[PYTHON] Economic development

# Install libraries !pip install --upgrade pip --quiet !pip install statsmodels --quiet !pip install linearmodels --quiet

import numpy as np import pandas as pd import matplotlib.pyplot as plt import seaborn as sns sns.set_style('darkgrid') sns.set_context('paper', font_scale= 1.3) import statsmodels.formula.api as smf import statsmodels.api as sm import statsmodels.base.model as smclass import linearmodels as plm

df: Polled cross-section

# import data #df = pd.read_csv("https://raw.githubusercontent.com/quarcs-lab/mendez2020-convergence-clubs-code-data/master/assets/dat.csv") df = pd.read_csv("https://bit.ly/mendez2020") df

df1: Panel data ready (muti-index)

# import data with sorted multi-index for country and year df1 = pd.read_csv("https://raw.githubusercontent.com/quarcs-lab/mendez2020-convergence-clubs-code-data/master/assets/dat.csv", index_col=["country", "year"]).sort_index() df1

dataDefinitions

# import data definitions #dataDefinitions = pd.read_csv("https://raw.githubusercontent.com/quarcs-lab/mendez2020-convergence-clubs-code-data/master/assets/dat-definitions.csv") dataDefinitions = pd.read_csv("https://bit.ly/mendez2020d") dataDefinitions

df2: Determinants of growth

# All countries, all years, and proximate sources of economic growth df2 = df[['country', 'region', 'year', 'hi1990', 'GDPpc', 'lp', 'h', 'ky', 'TFP']] df2

df3: Determinants of growth 1990

# All countries, year 1990, and proximate sources of economic growth df3 = df.query('year == 1990').loc[:, ['country', 'region', 'hi1990', 'GDPpc', 'lp', 'h', 'ky', 'TFP']] df3

df4: Determinants of growth 2014

# All countries, year 2014, and proximate sources of economic growth df4 = df.query('year == 2014').loc[:, ['country', 'region', 'hi1990', 'GDPpc', 'lp', 'h', 'ky', 'TFP']] df4

df5: Determinants of growth 1990 in log form

# All countries, year 1990, and proximate sources of economic growth in log form (trend component) df5 = df.query('year == 1990').loc[:, ['country', 'region', 'hi1990', 'log_GDPpc', 'log_lp', 'log_h', 'log_ky', 'log_tfp']] df5

df6: Determinants of growth 2014 in log form

# All countries, year 2014, and proximate sources of economic growth in log form (trend component) df6 = df.query('year == 2014').loc[:, ['country', 'region', 'hi1990', 'log_GDPpc', 'log_lp', 'log_h', 'log_ky', 'log_tfp']] df6

df7: Wide panel for log lp

# Pivot panel data from long form to wide form df7 = df.pivot_table( index=['country', 'region', 'hi1990'], columns='year', values='log_lp').reset_index(drop=False)

# Make sure the column names are strings df7.columns = df7.columns.astype(str)

df7

df8: Wide panel for log ky

# Pivot panel data from long form to wide form df8 = df.pivot_table( index=['country', 'region', 'hi1990'], columns='year', values='log_ky').reset_index(drop=False) # Make sure the column names are strings df8.columns = df8.columns.astype(str)

df9: Wide panel for log h

# Pivot panel data from long form to wide form df9 = df.pivot_table( index=['country', 'region', 'hi1990'], columns='year', values='log_h').reset_index(drop=False) # Make sure the column names are strings df9.columns = df9.columns.astype(str)

df10: Wide panel for log tfp

# Pivot panel data from long form to wide form df10 = df.pivot_table( index=['country', 'region', 'hi1990'], columns='year', values='log_tfp').reset_index(drop=False) # Make sure the column names are strings df10.columns = df10.columns.astype(str)

# descriptive statistics for labor productivity df7.describe().round(2)

# descriptive statistics for the ratio physical capital to output df8.describe().round(2)

Exploratory analysis

Univariate plots

# cross-country distribution of log labor productivity in 1990 sns.displot(df7['1990'], kde = True);

sns.kdeplot(df7['1990']);

sns.kdeplot(df7['1990']) sns.kdeplot(df7['2014']);

