# Regression results contradict economic theory (GDP analysis)

I am currently doing a thesis on the impacts of oil prices on the macroeconomy. The literature suggests that there is a negative relationship between the oil price and the GDP. After running a basic regression, my results contradict this theory:

With GDP as my dependent variable, the coefficients of the independent variables are as followed:

Oil                  0.005558
CPI                 -0.216906
Balance of payments  0.068550
Money Supply        -0.141428
Federal Funds Rate   0.065686


I don't know why this is. Using Eviews I have induced stationarity for all variables and the coefficients are all significant. Any help/suggestions would be greatly appreciated!

My results are below.

• 0.005558 seems really small, as well as your other coefficients. I cannot see the image as I am through a secured network at work, but what is the scale of your data?
– Jon
Feb 16, 2017 at 23:59
• This should be moved to economics. Feb 17, 2017 at 0:03
• Can you include a plot of your data? Are you sure a linear fit for each variable is appropriate? Feb 17, 2017 at 0:12
• Sorry to sound snarky, but in the cities in the US that have raised the minimum wage, there is at least some evidence that this hasn't led to the level of job losses that economic theory would predict. Economic theory should never be viewed as ironclad. It makes a lot of simplifying assumptions. Your results could contradict theory for any number of reasons. Markets can be a lot less efficient than we think. Your variables might be measured imprecisely, or they might be measured in a biased fashion. your data are observational, and all else is not equal. Feb 17, 2017 at 2:59
• You need to give more information on your data! The data points concerne countries? Which countries? There might well be differences between rich and poor countries, oil-producers and consumers, for instance. Time series? How many years? ... Feb 17, 2017 at 10:26

Endogeneity
The macroeconomic variables you are studying are likely to affect each other. In particular, the variables on the right hand side of your model might be affected by the variable on the left hand side. That is known as the problem of endogeneity, which results in inconsistent estimates of model parameters and problems in interpreting the model.
A possible remedy to the problem is to use a (structural) VAR or (structural) VEC model that are estimated in reduced form that ensures endogeneity is not a problem. (I will not expand on how that works, but you will find plenty of material online.)

Unit roots
Are you using logged time series but not log-returns? Quite a few of the series are likely to be integrated (e.g. GPD and oil price, probably also others), which would result in nonstandard distribution of the coefficient estimates and/or the left-hand-side variable diverging from any linear combination of the right-hand-side variables.
To deal with that, you could investigate presence of cointegration and then either (in case of cointegration) proceed to VEC modelling or (in absence of cointegration) to VAR modelling on first-differenced data.

Autocorrelated errors
Another (smaller) problem might be autocorrelation in your model errors, which would lead to imprecise estimation of standard errors and statistical significance. Based on the model setup, I would guess there will be autocorrelation in the errors, and hence the problems.
That could be avoided by going for VAR or VEC model as mentioned above.

(I suppose, besides the VAR and VEC models there could be other solutions, too.
Also, there may be other problems with the model, while I am listing just the few obvious ones.)

• Thanks for your suggestions. I initially thought about using the VECM, however I was unsure about whether it's suitable. If i'm correct - you can only use VECM if you have I(1) variables? In my model, only GDP and Oil are I(1), the rest follow I(0). So does that mean that it's not possible to use VECM? Feb 21, 2017 at 23:04
• @Stewart22, I give an example how to do precisely that in this thread. Feb 22, 2017 at 6:12

Another slant: OLS regression techniques were developed to summarize cross-sectional data. Your data is time series, which requires the application of time series techniques. You might try using transfer functions (dynamic regressions) for a single equation model. Mr. Hardy's responses below are directly on target, particularly his last.

Countries that export oil probably respond differently than countries that don't. You could try to control for that.