filter

Filter disturbances through vector autoregression (VAR) model

Syntax

Y = filter(Mdl,Z)

Y = filter(Mdl,Z,Name,Value)

[Y,E] = filter(___)

Description

Y = filter(Mdl,Z) returns the multivariate response series Y, which results from filtering the underlying multivariate disturbance series Z. The Z series are associated with the model innovations process through the fully specified VAR(p) model Mdl.

Y = filter(Mdl,Z,Name,Value) uses additional options specified by one or more name-value arguments. For example, you can specify exogenous predictor data or whether to scale the disturbances by the lower triangular Cholesky factor of the model innovations covariance matrix.

[Y,E] = filter(___) returns the multivariate model innovations series E using any of the input arguments in the previous syntaxes.

Examples

Filter Disturbances Through VAR(4) Model

Fit a VAR(4) model to the consumer price index (CPI) and unemployment rate data. Then, simulate responses by filtering a random series of Gaussian distributed disturbances through the estimated model.

Load the Data_USEconModel data set.

load Data_USEconModel

Plot the two series on separate plots.

figure;
plot(DataTable.Time,DataTable.CPIAUCSL);
title('Consumer Price Index');
ylabel('Index');
xlabel('Date');

figure;
plot(DataTable.Time,DataTable.UNRATE);
title('Unemployment rate');
ylabel('Percent');
xlabel('Date');

Stabilize the CPI by converting it to a series of growth rates. Synchronize the two series by removing the first observation from the unemployment rate series. Create a new data set containing the transformed variables, and do not include any rows containing at least one missing observation.

rcpi = price2ret(DataTable.CPIAUCSL);
unrate = DataTable.UNRATE(2:end);
idx = all(~ismissing([rcpi unrate]),2);
Data = array2timetable([rcpi(idx) unrate(idx)],...
    'RowTimes',DataTable.Time(idx),'VariableNames',{'rcpi','unrate'});

Create a default VAR(4) model using the shorthand syntax.

Mdl = varm(2,4);

Estimate the model using the entire data set.

EstMdl = estimate(Mdl,Data.Variables);

EstMdl is a fully specified, estimated varm model object.

Generate a numobs-by-2 series of random Gaussian distributed values, where numobs is the number of observations in the data.

numobs = size(Data,1);
rng(1) % For reproducibility
Z = mvnrnd(zeros(Mdl.NumSeries,1),eye(Mdl.NumSeries),numobs);

To simulate responses, filter the disturbances through the estimated model.

Y = filter(EstMdl,Z);

Y is a 245-by-2 matrix of simulated responses. The first and second columns contain the simulated CPI growth rate and unemployment rate, respectively.

Plot the simulated and true responses.

figure;
plot(Data.Time,Y(:,1));
hold on;
plot(Data.Time,Data.rcpi)
ylabel('Growth rate')
xlabel('Date')
title('CPI Growth Rate');
legend('Simulation','True')

figure;
plot(Data.Time,Y(:,2));
hold on;
plot(Data.Time,Data.unrate)
ylabel('Percent');
xlabel('Date')
title('Unemployment Rate');
legend('Simulation','True')

Filter Multiple Disturbance Paths

Estimate a VAR(4) model of the consumer price index (CPI), the unemployment rate, and the gross domestic product (GDP). Include a linear regression component containing the current quarter and the last four quarters of government consumption expenditures and investment (GCE). Pass multiple multivariate Gaussian disturbance paths through the estimated model.

Load the Data_USEconModel data set. Compute the real GDP.

load Data_USEconModel
DataTable.RGDP = DataTable.GDP./DataTable.GDPDEF*100;

Plot all variables on separate plots.

figure;
subplot(2,2,1)
plot(DataTable.Time,DataTable.CPIAUCSL);
ylabel('Index');
title('Consumer Price Index');
subplot(2,2,2)
plot(DataTable.Time,DataTable.UNRATE);
ylabel('Percent');
title('Unemployment Rate');
subplot(2,2,3)
plot(DataTable.Time,DataTable.RGDP);
ylabel('Output');
title('Real Gross Domestic Product');
subplot(2,2,4)
plot(DataTable.Time,DataTable.GCE);
ylabel('Billions of $');
title('Government Expenditures');

Stabilize the CPI, GDP, and GCE by converting each to a series of growth rates. Synchronize the unemployment rate series with the others by removing its first observation.

inputVariables = {'CPIAUCSL' 'RGDP' 'GCE'};
Data = varfun(@price2ret,DataTable,'InputVariables',inputVariables);
Data.Properties.VariableNames = inputVariables;
Data.UNRATE = DataTable.UNRATE(2:end);

Expand the GCE rate series to a matrix that includes its current value and up through four lagged values. Remove the GCE variable from Data.

rgcelag4 = lagmatrix(Data.GCE,0:4);
Data.GCE = [];

Create a default VAR(4) model using the shorthand syntax.

