Econometrics 463 Term Project:

The Relationship Between

Economic Freedom and Prosperity Around The World

By David Veksler

December 10, 2002

Introduction

The collapse of the Soviet Union has been followed by worldwide economic liberalization and increasing international trade.  The increase in economic freedom has been followed by a global increase in prosperity, but not without setbacks.  The economic slowdown beginning in the late 1990’s has caused some to question the progress of economic liberalization as well as its benefits.  An analysis of relationship between economic freedom and prosperity may be useful in determining whether economic freedom increases prosperity and which specific factors have the greatest effect on wealth.  This information would be very useful to anyone seeking to determine the weakness of various governmental policies.  Governments may find it important to know which factors have the greatest effect on economic growth and investors may find it helpful to predict which countries are more likely to develop as potential markets. 

Methodology

For this paper, ten factors that measure different aspects of economic freedom were correlated against the 2001 per capita GDP of 155 nations.  There have been several studies on economic freedom since 1980 (Easton, 1997) by various think tanks.  The Heritage Institute results were used for this paper because they measured the largest number of factors, and had data for over 155 countries –more than the other studies.  For the GDP data, the 2001 results from the CIA Factbook were used because the CIA had the most complete and generally the latest GDP data.  The Heritage Institute’s 2003 report actually came out in November 2002, but the 2001 data was used because for the majority of countries, the latest available GDP data is still for 2000 or 2001.  A comparison of the Heritage Institute evaluations of economic freedom to those of the Fraser Institute (2002), showed that the reports all gave the same approximate evaluations of economic freedom.

There were ten independent variables tested:  (the abbreviations used in are parenthesis)

• Trade policy (Trade)

• Fiscal burden of government (FiscalBu)

• Government intervention in the economy (Governme)

• Monetary policy (Monetary)

• Capital flows and foreign investment (ForeignI)

• Banking and finance (BK)

• Wages and prices (Wagesand)

• Property rights (Property)

• Regulation (RG)

• Black market (BlackMar)

Each of these variables was a cardinal value from one to five, with one being the least government involvement and five being the most.   The dependent variable was indGDP, the 2001 per capita GDP.   Additional variables considered were Econ_Sco, the overall economic score and WorldRan, the ranking according to the overall score.

Regressions were run on the overall economic score and then the ten independent variable.  Initially a linear model was used, then all the variables were tested for significance under logarithmic, quadratic, and interactive relations, and variables that were not significant under any test at the 5% level were dropped from the model.

The CIA data had estimated GDP data for 236 regions -- almost every country and territory on earth, but the Heritage listed 161 countries.  Of these, 155 had data available for all ten variables, so 155 countries and 10 factors were used in estimating the model.  The software used for all the regressions was Gretl.


Procedure

The first step was to find out what overall relationships were present in the model.  For this regression, the indGDP and Econ_Sco variables were used.  A linear regression between them produced an unadjusted R-squared of 0.496862. 

  VARIABLE      COEFFICIENT        STDERROR       T STAT   2Prob(t > |T|)

  0)    const     34622.6             2160.89         16.022   < 0.00001 ***

  5) Econ_Sco     -8480.91             689.958       -12.292   < 0.00001 ***

  Mean of dependent variable = 8854.32

  Standard deviation of dep. var. = 9169.89

  Sum of squared residuals = 6.51533e+009

  Standard error of residuals = 6525.63

  Unadjusted R-squared = 0.496862

  Adjusted R-squared = 0.493573

A more closer match was found by adding sq_Econ to the model:

      VARIABLE      COEFFICIENT        STDERROR       T STAT   2Prob(t > |T|)

   0)    const     76270.6             5767.89         13.223   < 0.00001 ***

   5) Econ_Sco    -36958.0             3782.94         -9.770   < 0.00001 ***

  17) sq_Econ_      4575.07             600.351         7.621   < 0.00001 ***

  Mean of dependent variable = 8854.32

  Standard deviation of dep. var. = 9169.89

  Sum of squared residuals = 4.71419e+009

  Standard error of residuals = 5569.06

  Unadjusted R-squared = 0.635953

  Adjusted R-squared = 0.631163

The graph below shows the predicted model and the Econ_Sco variables:

The relationship here is clear: increasing government involvement lowers GDP at a rate of $36, 958 per index point.  The quadratic function seems to indicate that super-high levels of involvement actually raise GDP, but only about six countries seem to be part of that trend, with two having a per capita GDP above 5000: Iran and Libya.

