Dear all,
I am doing my dissertation and I having some problems with my empirical studies. Really need your helps !!! as the deadline is coming :(.
I have to test the impact of FDI on economic GROWTH in my country with time series analysis. Here a the problems:
1/ I use the Johansen VAR to find a cointegration relationship. I have 3 main endogenous variables: GROWTH (calculated by % increase of GDP), FDI inflows and initial GDP at beginning of every period. However I want to include some more variables which are argued to affect the FDI's impact on GROWTH which are HUMAN CAPITAL, TRADE OPENNESS, GOVERNMENT SPENDING. The problem is that I don't know whether to put these variables in the VAR as exogenous or endogenous.
2/ I have GROWTH, FDI, GDP all are I(1). If I put HUMAN CAPITAL, TRADE OPENNESS, GOV SPENDING variables as exogenous, do I have to test for unit root for them since normally if I put them as endogenous I have to make sure they are all I(1) as well. Moreover, as I know if I put them as exogenous in Eviews they will not appear in the long-run relationship. Thus, how can I conclude that their effects are positive or negative?
3/ I have GROWTH, FDI, GDP in quarterly data, but HUMAN CAPITAL in only yearly data but I still want to test effect of HUMAN CAPITAL. Is there any way to go around this?
Thank you in advance!
Best Regars
Bob
exogenous or endogenous for VAR
Moderators: EViews Gareth, EViews Moderator
Re: exogenous or endogenous for VAR
Some suggestions:
1. HUMAN CAPITAL and GOVERNMENT SPENDING are naturally exogenous in this model. If TRADE OPENNESS is measured as the share of X+M in GDP, then you have an endogeneity issue
2. Most economists consider that variables in a VAR should be I(0), NOT I(1). Typically, this is achieved by taking dlogs of the underlying level variables, be they endogenous or exogenous. Doing so will also enable you to get around the endogeneity problem in 1.
3. Use EViews automatic functions to interpolate (copy the annual data and paste to a new page set up as quarterly data)
Regards
Donihue
1. HUMAN CAPITAL and GOVERNMENT SPENDING are naturally exogenous in this model. If TRADE OPENNESS is measured as the share of X+M in GDP, then you have an endogeneity issue
2. Most economists consider that variables in a VAR should be I(0), NOT I(1). Typically, this is achieved by taking dlogs of the underlying level variables, be they endogenous or exogenous. Doing so will also enable you to get around the endogeneity problem in 1.
3. Use EViews automatic functions to interpolate (copy the annual data and paste to a new page set up as quarterly data)
Regards
Donihue
Re: exogenous or endogenous for VAR
Dear hieppk911
The following are my comments and suggestions:
1. According to the common practice, all endogenous variables must be I(1) if you want to use Johansen’s cointegration test. Here, I suspect that your growth rate of GDP is not I(1). Hence, the application of Johansen’s cointegration test may be inappropriate in this context. Regarding whether to specify HUMAN CAPITAL, TRADE OPENESS, and GOVERNMENT SPENDING as endogenous or exogenous is subjective. From my knowledge, I will specify the variables as endogenous instead of exogenous because all economic variables are inter-related in nature. Therefore, only deterministic components such as trend, constant, and dummies variables are usually treat as exogenous. However, this is only my rough guideline, and you have to ask yourself why do you think these variables are exogenous instead of endogenous and vice versa? There is no correct or wrong answer for this question; it is depend on how you justify it. Apart from that, you can also refer to other studies and learn how they treat and justify these variables if any.
2. You mentioned that GROWTH, FDI and GDP all are I(1). I am highly suspecting the reliability of your unit root findings. If your GROWTH is I(1), your GDP should be I(2) which is increase at the increasing rate over time, then only the growth rate of GDP (GROWTH) will increase from time to time. The finding of I(1) for GROWTH is a very unusual case and once yours supervisor or examiner look at this, they may have the same doubt as well. If you treat the variables as exogenous and you found that the variables are I(1), then you have to make them stationary (i.e. take a first difference) before put them in the exogenous box in Eviews. This is similar to Donihue's suggestion. Of course, the exogenous variables will not appear in long run because you presume that the variables are determined outside the system and it will only appears in the short run, while in the long run these variables are assumed to be independent from others.
