I think your specification is wrong. Of course, a higher capacity could be an incentive to increase export supply. But for capacity utilisation, the causality is likely to run into the other direction: If exports increase (decrease), more (less) is produced and CU goes up (down). Also, from a pure statistical point of view, it is dangerous to have these variables in one equation, because they depend on each other:

lnCapacity = lnProduction - lnCU

lnCU = lnProduction - lnCapacity

Therefore, one can expect crazy coefficients. Nevertheless, they should not be as crazy as your results. There seems to be a measurement problem behind. Maybe, that you find the problem, if you apply your estimated equation on the indendent variables and compare this ex-post-forecast with the dependent variable. 

Dear Professor Rainer

First, i have measured the capacity utilization by deviation from trend real GDP. i think it is correctly signed as an increase in the deviation from trend income means that the country production is lying below the production possibility curve. think it ok

Second, the production capacity has also correct sign in long run but only in the case of primary exports it has adverse effect in short run, it is pretty possible because Pakistan primary exports are mainly consist on raw material especially raw material, raw wool etc. An increase in production capacity means the country established more industries and expanding the existing ones, so in the case the more raw material are used domestically and less available for exports. so i think this is also ok

But my question why the coefficient of is so high in short run especially in the restricted_ECM please give some comments on this point i am very thankful

Regards

Irshad

Dear Irshad,

If you measure capcity utlisation as deviations from trend of real GDP, then for usual trend calculations, the deviations (absolute or percentage?) show differnt signs. Therefore a log-transformation is not possible. I assume that you use ln(GDP/trend GDP) in your estimation.  As capacity mearure, you seem to use real trend GDP. Because of the strong correlation between the variables, the coefficients are likely to be highly biased.  In fact, you seem to use the following equation (not complete):

dlnX =a*dln(TrendGDP)+b*dln(GDP/Trend)+c*ln(TrendGDP)+d*ln(GDP/Trend)+... or a similar specification. lnTrend is, of course near a linear time trend (for an exponential trend it is exactly linear), and llnTrend is (nearly) a constant.

If X has a strong trend, one can expect too high estimates for a and c and too low  ones for b and d (if not integrated into the third term). High random fluctuations of X are likely to go into a and b (partly compensating each other). Your short-run elasticities are far from any acceptable range, but I would also not accept hte high long-run ones.

From an economic (theoretical) point of view, I do not understand why capacity utilisation should have a strong (positive or negative) influenceon exports. I see much stronger causality in the other direction, higher exports lead to a higher CU.

I propose to abstain from ARDL and ECM and apply simple OLS. A good specification is much, much more important than a highly refined method (often simply used because it is modern).

Best regards,

Anton