Mdl = varm(3,4);
Mdl.SeriesNames = {'rcpi' 'unrate' 'rgdpg'};

Estimate the model using the entire sample. Specify the GCE matrix as data for the regression component.

EstMdl = estimate(Mdl,Data.Variables,'X',rgcelag4);

Generate 1000 paths of numobs observations from a 3-D Gaussian distribution. numobs is the number of observations in the data without any missing values.

numpaths = 1000;
numseries = Mdl.NumSeries;
idx = all(~ismissing([Data array2table(rgcelag4)]),2);
numobs = sum(idx);
rng(1);
Z = mvnrnd(zeros(Mdl.NumSeries,1),eye(Mdl.NumSeries),numobs*numpaths);
Z = reshape(Z,[numobs,3,numpaths]);

Filter the disturbances through the estimated model. Supply the predictor data. Return the innovations (scaled disturbances).

[Y,E] = filter(EstMdl,Z,'X',rgcelag4);

Y and E are 244-by-3-by-1000 matrices of filtered responses and scaled disturbances, respectively. The columns correspond to the CPI growth rate, unemployment rate, and GDP growth rate, respectively. filter applies the same predictor data to all paths.

For each time point, compute the mean vector of the filtered responses among all paths.

MeanFilt = mean(Y,3);

MeanFilt is a 244-by-3 matrix containing the average of the responses at each time point.

Plot the filtered responses, their averages, and the data.

Data = Data(idx,:);

figure;
for j = 1:Mdl.NumSeries
    subplot(2,2,j)
    plot(Data.Time,squeeze(Y(:,j,:)),'Color',[0.8,0.8,0.8])
    title(Mdl.SeriesNames{j});
    hold on
    h1 = plot(Data.Time,Data{:,j});
    h2 = plot(Data.Time,MeanFilt(:,j));
    hold off
end

hl = legend([h1 h2],'Data','Mean');
hl.Location = 'none';
hl.Position = [0.6 0.25 hl.Position(3:4)];

Input Arguments

Mdl — VAR model

varm model object

VAR model, specified as a varm model object created by varm or estimate. Mdl must be fully specified.

Z — Underlying multivariate disturbance series

numeric matrix | numeric array

Underlying multivariate disturbance series associated with the model innovations process, specified as a numobs-by-numseries numeric matrix or a numobs-by-numseries-by-numpaths numeric array.

numobs is the sample size. numseries is the number of disturbance series (Mdl.NumSeries). numpaths is the number of disturbance paths.

Rows correspond to sampling times, and the last row contains the latest set of disturbances.

Columns correspond to individual disturbance series for response variables.

Pages correspond to separate, independent paths. For a numeric matrix, Z is a single numseries-dimensional path of disturbance series. For a 3-D array, each page of Z represents a separate numseries-dimensional path. Among all pages, disturbances in corresponding rows occur at the same time.

The 'Scale' name-value pair argument specifies whether to scale the disturbances before filter filters them through Mdl. For more details, see Scale.

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: 'Scale',false,'X',X does not scale Z by the lower triangular Cholesky factor of the model covariance matrix before filtering, and uses the matrix X as predictor data in the regression component.

Y0 — Presample responses

numeric matrix | numeric array

Presample responses that provide initial values for the model Mdl, specified as the comma-separated pair consisting of 'Y0' and a numpreobs-by-numseries numeric matrix or a numpreobs-by-numseries-by-numprepaths numeric array.

numpreobs is the number of presample observations. numprepaths is the number of presample response paths.

Rows correspond to presample observations, and the last row contains the latest presample observation. Y0 must have at least Mdl.P rows. If you supply more rows than necessary, filter uses the latest Mdl.P observations only.

Columns must correspond to the response series in Y.

Pages correspond to separate, independent paths.

• If Y0 is a matrix, then filter applies it to each path (page) in Y. Therefore, all paths in Y derive from common initial conditions.

• Otherwise, filter applies Y0(:,:,j) to Y(:,:,j). Y0 must have at least numpaths pages, and filter uses only the first numpaths pages.

Among all pages, observations in a particular row occur at the same time.

By default, filter sets any necessary presample observations.

• For stationary VAR processes without regression components, filter uses the unconditional mean $$\mu ={\Phi }^{-1}(L)c.$$

• For nonstationary processes or models containing a regression component, filter sets presample observations to an array composed of zeros.