Categorical plots

# count countries by income groups in 1990 sns.countplot(x= 'hi1990', data= df7);

# log labor productivity differences across regions in 1990 sns.barplot(x= 'region', y='1990', data= df7);

# Standard deviation of log labor productivity differences across regions in 1990 sns.barplot(x= 'region', y='1990', data= df7, estimator=np.std);

# Labor productivity differences across and within continents in 1990 sns.boxplot(x= 'region', y='1990', data= df7);

# The relationship between labor productivity in 1990 and 2014 sns.jointplot(data=df7, x="1990", y="2014", kind="reg");

# The relationship between labor productivity in 1990 and 2014 using a bi-variate distribution sns.jointplot(data=df7, x="1990", y="2014", kind="kde");

# The relationship between labor productivity in 1990 and 2014 sns.jointplot(data=df7, x="1990", y="2014", hue= "region");

sns.lmplot(x="1990", y="2014", hue='region', ci=None, data=df7);

Persistence and mobility

# The relationship between labor productivity in 1990 and 2014 g= sns.jointplot(data=df7, x="1990", y="2014", kind="scatter") # Draw a line of x=y x0, x1 = g.ax_joint.get_xlim() y0, y1 = g.ax_joint.get_ylim() lims = [max(x0, y0), min(x1, y1)] g.ax_joint.plot(lims, lims, '-r');

g = sns.lmplot(x="1990", y="2014", ci=None, data=df7) # Draw a line of x=y x0, x1 = g.ax.get_xlim() y0, y1 = g.ax.get_ylim() lims = [max(x0, y0), min(x1, y1)] g.ax.plot(lims, lims, '-r');

# The relationship between labor productivity in 1990 and 2014 using a bi-variate distribution g = sns.jointplot(data=df7, x="1990", y="2014", kind="kde", fill="true", height=6, space=0, levels=5); # Draw a line of x=y x0, x1 = g.ax_joint.get_xlim() y0, y1 = g.ax_joint.get_ylim() lims = [max(x0, y0), min(x1, y1)] g.ax_joint.plot(lims, lims, '-r');

Evolution of disparities

Labor productivity (Output per worker)

# Create array object df7a = df7.loc[:,'1990':'2014'].values df7a

df7a.shape

# Create array of years. Not the upper limit. The data is only avilable up 2014, but we need to write 2015 years = np.arange(1990,2015) years

years.shape

# Sigma convergence analysis sigma_df7a = df7a.std(axis=0) plt.plot(years, sigma_df7a) plt.title("Sigma Convergence in (log) labor productivity") plt.ylabel('Stand. Dev.of log labor productivity');

# Beta convergence analysis # Compute the growth rate between 1990 and 2014 (log aproximation method) df7['g1990_2014'] = df7['2014']-df7['1990'] sns.lmplot(x="1990", y="g1990_2014", data=df7);

y = df7['g1990_2014'] X = df7['1990'] X_withconst = sm.add_constant(X) OLS = sm.OLS(y, X_withconst).fit() print(OLS.summary())

                            OLS Regression Results                            
==============================================================================
Dep. Variable:             g1990_2014   R-squared:                       0.026
Model:                            OLS   Adj. R-squared:                  0.017
Method:                 Least Squares   F-statistic:                     2.814
Date:                Thu, 14 Oct 2021   Prob (F-statistic):             0.0964
Time:                        03:48:20   Log-Likelihood:                -71.610
No. Observations:                 108   AIC:                             147.2
Df Residuals:                     106   BIC:                             152.6
Df Model:                           1                                         
Covariance Type:            nonrobust                                         
==============================================================================
                 coef    std err          t      P>|t|      [0.025      0.975]
------------------------------------------------------------------------------
const          1.2900      0.391      3.299      0.001       0.515       2.065
1990          -0.0684      0.041     -1.677      0.096      -0.149       0.012
==============================================================================
Omnibus:                        2.210   Durbin-Watson:                   1.702
Prob(Omnibus):                  0.331   Jarque-Bera (JB):                1.636
Skew:                           0.249   Prob(JB):                        0.441
Kurtosis:                       3.340   Cond. No.                         83.1
==============================================================================

Notes:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.

