This exercise, I am doing for writing research paper I have average monthly prices data for 6 markets (150 observations). I wanted to determine the speed of adjustment of the price series towards equilibrium as well as to confirm the short-run relationship by employing VECM in stata. i generate natural log of each market monthly data and follow following steps
3. identify the number of cointegrating relationships (vecrank); [CODE][vecrank ln_price_1 ln_price_2 ln_price_3 ln_price_4 ln_price_5 ln_price_6, max trend(constant) lags(2) ic] results of Johansen co-integration rank test in Stata (Eigenvalue, trace statistic, max statistic, SBIC, HQIC) indicated the existence of 6 co-integrating equations
4. fit a VECM (vec): fitted several VECMs. Because we are uncertain about the number of lags and the constant and trend terms. We used Likelihood Ratio test (Assumption: reduced model nested in full model) and finally selected
[CODE][vec ln_price_1 ln_price_2 ln_price_3 ln_price_4 ln_price_5 ln_price_6, trend(rtrend) rank(5) lags(3)]
I am finding difficulty in interpreting the results.
5. test for stability and white-noise residuals (vecstable, veclmar, and vecnorm);
veclmar
* as p-value is greater than 0.05, based on the Lagrange-Multiplier test results, there is no strong evidence of serial correlation in the residuals of the VEC model at lags 1 and 2.
vecnorm: The overall conclusion is that the residuals in most equations (D_ln_price_1 through D_ln_price_6) of the VEC model are not normally distributed. While skewness test didnt rejected null hypothersis for D_ln_price_1 through D_ln_price_4. The combined test (ALL) indicates that there is strong evidence against the null hypothesis of symmetrically distributed residuals for the entire system.
vecstable : The VECM specification imposes a unit modulus.
- confirm that the variables contain a unit root. using ADF and KPSS test, all variables are not-stationary at level , but stationary at difference.
- identify the number of lags to include in the VECM (varsoc); [CODE][#forval i = 1/6 {
varsoc d.ln_price_`i'
}]
3. identify the number of cointegrating relationships (vecrank); [CODE][vecrank ln_price_1 ln_price_2 ln_price_3 ln_price_4 ln_price_5 ln_price_6, max trend(constant) lags(2) ic] results of Johansen co-integration rank test in Stata (Eigenvalue, trace statistic, max statistic, SBIC, HQIC) indicated the existence of 6 co-integrating equations
4. fit a VECM (vec): fitted several VECMs. Because we are uncertain about the number of lags and the constant and trend terms. We used Likelihood Ratio test (Assumption: reduced model nested in full model) and finally selected
[CODE][vec ln_price_1 ln_price_2 ln_price_3 ln_price_4 ln_price_5 ln_price_6, trend(rtrend) rank(5) lags(3)]
I am finding difficulty in interpreting the results.
5. test for stability and white-noise residuals (vecstable, veclmar, and vecnorm);
veclmar
* as p-value is greater than 0.05, based on the Lagrange-Multiplier test results, there is no strong evidence of serial correlation in the residuals of the VEC model at lags 1 and 2.
vecnorm: The overall conclusion is that the residuals in most equations (D_ln_price_1 through D_ln_price_6) of the VEC model are not normally distributed. While skewness test didnt rejected null hypothersis for D_ln_price_1 through D_ln_price_4. The combined test (ALL) indicates that there is strong evidence against the null hypothesis of symmetrically distributed residuals for the entire system.
vecstable : The VECM specification imposes a unit modulus.
Comment