Wald test of model specification
returns
a logical value (h
= waldtest(r
,R
,EstCov
)h
) with the rejection decision
from conducting a Wald
test of model specification.
waldtest
constructs the test statistic using
the restriction function and its Jacobian, and the value of the unrestricted
model covariance estimator, all evaluated at the unrestricted parameter
estimates (r
, R
, and EstCov
,
respectively).
If any input argument is a cell vector of length k >
1, then the other input arguments must be cell arrays of length k. waldtest
(r
,R
,EstCov
)
treats each cell as a separate, independent test, and returns a vector
of rejection decisions.
If any input argument is a row vector, then the software returns output arguments as row vectors.
Check for significant lag effects in a time series regression model.
Load the U.S. GDP data set.
load Data_GDP
Plot the GDP against time.
plot(dates,Data) datetick
The series seems to increase exponentially.
Transform the data using the natural logarithm.
logGDP = log(Data);
logGDP
is increasing in time, so assume that there is a significant lag 1 effect. To use the Wald test to check if there is a significant lag 2 effect, you need the:
Estimated coefficients of the unrestricted model
Restriction function evaluated at the unrestricted model coefficient values
Jacobian of the restriction function evaluated at the unrestricted model coefficient values
Estimated, unrestricted parameter covariance matrix.
The unrestricted model is
Estimate the coefficients of the unrestricted model.
LagLGDP = lagmatrix(logGDP,1:2); UMdl = fitlm(table(LagLGDP(:,1),LagLGDP(:,2),logGDP));
UMdl
is a fitted LinearModel
model. It contains, among other things, the fitted coefficients of the unrestricted model.
The restriction is . Therefore, the restiction function (r) and Jacobian (R) are:
Specify r, R, and the estimated, unrestricted parameter covariance matrix.
r = UMdl.Coefficients.Estimate(3); R = [0 0 1]; EstParamCov = UMdl.CoefficientCovariance;
Test for a significant lag 2 effect using the Wald test.
[h,pValue] = waldtest(r,R,EstParamCov)
h = 1 pValue = 1.2521e-07
h = 1
indicates that the null, restricted hypothesis (
) should be rejected in favor of the alternative, unrestricted hypothesis. pValue
is quite small, which suggests that there is strong evidence for this result.
Test whether there are significant ARCH effects in a simulated response series using waldtest
.
Suppose that the model for the simulated data is AR(1) with an ARCH(1) variance. Symbolically, the model is
where
is Gaussian with mean 0 and variance 1.
Specify the model for the simulated data.
VarMdl = garch('ARCH',0.5,'Constant',1); Mdl = arima('Constant',0,'Variance',VarMdl,'AR',0.9);
Mdl
is a fully specified AR(1) model with an ARCH(1) variance.
Simulate presample and effective sample responses from Mdl
.
T = 100; rng(1); % For reproducibility n = 2; % Number of presample observations required for the Jacobian [y,epsilon,condVariance] = simulate(Mdl,T + n); psI = 1:n; % Presample indices esI = (n + 1):(T + n); % Estimation sample indices
epsilon
is the random path of innovations from VarMdl
. The software filters epsilon
through Mdl
to yield the random response path y
.
Specify the unrestricted model assuming that the conditional mean model is
where
. Fit the simulated data (y
) to the unrestricted model using the presample observations.
UVarMdl = garch(0,1); UMdl = arima('ARLags',1,'Variance',UVarMdl); [UEstMdl,UEstParamCov] = estimate(UMdl,y(esI),'Y0',y(psI),... 'E0',epsilon(psI),'V0',condVariance(psI),'Display','off');
UEstMdl
is the the fitted, unrestricted model, and UEstParamCov
is the estimated parameter covariance of the unrestricted model parameters.
The null hypothesis is that , i.e., the restricted model is AR(1) with Gaussian innovations that have mean 0 and constant variance. Therefore, the restriction function is , where . The components of the Wald test are:
The restriction function evaluated at the unrestricted parameter estimates is .
The Jacobian of r evaluated at the unrestricted model parameters is .
The unrestricted model estimated parameter covariance matrix is UEstParamCov
.
Specify r and R.
r = UEstMdl.Variance.ARCH{1}; R = [0, 0, 0, 1];
Test the null hypothesis that
at the 1% significance level using waldtest
.
