I am writing my dissertation about the effects of economic reforms on economic growth in the case of Egypt in 2016 and I need suggestions for what econometric models are usually used in testing such effects using indices like GDP, inflation and so on.
The choice of econometrics model depends on the DGP of the variables and the understanding of the theoretical undertone of the topic.. Some works are better modeled non linearly while some are good linearly. Subject the variables to different pretty estimation test to know the hidden structure of the variables. Then you can decide the best antidote for the modeling.
Great question. Long time ago, I decided to return to estimating econometric models because I was surprised by the conclusions that some macroeconomists drew from their results: with structural reforms, sky was the limit...
Biggest technical issues at that time were linked to endogeneity: First, how to measure a particular component of reform through indicators that are not the results of the reform itself? (e.g., trade reform should not be measured by an increase in the trade/GDP ratio but by changes in tariff rates and non-tariff barriers) Second, how to make sure the estimation technique limits the endogeneity issue?
The other big issue was a more fundamental one: how to specify the model? (I used Barro (1997), “Determinants of economic growth: a cross country empirical study”, Cambridge: MIT Press
I worked with the reform indices build by Sam Morley and Igor Paunovic on Latin America, and the result is a paper The Impact of Structural Reforms on Growth in Latin America and the Caribbean: An Empirical Estimation is available in RGate at 10.2139/ssrn.970012 But I mention this paper only to illustrate my doubts about the econometrics (as a statistician, I don't recommend it: I don't trust applied econometrical models, as they usually violate so many statistical preconditions that they mislead more than they guide).
At that time, the econometric software I used were quite limited and I struggled with a Barro-type model of endogenous growth with pooled cross-sectional across time series and fixed effects. Today, I would use GMM panel or, better Difference-in-Differences.
Difference-In-Differences approaches seem to be better grounded in statistics, but you need to have a good control group. Not easy to find good control groups in macroeconomics. I would try to match similar countries based on macro and structural characteristics, some implementing reforms, others not implementing...You still need to have a good idea of when the reform was implemented and started to have an impact. Not always easy: effects can be delayed or, at the contrary, anticipated...
In my opinion, GMM is just a sophistication that looks fine on the paper (and will help you having your paper accepted by a Journal) but does not solve the real specification issue as mentioned in Leamer (1983) “Lets take the con out of econometrics”, American Economic Review 73, 31-43. To use Hamid Muili terms, we do not know what is the actual DGP (what is the actual model).
For my part, I compared various combinations of potential "exogenous" variables. Indeed, only by running various combinations of variables are we able to identify collinearities. But I wasn't as patient as Sala-I-Martin (1997), “I just ran two million regressions”, American Economic Review 87, 178-183.
Difference-in Differences avoids the Leamer issue, at least in my opinion and reconciliated me with econometrics. I played with it when I returned to econometrics (a paper on the impact of WTO membership, with V. Chemutai).
Sorry for the very long answer, which does not provide you with an answer by the way (not sure there is one). Good luck...
711 / 5000
Resultados de tradução
To measure economic impacts you can do 2 different things. (1) create a macro structural model with endogenous, exogenous variables and specify "all" the important causal relationships of the model. Such economic models can take different forms (systems of structural equations, structural var and vec ...). (2) another possibility is to use quasi-experimental techniques, where causality relationships are estimated by synthetic control, dif-in-dif, panel data, regression discontinuity (RDD), before and after, etc. In this second approach, the idea is to compare what is observed and a valid counterfactal (which eliminates endogeneity) without having to estimate all the existing relations of the economic system.
Dear Revana Hazem, if you are using the time series data in your research, you can use OLS, GMM, ARDL, or NARDL. I will suggest you to use the ARDL bounds testing approach.
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Writing