03. IEF: The Gravity Model of Trade - Theoretical Introduction and Estimation in Eviews and R

[@elwirabubel1181]

Very accessible presentation of a rather complex problem. Thank you very much!!!

[@pashton]

that XLXX made me crack a laugh too, thank you for your effort awesome videos.

[@mohammedalnour318]

Professional presentation, MANY THANKS dear Dr

[@renatahuseynova7117]

Thanks for the explanation, the video is explained in a very clear and easy way to understand.

[@MichalCiara]

Great video :)

[@deepanshulakhwan3809]

Quality material, loved the variable part. 💬✅

[@milenabarka7955]

Very helpful in writing a resit paper on gravity model of trade ;P

[@sho1715]

Nice video! Very helpful to understand the theory

[@lazarskiopencourses4059]

thank you :)

[@robiyausmanova6444]

Thank you very much! Great explanation and it was really helpful for my school assignment

[@lazarskiopencourses4059]

thank you for the kind words :)

[@mariakrowka3806]

Awesome video ! Very easy to understand :)

[@user-ee8qw1uo3w]

thank youuuu much

[@lazarskiopencourses4059]

Our pleasure :)

[@MichalCiara]

👍👍👍👍👍

[@SG-lj8uu]

Thank you so much

[@lazarskiopencourses4059]

Our pleasure :)

[@barborapavlikova6610]

Very interesting, I have read your paper, too. I would only like to ask you if you could share a simple explanation on the transformation of nominal exports and imports in real terms. Thank you!

[@lazarskiopencourses4059]

Hi, because the data was expressed in US dollars (its from IMF Directions of Trade) I used USA CPI to get the real values.

[@barborapavlikova6610]

Do you mean price level of export and import from PWT right?

[@lazarskiopencourses4059]

@@barborapavlikova6610 Hi, I have written this paper long time ago please let me check in my files and get back to you.

[@lazarskiopencourses4059]

@@barborapavlikova6610 Yes

[@Dr_Coffie]

You did not show how the trade volume is calculated

[@lazarskiopencourses4059]

It is the sum of real exports and imports

[@hieuvuminh8761]

can it be used on cities instead of countries, from cities A to cities B perhaps?

[@lazarskiopencourses4059]

Hi, in principle yes, however, in practice it is really hard to find data on trade between cities. Therefore, researchers use data on regional trade (like provinces, states, counties, or NUTS region in the European Union). In this case you use capital cities in order to measure the distance between the regions. Please, let me know if this does not answer you question. Best regards

[@sthmdy8859]

Please explain why the signs of alpha and beta are interpreted in reverse ?

[@lazarskiopencourses4059]

Please notice equation (1) at the beginning of the video. Alpha is an exponent in the numerator while Beta is an exponent in the denominator. Consequently, the higher the Alpha the bigger the trade and the bigger the Beta the lower the trade. This result is in tact with the intuition of the model explained in the video.

[@sthmdy8859]

@@lazarskiopencourses4059 Thanks a lot. I am studying again.

[@MrAndri9]

Thank you so much for the video🙏 I'm wondering how you would calculate the home bias in trade by adding the dummy variable for "home" that takes 1 if i = j , how would you calculate the domestic trade ? I know that in shang jin wei paper (1996) [link below] argues that intranational trade = domestic production−exports, but what is domestic production in that sense because it can't just be GDP right ?

[@lazarskiopencourses4059]

Domestic production = GDP-imports. Going back to you first question, to calculate domestic trades you need to have data on regions within countries. Then calculate the model where regions trade with each other and then you can assess home bias by comparison of trade between the regions from the same countries with the trade between the regions from different countries. The seminal work on the subject are: www.hec.unil.ch/mbrulhar/Paris/mccallum_95.pdf and artnet.unescap.org/tid/artnet/mtg/gravity10_reading3.pdf I hope this answer helps. Please let me know if you need any additional information :)

[@MrAndri9]

@@lazarskiopencourses4059 Thank you for your answer, it really helped :)

[@lazarskiopencourses4059]

@@MrAndri9 Our pleasure :)

[@esamaulina]

Nice explanation video, Please can you tell me for some a book or e-book related to this gravity model? my constraints on references are hard to collect. Thank you so much

[@lazarskiopencourses4059]

