters. For example, the marginal posterior distribution of IPI is t with a mean of 4.39, a standard deviation of 0.14, and 68 degrees of freedom.

Access properties of the posterior distribution using dot notation. For example, display the marginal posterior means by accessing the Mu property.

PosteriorMdl.Mu

ans = 4×1

  -24.2494
    4.3913
    0.0011
    2.4683

Estimate Conditional Posterior Distribution

Consider the linear regression model in Create Normal-Inverse-Gamma Conjugate Prior Model.

Create a normal-inverse-gamma conjugate prior model for the linear regression parameters. Specify the number of predictors p, and the names of the regression coefficients.

p = 3;
PriorMdl = bayeslm(p,'ModelType','conjugate','VarNames',["IPI" "E" "WR"])

PriorMdl = 
  conjugateblm with properties:

    NumPredictors: 3
        Intercept: 1
         VarNames: {4x1 cell}
               Mu: [4x1 double]
                V: [4x4 double]
                A: 3
                B: 1

 
           |  Mean     Std            CI95         Positive       Distribution     
-----------------------------------------------------------------------------------
 Intercept |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 IPI       |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 E         |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 WR        |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 Sigma2    | 0.5000   0.5000    [ 0.138,  1.616]     1.000   IG(3.00,    1)        
 

Load the Nelson-Plosser data set. Create variables for the response and predictor series.

load Data_NelsonPlosser
X = DataTable{:,PriorMdl.VarNames(2:end)};
y = DataTable{:,'GNPR'};

Estimate the conditional posterior distribution of $$\beta$$ given the data and $${\sigma }^2=2$$, and return the estimation summary table to access the estimates.

[Mdl,Summary] = estimate(PriorMdl,X,y,'Sigma2',2);

Method: Analytic posterior distributions
Conditional variable: Sigma2 fixed at   2
Number of observations: 62
Number of predictors:   4
 
           |   Mean      Std          CI95         Positive     Distribution    
--------------------------------------------------------------------------------
 Intercept | -24.2494  1.8695  [-27.914, -20.585]    0.000   N (-24.25, 1.87^2) 
 IPI       |   4.3913  0.0301   [ 4.332,  4.450]     1.000   N (4.39, 0.03^2)   
 E         |   0.0011  0.0001   [ 0.001,  0.001]     1.000   N (0.00, 0.00^2)   
 WR        |   2.4683  0.0743   [ 2.323,  2.614]     1.000   N (2.47, 0.07^2)   
 Sigma2    |    2       0       [ 2.000,  2.000]     1.000   Fixed value        
 

estimate displays a summary of the conditional posterior distribution of $$\beta$$. Because ${\sigma }^2$ is fixed at 2 during estimation, inferences on it are trivial.

Extract the mean vector and covariance matrix of the conditional posterior of $\beta$ from the estimation summary table.

condPostMeanBeta = Summary.Mean(1:(end - 1))

condPostMeanBeta = 4×1

  -24.2494
    4.3913
    0.0011
    2.4683

CondPostCovBeta = Summary.Covariances(1:(end - 1),1:(end - 1))

CondPostCovBeta = 4×4

    3.4950    0.0350   -0.0001    0.0241
    0.0350    0.0009   -0.0000   -0.0013
   -0.0001   -0.0000    0.0000   -0.0000
    0.0241   -0.0013   -0.0000    0.0055

Display Mdl.

Mdl

Mdl = 
  conjugateblm with properties:

    NumPredictors: 3
        Intercept: 1
         VarNames: {4x1 cell}
               Mu: [4x1 double]
                V: [4x4 double]
                A: 3
                B: 1

 
           |  Mean     Std            CI95         Positive       Distribution     
-----------------------------------------------------------------------------------
 Intercept |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 IPI       |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 E         |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 WR        |  0      70.7107  [-141.273, 141.273]    0.500   t (0.00, 57.74^2,  6) 
 Sigma2    | 0.5000   0.5000    [ 0.138,  1.616]     1.000   IG(3.00,    1)        
 

Because estimate computes the conditional posterior distribution, it returns the original prior model, not the posterior, in the first position of the output argument list.

Estimate Posterior Probability Using Monte Carlo Simulation

Consider the linear regression model in Estimate Marginal Posterior Distributions.

Create a prior model for the regression coefficients and disturbance variance, then estimate the marginal posterior distributions.

p = 3;
PriorMdl = bayeslm(p,'ModelType','conjugate','VarNames',["IPI" "E" "WR"]);

load Data_NelsonPlosser
X = DataTable{:,PriorMdl.VarNames(2:end)};
y = DataTable{:,'GNPR'};

PosteriorMdl = estimate(PriorMdl,X,y);

Method: Analytic posterior distributions
Number of observations: 62
Number of predictors:   4
Log marginal likelihood: -259.348
 
           |   Mean      Std          CI95        Positive       Distribution      
-----------------------------------------------------------------------------------
 Intercept | -24.2494  8.7821  [-41.514, -6.985]    0.003   t (-24.25, 8.65^2, 68) 
 IPI       |   4.3913  0.1414   [ 4.113,  4.669]    1.000   t (4.39, 0.14^2, 68)   
 E         |   0.0011  0.0003   [ 0.000,  0.002]    1.000   t (0.00, 0.00^2, 68)   
 WR        |   2.4683  0.3490   [ 1.782,  3.154]    1.000   t (2.47, 0.34^2, 68)   
 Sigma2    |  44.1347  7.8020   [31.427, 61.855]    1.000   IG(34.00, 0.00069)     
 

Extract the posterior mean of $\beta$ from the posterior model, and the posterior covariance of $\beta$ from the estimation summary returned by summarize.

estBeta = PosteriorMdl.Mu;
Summary = summarize(PosteriorMdl);
estBetaCov = Summary.Covariances{1:(end - 1),1:(end - 1)};

Suppose that if the coefficient of real wages (WR) is below 2.5, then a policy is enacted. Although the posterior distribution of WR is known, and so you can calculate probabilities directly, you can estimate the probability using Monte Carlo simulation instead.

