Problem Solving 02

Beka Modebadze 2019 - https://github.com/bexxmodd/econ-papers-reproduction

---

Replication of a paper published by Sam Jones and Finn Tarp - "Does foreign aid harm political institutions?"

Original paper, published by Journal of Development Economics in 2015. You can download it from here

---

For the second problem solving I will replicate a paper that examines the effect of the financial aid on governmental institutions. Here is the abstract from the original paper:

The notion that foreign aid harms the institutions of recipient governments remains prevalent. We combine new disaggregated aid data and various metrics of political institutions to re-examine this relationship. Long run cross-section and alternative dynamic panel estimators show a small positive net effect of total aid on political institutions. Distinguishing between types of aid according to their frequency domain and stated objectives, we find that this aggregate net effect is driven primarily by the positive contribution of more stable inflows of ‘governance aid’. Weconclude that the data do not support the view that aid has had a systematic negative effect on political institutions

%%capture
! pip install seaborn
import seaborn as sns
import numpy as np
import pandas as pd

%matplotlib inline
import matplotlib as mpl 
import matplotlib.pyplot as plt

from matplotlib.ticker import StrMethodFormatter

lets import the data and observe first and last five rows

path = "PS04-5-Jones2016.dta"
df = pd.read_stata(path)
df.head(5)

	year	recipientCode	recipient	recipientRegion	weight	pop_weight	calpha	polity	xconst	ptss	...	missing	rc_ht_region	rc_ht_colonial	recipientIncL	recipientIncLM	recipientIncUM	recipientIncH	bilateral	devFund	devBank
0	1983-01-01	2	AGO	Sub-Saharan Africa	1	8.491680	-128.401306	-146.757950	-51.031757	-200.921722	...	0	4	7	0	0	0	0	0.854167	0	0
1	1983-01-01	3	ALB	Europe & Central Asia	1	2.831682	-146.175674	-163.987503	-148.354706	-13.572638	...	0	1	0	0	0	0	0	0.947368	0	0
2	1983-01-01	4	ARG	Latin America & Caribbean	1	29.447966	43.978588	97.327393	143.614151	-13.572638	...	0	2	2	0	0	0	0	0.854167	0	0
3	1983-01-01	6	BDI	Sub-Saharan Africa	1	4.542465	-107.733826	-118.042023	-148.354706	80.101906	...	0	4	8	0	0	0	0	0.875000	0	0
4	1983-01-01	7	BEN	Sub-Saharan Africa	1	3.917942	-111.687050	-132.399979	-148.354706	80.101906	...	0	4	6	0	0	0	0	0.912500	0	0

5 rows × 33 columns

df.tail(5)

	year	recipientCode	recipient	recipientRegion	weight	pop_weight	calpha	polity	xconst	ptss	...	missing	rc_ht_region	rc_ht_colonial	recipientIncL	recipientIncLM	recipientIncUM	recipientIncH	bilateral	devFund	devBank
2838	2010-01-01	111	VNM	East Asia & Pacific	1	86.927773	-106.729301	-118.042023	-51.031757	-13.572638	...	0	7	6	0	1	0	0	0.916667	0	0
2839	2010-01-01	112	VUT	East Asia & Pacific	1	0.239651	213.432800	122.915260	NaN	80.101906	...	1	9	9	0	1	0	0	0.984375	0	0
2840	2010-01-01	113	YEM	Middle East & North Africa	1	24.052523	-114.155952	-60.610176	-99.693237	-107.247177	...	0	3	5	0	1	0	0	0.923913	0	0
2841	2010-01-01	114	ZAF	Sub-Saharan Africa	1	49.991310	70.207100	134.658081	143.614151	-13.572638	...	0	4	5	0	0	1	0	0.916667	0	0
2842	2010-01-01	115	ZMB	Sub-Saharan Africa	1	12.926404	-20.851480	74.354675	46.291199	-107.247177	...	0	4	5	0	1	0	0	0.916667	0	0

5 rows × 33 columns

I'll rename column 'recipient' to more familiar name: 'country'

df.rename(columns={'recipient':'country'}, inplace=True)
df.head(2)

	year	recipientCode	country	recipientRegion	weight	pop_weight	calpha	polity	xconst	ptss	...	missing	rc_ht_region	rc_ht_colonial	recipientIncL	recipientIncLM	recipientIncUM	recipientIncH	bilateral	devFund	devBank
0	1983-01-01	2	AGO	Sub-Saharan Africa	1	8.491680	-128.401306	-146.757950	-51.031757	-200.921722	...	0	4	7	0	0	0	0	0.854167	0	0
1	1983-01-01	3	ALB	Europe & Central Asia	1	2.831682	-146.175674	-163.987503	-148.354706	-13.572638	...	0	1	0	0	0	0	0	0.947368	0	0

