ECONOMIC GROWTH
(Barro.dta)
A. Introduction
Economic theory does not have a lot of answers regarding economic growth and the answers do not tend to have empirical corroboration. The data sets tend to have few observations, so it is hard to come up with strong conclusions.
Doing empirical work on macro variables often requires considerable ability to tolerate ambiguity. The Barro data set is particularly rich and offers a unique opportunity to test many far flung hypotheses. For cross country observations, this is an unusually large and comprehensive data set, yet it only has 114 countries and countries are not uniform commodities. This is considerably different from having thousands of observations of one stock price or hundreds of observations of wheat prices. So, in comparison to many micro econometric studies, the empirical results are never very persuasive and controversies remain unresolved.
In this section, I follow a different format close to that of the Barro article. I consider various simple regressions first (each devoted to a different idea). Then at the end of the section on growth rates, I present the multiple regression that I would have chosen before looking at the data.
B. Regressions: Growth rates
1. Growth convergence
Low income countries (other things being equal) should have higher growth rates since marginal productivity of capital is higher. That is, there should be convergence of countries over time. Past studies have not supported. this hypothesis. We can do a quick simple regression to get the flavor of such results (see the previous page for the diagram):
scat(R) GR6085 GDP60[1]
If the relationship is positive, it says that countries with lower GDP in 1960 had higher growth rates from 1960 until 1985. The scatter diagram shows that there is only very mild support for this hypothesis.
2. More human capital implies higher growth rates
This is the endogenous economic growth model. A simple endogenous growth model is that human capital can be passed on to future generation (and not only to own children) with little cost or depreciation. Therefore human capital is under-invested unless government policy to subsidize education. If government policy promotes human capital, then there will be high growth. This says that countries with more education should grow faster. There is some evidence to support this argument. Remember that should have lagged values. That is, education rates in 1960 determine growth rates in 1970s and 1980s. Also we may want to control for quality of education. Again, we can initially try a simple regression such as the following:
scat(R) GR6085 LIT60
The scatter diagram shows some evidence of a positive relationship.
3. Stability of property rights
If you do not know whether investment will be returned to you because property rights are insecure, then you will not invest and there will be no growth. This explains why countries in revolution do not have high growth rates, but cannot explain differential between Japan and U.S. Again, it is useful to start with a simple regression:
scat(R) GR6085 REVCOUP
The ocular test suggests that there is empirical support for this theory.
4. Fertility and population growth rate per capita.
Hypothesis: more people, lower growth rate per capita. Should you invest in more children or fewer children with higher productivity? We will not investigate this hypothesis.
5. Government expenditures: good or bad?
The a priori hypothesis depends on whether you are a Republican or a Democrat. May want to break down into defense and non-defense expenditures.
genr GOVOTHER = HSGOV - GDE - GEETOT
ls GR6085 c GDE GEETOT GOVOTHER
This regression states that the growth rate of per capita GDP depends on the ratio of government expenditures on defense to GDP (GDE), the ratio of government expenditures on education to GDP (GEETOT), and the ratio of other government expenditures to GDP (GOVOTHER).
The coefficients are expected to be negative for GDP, positive for GEETOT, and positive (negative) for GOVOTHER if you believe that government expenditures are efficient (inefficient).
LS // Dependent Variable is GR6085
Date: 6/6/93 / Time: 9:18
SMPL range: 1901 - 2018
Observations excluded because of missing data
Number of observations: 98
______
VARIABLE COEFFICIENT STD. ERROR T-STAT. 2-TAIL SIG.
______
C 0.0311348 0.0060915 5.1111623 0.000
GDE -0.0193424 0.0543865 -0.3556466 0.723
GEETOT 0.1327473 0.1196847 1.1091419 0.271
GOVOTHER -0.1317727 0.0332273 -3.9658007 0.000
______
R-squared 0.148075 Mean of dependent 0.022032
Adjusted R-squared 0.120886 S.D. of dependent 0.018518
S.E. of regression 0.017363 Sum of squared resid 0.028337
Durbin-Watson stat 1.796224 F-statistic 5.446134
Log likelihood 260.2224
The results are only very weakly supportive of the first two hypotheses.