[h,pValue,stat,cValue] = waldtest(r,R,UEstParamCov,0.01)
h = 0 pValue = 0.0549 stat = 3.6846 cValue = 6.6349
h = 0
indicates that the null, restricted model should not be rejected in favor of the alternative, unrestricted model. This result is consistent with the model for the simulated data.
Assess model specifications by testing down among multiple restricted models using simulated data. The true model is the ARMA(2,1)
where is Gaussian with mean 0 and variance 1.
Specify the true ARMA(2,1) model, and simulate 100 response values.
TrueMdl = arima('AR',{0.9,-0.5},'MA',0.7,... 'Constant',3,'Variance',1); T = 100; rng(1); % For reproducibility y = simulate(TrueMdl,T);
Specify the unrestricted model and the names of the candidate models for testing down.
UMdl = arima(2,0,2); RMdlNames = {'ARMA(2,1)','AR(2)','ARMA(1,2)','ARMA(1,1)',... 'AR(1)','MA(2)','MA(1)'};
UMdl
is the unrestricted, ARMA(2,2) model. RMdlNames
is a cell array of strings containing the names of the restricted models.
Fit the unrestricted model to the simulated data.
[UEstMdl,UEstParamCov] = estimate(UMdl,y,'Display','off');
UEstMdl
is the fitted, unrestricted model, and UEstParamCov
is the estimated parameter covariance matrix.
The unrestricted model has six parameters. To construct the restriction function and its Jacobian, you must know the order of the parameters in UEstParamCov
. For this arima
model, the order is
.
Each candidate model corresponds to a restriction function. Put the restriction function vectors into separate cells of a cell vector.
rf1 = UEstMdl.MA{2}; % ARMA(2,1) rf2 = cell2mat(UEstMdl.MA)'; % AR(2) rf3 = UEstMdl.AR{2}; % ARMA(1,2) rf4 = [UEstMdl.AR{2};UEstMdl.MA{2}]'; % ARMA(1,1) rf5 = [UEstMdl.AR{2};cell2mat(UEstMdl.MA)']; % AR(1) rf6 = cell2mat(UEstMdl.AR)'; % MA(2) rf7 = [cell2mat(UEstMdl.AR)';UEstMdl.MA{2}]; % MA(1) r = {rf1;rf2;rf3;rf4;rf5;rf6;rf7};
r
is a 7-by-1 cell vector of vectors corresponding to the restriction function for the candidate models.
Put the Jacobian of each restriction function into separate, corresponding cells of a cell vector. The order of the elements in the Jacobian must correspond to the order of the elements in UEstParamCov
.
J1 = [0 0 0 0 1 0]; % ARMA(2,1) J2 = [0 0 0 1 0 0; 0 0 0 0 1 0]; % AR(2) J3 = [0 1 0 0 0 0]; % ARMA(1,2) J4 = [0 1 0 0 0 0; 0 0 0 0 1 0]; % ARMA(1,1) J5 = [0 1 0 0 0 0; 0 0 0 1 0 0; 0 0 0 0 1 0]; % AR(1) J6 = [1 0 0 0 0 0; 0 1 0 0 0 0]; % MA(2) J7 = [1 0 0 0 0 0; 0 1 0 0 0 0; 0 0 0 0 1 0]; % MA(1) R = {J1;J2;J3;J4;J5;J6;J7};
R
is a 7-by-1 cell vector of vectors corresponding to the restriction function for the candidate models.
Put the estimated parameter covariance matrix in each cell of a 7-by-1 cell vector.
EstCov = cell(7,1); % Preallocate for j = 1:length(EstCov) EstCov{j} = UEstParamCov; end
Apply the Wald test at a 1% significance level to find the appropriate, restricted model specifications.
alpha = .01; h = waldtest(r,R,EstCov,alpha); RestrictedModels = RMdlNames(~h)
RestrictedModels = 'ARMA(2,1)' 'ARMA(1,2)' 'ARMA(1,1)' 'MA(2)' 'MA(1)'
RestrictedModels
lists the most appropriate restricted models.
You can test down again, but use ARMA(2,1) as the unrestricted model. In this case, you must remove MA(2) from the possible restricted models.
Test whether the parameters of a nested model have a nonlinear relationship.
Load the Deutschmark/British Pound bilateral spot exchange rate data set.
load Data_MarkPound
The data set (Data
) contains a time series of prices.