sns.lmplot(x="1990", y="g1990_2014", hue='region', ci=None, data=df7);

for row in df7a: plt.plot(years, row)

sns.kdeplot(df7['1990'], label="1990") sns.kdeplot(df7['2014'], label="2014") plt.legend() plt.xlabel('Log labor productivity (output per worker)');

# Standardize each observation by its cross-sectional mean rel_df7a = df7a / df7a.mean(axis=0)

sns.kdeplot(rel_df7a[:,1], label="1990") sns.kdeplot(rel_df7a[:,24], label="2014") plt.legend() plt.xlabel('Relative (log) labor productivity');

# Distributional mobility and convergence g = sns.jointplot(data=rel_df7a, x=rel_df7a[:,1], y=rel_df7a[:,24], kind="kde", fill="true", height=6, space=0, levels=5, ratio=3) g.plot_joint(sns.regplot, scatter=False, truncate=False, color="b", line_kws={"lw":4}) # Draw a line of x=y x0, x1 = g.ax_joint.get_xlim() y0, y1 = g.ax_joint.get_ylim() lims = [max(x0, y0), min(x1, y1)] g.ax_joint.plot(lims, lims, '-r', linestyle='--') g.set_axis_labels("Relative (log) labor productivity in 1990", "Relative (log) labor productivity in 2014") g.ax_marg_x.set_xlim(0.65, 1.3) g.ax_marg_y.set_ylim(0.65, 1.3);

/shared-libs/python3.7/py-core/lib/python3.7/site-packages/ipykernel_launcher.py:8: UserWarning: linestyle is redundantly defined by the 'linestyle' keyword argument and the fmt string "-r" (-> linestyle='-'). The keyword argument will take precedence.

Physical capital ratio

# Create array object df8a = df8.loc[:,'1990':'2014'].values years = np.arange(1990,2015)

sigma_df8a = df8a.std(axis=0) plt.plot(years, sigma_df8a) plt.title("Sigma Convergence in (log) capital-output ratio") plt.ylabel('Stand. Dev');

# Beta convergence analysis # Compute the growth rate between 1990 and 2014 (log aproximation method) df8['g1990_2014'] = df8['2014']-df8['1990'] sns.lmplot(x="1990", y="g1990_2014", data=df8);

y = df8['g1990_2014'] X = df8['1990'] X_withconst = sm.add_constant(X) OLS = sm.OLS(y, X_withconst).fit() print(OLS.summary())

                            OLS Regression Results                            
==============================================================================
Dep. Variable:             g1990_2014   R-squared:                       0.542
Model:                            OLS   Adj. R-squared:                  0.538
Method:                 Least Squares   F-statistic:                     125.4
Date:                Thu, 14 Oct 2021   Prob (F-statistic):           1.12e-19
Time:                        03:48:24   Log-Likelihood:                -42.443
No. Observations:                 108   AIC:                             88.89
Df Residuals:                     106   BIC:                             94.25
Df Model:                           1                                         
Covariance Type:            nonrobust                                         
==============================================================================
                 coef    std err          t      P>|t|      [0.025      0.975]
------------------------------------------------------------------------------
const          0.9665      0.056     17.170      0.000       0.855       1.078
1990          -0.6692      0.060    -11.197      0.000      -0.788      -0.551
==============================================================================
Omnibus:                       10.022   Durbin-Watson:                   1.907
Prob(Omnibus):                  0.007   Jarque-Bera (JB):               10.862
Skew:                          -0.585   Prob(JB):                      0.00438
Kurtosis:                       4.022   Cond. No.                         2.89
==============================================================================

Notes:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.

sns.lmplot(x="1990", y="g1990_2014", hue='region', ci=None, data=df8);

for row in df8a: plt.plot(years, row)

sns.kdeplot(df8['1990'], label="1990") sns.kdeplot(df8['2014'], label="2014") plt.legend() plt.xlabel('Log physical capital ratio');

# Standardize each observation by its cross-sectional mean rel_df8a = df8a / df8a.mean(axis=0)

sns.kdeplot(rel_df8a[:,1], label="1990") sns.kdeplot(rel_df8a[:,24], label="2014") plt.legend() plt.xlabel('Relative (log) physical capital ratio');