Draw 1e6 samples from the marginal posterior distribution of $$\beta$$.

NumDraws = 1e6;
rng(1);
BetaSim = simulate(PosteriorMdl,'NumDraws',NumDraws);

BetaSim is a 4-by- 1e6 matrix containing the draws. Rows correspond to the regression coefficient and columns to successive draws.

Isolate the draws corresponding to the coefficient of WR, and then identify which draws are less than 2.5.

isWR = PosteriorMdl.VarNames == "WR";
wrSim = BetaSim(isWR,:);
isWRLT2p5 = wrSim < 2.5;

Find the marginal posterior probability that the regression coefficient of WR is below 2.5 by computing the proportion of draws that are less than 2.5.

probWRLT2p5 = mean(isWRLT2p5)

probWRLT2p5 = 
0.5362

The posterior probability that the coefficient of real wages is less than 2.5 is about 0.54.

The marginal posterior distribution of the coefficient of WR is a $$t_{68}$$, but centered at 2.47 and scaled by 0.34. Directly compute the posterior probability that the coefficient of WR is less than 2.5.

center = estBeta(isWR);
stdBeta = sqrt(diag(estBetaCov));
scale = stdBeta(isWR);
t = (2.5 - center)/scale;
dof = 68;
directProb = tcdf(t,dof)

directProb = 
0.5361

The posterior probabilities are nearly identical.

Copyright 2018 The MathWorks, Inc.

Forecast Responses Using Posterior Predictive Distribution

Consider the linear regression model in Estimate Marginal Posterior Distributions.

Create a prior model for the regression coefficients and disturbance variance, then estimate the marginal posterior distributions. Hold out the last 10 periods of data from estimation so you can use them to forecast real GNP. Turn the estimation display off.

p = 3;
PriorMdl = bayeslm(p,'ModelType','conjugate','VarNames',["IPI" "E" "WR"]);

load Data_NelsonPlosser
fhs = 10; % Forecast horizon size
X = DataTable{1:(end - fhs),PriorMdl.VarNames(2:end)};
y = DataTable{1:(end - fhs),'GNPR'};
XF = DataTable{(end - fhs + 1):end,PriorMdl.VarNames(2:end)}; % Future predictor data
yFT = DataTable{(end - fhs + 1):end,'GNPR'};                  % True future responses

PosteriorMdl = estimate(PriorMdl,X,y,'Display',false);

Forecast responses using the posterior predictive distribution and using the future predictor data XF. Plot the true values of the response and the forecasted values.

yF = forecast(PosteriorMdl,XF);

figure;
plot(dates,DataTable.GNPR);
hold on
plot(dates((end - fhs + 1):end),yF)
h = gca;
hp = patch([dates(end - fhs + 1) dates(end) dates(end) dates(end - fhs + 1)],...
    h.YLim([1,1,2,2]),[0.8 0.8 0.8]);
uistack(hp,'bottom');
legend('Forecast Horizon','True GNPR','Forecasted GNPR','Location','NW')
title('Real Gross National Product');
ylabel('rGNP');
xlabel('Year');
hold off

yF is a 10-by-1 vector of future values of real GNP corresponding to the future predictor data.

Estimate the forecast root mean squared error (RMSE).

frmse = sqrt(mean((yF - yFT).^2))

frmse = 
25.5397

The forecast RMSE is a relative measure of forecast accuracy. Specifically, you estimate several models using different assumptions. The model with the lowest forecast RMSE is the best-performing model of the ones being compared.

Copyright 2018 The MathWorks, Inc.

More About

Bayesian Linear Regression Model

A Bayesian linear regression model treats the parameters β and σ² in the multiple linear regression (MLR) model y_t = x_tβ + ε_t as random variables.

For times t = 1,...,T:

• y_t is the observed response.

• x_t is a 1-by-(p + 1) row vector of observed values of p predictors. To accommodate a model intercept, x_1t = 1 for all t.

• β is a (p + 1)-by-1 column vector of regression coefficients corresponding to the variables that compose the columns of x_t.

• ε_t is the random disturbance with a mean of zero and Cov(ε) = σ²I_T×T, while ε is a T-by-1 vector containing all disturbances. These assumptions imply that the data likelihood is

  $$\ell \left(\beta ,{\sigma }^2|y,x\right)={\displaystyle \prod _{t=1}^T\varphi \left(y_t;x_t\beta ,{\sigma }^2\right).}$$

  ϕ(y_t;x_tβ,σ²) is the Gaussian probability density with mean x_tβ and variance σ² evaluated at y_t;.

Before considering the data, you impose a joint prior distribution assumption on (β,σ²). In a Bayesian analysis, you update the distribution of the parameters by using information about the parameters obtained from the likelihood of the data. The result is the joint posterior distribution of (β,σ²) or the conditional posterior distributions of the parameters.

Algorithms

You can reset all model properties using dot notation, for example, PriorMdl.V = diag(Inf(3,1)). For property resets, conjugateblm does minimal error checking of values. Minimizing error checking has the advantage of reducing overhead costs for Markov chain Monte Carlo simulations, which results in efficient execution of the algorithm.

Alternatives

The bayeslm function can create any supported prior model object for Bayesian linear regression.

Version History

Introduced in R2017a