Hi everyone, I hope I can get some help. My group is running an ARDL bounds model to test the Environmental Kuznets Curve in Cyprus. We're trying to perform it on Eviews, but we can't seem to be able to run the model - we get a "singular matrix" error. We are using time series from 1975-2016 with the variables of GDP, GDP squared, FDI, energy consumption, and CO2 emissions as our dependent variable. We can run the model with either GDP or GDP squared, but not both at the same time. Eviews returns a "singular matrix" error. Sometimes also a "regressors are perfectly collinear" error message pops up. If this is the case, can anyone point to a solution? Our project is due in 2 weeks and we've been mulling over this for days.
I am attaching the data we're using. There's also a renewable energy column, but we dropped that on account of the missing values.
Thank you!
A lot of discussions have been on this issue in RG:
See, for example:
https://www.researchgate.net/post/I-am-doing-a-panel-data-regression-analysis-and-I-am-having-this-warning-near-singular-matrix-what-does-this-mean-and-how-do-i-solve-this-error
Sometimes, the singular matrix might not be due to collinearity in regressors but due to high lag selected. You might try to reduce the lag from 4 chosen automatically to 3 or 2. Just try to play with the lags ..
If you have a singular matrix in your estimation then some of the variables in the model must be collinear. If this is so your model is ill-defined and can not be estimated. By any chance are you using the log of GDP and the log of (GDP squared)? These will be perfectly collinear.
when I look at your variables I would be worried that variables such as energy consumption and perhaps some others are endogenous (simultaneity). If this is so the ARDL approach is not valid.
Thanks for your answers! Yes we were using the log of GDP squared. We found a solution though: squaring the logged (GDP) made the model run.
- Badges
- Science topic
- Similar topics
- Mathematics
- Algebra
- Matrix