After a simple one-factor regression, a regression with all ten factors (but not Econ_Sco) was attempted.  Initially, a simple linear regression was used with all 155 variables and all 10 factors, without any quadratic or interacting variables:

    VARIABLE      COEFFICIENT        STDERROR       T STAT   2Prob(t > |T|)

   0)    const     20847.5             2372.98          8.785   < 0.00001 ***

   6)    Trade     -1209.12             445.846        -2.712    0.007503 ***

   7) FiscalBu      1576.36             481.540         3.274    0.001330 ***

   8) Governme       976.563            594.194         1.644    0.102459

   9) Monetary      -609.650            333.642        -1.827    0.069732 *

  10) ForeignI       194.577            720.675         0.270    0.787553

  11)       BK      -826.296            689.728        -1.198    0.232884

  12) Wagesand      1382.06             720.518         1.918    0.057071 *

  13) Property     -2068.62             739.690        -2.797    0.005870 ***

  14)       RG      -213.362            838.028        -0.255    0.799395

  15) BlackMar     -2863.18             535.661        -5.345   < 0.00001 ***

  Mean of dependent variable = 8854.32

  Standard deviation of dep. var. = 9169.89

  Sum of squared residuals = 3.43621e+009

  Standard error of residuals = 4884.93

  Unadjusted R-squared = 0.734643

  Adjusted R-squared = 0.716216

  F-statistic (10, 144) = 39.8665 (p-value < 0.00001)

  Durbin-Watson statistic = 2.14011

  First-order autocorrelation coeff. = -0.071209

(Higher coefficients indicate higher government involvement in the various areas.)

A test for quadratic variables was used:

      VARIABLE      COEFFICIENT        STDERROR       T STAT   2Prob(t > |T|)

   0)    const     25616.2             5818.15          4.403    0.000022 ***

   6)    Trade      3241.39            1794.43          1.806    0.073105 *

   7) FiscalBu     -4843.72            2560.55         -1.892    0.060693 *

   8) Governme       596.296           2228.13          0.268    0.789402

   9) Monetary      -263.771           1262.99         -0.209    0.834885

  10) ForeignI      -867.630           2083.24         -0.416    0.677725

  11)       BK     -2086.66            1858.50         -1.123    0.263547

  12) Wagesand     -2955.80            2517.71         -1.174    0.242476

  13) Property     -3070.39            1991.59         -1.542    0.125510

  14)       RG     -1281.77            2795.31         -0.459    0.647305

  15) BlackMar     -7242.81            1796.90         -4.031    0.000093 ***

  16) sq_Trade      -423.008            265.461        -1.593    0.113407

  17) sq_Fisca       744.970            378.890         1.966    0.051343 *

  18) sq_Gover      -149.109            386.857        -0.385    0.700524

  19) sq_Monet       121.290            207.138         0.586    0.559160

  20) sq_Forei        88.0300           369.384         0.238    0.812000

  21)    sq_BK       338.051            314.633         1.074    0.284561

  22) sq_Wages       216.479            414.127         0.523    0.602022

  23) sq_Prope       626.785            334.349         1.875    0.063019 *

  24)    sq_RG       135.053            428.646         0.315    0.753199

  25) sq_Black      1210.16             262.880         4.603   < 0.00001 ***

  Unadjusted R-squared = 0.396284

  Adjusted R-squared = 0.306177

Test statistic: TR^2 = 61.424005,

with p-value = prob(Chi-square(10) > 61.424005) = 0.000000

And a test for logs:

      VARIABLE      COEFFICIENT        STDERROR       T STAT   2Prob(t > |T|)

