Certainly it's an unusually large correction, but not actually an unstable one (

If the coefficient on the error correction term is negative and statistically significant, is consistent. with error correcting behavior. The bigger the (negative) statistically significant coefficient, the more rapid is the correction. Desirable values of ECM (-1 to 0). Should not worry much. Positive greater than 1 values may indicate autocorrelation.

The coefficient of is negative (-1.5) and significant meaning that system corrects its previous period disequilibrium at a speed of 150% and it indicates the sizable speed of adjustment of disequilibrium correction for reaching long run equilibrium steady state position.

Narayan and Smyth (2006) state that an error correction term that lies between -1 and -2 means that the equilibrium is achieved in dicreasingly flactuating form. Details can be found in the paper.

Article What Determines Migration Flows from Low-Income to High-Inco...

 I agree with Cushman, but I do not agree with Narayan and Smyth. It is clear evidence of a misspecification, not captured by the diagnostics tests they reported. Mousumi is correct about the desirable values, she noted above.  Footnote 9 in Narayan and Smyth also incorrect. The ARDL estimator cannot be applied I(2) situation. 

 “-1

I guess, I have said all that I need to say about this. Also, Abdul is correct.

From the theoretical basis and derivation of the error correction term, the maximum value is 1, ie 100% and no more, suggesting full return to long run equilibrium from a possible short run distortion. I have to seen in a standard textbook suggesting otherwise as at now. My submission is that the theoretical structure of ECM does not allow for higher that 100% return to LR equilibrium. Regards.

It can be lower than -1 and its dynamically stable. It suggests oscillatory convergence. See this paper by Loayza and Ranciere (page 12).

https://www.imf.org/external/pubs/ft/wp/2005/wp05170.pdf

Thank you so much to everyone.

i have same for my time series growth equation and wondering if i should report it in the results and explain or discard the results altogether

mine is -1.122

This type of coefficient appear in annual data with small size and it is claimed to be the result of oscillation which is wrong. This is because the researcher did not spend more time on their model. If they did they will see that it is a misspecification error. Diagnostic test may not capture everything within a data generating process. A careful researcher spends a lot time on his model and that is why it is called research. It is not meant to be easy and one can not blindly follow a computer flow chart.

Check very well and ensure there is no mis specification error in the model

I agree with Frank Ogbeide and Abdul Rasheed Sithy Jesmy arguments , the error correction residual ( ecmt-1) is important variable in explaining the speed of adjustment of short run disturbances in the long-run.

For example, the coefficient of error residual was found to be negative (-0.683429) and statistically significant at 5% level of significant with p-value 0.0004. This means that, the system corrects its previous period at a speed of convergence of 68 percent per annum.

John Mganga, if am correct, the significance level is 1%. Furthermore, you can go ahead to compute the duration for adjustment from the disequilibrium state to a steady-state by the formula: 1/ECT * 100. In this case, 1/0.683429*100= 1.46, meaning it will take the model the period of 1 year 5 months to adjust back to the equilibrium state.

I completely agree with the idea claiming that the correction value should not be less than -1, despite having seen values in some top-line journals both less than -1 and even positive. As far as I know those studies aren't used for typical explanation purposes, but rather for forecast validation etc.

I think under the assumption of theoretical consistency, the value being less than -1 should be interpreted as specification error. Try estimating exactly the same equation using two different methods: one with original dependent variable and the other with the rate at which the dependent variable changes (for example, GDP and GDP growth). Under many circumstances you will get much greater value for error-correction term when estimating with a change rate. That's why many researchers use all the variables as change rates or as original observation value.

Model misspecification is a serious problem even when an estimation returns with an error-correction value between (-1 - 0). But it is a good thing that most of the time the value is out of range. I remember one time getting -4.28 with well diagnostics.

With Respect, Çağlar Karaduman how is it a good thing that MOST of the time the value is out of range. of (-1. 0)?

I think statistically is right but the most important question here in my opinion is the speed of adjustment really at this speed in economics??? I think ECT higher than absolute 1 doesn't appear in economics but can appear statistically

it is all right to have -1.5 as ECM. this simply means that the correction mechanism is oscillatory. in other words, the speed of adjustment fluctuates forward before settling to equilibrium. (note, the frequency of your data as well as outliers can play a role in this but by no means your ECM is wrong)

Sorry to all and Chuck A Arize , my statement above isn't clear. What I meant is;

"it is a good thing that most of the time the value of ECM turns out to be out of range when something about the estimation isn't right"

because it indicates that either the model is misspecified and/or there is no cointegration.

@çağlar. When dealing with financial data, it can make perfect sense to get an ECM lower than -1. this is because financial/monetary variables can be subject to overshooting theory.

I have taken a annual data of GDP and ECT comes out to be -1.45 which is quite higher.

If I change the frequency to quarter will it be possible to lower the ECT.

Or I can quote the same

Range of error correction coefficient is from 0 to less than 2 (with negative sign and significant). If the value is less than 1 it means equilibrium will be adjust monotonically and if value is greater than 1 and less than 2 (with negative sign and significant) it shows equilibrium will be adjust in a dampening manner ( the error correction process fluctuates around the long-run value ).

See YouTube Link: https://youtu.be/1oasRhnt5AI

Also see: Narayan, P. K., & Smyth, R. (2006). What determines migration flows from low‐income to high‐income countries? An empirical investigation of Fiji–Us migration 1972–2001. Contemporary Economic Policy, 24(2), 332-342.

ECM can be statistically less than -1. But how true is this in real life? Will deviations from balance stabilize in less than 1 year (if data is annual)? If the dynamics of the country of study allow this, the analysis continues. In conclusion, economically, this result should be explained in detail. For quality journals, it is preferred that ecm is greater than -1.

What is specification test is good, normality is good, serial correlation, heteroskedasticity,stability, all good but the error correction term is greater than 2 even though negative and significant???

Hello

I am studying the link between economic growth and female employment with a model ARDL, all the results and tests are good except that the coefficient of the adjustment force is equal to -3 and significant, I don't know if it's good or the model is badly estimated

THANKS