2 rows × 33 columns

Part 1:

---

1.a) First, we'll look at the number of observations for each country

df.groupby('country').count().year.to_frame()

	year
country
AGO	28
ALB	28
ARG	28
AZE	19
BDI	28
...	...
VNM	28
VUT	28
YEM	21
ZAF	28
ZMB	28

106 rows × 1 columns

We see that we have a total of 2843 observations from 106 countries but the number of observations (years) for each country varies

tab = df.groupby(['country', 'year'])
tab.first()

		recipientCode	recipientRegion	weight	pop_weight	calpha	polity	xconst	ptss	indjud	checks	...	missing	rc_ht_region	rc_ht_colonial	recipientIncL	recipientIncLM	recipientIncUM	recipientIncH	bilateral	devFund	devBank
country	year
AGO	1983-01-01	2	Sub-Saharan Africa	1	8.491680	-128.401306	-146.757950	-51.031757	-200.921722	13.868533	-82.347115	...	0	4	7	0	0	0	0	0.854167	0	0
	1984-01-01	2	Sub-Saharan Africa	1	8.786655	-114.635727	-146.757950	-51.031757	-13.572638	-123.484238	-82.347115	...	0	4	7	0	0	0	0	0.920455	0	0
	1985-01-01	2	Sub-Saharan Africa	1	9.065557	-128.401306	-146.757950	-51.031757	-200.921722	13.868533	-82.347115	...	0	4	7	0	0	0	0	0.912500	0	0
	1986-01-01	2	Sub-Saharan Africa	1	9.322552	-166.218903	-146.757950	-51.031757	-200.921722	-123.484238	-82.347115	...	0	4	7	0	0	0	0	0.920455	0	0
	1987-01-01	2	Sub-Saharan Africa	1	9.562799	-166.218903	-146.757950	-51.031757	-200.921722	-123.484238	-82.347115	...	0	4	7	0	0	0	0	0.912500	0	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
ZMB	2006-01-01	115	Sub-Saharan Africa	1	11.750110	53.964039	57.125122	46.291199	-13.572638	13.868533	91.445068	...	0	4	5	1	0	0	0	0.923913	0	0
	2007-01-01	115	Sub-Saharan Africa	1	12.055376	38.013863	57.125122	46.291199	-13.572638	13.868533	33.514343	...	0	4	5	1	0	0	0	0.923913	0	0
	2008-01-01	115	Sub-Saharan Africa	1	12.379619	46.710915	88.712616	46.291199	-13.572638	13.868533	33.514343	...	0	4	5	1	0	0	0	0.916667	0	0
	2009-01-01	115	Sub-Saharan Africa	1	12.723741	42.757710	74.354675	46.291199	-13.572638	13.868533	33.514343	...	0	4	5	1	0	0	0	0.916667	0	0
	2010-01-01	115	Sub-Saharan Africa	1	12.926404	-20.851480	74.354675	46.291199	-107.247177	-123.484238	33.514343	...	0	4	5	0	1	0	0	0.916667	0	0

2843 rows × 31 columns

1.b) This paper uses five variables of interest (depended) to try and explain the effect of the financial aid on the institutions. These variables are indices that measure a variety of institutions.

• We'll summarize and then visualize the distribution of those variables

This is the summary of the institutional variables:

Variables	Source	var. name	Notes
Democracy	Freedom House	polity	Captures the extent of democracy. Original data, taken directly from the QoG database, ranges from: 0 = “least democratic”; to 10 = “most democratic”. Following (Hadenius and Teorell, 2005) this variable is calculated as the average of the Freedom House polity measure (the mean of political rights and civil liberty indexes, fh_pr and fh_cl) and the Polity IV revised combined polity score (p_polity2), both converted to a 0–10 scale.
Checks	DPI	checks	Reflects the number of veto players over political decisions, which is the number actors whose approval is necessary for a shift in policy. Original data ranges from 1 (lowest) to 18 (highest).
Executive constraints	Polity IV	xconst	Captures the extent of institutionalized constraints on the decision-making powers of chief (political) executives, whether individuals or collectivities. Original data ranges from: 1 = “unlimited authority”; to 7=“executive parity or subordination”.
Political terror (–)	WG10	ptss	A reverse scale of political terror. Original data is collated from annual reports on human rights published by the U.S. State Department and ranges from: 1 =“countries under a secure rule of law”; to 5=“terror has expanded to the whole population”. This original scale is reversed for the present analysis.
Judicial independence	CR10	injud	Indicates the extent to which the judiciary is independent of control from other sources, such as another branch of the government or the military. Original data ranges from: 0 = “not independent”; to 2=“generally independent”.