Convert the prices to returns, and plot the return series.
returns = price2ret(Data); figure plot(returns) axis tight ylabel('Returns') xlabel('Days, 02Jan1984 - 31Dec1991') title('{\bf Deutschmark/British Pound Bilateral Spot Exchange Rate}')
The returns series shows signs of heteroscedasticity.
Suppose that a GARCH(1,1) model is an appropriate model for the data. Fit a GARCH(1,1) model to the data including a constant.
Mdl = garch(1,1); [EstMdl,EstParamCov] = estimate(Mdl,returns); g1 = EstMdl.GARCH{1}; a1 = EstMdl.ARCH{1};
GARCH(1,1) Conditional Variance Model: ---------------------------------------- Conditional Probability Distribution: Gaussian Standard t Parameter Value Error Statistic ----------- ----------- ------------ ----------- Constant 1.05346e-06 3.50483e-07 3.00575 GARCH{1} 0.806576 0.0129095 62.4794 ARCH{1} 0.154357 0.0115746 13.3358
g1
is the estimated GARCH effect, and a1
is the estimated ARCH effect.
The following might represent relationships between the GARCH and ARCH coefficients:
where is the GARCH effect and is the ARCH effect. Specify these relationships as the restriction function , evaluated at the unrestricted model parameter estimates. This specification defines a nested, restricted model.
r = [g1*a1; g1+a1] - 1;
Specify the Jacobian of the restriction function vector.
R = [0, a1, g1;0, 1, 1];
Conduct a Wald test to assess whether there is sufficient evidence to reject the restricted model.
[h,pValue,stat,cValue] = waldtest(r,R,EstParamCov)
h = 1 pValue = 0 stat = 1.4594e+04 cValue = 5.9915
h = 1
indicates that there is sufficient evidence to reject the restricted model in favor of the unrestricted model. pValue = 0
indicates that the evidence for rejecting the restricted model is strong.
r
— Restriction functionsscalar | vector | cell vector of scalars or vectorsRestriction functions corresponding to restricted models, specified as a scalar, vector, or cell vector of scalars or vectors.
If r
is a q-vector
or a singleton cell array containing a q-vector,
then the software conducts one Wald test. q must
be less than the number of unrestricted model parameters.
If r
is a cell vector of length k >
1, and cell j contains a q_{j}-vector, j =
1,...,k, then the software conducts k independent
Wald tests. Each q_{j} must
be less than the number of unrestricted model parameters.
Data Types: double
| cell
R
— Restriction function Jacobiansrow vector | matrix | cell vector of row vectors or matricesRestriction function Jacobians, specified as a row vector, matrix, or cell vector of row vectors or matrices.
Suppose r_{1},...,r_{q} are the q restriction functions, and the unrestricted model parameters are θ_{1},...,θ_{p}. Then, the restriction function Jacobian is
$$R=\left(\begin{array}{ccc}\frac{\partial {r}_{1}}{\partial {\theta}_{1}}& \dots & \frac{\partial {r}_{1}}{\partial {\theta}_{p}}\\ \vdots & \ddots & \vdots \\ \frac{\partial {r}_{q}}{\partial {\theta}_{1}}& \cdots & \frac{\partial {r}_{q}}{\partial {\theta}_{p}}\end{array}\right).$$
If R
is a q-by-p matrix
or a singleton cell array containing a q-by-p matrix,
then the software conducts one Wald test. q must
be less than p, which is the number of unrestricted
model parameters.
If R
is a cell vector of length k >
1, and cell j contains a q_{j}-by-p_{j} matrix, j =
1,...,k, then the software conducts k independent
Wald tests. Each q_{j} must
be less than p_{j}, which is
the number of unrestricted parameters in model j.
Data Types: double
| cell
EstCov
— Unrestricted model parameter covariance estimatematrix | cell vector of matricesUnrestricted model parameter covariance estimates, specified as a matrix or cell vector of matrices.
If EstCov
is a p-by-p matrix
or a singleton cell array containing a p-by-p matrix,
then the software conducts one Wald test. p is
the number of unrestricted model parameters.
If EstCov
is a cell vector of length k >
1, and cell j contains a p_{j}-by-p_{j} matrix, j =
1,...,k, then the software conducts k independent
Wald tests. Each p_{j} is the
number of unrestricted parameters in model j.
Data Types: double
| cell
alpha
— Nominal significance levels0.05 (default) | scalar | vectorNominal significance levels for the hypothesis tests, specified as a scalar or vector.