Thank you for the kind words. The video is prepared based on a paper you can find here: www.researchgate.net/publication/344617702_What_drives_international_trade_Robust_analysis_for_the_European_Union There you can find references to other papers on the subject. I also recommend this paper: www.researchgate.net/publication/321050853_Bayesian_model_averaging_and_jointness_measures_Theoretical_framework_and_application_to_the_gravity_model_of_trade The book that gives the most thorough dive into the gravity model of trade is: Feenstra, R., "Advanced International Trade: Theory and Evidence" Please let me know if you need any other information. Best regards :)

[@karlmags2274]

hey thanks for the video, it really helped me with understanding what this is all about. however, how do we interprete the coefficients of the dummy variables? i doubt its the same like for normal variables, right?

[@lazarskiopencourses4059]

Hi, because here we have a case where dependent variable is in logarithm we need to use special formula. For example, the coefficient on border dummy is approximately 0.35. We cannot interpret it as: having common border is associated with trade higher, on average, by 35 percent - this is a very common mistake made even by nobel prize winners (You can find references to famous economist making this mistake in a paper in the link at the end). If the coefficient on dummy variable is b, then appropriate interpretation is: having common border is associated with trade higher, on average, by (e^b-1)*100=(e^0.35-1)*100=42 percent. The reference to the formula and examples of famous economists making mistakes can be found here: fvela.files.wordpress.com/2010/11/dummyinterpretation.pdf

[@emiliaborko7016]

Thank you Prof. for this video. Could you explain how to capture multilateral resistance term? In the article of van Wincoop (2003) I read about price indices. I read that the other way are country-year fixed effects. How to implement them? Is it ok if I just run fixed effect regression in stata? I work with panel data

[@lazarskiopencourses4059]

Hi, the beginners information how to estimate multilateral resistance you can find here: artnet.unescap.org/tid/artnet/mtg/cbcam_d2s3.pdf More detailed information is here www.st-andrews.ac.uk/~wwwecon/CDMA/papers/cp0702.pdf apcz.umk.pl/DEM/article/view/DEM.2014.002 Finally, I also recommend seeing this if you want to try estimations in R: www.unescap.org/sites/default/files/Gravity-model-in-R_1.pdf

[@emiliaborko7016]

@@lazarskiopencourses4059 Thank you Prof. It helped me a lot. I wanna investigate the effect on GDP and exchange rates. I found in the guide of Shepherd that we should take the total nominal GDP. Should we take nominal exchange rates as well? I didn't find the information about it on the Internet

[@lazarskiopencourses4059]

@@emiliaborko7016 Hi, it really depends on what exactly are you trying to examine. If you are examining impact on nominal exchange rates that should be the best choice. Please, let me know if this is answering your question. Or please provide me with additional questions.

[@emiliaborko7016]

@@lazarskiopencourses4059 Thank you for your answer. I regress export at the sectoral level. I obtained GDP and exchange rates elasticities for different sectors. When drawing a histogram for GDP it doesn't look bad. However, when drawing it for exchange rates it looks like the exchange rates elasticities are extremely poor behaved. I thought maybe it is the case that I used nominal exchange rates. But it was only my guess. However, it looks like there is no recommendation what should I use.

[@emiliaborko7016]

@@lazarskiopencourses4059 Dear Prof. I've just came up with the idea that maybe my data is inappropriate. I have data on bilateral trade for more than 6000 sectors. I'm doing regression for every sector to see the elasticities of GDP and exchange rates. From my histogram it looks like they are extremely poor behaved. It might be the case that I have insufficient observations for this 6000 sectors. When regressing in stata I had to add capture command in my loop since some regressions failed due to insufficient observations. I think I have too many sectors and it should be more general. On the other side it takes my stata 11 hours to calculate the elasticities. The task is kinda frustrating when trying different models. Could you recommend me a database on bilateral trade when sectors would be more generalized so that I would do the regressions for less sectors?

[@nirmuk6217]

While using panel data .. distance between two countries will be same in different year . So resultof p value is not good . Can you suggest how to take distance?

[@lazarskiopencourses4059]

Hi, in panel data setting you can just skip the distance cause it can be taken into account through fixed country effects. However, if you add this variable it will have amazingly small value, so regardless of the specific form of the distribution will be significant. The coefficient will remain unchanged so the interpretations based on this coefficient are still correct.