# Distributional mobility and convergence g = sns.jointplot(data=rel_df8a, x=rel_df8a[:,1], y=rel_df8a[:,24], kind="kde", fill="true", height=6, space=0, levels=5, ratio=3); g.plot_joint(sns.regplot, scatter=False, truncate=False, color="b", line_kws={"lw":4}) # Draw a line of x=y x0, x1 = g.ax_joint.get_xlim() y0, y1 = g.ax_joint.get_ylim() lims = [max(x0, y0), min(x1, y1)] g.ax_joint.plot(lims, lims, '-r', linestyle='--') g.set_axis_labels("Relative (log) physical capital ratio in 1990", "Relative (log) physical capital ratio in 2014") g.ax_marg_x.set_xlim(-3, 3) g.ax_marg_y.set_ylim(-3, 3);

/shared-libs/python3.7/py-core/lib/python3.7/site-packages/ipykernel_launcher.py:8: UserWarning: linestyle is redundantly defined by the 'linestyle' keyword argument and the fmt string "-r" (-> linestyle='-'). The keyword argument will take precedence.

/shared-libs/python3.7/py-core/lib/python3.7/site-packages/ipykernel_launcher.py:8: UserWarning: linestyle is redundantly defined by the 'linestyle' keyword argument and the fmt string "-r" (-> linestyle='-'). The keyword argument will take precedence.

Human capital per worker

# Create array object df9a = df9.loc[:,'1990':'2014'].values years = np.arange(1990,2015)

sigma_df9a = df9a.std(axis=0) plt.plot(years, sigma_df9a) plt.title("Sigma Convergence in (log) human capital per worker") plt.ylabel('Stand. Dev');

# Beta convergence analysis # Compute the growth rate between 1990 and 2014 (log aproximation method) df9['g1990_2014'] = df9['2014']-df9['1990'] sns.lmplot(x="1990", y="g1990_2014", data=df9);

y = df9['g1990_2014'] X = df9['1990'] X_withconst = sm.add_constant(X) OLS = sm.OLS(y, X_withconst).fit() print(OLS.summary())

                            OLS Regression Results                            
==============================================================================
Dep. Variable:             g1990_2014   R-squared:                       0.192
Model:                            OLS   Adj. R-squared:                  0.185
Method:                 Least Squares   F-statistic:                     25.21
Date:                Thu, 14 Oct 2021   Prob (F-statistic):           2.09e-06
Time:                        03:48:28   Log-Likelihood:                 106.32
No. Observations:                 108   AIC:                            -208.6
Df Residuals:                     106   BIC:                            -203.3
Df Model:                           1                                         
Covariance Type:            nonrobust                                         
==============================================================================
                 coef    std err          t      P>|t|      [0.025      0.975]
------------------------------------------------------------------------------
const          0.3572      0.030     11.968      0.000       0.298       0.416
1990          -0.1510      0.030     -5.021      0.000      -0.211      -0.091
==============================================================================
Omnibus:                        0.500   Durbin-Watson:                   1.842
Prob(Omnibus):                  0.779   Jarque-Bera (JB):                0.639
Skew:                           0.051   Prob(JB):                        0.727
Kurtosis:                       2.637   Cond. No.                         6.65
==============================================================================

Notes:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.

sns.lmplot(x="1990", y="g1990_2014", hue='region', ci=None, data=df9);

for row in df9a: plt.plot(years, row)

sns.kdeplot(df9['1990'], label="1990") sns.kdeplot(df9['2014'], label="2014") plt.legend() plt.xlabel('Log human capital per worker');

# Standardize each observation by its cross-sectional mean rel_df9a = df9a / df9a.mean(axis=0)

sns.kdeplot(rel_df9a[:,1], label="1990") sns.kdeplot(rel_df9a[:,24], label="2014") plt.legend() plt.xlabel('Relative (log) human capital per worker');

# Distributional mobility and convergence g = sns.jointplot(data=rel_df9a, x=rel_df9a[:,1], y=rel_df9a[:,24], kind="kde", fill="true", height=6, space=0, levels=5, ratio=3); g.plot_joint(sns.regplot, scatter=False, truncate=False, color="b", line_kws={"lw":4}) # Draw a line of x=y x0, x1 = g.ax_joint.get_xlim() y0, y1 = g.ax_joint.get_ylim() lims = [max(x0, y0), min(x1, y1)] g.ax_joint.plot(lims, lims, '-r', linestyle='--') g.set_axis_labels("Relative (log) human capital in 1990", "Relative (log) human capital in 2014") g.ax_marg_x.set_xlim(0, 1.75) g.ax_marg_y.set_ylim(0, 1.75);

