8
$\begingroup$

For my master thesis I basically want to find out, why developing countries are stagnating. Next to theoretical aspects I also want to make a regression. I want to regress GDP or GDP growth as dependent variable on many independent variables, such as tenure of head of state, life expectancy, working hours restriction, adult literacy and population growth and some (other) institutional variables over five years. My question is: Does it make more sense to regress GDP-Growth (in %) on the independent variables or should I use the real GDP-Value (e.g. in $)?

$\endgroup$
2
  • 4
    $\begingroup$ "For my master thesis I basically want to find out, why developing countries are stagnating." that is a rather ambitious goal! $\endgroup$ May 18 '15 at 12:00
  • 2
    $\begingroup$ @kjetilbhalvorsen that's what grad theses are supposed to do: hail Mary moonshots! $\endgroup$
    – Aksakal
    May 18 '15 at 13:03
4
$\begingroup$

Basically, there's no ambiguity here: you must use the differences in the regression. There could be some discussion whether it's a simple difference or a log difference, but the latter is more common in the literature. If you had a cross-section, then this wouldn't matter. For a short period of time it probably doesn't matter that much either, but you're probably going to include countries with high growth in your sample, so the rate of change would be more appropriate.

GDP is not stationary even when you suspect or observe that it's stagnant, i.e. not growing over a sample period.

I think most researchers would take it that GDP is an exponential random walk process: $\Delta \ln GDP_t=g_t \Delta t+\varepsilon_t$. However, you can google to find that some would argue that it's an exponential trend process such as $GDP_t=GDP_0 e^{g_t\Delta t}+\varepsilon_t$. In the former case you see that the growth rate is stochastic, while in the latter it's deterministic and the error simply adds up. Both are nonstationary, hence, the growth rate must be used in any case.

$\endgroup$
4
  • $\begingroup$ I already have the growth rates, should I even take the first difference of the growth rates itself or would that only be for the gdp value? $\endgroup$ May 18 '15 at 12:49
  • $\begingroup$ If you take the difference in growth rates you'll deal with difference-in-differences study as they call it in econometrics. I'm not a macroeconomist, but doubt that they have theories of unstable growth rates. In other words I don't think they say whether growth rates themselves should be growing or declining. Hence, I think you should simply use the growth rates themselves. Your hypothesis would be whether they are too low or negative. $\endgroup$
    – Aksakal
    May 18 '15 at 13:01
  • 2
    $\begingroup$ Difference in differences refers to something else, though: en.m.wikipedia.org/wiki/Difference_in_differences $\endgroup$ May 18 '15 at 15:12
  • 1
    $\begingroup$ @ChristophHanck, you are right. I thought OP wants to see how GDP growth rate starts declining in some countries, or something like that. $\endgroup$
    – Aksakal
    May 19 '15 at 15:12
5
$\begingroup$

You are interested in why countries are "stagnating"; stagnation is about a lack of growth, so, GDP growth seems logical.

A couple notes: 1. You will need data on countries that are growing at different rates. 2. If you are using data from different years for the same countries (e.g Ghana in 2011, Ghana in 2012, Ghana in 2013...) then a regular regression is probably inappropriate, Your errors are unlikely to be independent.

$\endgroup$
2
  • $\begingroup$ Yes, but I use the data for many countries over many years (at least 5), so its a paneldata. Would this solve the problem of the dependent error term? $\endgroup$ May 18 '15 at 12:41
  • $\begingroup$ No, it would not. You cannot use standard regression for this. Some ideas are multilevel models or genearalized estimating equations. $\endgroup$
    – Peter Flom
    May 18 '15 at 15:05
3
$\begingroup$

The excellent textbook by Barro and Sala-i-Martin (Economic Growth, MIT press, 2004), can help you to choose your model. However, as Peter Flom said, be careful with cross-section regression, it can be misleading; you might need to apply a panel data methodology (see the paper by Islam, 1995, on The Quarterly Journal of Economics 110(4), 1127-1170). Again, see Barro and Sala-i-Martin (2004) for almost all references you might need.

It might be useful for you also to check at some classical (but old!) papers on growth economics such as Sala-i-Martin (1994, European Economic Review 38(3-4), 739-747) , Islam (1995, see above) and Baumol (1986, The American Economic Review 76(5), 1072-1085) among many, many others.

$\endgroup$
4
  • 1
    $\begingroup$ Please include full citations for articles and textbooks. While the link works today, it may not work tomorrow. The goal of SE sites is to build a durable repository of knowledge for future visitors. $\endgroup$
    – Sycorax
    May 19 '15 at 12:38
  • 1
    $\begingroup$ That's right. I've revised the citations. $\endgroup$
    – utobi
    May 19 '15 at 13:44
  • 1
    $\begingroup$ This is a step in the right direction, but an author's last name and a year are not full citations. Searching for "Islam 1995" won't help anyone locate the document. Please include the author's full name, the article or chapter title, the full journal or book title, the volume and issue number and the year. Complete information about how to cite an academic source can be found in a style manual such as this. owl.english.purdue.edu/owl/resource/747/08 $\endgroup$
    – Sycorax
    May 19 '15 at 13:49
  • $\begingroup$ This should be more than enough to locate the documents. $\endgroup$
    – utobi
    May 19 '15 at 15:08
2
$\begingroup$

A very important question is availability of data. Example, you can get the data from wto.org.

A very realistic answer to your question is that you should build your model in both ways and compare the accuracy prediction parameters e.g. $R^2$ , adjusted $R^2$, and residual graphs. Then start validating your hypotheses for which form of GDP you should take.

My hypotheses on this is that it is just equivalent to the transformation of the GDP variable.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.