VAR, SVAR, VECM

Eric Amoo Bondzie (PhD)

 Introduction to VAR Models
☞The VAR is a workhouse multivariate time series model that
relates current observations of a variable with past observations
of itself and past observations of other variables in the system.
☞It popularised by Sims(1980) as a natural generalisation of
univariate autoregressive models.
☞Sims (1980) advocated the use of VAR models over structural
simultaneous equations models.
☞This due to the distinction between endogenous and exogenous
variables does not have to be made a priori, and ‘arbitrary’
constraints to ensure identification are not required.
 Introduction to VAR Models
☞Variables should be treated symmetrically to avoid
‘incredible identification restrictions’
☞Let the data speak for itself i.e., no apriori
assumption about exogeneity of variables.
☞Very helpful in identifying the relationship among a
set of macroeconomic models.
☞Multi-equation time series model Considers a
number of interrelated variables
 Introduction to VAR Models
☞Imposes zero restrictions on estimation of
parameters.
☞Like a reduced form a VAR is always identified.
☞Atheoretical i.e., no strict reliance on theory to
formulate the model.
☞‘Everything causes everything’
☞However, the number of estimated parameters
makes the model difficult to interpret.
 Advantages & Disadvantages of VAR Models
☞Every variable is endogenous (no incredible
exogeneity assumptions).

☞Every variable depends on the others (no incredible

exclusion restrictions).

☞Simple to estimate and use.

 Advantages & Disadvantages of VAR Models
☞It is a reduced form model; no economic
interpretation of the dynamics is possible.

☞Potentially difficult to relate VAR dynamics with

DSGE dynamics (which have an ARMA structure).

☞Can’t be used for certain policy analyses (Lucas'

critique).
 VAR Models
☞Consider a VAR of order P, denoted as VAR(P), is as
follows:

☞Where is a vector of variables, each modeled as a

function of -lags of those variables and, optionally, a
set of exogenous variables .
☞We assume that and
☞Estimations of VAR parameters requires that
variables in are covariance stationary.
 VAR Models
☞If variables are not stationary at levels, but their first
difference tend to be stationary, they may be modelled with
a vector error correction model or VECM.
☞In the absence of exogenous variables, the disturbance
variance-covariance matrix contains all relevant
information about the contemporaneous correlation among
the variables in
☞In such a situation the VAR model may specified as:
 VAR Models
is vector of I(0) variables
is a vector of constants, and
are matrices of parameters,
is the appropriate lag length of the model
is a vector of normally distributed error terms.
 Properties of VAR
☞The variables are stationary.
☞Error terms are white noise disturbances with a
constant variance.
☞Error terms are not serially correlated.
☞The structure of the system allows for feedback effects.
☞If contemporaneous effects are assumed to be zero, the
VAR is said to be in standard or reduce form and
estimation can proceed using OLS.
 Types of VAR Models