[@mukeshniroula8209]

Thank you for your video.. I am using panel data gravity model for tourism demand, while using distance I got tood bad result on p value. How to choose distance between two countries?

[@lazarskiopencourses4059]

Hi, we usually take the distance between the capitals. The database that contains information about that is: www.cepii.fr/CEPII/en/bdd_modele/presentation.asp?id=8 while the description of the data can be found here: www.cepii.fr/DATA_DOWNLOAD/gravity/doc/Gravity_documentation.pdf

[@mukeshniroula8209]

@@lazarskiopencourses4059 thank you alot

[@nl7287]

Thank you sir, sir could you please make a video how to analyse gravity model using PPML and Heckman model in eviews

[@lazarskiopencourses4059]

Please wait a while, and we'll try :)

[@nl7287]

Thank you 🙏

[@lydiaexplains]

@@lazarskiopencourses4059 thank you for this great video, I am also trying to to a PPML gravity estimation and would love to see a video on that :)

[@mukeshniroula8209]

In a gravity model should all the independent variables must be in pair?

[@lazarskiopencourses4059]

Yes, this is a model that uses bilateral data.

[@minyarkhayat40]

please could you answer my question. I work with panel data on bilateral trade of 54 countries trading 100 specific products using a structural gravity model. I'm still biggener. How can I set up the database in Excel? I can sort the trade by country pairs maybe but how I will sort it by product in parallel. Please clarify my concerns if you can thank you

[@lazarskiopencourses4059]

Well this question is problematic. Are you trying to investigate total bilateral trade or for example intra-industry trade? the issue here is associated with the fact that within a given product group the reference GDPs are the same, as well as distances. This will heavily weight on the results. Can describe in more detail what are you trying to do, and maybe I'll be able to help a little :)

[@minyarkhayat40]

Thank you for you answer, In fact, i 'm doing an analyse of the trade of enviromental goods between africain countries. So, my goal is to run a structural gravity model to detect constraints of exchange of such goods amongst those countries and i' m seeking to evaluate the potential of exports compared tp the actual trade. So i think i have to interptet resultats individually according to each specific product not for a groupe of products and for specific couple of countries.

[@MohamedElAnya-nn3yd]

hi, good explanation, I have a question, I am doing an empirical study by gravity model on a sample of 47 countries, to estimate the impact of the free trade agreements concluded by Morocco to capture the effects of creation and the diversion of exchanges on a panel of 47 countries for 4 agreements, I have two equations, the first equation contains the following dependent variable the exports from country i to country j, the independent variables contain the gdp of country i and j, the distance, the population, the bilateral exchange rate, and dummy variables to capture the creation and the diversion under its agreements, the problem is that I have no idea how to calculate the diversion under agreements, the creation as for it I will calculate without problem namely ALE (ijk) t: Dummy variable takes the value 1 if the two countries belong to the same ALEk at date t and 0 otherwise. the diversion of trade exALE (ijk) t: Dummy variable takes the value 1 when the exports of the country i belong to the FTA k on the date t without the country j participating, 0 otherwise. it is here that j I have the problem, I don't know how to calculate it on excel this dummy variable of the diversion, if you know what to do tell me thank you. the second equation contains the same independent variable, the only change is in the dependent variable Mij: import of country i from country j with dummy variable imALE (ijk) t: Dummy variable takes the value 1 when imports from country j belong to FTA k at date t without country i participating, 0 otherwise. same problem I don't know how to calculate i

[@lazarskiopencourses4059]

Hi, there is a very simple trick to do this. Please check this paper for the detailed discussion: www.sciencedirect.com/science/article/abs/pii/S0022199603000588

[@MohamedElAnya-nn3yd]

Thank you i have another question

[@MohamedElAnya-nn3yd]

In my panel i have multicolinearity ,autocolleration ,and heteroskedacity what can i do to fix this problems

[@lazarskiopencourses4059]

@@MohamedElAnya-nn3yd multicolinearity is not an issue you have to worry about. The easiest way to account for autocorrelation and heteroscedasticity is to use a robust covariance matrix - Newey-West is the most popular option that is available in virtually all econometric software. Alternatively, fixed effects usually to the trick when you want to account for heteroscedasticity and the use of dynamic panel (introduction of lagged dependent variable) can help you with the autocorrelation.