   0)    const     -1873.56            2260.44         -0.829    0.408665

   6)    Trade     -1990.24            1334.03         -1.492    0.138075

   7) FiscalBu      4059.26            2287.05          1.775    0.078187 *

   8) Governme      -506.908           1945.41         -0.261    0.794827

   9) Monetary       804.378           1058.59          0.760    0.448676

  10) ForeignI       192.145           1865.56          0.103    0.918120

  11)       BK      1984.84            1670.56          1.188    0.236883

  12) Wagesand       847.905           2099.09          0.404    0.686902

  13) Property      3116.16            1779.32          1.751    0.082179 *

  14)       RG       343.683           2304.40          0.149    0.881666

  15) BlackMar      6442.66            1340.24          4.807   < 0.00001 ***

  17)  l_Trade      6925.99            3734.78          1.854    0.065872 *

  18) l_Fiscal    -12569.4             7106.97         -1.769    0.079236 *

  19) l_Govern       978.852           4892.75          0.200    0.841736

  20) l_Moneta      -960.438           2574.34         -0.373    0.709678

  21) l_Foreig     -1249.52            4273.17         -0.292    0.770425

  22)     l_BK     -4921.49            3858.68         -1.275    0.204362

  23) l_Wagesa     -6570.98            5447.23         -1.206    0.229828

  24) l_Proper     -6491.89            4115.55         -1.577    0.117061

  25)     l_RG     -2554.60            6290.98         -0.406    0.685337

  26) l_BlackM    -16450.3             3732.15         -4.408    0.000021 ***

  Unadjusted R-squared = 0.411176

  Adjusted R-squared = 0.323292

Test statistic: TR^2 = 63.732241,

with p-value = prob(Chi-square(10) > 63.732241) = 0.000000

It appears that sq_Fisca, sq_Prope,  sq_Black, and l_Trade and l_BlackM are significant, so these variables were created:

Model 2: OLS estimates using the 155 observations 1-155

Dependent variable: indGDP

      VARIABLE      COEFFICIENT        STDERROR       T STAT   2Prob(t > |T|)

   0)    const     34351.3             9683.62          3.547    0.000531 ***

   6)    Trade     -2750.11            1190.66         -2.310    0.022376 **

   7) FiscalBu     -2757.13            2424.35         -1.137    0.257385

   8) Governme       964.732            488.511         1.975    0.050267 *

   9) Monetary      -109.990            271.083        -0.406    0.685555

  10) ForeignI      -190.634            582.886        -0.327    0.744120

  11)       BK      -965.605            556.143        -1.736    0.084735 *

  12) Wagesand       -63.7690           603.954        -0.106    0.916063

  13) Property     -5953.28            1787.79         -3.330    0.001112 ***

  14)       RG      -867.724            679.755        -1.277    0.203899

  15) BlackMar      1615.02            9421.44          0.171    0.864143

  16)  l_Trade      5907.95            3327.77          1.775    0.078029 *

  17) l_BlackM    -15172.6            11337.6          -1.338    0.182998

  18) sq_Fisca       651.370            362.476         1.797    0.074507 *

  19) sq_Prope       800.209            284.262         2.815    0.005586 ***

  20) sq_Black       233.137            848.323         0.275    0.783861

  Mean of dependent variable = 8854.32

  Standard deviation of dep. var. = 9169.89

  Sum of squared residuals = 2.08499e+009

  Standard error of residuals = 3872.98

  Unadjusted R-squared = 0.838989

  Adjusted R-squared = 0.821614

  F-statistic (15, 139) = 48.2863 (p-value < 0.00001)

  Durbin-Watson statistic = 1.84585

  First-order autocorrelation coeff. = 0.0767549

Taking out the non-significant variables, the resulting regression is:

Model 3: OLS estimates using the 155 observations 1-155

Dependent variable: indGDP

      VARIABLE      COEFFICIENT        STDERROR       T STAT   2Prob(t > |T|)