Each element of alpha
must be greater than
0 and less than 1.
When conducting k > 1 tests,
If alpha
is a scalar, then the
software expands it to a k-by-1 vector.
If alpha
is a vector, then it must
have length k.
Data Types: double
h
— Test rejection decisionslogical | vector of logicalsTest rejection decisions, returned as a logical value or vector of logical values with a length equal to the number of tests that the software conducts.
h = 1
indicates rejection of the
null, restricted model in favor of the alternative, unrestricted model.
h = 0
indicates failure to reject
the null, restricted model.
pValue
— Test statistic p-valuesscalar | vectorTest statistic p-values, returned as a scalar or vector with a length equal to the number of tests that the software conducts.
stat
— Test statisticsscalar | vectorTest statistics, returned as a scalar or vector with a length equal to the number of tests that the software conducts.
cValue
— Critical valuesscalar | vectorCritical values determined by alpha
, returned
as a scalar or vector with a length equal to the number of tests that
the software conducts.
The Wald test compares specifications of nested models by assessing the significance of q parameter restrictions to an extended model with p unrestricted parameters.
The test statistic is
$$W={r}^{\prime}{\left(R{\Sigma}_{\widehat{\theta}}{R}^{\prime}\right)}^{-1}r,$$
where
r is the restriction function that specifies restrictions of the form r(θ) = 0 on parameters θ in the unrestricted model, evaluated at the unrestricted model parameter estimates. In other words, r maps the p-dimensional parameter space to the q-dimensional restriction space.
In practice, r is a q-by-1 vector, where q < p.
Usually, $$r=\widehat{\theta}-{\theta}_{0}$$, where $$\widehat{\theta}$$ is the unrestricted model parameter estimates for the restricted parameters and θ_{0} holds the values of the restricted model parameters under the null hypothesis.
R is the restriction function Jacobian evaluated at the unrestricted model parameter estimates.
$${\widehat{\Sigma}}_{\widehat{\theta}}$$ is the unrestricted model parameter covariance estimator evaluated at the unrestricted model parameter estimates.
W has an asymptotic chi-square distribution with q degrees of freedom.
When W exceeds a critical value in its asymptotic distribution, the test rejects the null, restricted hypothesis in favor of the alternative, unrestricted hypothesis. The nominal significance level (α) determines the critical value.
Note: Wald tests depend on the algebraic form of the restrictions. For example, you can express the restriction ab = 1 as a – 1/b = 0, or b – 1/a = 0, or ab – 1 = 0. Each formulation leads to different test statistics. |
Estimate unrestricted univariate linear time series
models, such as arima
or garch
,
or time series regression models (regARIMA
) using estimate
.
Estimate unrestricted multivariate linear time series models using vgxvarx
.
estimate
and vgxvarx
return
parameter estimates and their covariance estimates, which you can
process and use as inputs to waldtest
.
If you cannot easily compute restricted parameter
estimates, then use waldtest
. By comparison:
lratiotest
requires both restricted
and unrestricted parameter estimates.
lmtest
requires restricted parameter
estimates.
waldtest
performs multiple, independent
tests when the restriction function vector, its Jacobian, and the
unrestricted model parameter covariance matrix (r
, R
,
and EstCov
, respectively) are equal-length cell
vectors.
If EstCov
is the same for all tests,
but r
varies, then waldtest
"tests
down" against multiple restricted models.
If EstCov
varies among tests, but r
does
not, then waldtest
"tests up" against
multiple unrestricted models.
Otherwise, waldtest
compares
model specifications pair-wise.
alpha
is nominal in that it specifies
a rejection probability in the asymptotic distribution. The actual
rejection probability is generally greater than the nominal significance.
The Wald test rejection error is generally greater than the likelihood ratio and Lagrange multiplier test rejection errors.
[1] Davidson, R. and J. G. MacKinnon. Econometric Theory and Methods. Oxford, UK: Oxford University Press, 2004.
[2] Godfrey, L. G. Misspecification Tests in Econometrics. Cambridge, UK: Cambridge University Press, 1997.
[3] Greene, W. H. Econometric Analysis. 6th ed. Upper Saddle River, NJ: Pearson Prentice Hall, 2008.
[4] Hamilton, J. D. Time Series Analysis. Princeton, NJ: Princeton University Press, 1994.