6. Cultural dummy variables
AFRICA, Latin America, ASIAN, pacific rim. Economists generally frown on the use of such data.
7. Cross section or time series?
This data is basically cross section.
8. Multiple regressions
Now that we have some simple ideas we may want to combine them in a multiple regression:
ls GR6085 c GDE GEETOT HSINV
LS // Dependent Variable is GR6085
Date: 6/6/93 / Time: 9:19
SMPL range: 1901 - 2018
Observations excluded because of missing data
Number of observations: 98
______
VARIABLE COEFFICIENT STD. ERROR T-STAT. 2-TAIL SIG.
______
C 0.0013718 0.0047365 0.2896291 0.773
GDE -0.0068566 0.0447597 -0.1531871 0.879
GEETOT -0.3049850 0.1097691 -2.7784227 0.007
HSINV 0.1713973 0.0210204 8.1538609 0.000
______
R-squared 0.417520 Mean of dependent 0.022032
Adjusted R-squared 0.398930 S.D. of dependent 0.018518
S.E. of regression 0.014357 Sum of squared resid 0.019375
Durbin-Watson stat 1.725377 F-statistic 22.45961
Log likelihood 278.8523
The last variable is the ratio of investment to GDP. Generally, one would expect that this would increase growth. The results show a very clear role for investment, but investment in human capital by governments (education) has the wrong sign!
ls GR6085 c INV GDE SEC85 OIL
LS // Dependent Variable is GR6085
Date: 6/15/93 / Time: 4:48
SMPL range: 1901 - 2018
Observations excluded because of missing data
Number of observations: 99
______
VARIABLE COEFFICIENT STD. ERROR T-STAT. 2-TAIL SIG.
______
C -0.0021468 0.0043470 -0.4938672 0.623
INV 0.0957256 0.0279977 3.4190561 0.001
GDE -0.0315340 0.0488201 -0.6459214 0.520
SEC85 0.0124667 0.0073909 1.6867639 0.096
OIL 0.0032889 0.0067544 0.4869257 0.628
______
R-squared 0.299248 Mean of dependent 0.021715
Adjusted R-squared 0.269429 S.D. of dependent 0.018642
S.E. of regression 0.015934 Sum of squared resid 0.023866
Durbin-Watson stat 1.744861 F-statistic 10.03542
Log likelihood 271.8817
In this equation, we use a slightly different measure of INV, and we use an alternative approach to measuring education (SEC85). SEC85 measures the ratio of children enrolled in secondary education to the total number in that age bracket. OIL stands for a country that is a member of OPEC. The coefficient is expected to be positive. All of the coefficients are in the right direction, but not all of them are significant.
Finally, I run the equation that I would have run originally if I had not wanted to go step by step through the various issues:
ls GR66085 c GDP60 INV LIT60 REVCOUP SOC
LS // Dependent Variable is GR6085
Date: 5/15/94 / Time: 4:16
SMPL range: 1901 - 2018
Observations excluded because of missing data
Number of observations: 111
______
VARIABLE COEFFICIENT STD. ERROR T-STAT. 2-TAIL SIG.
______
C 0.0039055 0.0046259 0.8442587 0.401
GDP60 -0.0041071 0.0011739 -3.4987569 0.001
INV 0.1202557 0.0239137 5.0287323 0.000
LIT60 0.0118513 0.0068373 1.7333349 0.087
REVCOUP -0.0120364 0.0063868 -1.8845583 0.063
SOC -0.0116434 0.0042893 -2.7144921 0.008
______
R-squared 0.431730 Mean of dependent 0.020587
Adjusted R-squared 0.404670 S.D. of dependent 0.018732
S.E. of regression 0.014453 Sum of squared resid 0.021934
Durbin-Watson stat 1.552912 F-statistic 15.95427
Log likelihood 315.8703
C. Regressions: opportunistic empiricism
With this data set, we need not restrict our investigation to the determination of growth rates. We can ask many other questions, such as the determination of fertility and the causes of revolution (although the political data may be a bit strange).