   0)    const     31596.2             2336.61         13.522   < 0.00001 ***

   6)    Trade     -2970.81            1199.32         -2.477    0.014387 **

   8) Governme       756.926            481.722         1.571    0.118279

  11)       BK     -1047.66             493.696        -2.122    0.035519 **

  13) Property     -8954.56            1615.13         -5.544   < 0.00001 ***

  16)  l_Trade      6379.73            3351.79          1.903    0.058959 *

  17) l_BlackM     -7828.71            1176.19         -6.656   < 0.00001 ***

  18) sq_Fisca       242.486             56.5346        4.289    0.000032 ***

  19) sq_Prope      1273.94             249.607         5.104   < 0.00001 ***

  Mean of dependent variable = 8854.32

  Standard deviation of dep. var. = 9169.89

  Sum of squared residuals = 2.3127e+009

  Standard error of residuals = 3980

  Unadjusted R-squared = 0.821405

  Adjusted R-squared = 0.811619

  F-statistic (8, 146) = 83.9364 (p-value < 0.00001)

  Durbin-Watson statistic = 1.97101

  First-order autocorrelation coeff. = 0.0129411

Government and log(Trade) do not appear to be significant in this model, so they were dropped:

        VARIABLE      COEFFICIENT        STDERROR       T STAT   2Prob(t > |T|)

   0)    const     32948.4             2093.65         15.737   < 0.00001 ***

   6)    Trade      -701.47             352.948        -1.987    0.048715 **

  11)       BK      -930.70             492.710        -1.889    0.060856 *

  13) Property     -9310.0             1571.14         -5.926   < 0.00001 ***

  18) l_BlackM     -7606.7             1174.58         -6.476   < 0.00001 ***

  19) sq_Fisca       255.859             56.3419        4.541    0.000012 ***

  20) sq_Prope      1329.27             242.992         5.470   < 0.00001 ***

  Mean of dependent variable = 8854.32

  Standard deviation of dep. var. = 9169.89

  Sum of squared residuals = 2.40055e+009

  Standard error of residuals = 4027.4

  Unadjusted R-squared = 0.814621

  Adjusted R-squared = 0.807105

  F-statistic (6, 148) = 108.394 (p-value < 0.00001)

  Durbin-Watson statistic = 2.01359

  First-order autocorrelation coeff. = -0.00986223

The estimated model was thus:

indGDP=32948.4 -701.47 *Trade  -930.70 BK--9310.0 *Property-7606.7      *log(BlackM)+ 255.859 *Fiscal^2 + 1329.27 *Property^2                                                                                              

(The mean of indGDP was 8854.323 and S.D. was 9169.893 , and the mean of Econ_Sco was 3.0384 and S.D. of 0.76215.)


Below are the observed versus fitted ingGDP residuals:

Summary and Discussion

The most obvious evidence shown by the data is that the average value of economic freedom has a strong relationship with per capita GDP.  The R-squared value in the quadratic model with Econ_Sco is 0.635953 and at least up to a freedom factor of about 4.2, where the trend suddenly reverses.  This may be because the three outliers at the 4.5 range (Libya (capita GDP of 7600) Iran (capita GDP of 6400), and Iraq (capita GDP of 2500)) are socialist economies that maintain unusually high incomes because they derive most of their GDP from oil exports.  However, the quadratic relation also suggests that increasing government involvement is progressively less harmful to indGDP.  On the other hand, this also means that decreasing government involvement is exponetial more beneficial to GDP. 

The final model derived from the regressions was:

indGDP=32948.4 -701.47 *Trade  -930.70 BK--9310.0 *Property-7606.7      *log(BlackM)+ 255.859 *Fiscal^2 + 1329.27 *Property^2                                                                                              

This model has several interesting properties.  First, it is surprisingly accurate at predicting per capita GDP.  With an R-squared of 0.814621, it indicates that 81 percent of variation in wealth between countries is caused by their economic policies.  This makes the fact that increased economic freedom leads to more prosperity is hard to dispute.  The model also shows that other than the small positive coefficient on fiscal burden, there are strong negative correlations between the above factors and prosperity: that is, free trade, strong property rights, and low black market activity lead to higher prosperity. The question of what exact effect the factors had on GDP and which factors were most influences was more complicated however. 