3. Yes, you can use the interpolation technique provided by Eviews to derive the quarterly series from annual. Eviews provide various interpolation techniques (Please refer to Eviews User Guide), thus you need to chose one and justify it why this is your choice. In addition, you also need to understand the basic idea of the selected interpolation technique. In other words, how the data is interpolated instead of apply it blindly. Strictly speaking, I am not encouraging you to do it because now you mix the original quarterly data with the interpolated quarterly data which is not very inappropriate. Usually, researchers will use only one kind of data that is either interpolated or original. This may become a question later if you mix it. If annual data is sufficient, why not directly use annual data. Sometime annual data is better because you may less seasonality, non-normality, ARCH and others technical problems. Moreover, cointegration is a long run concept, thus using long span of annual data is more reasonable than just high frequency but short data span.
Finally, I hope that the above commons and suggestions are helpful to you to improve the quality of your research and Good luck :). Please check the degree of integration carefully with a proper choice of lags order and deterministic components in unit root test as well as in the Johansen test. Many Monte Carlo experiments have proven that these tests are very sensitive to the choice of lag order and also the deterministic component in the model. Instead of unit root tests, you can also plot the series at level and at the first difference to confirm the degree of integration. Sometime the simple graphical method tells more than complicated statistical tests.
Warmest regards,
tcfoon
The following are my comments and suggestions:
1. According to the common practice, all endogenous variables must be I(1) if you want to use Johansen’s cointegration test. Here, I suspect that your growth rate of GDP is not I(1). Hence, the application of Johansen’s cointegration test may be inappropriate in this context. Regarding whether to specify HUMAN CAPITAL, TRADE OPENESS, and GOVERNMENT SPENDING as endogenous or exogenous is subjective. From my knowledge, I will specify the variables as endogenous instead of exogenous because all economic variables are inter-related in nature. Therefore, only deterministic components such as trend, constant, and dummies variables are usually treat as exogenous. However, this is only my rough guideline, and you have to ask yourself why do you think these variables are exogenous instead of endogenous and vice versa? There is no correct or wrong answer for this question; it is depend on how you justify it. Apart from that, you can also refer to other studies and learn how they treat and justify these variables if any.
2. You mentioned that GROWTH, FDI and GDP all are I(1). I am highly suspecting the reliability of your unit root findings. If your GROWTH is I(1), your GDP should be I(2) which is increase at the increasing rate over time, then only the growth rate of GDP (GROWTH) will increase from time to time. The finding of I(1) for GROWTH is a very unusual case and once yours supervisor or examiner look at this, they may have the same doubt as well. If you treat the variables as exogenous and you found that the variables are I(1), then you have to make them stationary (i.e. take a first difference) before put them in the exogenous box in Eviews. This is similar to Donihue's suggestion. Of course, the exogenous variables will not appear in long run because you presume that the variables are determined outside the system and it will only appears in the short run, while in the long run these variables are assumed to be independent from others.
3. Yes, you can use the interpolation technique provided by Eviews to derive the quarterly series from annual. Eviews provide various interpolation techniques (Please refer to Eviews User Guide), thus you need to chose one and justify it why this is your choice. In addition, you also need to understand the basic idea of the selected interpolation technique. In other words, how the data is interpolated instead of apply it blindly. Strictly speaking, I am not encouraging you to do it because now you mix the original quarterly data with the interpolated quarterly data which is not very inappropriate. Usually, researchers will use only one kind of data that is either interpolated or original. This may become a question later if you mix it. If annual data is sufficient, why not directly use annual data. Sometime annual data is better because you may less seasonality, non-normality, ARCH and others technical problems. Moreover, cointegration is a long run concept, thus using long span of annual data is more reasonable than just high frequency but short data span.
