EViews.com

Hello again,

I am currently testing for endogeneity for my two explanatory variables. If I do the test for all the regressors at once then my model suffers from endogeneity. However if I do the test for hcipp-2 and ygapp seperately then the test concludes they are both exogenous.

But when I tried to perform a weak instrument test an error occured stating that the critial F value was not available for a model including two or more endogenous variables.

Should I test all the regressors at once instead of seperately at the endogenous test?

My model:

stn = c + hcipp-2 + ygapp + r*c + r*(hcipp-2) + r*ygapp

stn = nominal interest rate
hcipp = inflation rate
ygapp = output gap
r = dummy variable

> formula object is tr_white_ar


Kind regards,
Ties

I am not sure how you are testing the endogeneity or which method you are using, but I suggest "Hausman test". It is quite straightforward and tests endogeneity of your explanatory variables vis-a-vis the dependent variable (stn).

Since you have already estimated a TSLS, you can also easily perform Donald-Wu test for the regressor endogeneity:

I am using the Regressor Endogeneity Test which can be find under view > IV Diagnostics & Tests > Regressor Endogeneity Test. I think that is the same test as you describe as Donald-Wu Test.

However the problem occurs when performing a weak instrument test with the Cragg Donald F statistic (view > IV Diagnostics & Tests > Weak Instrument Diagnostics). This test fails because it states that there are two or more endogenous variables in my model and therefore there is no critical F statistic. How is that possible as the endogeneity test suggest that both of my regressors are exogenous?

The test output:

Weak Instrument Diagnostics
Equation: TR_WHITE_AR

Cragg-Donald F-statistic: 0.169861

Stock-Yugo bias critical values not available for
models with more than 3 endogenous
variables.

Stock-Yugo size critical values not available for
models with more than 2 endogenous
variables.

Moment selection criteria:

SIC-based: 1.783998
HQIC-based: 5.492177
Relevant MSC: 32.11612

You'll have to revise your model specification. Be careful with the inclusion of dummy variable. The resulting model significantly looses its explanatory power and diagnostics become worse. Weak instrument test measures the validity of your instruments. Finding proper instruments is of crucial importance in this type of modeling and nobody can help you in that respect, I am afraid. If you drop the dummy variable and corresponding cross-terms, then model becomes less ill-defined. You can start working from there and may consider adding the lagged/lead values of your (in)dependent variables.