Ananadita

As I understand it ovtest ( that is Ramsey's RESET test) tests for possible non-linearity in the included predictor variables. According to Woolridge (Econometric analysis) it is generally a poor test for omitted variables (despite claim and indeed the stata name ) after all it is only predicted yhat regressed on a polynomial of yhat.- so it is about linearity rather than omission of variables.

So why in your case would non-linearity appear when country time-varying dummies are included?

It MAY be that the dummies are actually having a different effect in different countries and that you are not fully capturing this particular form of parameter heterogeneity. These two papers - both on Research-gate in pre-publication form illustrate this issue; and how important it can be to do so:

Andrew Bell, Kelvyn Jones

Explaining Fixed Effects: Random Effects modelling of Time-Series Cross-Sectional and Panel Data

Andrew Bell, Ron Johnston, Kelvyn Jones

Stylised fact or situated messiness? The diverse effects of increasing debt on national economic growth

Unlike Rajesh - we argue for random effects in this sort of analysis- it will be interesting if you do this ,to report what you find.

Happy modelling and I would go with theory - that is include the meaningful dummies.

Kelvyn

Anandita,

You may want to look at the two papers below for trade resistance modelling using gravity flow estimation.

(1) Kalirajan, K., 2007. “Regional Cooperation and Bilateral Trade Flows: An Empirical

Measurement of Resistance Model,” The International Trade Journal, 21(2).

(2) Adil Khan Miankhel, Shandre Thangavelu, Kalirajan Kaliappa. 2009.

“On Modeling and Measuring Potential Trade: Quantitative Approaches to Public Policy” Fourth Annual International Conference on Public Policy and Management, Indian Institute of Management Bangalore (IIMB).

The authors in the second paper look at various categories of exports using a trade gravity Stochastic Frontier model for Australia and try to measure the so-called “behind the border measures”, with a parameter estimate. They show that if sigma_U has a value much greater than zero, then country-specific “behind the border measures” are important. This indicates that that there is very high trade resistance (behind the border restriction) for a given export basket.

You may want to start with these two papers and then extend the methods to your research and see how to estimate multilateral trade resistance.

Barnabas Kiiza

Makerere University

Trade models are not my expertise, but, since your data includes a relatively large amount of years, you should consider to initially study if your response variable is stationary, since the first regression might even result as spurious. Due to that, maybe, when including the country-time regressors, this kind of «noise» arises in your model.