Finally, I hope that the above commons and suggestions are helpful to you to improve the quality of your research and Good luck :). Please check the degree of integration carefully with a proper choice of lags order and deterministic components in unit root test as well as in the Johansen test. Many Monte Carlo experiments have proven that these tests are very sensitive to the choice of lag order and also the deterministic component in the model. Instead of unit root tests, you can also plot the series at level and at the first difference to confirm the degree of integration. Sometime the simple graphical method tells more than complicated statistical tests.
Warmest regards,
tcfoon
Re: exogenous or endogenous for VAR
Dear Donihue and tcfoon,
Thank you very much for your response! I really appreciate your helps.
@Donihue: Sorry but I do not clearly understand your 2/ suggestion. Could you explain it a little more? Btw, as I know VAR can be used for I(1) variables in order to find the cointegrating vectors between them.
@tcfoon: According to papers I read, HUMAN CAPITAL, TRADE OPENNESS and GOV SPEND is argued to affect the FDI's effect on growth however the papers do not conclude about effect of FDI or GROWTH on these terms. But I think I will put them as endogenous as you and Donihue suggest.
I do not enough annual data to do the test. I only have data of 20 years, that's why I think of quarterly. I double checked the oder of integration of GROWTH, if I test with log(GROWTH) then it is I(0) and with GROWTH it is I(1). I don't know which to choose, could you please explain this for me?
If i choose log(GROWTH) then it is I(0) and with FDI and initial GDP are I(1) how can I test the relationship between them? Moreover, I only have HUMAN CAPITAL in quarterly but I still want to test its effect on growth-enhancing impact of FDI. How can I solve this problem?
As I only study Johansen technique then I only know that in order to test for cointegration all variables need same order of integrations, I don't know how to do in another cases.
Anyway, my object is to test the effect of FDI on ECONOMIC GROWTH while also consider at HUMAN CAPITAL, TRADE OPENNESS and GOV SPEND, could you suggest me how can I do the test within the time-series analysis framework.
Again thank you all very much!!!
Best Regards
Hiep Pham
Thank you very much for your response! I really appreciate your helps.
@Donihue: Sorry but I do not clearly understand your 2/ suggestion. Could you explain it a little more? Btw, as I know VAR can be used for I(1) variables in order to find the cointegrating vectors between them.
@tcfoon: According to papers I read, HUMAN CAPITAL, TRADE OPENNESS and GOV SPEND is argued to affect the FDI's effect on growth however the papers do not conclude about effect of FDI or GROWTH on these terms. But I think I will put them as endogenous as you and Donihue suggest.
I do not enough annual data to do the test. I only have data of 20 years, that's why I think of quarterly. I double checked the oder of integration of GROWTH, if I test with log(GROWTH) then it is I(0) and with GROWTH it is I(1). I don't know which to choose, could you please explain this for me?
If i choose log(GROWTH) then it is I(0) and with FDI and initial GDP are I(1) how can I test the relationship between them? Moreover, I only have HUMAN CAPITAL in quarterly but I still want to test its effect on growth-enhancing impact of FDI. How can I solve this problem?
As I only study Johansen technique then I only know that in order to test for cointegration all variables need same order of integrations, I don't know how to do in another cases.
Anyway, my object is to test the effect of FDI on ECONOMIC GROWTH while also consider at HUMAN CAPITAL, TRADE OPENNESS and GOV SPEND, could you suggest me how can I do the test within the time-series analysis framework.
Again thank you all very much!!!
Best Regards
Hiep Pham
Re: exogenous or endogenous for VAR
Re your query on my point 2, it is in line with tcfoon's point 1 (GROWTH is very unlikely to be I(1)). If you want to estimate a VAR, use I(0) variables; if you want to estimate cointegrating relationships, you need all variables to be of the same degree of integration (typically I(1)),
Regards
Donihue
Regards
Donihue
Re: exogenous or endogenous for VAR
Dear hieppkk911,
You are most welcome:
After I read your text, regret to inform that I am really suspect the dataset that you used to perform the analysis. If you don’t mind, can you please tell me which country you analyse, what theoretical framework you employ to perform the analysis and how you to compute the GROWTH variable? If possible attached the dataset as well. You can’t simply put whatever form of variables you like without strong support by at least one theory. Otherwise, your findings are spurious even though you found that the variables are cointegrated.