☞Reduced form or Standard VAR models

☞Recursive VAR
☞Structure VAR
 Reduced Form
☞A reduced form VAR expresses each variable as a linear
function of its own past values.
☞The past values of all other variables being considered, and a
serially uncorrelated error term.
☞Each equation is estimated by ordinary least squares (OLS)
regression.
☞The number of lagged values to include in each equation can be
determined by several different methods
☞The errors terms in these regressions are the “surprise”
movements in the variables, after taking its past values into
account.
 Recursive & Structural
☞Constructs the error terms in each regression equation to be
uncorrelated with the error in the preceding equations.
☞This is done by judiciously including some contemporaneous values
as regressors.
☞A structural VAR uses economic theory to sort out the
contemporaneous links between the variables (Bernanke, 1986;
Blanchard and Watson, 1986; Sims, 1986).
☞Structural VARs require “identifying assumptions” that allow
correlations to be interpreted causally.
☞These identifying assumptions can involve the entire VAR, so that all
the causal links in the model are spelled out, or just a single equation,
so that only a specific causal link is identified.
 Recursive & Structural
☞This produces instrumental variables which permit the
contemporaneous links to be estimated using instrumental
variables regression.
☞The number of structural VARs is limited only by the
inventiveness of the researcher.
Implementing Vector
Autoregressive (VAR)
 Choosing variables for VAR models
☞By institutional knowledge: institutions doing macroeconomic
modelling develop intuition over what variables are needed to
adequately model the system.
☞ For a small open economy, it would be hard not to have the real exchange rate
and foreign output in the list of variables appearing in the VAR.
☞Including a variable such as GDP in the VAR, it can be useful to add in a
variable like Gross National Expenditure (GNE), with the latter playing
the role of the “absorption" variable that appears in theoretical models
of open economies.
☞From theoretical models: from the New Keynesian perspective, one
would choose and .
☞ One difficulty that often arises however is that such theoretical models often
incorporate variables that are not easily measured.
 Stationarity and VAR models
☞Brookes (2010): it is important that all the variables in the
VAR process be stationary otherwise hypothesis are invalid.
☞Argue that the objective of VAR analysis is to determine
interrelationships among variables and not to determine
parameter estimates.
☞Canova (2005): If we want a constant coefficient VAR, we
need stationarity of the variables. If non-stationarities are
present a VAR representation exists, but with time varying
coefficients.
 Stationarity and VAR models
☞The way forward:
☞It depends on what the VAR is going to be used for.
☞For FORECASTING purposes, we must avoid potential
spurious regressions that may results in in spurious
forecasts.
☞We use OLS to estimate VAR, so if we are interested in
using the model for forecasting the we need that all
variables are stationary or cointegrated to avoid spurious
regression problem associated with unit roots.
 Stationarity and VAR models
☞The way forward:
☞For IDENTIFYING SHOCKS, we must be careful about the stability of
the VAR (whether it can be inverted), the reliability of our impulse
response functions and the statistical properties of the residuals.
☞For structural identifications, we are interested in consistent
coefficient estimates, so we can use the variables in levels (Sims et al.
1990).
☞According to Sims et al (1990), potential non-stationarity in VAR
under investigation should not affect the model selection.
☞Also, SVAR uses maximum likelihood estimation procedure hence
cointegration restrictions are ignored to avoid too many restrictions.
 LAG LENGTH OF VAR models
☞Ways to determine the appropriate lag length/ order:
☞By using some theoretical model.
☞By using a rule of thumb: people choosing p = 4 when
working with quarterly data and p = 6 with monthly data.
Provided n is small these are probably upper limits to the
likely order.
☞By using statistical criteria that trade off fit against the
number of parameters fitted.
 LAG LENGTH OF VAR models
☞To determine the appropriate lag length using statistical
criteria
☞ Akaike Information Criterion (AIC)
☞ Schwartz Information Criterion (SIC)
☞ Likelihood Ratio test (LR)
☞ Final Prediction Error (FPE)
☞ HQ Information Criterion (HIC)
☞ Maximum lag for autoregressive models
☞You choose the lag length that soaks or expunges serial
correlation in the residuals
☞If maximum lag length is p, then it’s a VAR(p).
 LAG LENGTH OF VAR models
☞The correct Lag length will depend on the criteria or measure we use.
☞Researchers often use the criterion most convenient to their needs.
☞Ivanov and Kilian (2005), show that the choice of information
criterion depends on frequency of data and type of model.
☞DIFFERENCES
☞AIC is inconsistent and overestimates the true lag order with positive
probability.
☞Both SC and HQ are consistent.
☞The SC criterion is generally more conservative in terms of lag length
than AIC. i.e.: SC select shorter Lag than the other criteria.
☞HQ is typically appropriate for quarterly and monthly data.
An Example of VAR
 Diagnostic or POST estimation test
Residual tests:
Portmanteau Autocorrelation test: computes the
multivariate Box-Pierce/Ljung Box Q statistics for serial
correlation up to a specified order. Eviews reports both
tests under the null hypothesis of no serial correlation
Autocorrelation LM test: reports the multivariate LM
test statistics for residual serial correlation. Under the
null hypothesis of no serial correlation of order h the
LM test is asymptotically Chi Square distributed with
degrees of freedom.
 Diagnostic or POST estimation test
Normality tests:
 Reports the Multivariate extension for the Jarque Bera test
 For the multivariate test you must choose a factorisation of residuals that are orthogonal
to each other.
 Eviews reports the test statistics for each othorgonal component
Whites heteroskedasticity test:
Is an extension of the white’s 1980 test
 No cross terms: uses only levels and squares of regressors
 Cross terms: includes all non redundant cross products of regressors (heteroskedasticity
of an unknown form).
Stability:
Since we want to obtain a VMA from VAR, we must ensure that VAR is stable.
It is easy to check for the stability of the model using AR Roots Table or
Graph.
 Solving POST estimation errors
Autocorrelation:
Try adding further lags to the VAR.
Stability:
If the VAR is Unstable and Non-normal, adding a time trend as an
exogenous variable helps the stability of the VAR.
Normality
We may include exogenous variables, dummy variable or time trend
 VAR models - OUTPUTS
2 useful outputs from VAR models are:

Forecast Error Variance Decomposition (FEVD) or

Variance Decomposition for short.
Impulse Response Functions (IRFs)
 VAR models - Variance decompositions
FEVD separate the variation in an endogenous variable into the
contributions explained by the component shocks in the VAR.
That is, it tells us the proportion of the movement in a variable due
to its “own” shocks verses shocks to the other variables.
Enders (2010): enables us to study the variation in Y that is due to its
own shocks versus the component of the variation that is due to
shocks in other variables
FEVD also help to determine the relative importance of each
innovation in explaining the variables in the system.
It provides information about the relative importance of each
(structural) shocks in affecting the variables in the VAR.
 VAR models - Variance decompositions
To conduct variance decompositions, the AR process is inverted into an
MA process of the errors using Walds Decomposition Theorem.
Wald's Decomposition Theorem rewrites the AR process into MA
process:

into
 VAR models - Variance decompositions
If the forecast error variance is explained by shocks in the
variable itself, then the variable is exogenous
It is typical for a variable to explain almost all its forecast
error variance for short horizons and smaller proportions
at longer horizons (Enders, 2010).
It is also subject to an under-identification problem as is
the impulse response function, thus there might be need to
place additional restrictions on the system in order to
obtain the decomposition and impulse responses.
 VAR models - Variance decompositions
One such restriction is the Choleski decomposition
☞The contemporaneous value of Y has no contemporaneous effect
on X.
☞This implies an ordering of the variables
Brooks and Tsolacos (1998) and Enders (2010) - the Choleski
ordering of the variables has important ramifications on the
resulting impulse responses and variance decompositions and is
equivalent to an identifying restriction on the VAR.
VAR models - Variance decompositions
Variance Decom pos ition of GDP:
Period S.E. GDP CPI

1 0.010434662... 100 0
2 0.019854780... 99.96213939... 0.037860603...
3 0.028857777... 99.90578994... 0.094210059...
4 0.037233104... 99.85880332... 0.141196675...
5 0.044969145... 99.83029345... 0.169706540...
6 0.052124967... 99.81909349... 0.180906509...
7 0.058778024... 99.82012292... 0.179877077...
8 0.065003846... 99.82812272... 0.171877275...
9 0.070869061... 99.83916409... 0.160835907...
10 0.076429957... 99.85083276... 0.149167230...

Variance Decom pos ition of CPI:

Period S.E. GDP CPI

1 0.056077939... 0.059553773... 99.94044622...

2 0.094466846... 0.267189999... 99.73281000...
3 0.119951284... 0.698189521... 99.30181047...
4 0.134594978... 1.446056352... 98.55394364...
5 0.142060353... 2.576770222... 97.42322977...
6 0.145688445... 4.084999848... 95.91500015...
7 0.147738763... 5.881659425... 94.11834057...
8 0.149375375... 7.833843862... 92.16615613...
9 0.151016002... 9.823663778... 90.17633622...
10 0.152719587... 11.77906723... 88.22093276...

Choles ky Ordering: GDP CPI

VAR models - Impulse response functions (IRF)
Impulse response functions are a practical tool which aid in visualizing the
behaviour of the variables understudy in response to various shocks.
IRFs show how the different variables in the system respond to (identified)
shocks.
With a structural (an identified) VAR, the IRF will be depicting the responses
to the structural shocks.
Impulse responses allows for tracing the time profile of various shocks on the
variables in the VAR system.
They show the dynamics of transmission of shocks, direction and magnitude
of the shocks.
IRFs in practice are more informative than the structural parameter estimates.
In practice you should always plot your impulse responses together with
their standard deviation bands
 VAR models - Impulse response functions
Response to Cholesky One S.D. (d.f. adj usted) Innovati ons ± 2 S.E.