The BlackM factor, or the amount of black market activity shows a high negative, suggesting that the more market activity is conducted underground, the lower the level of GDP.  This is highly intuitive, as illegal and hidden activities are bound to have significantly higher costs because of the usual costs risk associated with operating against the law.  These factors may be a good proxy for how difficult it is to run a business legally in any given country.  Of course when a large percentage of business is underground, most of the legal restrictions on businesses do not apply, which may diminish the accuracy of the other variables at high levels of black market activity.  There are 45 countries with the highest level of black market activities, with an average economic freedom ranking of about 124, and average capita GDP of $2,746, significantly below the world average of $8,854.

Trade was another major factor.  The quadratic relationship suggests that increasing free trade yields diminishing returns, but the overall pattern was clear:  the 34 nations with the highest level of trade restrictions have a capita GDP of $3,578.53 and the top eight nations with the highest level of trade freedom have an average capita GDP of $11,577.5.  Since the grading system uses cardinal rather than quantitative rankings, it was hard to get more precise estimates of the effect of trade restrictions, but free trade nevertheless seems to be an important factor for GDP growth.  Economic theory would suggest this outcome, since foreign investment is key in the growth of developing nations.  A time series study on the growth in GDP versus trade restrictions may clarify this theory.

Property rights had the biggest coefficient out of all the other factors, which is not at all surprising, considering that private property ownership is at the root of capitalism.  However the relationship was not linear, as the plot of the actual and fitted indGDP versus Property bellow shows:

Despite the non-linear relationship, the trend in the quadratic relationship reverses itself only in unfree countries, which may be an indication that property rights are less effective in nations that already have weak property rights protections.  Indeed, the average per capita income in the 14 nations with the lowest level of property rights is $2,563.08 and their average ranking is only 144.6 out of 155.

An unexpected result of the model was that increasing fiscal burden (which is defined as “tax rates and the level of government expenditures” (Heritage p53)) seems to actually raise per capita GDP.  This may be explained by the fact that as nations get wealthier, increasing profits allow higher taxes to be raised.  Nevertheless, taxes do not seem to have a significant impact on GDP, and are probably not the first thing a country should look to cut if it desires economic growth

It is unclear why a number of variables (like foreign investment) that are clearly significant individually were not significant in the full model.  When a regression was done on the individual variables, nearly all (other than fiscal burden) variables show significant negative correlations between more government and per capita GDP.  This suggests that there is some degree of collinearity in the variables, which is not surprising considering that each factor attempts to isolate certain aspects of bureaucratic policy from a single structure of regulation.

While the particular relationship between the ten factors used is less than clear, the basic conclusion from the data is beyond question: increasing levels of economic freedom are highly correlated with increasing levels of per capita GDP, and the variation in economic freedom explains most of the differences in wealth around the world.  This outcome would be surprising to many individuals who attribute factors like natural resources, population growth, or income distributions to differences in wealth, and holds many lessons for anyone attempting to stimulate growth in their own country or companies looking for growth opportunities around the world.


References

1. Easton, S. T., and Walker, M.A., eds. (1992) Rating Global Economic Freedom. Vancouver, B.C., Canada: The Fraser Institute.

2. Hanke, S.H., and Walters, S.J.K. (1997) Liberty, Equality, Prosperity. A Report to the Senate Joint Economic Committee. Washington, D.C.: U.S. Senate.

3. Gerald P. O’Driscoll, Jr., Kim R. Holmes, (2002) 2002 Index of Economic Freedom. Washington, D.C.: The Heritage Foundation.

3. Gerald P. O’Driscoll, Jr., Edwin J. Feulner, (2002) 2003 Index of Economic Freedom. Washington, D.C.: The Heritage Foundation.

4. CIA World Factbook 2002. 

http://www.cia.gov/cia/publications/factbook/

5. Economic Freedom of the World, 2000. (2002) Fraser Institute.

http://www.freetheworld.com

6. Gwartney, James, and Lawson, Robert. Capital University Economic Freedom of the World Annual Report 2002: Cato Institute.

http://www.cato.org/economicfreedom


Page used 24 queries in 1.004 seconds.