Good luck...
Warmest regards,
tcfoon
You are most welcome:
After I read your text, regret to inform that I am really suspect the dataset that you used to perform the analysis. If you don’t mind, can you please tell me which country you analyse, what theoretical framework you employ to perform the analysis and how you to compute the GROWTH variable? If possible attached the dataset as well. You can’t simply put whatever form of variables you like without strong support by at least one theory. Otherwise, your findings are spurious even though you found that the variables are cointegrated.
Good luck...
Warmest regards,
tcfoon
Re: exogenous or endogenous for VAR
Dear Donihue and tcfoon,
Thank you very much!
@Donihue: I got your point. Thank you so much!
@tcfoon: I am investigating my country - Vietnam with data from World Bank Indicators which is only annually. (However, I am expecting to have quarterly data on GROWTH, FDI, GOV SPEND, OPENNESS from my country. I have some friends work for government and I am asking them to collect it for me)
The theoretical framework I use is from Alfaro (2003), "Foreign Direct Investment and Growth: Does the Sector Matter?". In his paper he suggest the equation:
GROWTH = B0 + B1*INITIAL GDP + B2*CONTROLS + B3*FDI + error.
With controls variables he use many variables like: domestic investment, trade openness, government spending, human capital, inflation, etc. I intend to include OPENNESS, GOV SPENDING and HUMAN CAPITAL in my study. However the problem is he using cross-country data for 47 countries but not time-series data for only 1 country.
I use GDP per capita annual growth rate for GROWTH variables. OPENNESS = share of X+M in GDP. GOV SPENDING = share of government consumption expenditure in GDP. HUMAN CAPITAL = primary school enrollment rate. Since the school enrollment rate as far as I know only reported in annual data, that is why I think I can't have quarterly data for HUMAN CAPITAL.
Almost all of FDI study using panel or cross-country data, very small number of them looking at particular country with time-series data. As I see in few study with particular country, they do not use GROWTH as dependent variable, instead they use GDP and exclude initial GDP in the right hand side of the equation. For example, a study of Sri Lanka use the following equations:
GDP = B0 + B1*FDI + B2*DOMESTIC INVESTMENT + B3*OPENNESS + error.
where GDP is in log form, FDI as percentage of GDP, Domestic Investment in log form, OPENNESS in log form as well. I guess they use GDP instead of GROWTH as GROWTH can't get I(1) then they can not use cointegrating technique.
I am confused whether to use log form or not, since some study use log for FDI and some do not. And in some study FDI is used as percentage of GDP, some just use FDI inflows in US$.
In sum, I have main problems:
1/I am no sure which model is appropriate since economists generally use model with GROWTH, but with GROWTH which is I(0) I can't use cointegrating technique ( the only method I am used to). If i use GDP, I do not know how to justify since in studies using GDP they do not explain it clearly either.
2/ It is easier if I use annually data, there will be a problem with just only 25 observations. If I use quarterly, I do not have HUMAN CAPITAL, moreover I am not sure I will get the quarterly data from my country yet.
I attached the data that I got from World Bank Indicators with only around 25 years of data and some missing data in several years.
Really hope that you can suggest me some solutions. Thank you very much in advance!
Best wishes
hieppk911
Thank you very much!
@Donihue: I got your point. Thank you so much!
@tcfoon: I am investigating my country - Vietnam with data from World Bank Indicators which is only annually. (However, I am expecting to have quarterly data on GROWTH, FDI, GOV SPEND, OPENNESS from my country. I have some friends work for government and I am asking them to collect it for me)
The theoretical framework I use is from Alfaro (2003), "Foreign Direct Investment and Growth: Does the Sector Matter?". In his paper he suggest the equation:
GROWTH = B0 + B1*INITIAL GDP + B2*CONTROLS + B3*FDI + error.
