The RATS Software Forum

Hi, I am trying to recover the error correction term after having done FMOLS and panel granger causality tests to see how long it takes for each variable to go back to the long run equilibrium. I have been using the example given on the website but I want to make sure I am doing it right and interpreting it right as well. Here is my input/output

@johmle(lags=3,det=rc,cv=cv1)
# lgdp lcap llabor lre

Likelihood Based Analysis of Cointegration
Variables: LGDP LCAP LLABOR LRE
Estimated from 1//1992:01 to 15//2011:01
Data Points 285 Lags 3 with Constant restricted to Cointegrating Vector

Unrestricted eigenvalues and -T log(1-lambda)
Rank EigVal Lambda-max Trace Trace-95% LogL
0 1585.9395
1 0.2461 80.5073 111.6744 53.9400 1626.1931
2 0.0854 25.4521 31.1672 35.0700 1638.9192
3 0.0116 3.3302 5.7151 20.1600 1640.5843
4 0.0083 2.3849 2.3849 9.1400 1641.7767

Cointegrating Vector for Largest Eigenvalue
LGDP LCAP LLABOR LRE Constant
-1.201889 0.801104 0.394960 0.016912 6.845360

*
* Create the equation that describes the cointegrating vector
*
equation(coeffs=cv1) cv *
# lgdp lcap llabor lre constant
*
* Setup and estimate the ECM
*
system(model=varmodel)
var lgdp lcap llabor lre
lags 1 to 3
ect cv
end(system)
*
estimate

VAR/System - Estimation by Cointegrated Least Squares
Panel(22) of Annual Data From 1//1993:01 To 15//2011:01
Usable Observations 285
Skipped/Missing (from 327) 42

Dependent Variable LGDP
Mean of Dependent Variable 0.0373342517
Std Error of Dependent Variable 0.0377138824
Standard Error of Estimate 0.0358359546
Sum of Squared Residuals 0.3544435168
Durbin-Watson Statistic 2.0361

Variable Coeff Std Error T-Stat Signif
************************************************************************************
1. D_LGDP{1} 0.447696066 0.088888787 5.03659 0.00000086
2. D_LGDP{2} -0.116098693 0.087535352 -1.32631 0.18583470
3. D_LCAP{1} -0.064991199 0.024470552 -2.65589 0.00836992
4. D_LCAP{2} 0.007083455 0.024688163 0.28692 0.77439111
5. D_LLABOR{1} 0.141104100 0.111817733 1.26191 0.20804594
6. D_LLABOR{2} -0.058125634 0.112142712 -0.51832 0.60465139
7. D_LRE{1} 0.001916404 0.009023333 0.21238 0.83196493
8. D_LRE{2} 0.010063824 0.009015784 1.11625 0.26528814
9. EC1{1} 0.013698179 0.002122739 6.45307 0.00000000

Dependent Variable LCAP
Mean of Dependent Variable 0.0520195954
Std Error of Dependent Variable 0.1329759575
Standard Error of Estimate 0.1298165007
Sum of Squared Residuals 4.6512413803
Durbin-Watson Statistic 2.0602

Variable Coeff Std Error T-Stat Signif
************************************************************************************
1. D_LGDP{1} 1.122947409 0.322001505 3.48740 0.00056736
2. D_LGDP{2} -0.311749046 0.317098656 -0.98313 0.32640472
3. D_LCAP{1} -0.084485712 0.088645092 -0.95308 0.34138434
4. D_LCAP{2} -0.071879667 0.089433389 -0.80372 0.42224866
5. D_LLABOR{1} 0.618025550 0.405062093 1.52576 0.12821595
6. D_LLABOR{2} -0.197091214 0.406239336 -0.48516 0.62794747
7. D_LRE{1} -0.001554286 0.032687215 -0.04755 0.96210905
8. D_LRE{2} 0.025081399 0.032659867 0.76796 0.44316880
9. EC1{1} 0.009703526 0.007689666 1.26189 0.20805309

Dependent Variable LLABOR
Mean of Dependent Variable 0.0192347341
Std Error of Dependent Variable 0.0207175527
Standard Error of Estimate 0.0186951994
Sum of Squared Residuals 0.0964648930
Durbin-Watson Statistic 2.0798

Variable Coeff Std Error T-Stat Signif
************************************************************************************
1. D_LGDP{1} 0.009699931 0.046372243 0.20918 0.83446565
2. D_LGDP{2} 0.037281824 0.045666172 0.81640 0.41497594
3. D_LCAP{1} -0.004142480 0.012766002 -0.32449 0.74581050
4. D_LCAP{2} -0.020028281 0.012879526 -1.55505 0.12108055
5. D_LLABOR{1} 0.336598245 0.058334007 5.77019 0.00000002
6. D_LLABOR{2} 0.224096298 0.058503544 3.83047 0.00015836
7. D_LRE{1} 0.003468403 0.004707368 0.73680 0.46186798
8. D_LRE{2} 0.003741676 0.004703429 0.79552 0.42699426
9. EC1{1} 0.003775465 0.001107408 3.40928 0.00074858

Dependent Variable LRE
Mean of Dependent Variable 0.0344531311
Std Error of Dependent Variable 0.2428647311
Standard Error of Estimate 0.2261516740
Sum of Squared Residuals 14.115903979
Durbin-Watson Statistic 2.3323

Variable Coeff Std Error T-Stat Signif
************************************************************************************
1. D_LGDP{1} 0.085977289 0.560954724 0.15327 0.87829769
2. D_LGDP{2} 0.461582826 0.552413534 0.83557 0.40411690
3. D_LCAP{1} -0.070092710 0.154427487 -0.45389 0.65026648
4. D_LCAP{2} -0.026593691 0.155800769 -0.17069 0.86459230
5. D_LLABOR{1} 0.229272000 0.705653518 0.32491 0.74549735
6. D_LLABOR{2} 0.291048698 0.707704378 0.41126 0.68120292
7. D_LRE{1} -0.370749891 0.056943982 -6.51078 0.00000000
8. D_LRE{2} -0.268011826 0.056896339 -4.71053 0.00000392
9. EC1{1} 0.015968478 0.013396069 1.19203 0.23427452



So for each variable, the EC{1} term is the error correction term?

Thanks!!

EC{1} is the error correction term. However:

1. The Johansen procedure isn't designed for panel data so the test statistics are likely invalid.
2. I assume those are trending series (GDP in particular). If so, then DET=RC is the wrong model. You need DET=CONSTANT to allow for trend.
3. Even if the procedure were valid for panel data, why would you expect that the constants would be the same? Is GDP (or any of the others) measured in local currency? If so, the level of log GDP will vary based upon that. If you do the more correct DET=CONSTANT, the way you're trying to estimate this forces all the countries to have the same trend rate.

So should I try to switch my code to using panel FMOLS values instead of the Johanssen results? Would that get rid of the issue?

Thanks,

celineboulenger13 wrote:So should I try to switch my code to using panel FMOLS values instead of the Johanssen results? Would that get rid of the issue?

Thanks,

What you need to do first is to figure out precisely what form your model is taking. What's heterogeneous and what (if anything) is homogeneous? You might want to read the introduction to the panel data e-course. Deciding on a model for panel data is a common problem, but you can't decide what tool to use until you've picked the desired behavior.

I have a heterogenous panel that is cointegrated. I used panel FMOLS to find the cointegration relationship and looked at granger causality tests to look at the short run relationships and the last thing I need to do is recovering these error correction terms, however I cannot find any examples of people doing that with panel data for RATS.
What would you do?

Thanks,