[@MohamedElAnya-nn3yd]

@@lazarskiopencourses4059 thank you you helped me a lot ,i think there is another solution ,by using cross section weights EGLS

[@tokofora8783]

I did my gravity model to analyze bilateraltrade. I modified it with some variables using gretl as you did. But i'm confuse on how to interpret the OLS result. Can you help me?

[@lazarskiopencourses4059]

Can you tell me which variables you want me to help interpret. Cause, some of the interpretations are already given in the video.

[@tokofora8783]

@@lazarskiopencourses4059 : can i send you an email? It is too long to explain in here

[@simonasimovicova1813]

@@tokofora8783 Hi, i would like to know if you did it? I am also working on paper which is about bilateral trade using the gravity model. Can I ask you some questions:?

[@tokofora8783]

@@simonasimovicova1813 yes

[@simonkasimovicova9294]

Can i contact you somewhere via message? Can you give me your email or another contact information. Thank you!

[@nanditadasgupta2049]

Can I get an access to your data table to understand the process?

[@lazarskiopencourses4059]

Hi, the link to the data is in the description of the video: www.researchgate.net/publication/344452737_Replication_data_for_Beck_K_2020_What_drives_international_trade_Robust_analysis_for_the_European_Union_Journal_of_International_Studies_133_68-84_doi_10142542071-8330202013-35

[@collinsjohn1468]

Hello Prof. awesome material. could you post the codes to replicate this result on r?

[@lazarskiopencourses4059]

Of course. I'll add the link to the description today.

[@collinsjohn1468]

@@lazarskiopencourses4059 thanks Prof. that will be really appreaciated.

[@lazarskiopencourses4059]

@@collinsjohn1468 hi, the link to the code has been added to the description. If you have any more detailed question, please do not hesitate to contact me here: www.researchgate.net/profile/Krzysztof-Beck

[@georgegeorge6411]

Nice work! what about heteroskedasticity? Did you have issues with that?

[@lazarskiopencourses4059]

Yes, in some of the models there is heteroskedasticity issue. However it is easy to overcome with robust covariance matrix (for example Newey-West in Eviews).

[@georgegeorge6411]

@@lazarskiopencourses4059 Thank you for your response! I tried this method but it doesn't work. Can you suggest an alternative solution?

[@lazarskiopencourses4059]

@@georgegeorge6411 Could you be more specific? What is exactly happening? What type of data are you using? What is the model specification?

[@georgegeorge6411]

@@lazarskiopencourses4059 my objective is to analyze the exports of UK from 2000 to 2019. So my independent variable is lnexports( i logged every variable except dummies). My data are panel. I have 6 dependent variables gdpi gdpj geographical distance gdppercapitai and gdppercapitaj and eumemeber.All numbers are in thousands dollars. So the main problem is that my regression has heteroscedasticity and autocorrelation. I tried several techniques but nothing worked.

[@lazarskiopencourses4059]

​@@georgegeorge6411 Well in panel data setting situation becomes more troublesome. I thing the best thing you can do is to resort to cointegration with fixed effect (however in this will result in the elimination of distance) and a related error correction mechanism. Alternatively, if the variables are characterised by different degree of integration, the best solution would be to use autoregressive distributive lag (ARDL) model - you should perform bounds test before deciding on whether to relay on cointegration or ARDL. Gravity model of trade becomes very problematic when estimated in panel setting - for example due to apparent simultaneity between trade and GDP. So, if you do not want to rely on cointegration or ARDL, probably you should choose some instrumental variable approach. In this link you can find a lot of sources of information on the subject: www.tandfonline.com/doi/abs/10.1080/07474938.2016.1167857

[@bumbiierdene4059]

Can you show me in stata?

[@lazarskiopencourses4059]

Why not, just give me some time, I need to prepare more videos on other subjects.

[@MaroofKhan-cr9ky]

@@lazarskiopencourses4059 still waiting for stata commands use for estimating gravity model using ols, FE, RE and also ppml

[@lazarskiopencourses4059]

@@MaroofKhan-cr9ky the video on estimation of the model in stat is coming

[@lazarskiopencourses4059]

Hi, here is the link to the video describing estimation of the gravity model in STATA: kzbin.info/www/bejne/kJS7i31vgcygf6c&ab_channel=LUDepartmentofEconometrics

[@asiaplewko2190]

Great video :)