With controls variables he use many variables like: domestic investment, trade openness, government spending, human capital, inflation, etc. I intend to include OPENNESS, GOV SPENDING and HUMAN CAPITAL in my study. However the problem is he using cross-country data for 47 countries but not time-series data for only 1 country.
I use GDP per capita annual growth rate for GROWTH variables. OPENNESS = share of X+M in GDP. GOV SPENDING = share of government consumption expenditure in GDP. HUMAN CAPITAL = primary school enrollment rate. Since the school enrollment rate as far as I know only reported in annual data, that is why I think I can't have quarterly data for HUMAN CAPITAL.
Almost all of FDI study using panel or cross-country data, very small number of them looking at particular country with time-series data. As I see in few study with particular country, they do not use GROWTH as dependent variable, instead they use GDP and exclude initial GDP in the right hand side of the equation. For example, a study of Sri Lanka use the following equations:
GDP = B0 + B1*FDI + B2*DOMESTIC INVESTMENT + B3*OPENNESS + error.
where GDP is in log form, FDI as percentage of GDP, Domestic Investment in log form, OPENNESS in log form as well. I guess they use GDP instead of GROWTH as GROWTH can't get I(1) then they can not use cointegrating technique.
I am confused whether to use log form or not, since some study use log for FDI and some do not. And in some study FDI is used as percentage of GDP, some just use FDI inflows in US$.
In sum, I have main problems:
1/I am no sure which model is appropriate since economists generally use model with GROWTH, but with GROWTH which is I(0) I can't use cointegrating technique ( the only method I am used to). If i use GDP, I do not know how to justify since in studies using GDP they do not explain it clearly either.
2/ It is easier if I use annually data, there will be a problem with just only 25 observations. If I use quarterly, I do not have HUMAN CAPITAL, moreover I am not sure I will get the quarterly data from my country yet.
I attached the data that I got from World Bank Indicators with only around 25 years of data and some missing data in several years.
Really hope that you can suggest me some solutions. Thank you very much in advance!
Best wishes
hieppk911
- Attachments
-
- VietnamData_tcfoon.xls
- (26 KiB) Downloaded 1063 times
Re: exogenous or endogenous for VAR
Dear hieppk911,
1. I suggest you to change the model because there are voluminous researches that work on the role of FDI on economic growth using time series analysis. I suggest two work that you may refer to for your dissertation. One thing I would like to remind you is that model with more variables (multivariate) are not always superior to parsimonious model. As long as your model passed all the relevant diagnostic tests and supported with a strong theoretical background the estimation results should not be a problem.
(a) (2004) Foreign direction investment, economic growth, and financial sector development: A comparative analysis. ASEAN Economic Bulletin, 21(3), pp. 278-289.
(b) (2008) A re-examination of the role of foreign direct investment and exports in Malaysia’s economic growth: A time series analysis, 1970-2006. International Journal of Management Studies, 15(Bumper Issue), pp. 47-67.
2. The problem of natural logarithm or not and GROWTH or others will be solved after you look at the articles you may know what they used and follow them if relevant. For your information, for country-specific study, we seldom use per capita form because this is more suitable to use when we attempt to compare the results with different population size.
3. To make it simple, just use annual data with 25 observations is ok with me, but the variables in the model must be around 3 or 4 included your dependent variable. The ideal is 3. This is because it will consume the degree of freedom.
These are the solution that I can give to you, Good luck..
Warmest regards,
tcfoon
1. I suggest you to change the model because there are voluminous researches that work on the role of FDI on economic growth using time series analysis. I suggest two work that you may refer to for your dissertation. One thing I would like to remind you is that model with more variables (multivariate) are not always superior to parsimonious model. As long as your model passed all the relevant diagnostic tests and supported with a strong theoretical background the estimation results should not be a problem.
(a) (2004) Foreign direction investment, economic growth, and financial sector development: A comparative analysis. ASEAN Economic Bulletin, 21(3), pp. 278-289.