1. Civil Liberties and Political rights
Consider the determination of civil liberties and polright (higher numbers means fewer liberties or political liberties
ls CIVLIB c SEC85 GDP85
LS // Dependent Variable is CIVLIB
Date: 07/31/96 Time: 15:38
Sample: 1901 2018
Included observations: 116
Excluded observations: 2
______
Variable Coefficient Std. Error T-Statistic Prob.
______
C 5.749196 0.224557 25.60241 0.0000
SEC85 -1.884538 0.628434 -2.998783 0.0033
GDP85 -0.258946 0.054039 -4.791825 0.0000
______
R-squared 0.565609 Mean dependent var 3.963793
Adjusted R-squared 0.557921 S.D. dependent var 1.856429
S.E. of regression 1.234322 Akaike info criterion 0.446566
Sum squared resid 172.1613 Schwartz criterion 0.517779
Log likelihood -187.4977 F-statistic 73.56716
Durbin-Watson stat 1.606199 Prob(F-statistic) 0.000000
ls POLRIGHT c LIT60 GDP80
LS // Dependent Variable is POLRIGHT
Date: 07/31/96 Time: 15:39
Sample: 1901 2018
Included observations: 113
Excluded observations: 5
______
Variable Coefficient Std. Error T-Statistic Prob.
______
C 6.348897 0.204519 31.04307 0.0000
LIT60 -4.042056 0.461054 -8.766989 0.0000
GDP80 -0.094072 0.044627 -2.107961 0.0373
______
R-squared 0.641897 Mean dependent var 4.018584
Adjusted R-squared 0.635386 S.D. dependent var 2.047583
S.E. of regression 1.236397 Akaike info criterion 0.450594
Sum squared resid 168.1547 Schwartz criterion 0.523002
Log likelihood -182.7986 F-statistic 98.58720
Durbin-Watson stat 1.806916 Prob(F-statistic) 0.000000
In both regressions increases in GDP per capita improve civil liberties and political rights. That is, Economic welfare improves political welfare.
ls CIVLIB c SEC85 GDP85 REVCOUP
LS // Dependent Variable is CIVLIB
Date: 07/31/96 Time: 15:40
Sample: 1901 2018
Included observations: 116
Excluded observations: 2
______
Variable Coefficient Std. Error T-Statistic Prob.
______
C 5.237795 0.274056 19.11217 0.0000
SEC85 -1.784504 0.607446 -2.937715 0.0040
GDP85 -0.218555 0.053814 -4.061300 0.0001
REVCOUP 1.482641 0.486250 3.049136 0.0029
______
R-squared 0.598904 Mean dependent var 3.963793
Adjusted R-squared 0.588161 S.D. dependent var 1.856429
S.E. of regression 1.191358 Akaike info criterion 0.384062
Sum squared resid 158.9654 Schwartz criterion 0.479013
Log likelihood -182.8725 F-statistic 55.74505
Durbin-Watson stat 1.610851 Prob(F-statistic) 0.000000
Revolutions reduce civil liberties (then again maybe a reduction in civil liberties increases the number of revolutions and coups).
All of these regressions can also be redone with the 1985 data. Have there been any substantial changes? Why would this be?
2. Fertility
What determines fertility rates? One would expect that higher levels of education and higher levels of GDP per capita would reduce fertility. Our results show that we can explain 60% of the variability in fertility with just these two variables.
ls FERT65 c LIT60 GDP65
LS // Dependent Variable is FERT65
Date: 5/15/94 / Time: 4:18
SMPL range: 1901 - 2018
Observations excluded because of missing data
Number of observations: 112
______
VARIABLE COEFFICIENT STD. ERROR T-STAT. 2-TAIL SIG.