/shared-libs/python3.7/py-core/lib/python3.7/site-packages/ipykernel_launcher.py:8: UserWarning: linestyle is redundantly defined by the 'linestyle' keyword argument and the fmt string "-r" (-> linestyle='-'). The keyword argument will take precedence.

Total factor productivity

# Create array object df10a = df10.loc[:,'1990':'2014'].values years = np.arange(1990,2015)

sigma_df10a = df10.std(axis=0) plt.plot(years, sigma_df10a) plt.title("Sigma Convergence in (log) total factor productivity") plt.ylabel('Stand. Dev');

# Beta convergence analysis # Compute the growth rate between 1990 and 2014 (log aproximation method) df10['g1990_2014'] = df10['2014']-df7['1990'] sns.lmplot(x="1990", y="g1990_2014", data=df10);

y = df10['g1990_2014'] X = df10['1990'] X_withconst = sm.add_constant(X) OLS = sm.OLS(y, X_withconst).fit() print(OLS.summary())

                            OLS Regression Results                            
==============================================================================
Dep. Variable:             g1990_2014   R-squared:                       0.024
Model:                            OLS   Adj. R-squared:                  0.015
Method:                 Least Squares   F-statistic:                     2.627
Date:                Thu, 14 Oct 2021   Prob (F-statistic):              0.108
Time:                        03:48:32   Log-Likelihood:                -129.23
No. Observations:                 108   AIC:                             262.5
Df Residuals:                     106   BIC:                             267.8
Df Model:                           1                                         
Covariance Type:            nonrobust                                         
==============================================================================
                 coef    std err          t      P>|t|      [0.025      0.975]
------------------------------------------------------------------------------
const         -1.7533      0.639     -2.744      0.007      -3.020      -0.486
1990          -0.1608      0.099     -1.621      0.108      -0.358       0.036
==============================================================================
Omnibus:                       15.607   Durbin-Watson:                   1.533
Prob(Omnibus):                  0.000   Jarque-Bera (JB):               17.597
Skew:                           0.877   Prob(JB):                     0.000151
Kurtosis:                       3.912   Cond. No.                         54.2
==============================================================================

Notes:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.

sns.lmplot(x="1990", y="g1990_2014", hue='region', ci=None, data=df10);

for row in df10a: plt.plot(years, row)

sns.kdeplot(df10['1990'], label="1990") sns.kdeplot(df10['2014'], label="2014") plt.legend() plt.xlabel('Log total factor productivity');

# Standardize each observation by its cross-sectional mean rel_df10a = df10a / df10a.mean(axis=0)

sns.kdeplot(rel_df10a[:,1], label="1990") sns.kdeplot(rel_df10a[:,24], label="2014") plt.legend() plt.xlabel('Relative (log) total factor productivity');

# Distributional mobility and convergence g = sns.jointplot(data=rel_df10a, x=rel_df10a[:,1], y=rel_df10a[:,24], kind="kde", fill="true", height=6, space=0, levels=5, ratio=3); g.plot_joint(sns.regplot, scatter=False, truncate=False, color="b", line_kws={"lw":4}) # Draw a line of x=y x0, x1 = g.ax_joint.get_xlim() y0, y1 = g.ax_joint.get_ylim() lims = [max(x0, y0), min(x1, y1)] g.ax_joint.plot(lims, lims, '-r', linestyle='--') g.set_axis_labels("Relative (log) TFP in 1990", "Relative (log) TFP in 2014") g.ax_marg_x.set_xlim(0.6, 1.4) g.ax_marg_y.set_ylim(0.6, 1.4);

/shared-libs/python3.7/py-core/lib/python3.7/site-packages/ipykernel_launcher.py:8: UserWarning: linestyle is redundantly defined by the 'linestyle' keyword argument and the fmt string "-r" (-> linestyle='-'). The keyword argument will take precedence.