(b) (2008) A re-examination of the role of foreign direct investment and exports in Malaysia’s economic growth: A time series analysis, 1970-2006. International Journal of Management Studies, 15(Bumper Issue), pp. 47-67.
2. The problem of natural logarithm or not and GROWTH or others will be solved after you look at the articles you may know what they used and follow them if relevant. For your information, for country-specific study, we seldom use per capita form because this is more suitable to use when we attempt to compare the results with different population size.
3. To make it simple, just use annual data with 25 observations is ok with me, but the variables in the model must be around 3 or 4 included your dependent variable. The ideal is 3. This is because it will consume the degree of freedom.
These are the solution that I can give to you, Good luck..
Warmest regards,
tcfoon
Re: exogenous or endogenous for VAR
Dear tcfoon,
Thank you very much for your suggestion and the papers you give me! I will read the papers and try another model.
One more thing I would like to ask is about the number of lag included when test for cointegration using Johansen technique. I read some of your posts concerning this problem and I just want to confirm the right way to do it: Firstly, I run the VAR for the "1st difference" of the variables and choose to view the Lag Length Criteria. Then, the lag suggested in Lag Length Criteria from "1st difference" will be the number of lag included in cointegration test for variables at "levels". If I am wrong, please correct me.
Moreover, when I select Lag Lengthe Criteria, it asks for number of lag to include. As I see the number of lag to include will affect the lags suggested in the results. Could you please briefly explain it for me and what should be the right way to choose no of lag to include?
Best Regards
hieppk911
Thank you very much for your suggestion and the papers you give me! I will read the papers and try another model.
One more thing I would like to ask is about the number of lag included when test for cointegration using Johansen technique. I read some of your posts concerning this problem and I just want to confirm the right way to do it: Firstly, I run the VAR for the "1st difference" of the variables and choose to view the Lag Length Criteria. Then, the lag suggested in Lag Length Criteria from "1st difference" will be the number of lag included in cointegration test for variables at "levels". If I am wrong, please correct me.
Moreover, when I select Lag Lengthe Criteria, it asks for number of lag to include. As I see the number of lag to include will affect the lags suggested in the results. Could you please briefly explain it for me and what should be the right way to choose no of lag to include?
Best Regards
hieppk911
Re: exogenous or endogenous for VAR
Dear hieppk911,
Don't worry I think your way to determine the lag order for Johansen's cointegration test is correct.
Warmest regards,
tcfoon
Don't worry I think your way to determine the lag order for Johansen's cointegration test is correct.
Warmest regards,
tcfoon
Re: exogenous or endogenous for VAR
Thank you tcfoon!
I modify my model into:
LGDP= a0 + a1*LFDI + a2*OPENNESS +error.
Do you think it is ok?
Result from johansen technique shows that there is 1 cointegrating vector and a1, a2 are both positive.
Again thank you very much!
Best wishes
hieppk911
I modify my model into:
LGDP= a0 + a1*LFDI + a2*OPENNESS +error.
Do you think it is ok?
Result from johansen technique shows that there is 1 cointegrating vector and a1, a2 are both positive.
Again thank you very much!
Best wishes
hieppk911
Re: exogenous or endogenous for VAR
Dear hieppk911,
Now your model look better and simple.
Regarding the question of suitability, you better check with your supervisor because he/she is the right person to determine this and he/she has the final say.
Good luck.
Thanks you,
Warmest regards,
tcfoon
Now your model look better and simple.
Regarding the question of suitability, you better check with your supervisor because he/she is the right person to determine this and he/she has the final say.
Good luck.
Thanks you,
Warmest regards,
tcfoon
Re: exogenous or endogenous for VAR
Dear tcfoon,
I am very thankful. Really appreciate your help!
Best wishes
Hiep Pham
I am very thankful. Really appreciate your help!
Best wishes
Hiep Pham
Return to “Econometric Discussions”
Who is online
Users browsing this forum: No registered users and 1 guest