MathType

Sunday, 19 June 2016

ARDL Cointegration Test with Stata (Time Series)





Pesaran, Shin and Smith (PSS)(2001) developed a new approach to cointegration testing which is applicable irrespective of whether the regressor variables are \(I\left( 0 \right)\),\(I\left( 1 \right)\)  or mutually cointegrated.

The statistic underlying PSS procedure is the familiar Wald or F­-statistic  in a generalized Dickey-Fuller type regression used to test the significance of lagged of the variables under consideration in an unrestricted error correction regression.

The test show that the asymptotic distribution of both statistics are non-standard under the null hypothesis that there exist no relationship between the level of the included variables; irrespective of whether the regressor are \(I\left( 0 \right)\),\(I\left( 1 \right)\) or mutually cointegrated.

PSS provide two sets of asymptotic critical values for the two polar case: one which assumes that all the regressor are , \(I\left( 1 \right)\) and the other assuming that they are \(I\left( 0 \right)\) .

Since these two sets of critical values provide critical value bounds for all classifications of the regressors into \(I\left( 1 \right)\) and/or \(I\left( 0 \right)\) PSS propose a bound testing procedure.

If the computed Wald or F-statistic falls outside the critical value bounds, a conclusive inference can be drawn without needing to know whether the underlying regressor are \(I\left( 1 \right)\): cointegrated amongst themselves or individually \(I\left( 0 \right)\) .

The bound test methodology has a number features that many researcher feel give it some advantages over conventional cointegration testing;
1.       It can be used with mixture of \(I\left( 0 \right)\) and \(I\left( 1 \right)\) data.
2.       It involves just a single-equation set-up, making it simple to implement and interpret.
3.       Different variables can be assigned different lag-lengths as they enter the model.

There are the basic steps that we are going to follow to perform the bound test and the basic model form of ARDL model is;

\({{Y}_{t}}={{\beta }_{0}}+\sum\limits_{i=1}^{p}{{{\beta }_{1}}{{Y}_{t-i}}+\sum\limits_{i=0}^{{{q}_{1}}}{{{a}_{1}}{{X}_{1,t-i}}\sum\limits_{i=0}^{{{q}_{2}}}{{{a}_{2}}{{X}_{2,t-i}}+}}}\sum\limits_{i=0}^{{{q}_{3}}}{{{a}_{3}}{{X}_{3,t-i}}+}\sum\limits_{i=0}^{q4}{{{a}_{4}}{{X}_{4,t-i}}+}{{\varepsilon }_{t}}\)                (1)


Step 1

Make sure that none of the variables are \(I\left( 2 \right)\), as such data will invalidate the methodology.

Step 2

Formulate the following model;

\({{Y}_{t}}={{\beta }_{0}}+\sum\limits_{i=1}^{p}{{{\beta }_{1}}{{Y}_{t-i}}+\sum\limits_{i=0}^{{{q}_{1}}}{{{a}_{1}}{{X}_{1,t-i}}\sum\limits_{i=0}^{{{q}_{2}}}{{{a}_{2}}{{X}_{2,t-i}}+}}}\sum\limits_{i=0}^{{{q}_{3}}}{{{a}_{3}}{{X}_{3,t-i}}+}\sum\limits_{i=0}^{q4}{{{a}_{4}}{{X}_{4,t-i}}}+{{\theta }_{0}}{{Y}_{t-1}}+{{\theta }_{1}}{{X}_{1,t-1}}+{{\theta }_{2}}{{X}_{2,t-1}}+{{\theta }_{3}}{{X}_{3,t-1}}+{{\theta }_{4}}{{X}_{4,t-1}}+{{\varepsilon }_{t}}\)
(2)


Step 3

The range of summation in various term in Eq (2) are from 1 to p ,and 0 to q1, 0 to q2, 0 to q3 and 0 to q4 respectively. We need to select the appropriate values for maximum lags, p, q1, q2, q3 and q4.

Use the information criterion (AIC,SIC or Bayesian information) to select appropriate  lags. The smaller the value of an information criterion the better the results.