Response of GDP to GDP Response of GDP to GE Response of GDP to FDI

.03 .03 .03

.02 .02 .02

.01 .01 .01

.00 .00 .00

-.01 -.01 -.01

2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20

Response of GE to GDP Response of GE to GE Response of GE to FDI

.03 .03 .03

.02 .02 .02

.01 .01 .01

.00 .00 .00

-.01 -.01 -.01

2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20

Response of FDI to GD P Response of FD I to GE Response of FDI to FDI

.08 .08 .08

.06 .06 .06

.04 .04 .04

.02 .02 .02

.00 .00 .00

-.02 -.02 -.02

2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20
 Criticisms of the VAR
☞Many argue that the VAR approach is lacking in theory.
☞There is much debate on how the lag lengths should be
determined
☞It is possible to end up with a model including numerous
explanatory variables, with different signs, which has
implications for degrees of freedom.
☞Many of the parameters will be insignificant, this affects
the efficiency of a regression.
☞There is always a potential for multicollinearity with many
lags of the same variable
 SVAR Methodology
☞The main objective of SVAR models is to find out the
dynamic responses of economic variables to disturbances
by combining time series analysis and economic theory.
☞SVAR employs additional restrictions and estimation of
structural shocks to transform VAR errors into uncorrelated
structural Shocks. We begin with the SVAR specification:

where are the structural coefficients and the are the

orthonormal unobserved structural innovations with
 SVAR Methodology
☞It is easy to see the relationship between the SVAR
specification and the corresponding reduced-form VAR.
☞Assuming is invertible, we have:

☞The reduced form error structure is given by

☞Where
 SVAR Methodology
☞SVAR estimation uses estimates obtained from the reduced
form VAR, the short-run covariance relationship and any
restrictions in the above Equation and long-run
restrictions on the accumulated impulse responses to
identify and estimate the model
☞There are only free parameters in , given its symmetric
nature, only that many parameters may be estimated in the A
and B matrices.
☞As there are parameters in A and B, the order condition for
identification requires that restrictions be placed on the
elements of these matrices.
Move to EViews
 Vector Error-Correction (VEC)
Regressing a nonstationary variable upon a nonstationary variable may lead
to a so-called spurious regression, in which estimators and test statistics
are misleading.
Johansen and Jesilius enhance the VAR (p) by including the long run
components (cointegrating relations) in the VAR (p) process i.e. separating
permanent effects from transitory effects.
It specifies a VECM among variables.
The error correction model also known as the dynamics of adjustment are
estimated using the lagged differences of the data series and the lag of the
equilibrium error.
Given that:
 Vector Error-Correction (VEC)
Given that:

is the first difference lag operator

is vector of I(1) time series variables
is a set of endogenous variables
terms are stationary variables integrated of order zero I(0)
is a vector of constants, and
are matrices of parameters,
is the appropriate lag length of the model
is a vector of normally distributed shocks with a constant variance and
a zero mean.
 Vector Error-Correction (VEC)
The number of cointegration relationships will be given by the
rank of .
When the reduced rank (when the VAR(p) has a unit root) , the
there exist linearly independent sequences of that are
stationary.
Where is the cointegrating vector and is the number of
endogenous variables.
If , then no cointegration relation exist.
The matrix contains the cointegrating vectors and the speed of
adjustment to the long run equilibrium path.
 Vector Error-Correction (VEC) - FACTS
VECM is a VAR(p) that incorporates the error correction terms
to account for more fully short run dynamics.
If the VAR(p) process has unit root then is SINGULAR ( does
not exist).
If the is SINGULAR then it has reduced rank; that is .
 Testing for cointegration
JJ suggest five assumption on which test can be conducted
Given that:
1. No deterministic trends in the VAR system and the cointegrating
relationship has no intercept and no trend ();
2. No deterministic trends in the VAR system and the cointegrating
relationship has an intercept and no trend ();
3. Linear trend in the VAR system and the cointegrating relationship has
no trend but has an intercept ();
4. Linear trend in the VAR system and the cointegrating relationship only
has a deterministic trend (); and
5. A quadratic trend in the VAR and the cointegrating relationship has a
linear deterministic trend ().
 Testing for cointegration - JJ
Trace statistics tests the null hypothesis that the rank (i.e. no
cointegration) against the alternative that (i.e. there is one or more
cointegrating vectors).
The maximum Eigenvalue statistics on the other hand tests the null
hypothesis that the number of cointegrating vectors is against the
specific alternative of cointegrating vectors.
The critical values for the tests are obtained using Monte Carlo
approach
The distribution of statistics depends on two components:
The number of non stationary components under the null hypothesis
The form of the deterministic components, constant, trend or both-
has similarity with the Dickey fuller test.
 Testing for cointegration - JJ
Sometimes the two tests may give conflicting results.
Harris(1995) the maximum eigen value has a sharper alternative
hypothesis and is preferred to pin down the number of cointegrating
vectors.
The sequence of the Trace tests leads to a consistent procedure.
Cheung and Lai (1993) propose choosing cointegration rank based on
the Trace statistic. They state that the trace statistic is more robust to
skewness and excess kurtosis in residuals than the maximum Eigen
value statistic.
Enders (2010) concurs with these findings and states that when the
two tests for cointegration rank are in conflict the Trace statistic is
likely to give more reliable results.
Current practice is to only consider the Trace test.
 Practicals (1)
Pretest data
1. Pretest all variables to determine their order of
integration i.e. test for unit roots
2. Plot the variables to see if a linear time trend is likely to
appear in the data series.
 Estimating var and vec
Estimating an Unrestricted VAR
Estimating a VAR
Choosing the optimal lag length
Selecting the lag length
☞Asterisk indicates lag length selected by Information criteria.
☞If a long lag is required to make residuals white noise, reconsider the
choice of variables and look for another important explanatory variable to
include in the information set.
☞A summary of test statistics that measures the magnitude of the residual
autocorrelation in given b the Portmanteau test
☞Eviews uses Wald Lag exclusion tests to determine the default lag
☞SIC (a more stricter test)
☞AIC gives a generous lag length
☞HQ is a middle of the road
 Estimating var and vec
Estimation and Determination of Rank (Cointegration Test)
☞In unique circumstances will you consider a trend in the
Cointegration vector
☞Enter the chosen lag length
☞The test is done in specific order from the largest eigen value
to the smallest. We use the ‘Pantula Principle’ where we test
for significance until you no longer reject the null. The first
null is that there is non stationary relations in the data (r=0)…
So long as TS/MES> critical value reject the null. Use p =-values
to make decision.
If test indicate no cointegration then ESTIMATE VAR or otherwise
ESTIMATE VECM.
Vector Error Correction Estimates
Date: 02/22/20 Time: 16:39
Estimating vec  
 