Data Types: double

X — Predictor data

numeric matrix

Predictor data for the regression component in the model, specified as the comma-separated pair consisting of 'X' and a numeric matrix containing numpreds columns.

numpreds is the number of predictor variables (size(Mdl.Beta,2)).

Rows correspond to observations, and the last row contains the latest observation. X must have at least as many observations as Z. If you supply more rows than necessary, filter uses only the latest observations. filter does not use the regression component in the presample period.

Columns correspond to individual predictor variables. All predictor variables are present in the regression component of each response equation.

filter applies X to each path (page) in Z; that is, X represents one path of observed predictors.

By default, filter excludes the regression component, regardless of its presence in Mdl.

Data Types: double

Scale — Flag indicating whether to scale disturbances

true (default) | false

Flag indicating whether to scale disturbances by the lower triangular Cholesky factor of the model covariance matrix, specified as the comma-separated pair consisting of 'Scale' and true or false.

For each page j = 1,...,numpaths, filter filters the numobs-by-numseries matrix of innovations E(:,:,j) through the VAR(p) model Mdl, according to these conditions.

• If Scale is true, then E(:,:,j) = L*Z(:,:,j) and L = chol(Mdl.Covariance,'lower').

• If Scale is false, then E(:,:,j) = Z(:,:,j).

Example: 'Scale',false

Data Types: logical

Note

NaN values in Z, Y0, and X indicate missing values. filter removes missing values from the data by list-wise deletion.

1. If Z is a 3-D array, then filter horizontally concatenates the pages of Z to form a numobs-by-numpaths*numseries matrix.

2. If a regression component is present, then filter horizontally concatenates X to Z to form a numobs-by-(numpaths*numseries + numpreds) matrix. filter assumes that the last rows of each series occur at the same time.

3. filter removes any row that contains at least one NaN from the concatenated data.

4. filter applies steps 1 and 3 to the presample paths in Y0.

This process ensures that the filtered responses and innovations of each path are the same size and are based on the same observation times. In the case of missing observations, the results obtained from multiple paths of Z can differ from the results obtained from each path individually.

This type of data reduction reduces the effective sample size.

Output Arguments

Y — Filtered multivariate response series

numeric matrix | numeric array

Filtered multivariate response series, returned as a numobs-by-numseries numeric matrix or a numobs-by-numseries-by-numpaths numeric array. Y represents the continuation of the presample responses in Y0.

E — Multivariate model innovations series

numeric matrix | numeric array

Multivariate model innovations series, returned as a numobs-by-numseries numeric matrix or a numobs-by-numseries-by-numpaths numeric array. For details on the value of E, see Scale.

Algorithms

• filter computes Y and E using this process for each page j in Z.

  1. If Scale is true, then E(:,:,j) = L*Z(:,:,j), where L = chol(Mdl.Covariance,'lower'). Otherwise, E(:,:,j) = Z(:,:,j). Set e_t = E(:,:,j).

  2. Y(:,:,j) is y_t in this system of equations.

     $$y_t={\widehat{\Phi }}^{-1}(L)\left(\widehat{c}+\widehat{\delta }t+e_t\right).$$

     For variable definitions, see More About.

• filter generalizes simulate. Both functions filter a disturbance series through a model to produce responses and innovations. However, whereas simulate generates a series of mean-zero, unit-variance, independent Gaussian disturbances Z to form innovations E = L*Z, filter enables you to supply disturbances from any distribution.

• filter uses this process to determine the time origin t₀ of models that include linear time trends.

  • If you do not specify Y0, then t₀ = 0.

  • Otherwise, filter sets t₀ to size(Y0,1) – Mdl.P. Therefore, the times in the trend component are t = t₀ + 1, t₀ + 2,..., t₀ + numobs, where numobs is the effective sample size (size(Y,1) after filter removes missing values). This convention is consistent with the default behavior of model estimation in which estimate removes the first Mdl.P responses, reducing the effective sample size. Although filter explicitly uses the first Mdl.P presample responses in Y0 to initialize the model, the total number of observations in Y0 and Y (excluding missing values) determines t₀.

References

[1] Hamilton, James D. Time Series Analysis. Princeton, NJ: Princeton University Press, 1994.

[2] Johansen, S. Likelihood-Based Inference in Cointegrated Vector Autoregressive Models. Oxford: Oxford University Press, 1995.

[3] Juselius, K. The Cointegrated VAR Model. Oxford: Oxford University Press, 2006.

[4] Lütkepohl, H. New Introduction to Multiple Time Series Analysis. Berlin: Springer, 2005.

Version History

Introduced in R2017a