Step 4

Perform the bound testing. Again;

\({{Y}_{t}}={{\beta }_{0}}+\sum\limits_{i=1}^{p}{{{\beta }_{1}}{{Y}_{t-i}}+\sum\limits_{i=0}^{{{q}_{1}}}{{{a}_{1}}{{X}_{1,t-i}}\sum\limits_{i=0}^{{{q}_{2}}}{{{a}_{2}}{{X}_{2,t-i}}+}}}\sum\limits_{i=0}^{{{q}_{3}}}{{{a}_{3}}{{X}_{3,t-i}}+}\sum\limits_{i=0}^{q4}{{{a}_{4}}{{X}_{4,t-i}}}+{{\theta }_{0}}{{Y}_{t-1}}+{{\theta }_{1}}{{X}_{1,t-1}}+{{\theta }_{2}}{{X}_{2,t-1}}+{{\theta }_{3}}{{X}_{3,t-1}}+{{\theta }_{4}}{{X}_{4,t-1}}+{{\varepsilon }_{t}}\)                                                                                              (2=3)

Perform an “F-test” of the hypothesis \({{H}_{0}}:{{\theta }_{0}}={{\theta }_{1}}=...={{\theta }_{4}}=0\) against the alternative that \({{H}_{0}}\)  is not true. The rejection implies that we have a long-run relationship.

Because the distribution of F-test for Eq(4) is non-standard,  Pesaran et.al (2001) supply bounds on the critical values for the asymptotic distribution of the F-statistics.
In each case, the lower bounds is based on the assumption that all the variables are I(0), and the upper bound is based on the assumption that all the variables are I(1).

Step 5

If the bound test leads to conclusion of cointgration, we can meaningfully estimate the long-run equilibrium relationship between the variables:

\({{Y}_{t}}={{\beta }_{0}}+{{a}_{1}}{{X}_{1,t}}+{{a}_{2}}{{X}_{2,t}}+{{a}_{3}}{{X}_{3,t}}+{{a}_{4}}{{X}_{4,t}}+{{\varepsilon }_{t}}\)                                                                 (4)

as well as the usual ECM:

\(\Delta {{Y}_{t}}={{\beta }_{0}}+\sum\limits_{i=1}^{p}{{{\beta }_{1}}\Delta {{Y}_{t-i}}+\sum\limits_{i=0}^{{{q}_{1}}}{{{a}_{1}}\Delta {{X}_{1,t-i}}\sum\limits_{i=0}^{{{q}_{2}}}{{{a}_{2}}\Delta {{X}_{2,t-i}}+}}}\sum\limits_{i=0}^{{{q}_{3}}}{{{a}_{3}}\Delta {{X}_{3,t-i}}+}\sum\limits_{i=0}^{q4}{{{a}_{4}}\Delta {{X}_{4,t-i}}}+\delta {{Z}_{t-1}}+{{\varepsilon }_{t}}\)         (5)

where  \({{Z}_{t-1}}=\left( {{Y}_{t-1}}-{{a}_{0}}-{{a}_{1}}{{X}_{1,t-1}}-{{a}_{2}}{{X}_{2,t-1}}-{{a}_{3}}{{X}_{3,t-1}}-{{a}_{4}}{{X}_{4,t-1}} \right)\)

We use the data Macro.

Perform the unit root test (ADF test or PP test) and conform that none the variables  is \(I\left( 2 \right)\).

To test the bound test with an ardl method;

We want perform cointegration test between the variables gdp, pdi, and pce with the bound test.

* Case 1 - no intercept and no determistic trend
ardl pce pdi gdp, ec1 aic noconstant 
ardl,noctable btest
* Case 2 - restricted intercepts and no deterministic trends
ardl pce pdi gdp, ec1 aic constant restricted 
ardl,noctable btest

* Case 3 - unrestricted intercepts and no deterministic trends
ardl pce pdi gdp, ec1 aic constant
ardl,noctable btest

* Case 4 - unrestricted intercepts restricted deterministic trend
ardl pce pdi gdp, ec1 aic constant trendvar(year) restricted
ardl,noctable btest

* Case 5 - unrestricted intercepts and trends
ardl pce pdi gdp, ec1 aic trendvar(year)
ardl,noctable btest


 

 


The ardl results show that the optimal lags for our model ardl is ARDL (3,0,1).

The results for Case 3 show that the value of \({{F}_{s}}=4.259\) which is greater than the critical value of upper bound \({{F}_{ub}}=4.14\) at 10%  level which is null for no cointegration can be rejected.

It means that there is cointegration between the variables pce, pdi and gdp.















No comments:

Post a Comment