Sample (adjusted): 1975Q4 2010Q4  
Included observations: 141 after adjustments  Speed of adjustment.
Standard errors in ( ) & t-statistics in [ ]
         NB: even if we are only
       
Cointegrating Eq:  CointEq1     interested in the first
 
 
 
 
 
 
 
 
cointegrating
GDP(-1)  1.000000     relationship, both
 
GE(-1)
 
-1.959302
 
 
 
 
cointegration relationships
   (0.23560)     should enter that equation
 
 
[-8.31622]
 
 
 
 
 
separately.
FDI(-1)
 
 0.332508
 (0.08155)
 
 
 
 
 And approximately 0.6% of
  [ 4.07710]     deviations in the GDP from its
       
C  2.848080     long run equilibrium are
       
        cleared in the next quarter.
Error Correction: D(GDP) D(GE) D(FDI)
       
       
CointEq1 -0.005777  0.016669 -0.178022
   (0.01027)  (0.01132)  (0.04382)
  [-0.56259] [ 1.47246] [-4.06231]
       
D(GDP(-1))  0.551166 -0.006978 -0.041583
   (0.11846)  (0.13061)  (0.50559)
  [ 4.65267] [-0.05342] [-0.08225]
       
 Diagnostic or POST estimation test
Residual tests:
Portmanteau Autocorrelation test: computes the
multivariate Box-Pierce/Ljung Box Q statistics for serial
correlation upto a specified order. Eviews reports both
tests under the null hypothesis of no serial correlation
Autocorrelation LM test: reports the multivariate LM test
statistics for residual serial correlation. Under the null
hypothesis of no serial correlation of order h the LM test is
asymptotically Chi Square distributed with degrees of
freedom.
 Diagnostic or POST estimation test
Normality tests
Reports the Multivariate extension for the Jarque Bera test
For the multivariate test you must choose a factorisation of
residuals that are orthogonal to each other:
☞Cholesky
☞Inverse square root of residual correlation matrix Doornik and
Hansen (1994)
☞Inverse square root of residual covariance matrix Urza (1997)
☞Factorisation from identified VECM
Eviews reports the test statistics for each othorgonal component
 Diagnostic or POST estimation
test
Whites heteroskedasticity test
Is an extension of the white’s 1980 test
No cross terms: uses only levels and squares of regressors
Cross terms: includes all non redundant cross products of
regressors (heteroskedasticity of an unknown form).
 VEC models - OUTPUTS
2 useful outputs from VEC models are:

Impulse Response Functions (IRFs)

Forecast Error Variance Decomposition (FEVD) or
Variance Decomposition for short.