______
C 7.3301038 0.1800064 40.721350 0.000
LIT60 -2.6594358 0.4264934 -6.2355849 0.000
GDP65 -0.2539961 0.0690727 -3.6772257 0.000
______
R-squared 0.619956 Mean of dependent 5.442857
Adjusted R-squared 0.612983 S.D. of dependent 1.723380
S.E. of regression 1.072127 Sum of squared resid 125.2907
Durbin-Watson stat 1.443849 F-statistic 88.90443
Log likelihood -165.2008
One might also consider the role of mortality. If there is a high infant mortality rate, more children would be born to replace the ones lost.
File: BARROTSP.WK1
Source:
DATA APPENDIX FOR
ECONOMIC GROWTH IN A CROSS SECTION OF COUNTRIES
ROBERT J. BARRO
HARVARD UNIVERSITY
AND
HOLGER C. WOLF
MIT
NOVEMBER
1989
Source 1950 to 2000" , Geneva
SIPRI: SIPRI Yearbooks, various issues
UNESCO: UNESCO Statistical Yearbooks , various issues
WB: World Bank World Tables, various editions
Variable Name | Definition and Source
------
AFRICA Dummy for Sub-Sahara Africa
ASSASS Number of assassinations per million population per year (1960-1985 or sub sample) Source: Banks
AVAGExx Average age of labor force. Constructed by multiplying age interval midpoints by interval size. Final point 70 years. SOURCE: ILO
BENCH Dummy for Summers and Heston Benchmark countries Source: HS88
BIGSMPL Dummy for the 98 country sample
CIVLIB Index of civil liberties (1 = highest, 7 = lowest) Source: Gastil
CONSTCH Number of constitutional changes (1960 to 1985 or subsample). Source: Banks
COUP Number of coups per year (1960 to 1985 or subperiod) Source: Banks
CRISES Number of Government Crises per year (1960 to 1985 or subperiod) Source: Banks
FERTxx Total fertility rate (children per woman) (1965 and 1985) Source: WB
FERTAV Total fertility rate , average of FERTxx for 1965 and 1985 Source: WB
FERTNET FERTAV*(1-MORTAV)
FERTNETC FERTAV*(1-MORT04)
GDE Average from 1970 to 1985 of the ratio of nominal government expenditure on defense to nominal GDP. Source: GFS , SIPRI
GDPxx GDP per capita in real terms Source: HS88
GEECUR Average from 1970 to 1985 of the ratio of current nominal government expenditure on education to total nominal government expenditure on education. Source: UNESCO
GEETOT Average from 1970 to 1985 of the ratio of nominal government expenditure on education to nominal GDP. Source: UNESCO GFS
GGCFD Average from 1970 to 1985 of the ratio of gross real public domestic investment (using HS deflator for investment) to real GDP (deflated). Source: HS88 IFS GFS
GII Average from 1970 to 1985 of the ratio of real public domestic investment to real domestic investment (private plus public).(GII = GGCFD/HSINV) Source: HS88 IFS GFS
GOV Ratio of real government "consumption" expenditure to real GDP.Average from 1960 to 1985) Source: HS88
GPOPxxyy Growth rate of population from 19xx to 19yy
GRxxyy Growth rate of per capita GDP Source: HS88
GTRAN Nominal Government Transfer Payments as ratio to nominal GDP (Average 1970 to 1985) Source: GFS
HSGOV Ratio of real government "consumption" expenditure to real GDP.(Average from 1970 to 1985). SOURCE: HS88
HSGVXDXE Ratio of real government "consumption" expenditure net of spending on defense and on education to real GDP. (HSGVXDXE = HSGOV-GDE-GEETOT) NOTE: It would be preferable to adjust for GEECUR
HSINV Average from 1970 to 1985 of the ratio of real domestic investment (private plus public) to real GDP Source: HS88
INV Same as HSINV for 1960 to 1985 Source: HS88
LAAMER Dummy variable for Latin America
LIT60 Adult literacy rate in 1960. Source: WB