plt.figure(figsize=(20, 3))

plt.subplot(151)
plt.hist(df['polity'], color='green', rwidth=.95)
plt.title('Democracy')
plt.xlabel('polity')

plt.subplot(152)

plt.hist(df['checks'], color='purple', rwidth=.95)
plt.title('Checks')
plt.xlabel('checks')

plt.subplot(153)
plt.hist(df['xconst'], color='blue', rwidth=.95)
plt.title('Executive constraints')
plt.xlabel('xconst')

plt.subplot(154)
plt.hist(df['ptss'], color='red', rwidth=.95)
plt.title('Political terror (–)')
plt.xlabel('ptss')

plt.subplot(155)
plt.hist(df['indjud'], rwidth=.95)
plt.title('Judicial independence')
plt.xlabel('indjud')

plt.show()

• As we see those variables have an exceptionally large distribution from negative to positive and it will be hard to interpret its relationship with the success of the financial aids. Later we will convert those indices to a more acceptable scale

Part 2:

---

The main explanatory variable (independent) is aid which is overall aid, and its three components. We'll explore those variables

2.a) we start with summarizing three types of Aid and the GDP of recipient country:

df[['cmt_recip_GDP','econAid_GDP','governAid_GDP','otherAid_GDP']].describe()

	cmt_recip_GDP	econAid_GDP	governAid_GDP	otherAid_GDP
count	2843.000000	2843.000000	2843.000000	2843.000000
mean	5.237108	1.370980	0.463849	3.402287
std	7.946933	2.173830	1.735124	5.959953
min	0.000000	0.000000	0.000000	0.000000
25%	0.496056	0.082193	0.005398	0.243156
50%	2.605047	0.594106	0.068150	1.444073
75%	6.896747	1.770954	0.372156	4.221335
max	122.853050	21.441109	36.373505	110.735672

• We see that average Aid received by the country as a share of their GDP is 5.24\%
• Financial aid to support and stimulate the economy is 1.37\%
• Government aid to support countries' governments is 0.46\%
• Other types of Aid (usually military or humanitarian support) is 3.40\%

import matplotlib.ticker as mtick

plt.figure(figsize=(20, 3.5))
sns.set(font_scale=1.1)

plt.subplot(141)
ax = df['cmt_recip_GDP'].plot(kind='hist', color='green', rwidth=.95)
plt.title('Total Aid as a share of GDP')
ax.xaxis.set_major_formatter(mtick.PercentFormatter())
plt.xlim(xmin=0)
plt.ylim(ymin=0)

plt.subplot(142)
ax = df['econAid_GDP'].plot(kind='hist', color='purple', rwidth=.95)
plt.title('Econ Aid as a share of GDP')
ax.xaxis.set_major_formatter(mtick.PercentFormatter())
plt.xlim(xmin=0)
plt.ylim(ymin=0)

plt.subplot(143)
ax = df['governAid_GDP'].plot(kind='hist', color='blue', rwidth=.95)
plt.title('Gov Aid as a share of GDP')
ax.xaxis.set_major_formatter(mtick.PercentFormatter())
plt.xlim(xmin=0)
plt.ylim(ymin=0)

plt.subplot(144)
ax = df['otherAid_GDP'].plot(kind='hist', color='red', rwidth=.95)
plt.title('Other Aid as a share of GDP')
ax.xaxis.set_major_formatter(mtick.PercentFormatter())
plt.xlim(xmin=0)
plt.ylim(ymin=0)

plt.show()

• Distribution of each type of aid is strongly skewed to the right and are heavily concentrated in a small range
• After examining histogram, it is apparent that the majority of countries that receive Aid are of the fairly equal in terms of aid being the share of their GDP

grouped_rg = df.groupby(df['recipientRegion']).mean()[['cmt_recip_GDP', 'econAid_GDP', 'governAid_GDP', 'otherAid_GDP']]
grouped_rg

	cmt_recip_GDP	econAid_GDP	governAid_GDP	otherAid_GDP
recipientRegion
East Asia & Pacific	4.596575	1.520697	0.964490	2.111387
Europe & Central Asia	2.002812	0.576212	0.268140	1.158460
Latin America & Caribbean	2.398856	0.539653	0.224128	1.635075
Middle East & North Africa	2.628124	0.640328	0.154621	1.833176
South Asia	3.678910	1.801623	0.181697	1.695591
Sub-Saharan Africa	8.572269	2.066255	0.596686	5.909328

ax = grouped_rg.plot(kind='barh', width=.80 , figsize=(12, 8))
ax.xaxis.set_major_formatter(mtick.PercentFormatter())
aid = ['Total Aid', 'Economic Aid', 'Governmental Aid', 'Other Aid']
plt.legend(aid, loc=4)
plt.style.use('seaborn-whitegrid')

• When breaking down Aids received by the region, we see that Sub-Saharan Africa received the biggest share - compared to their GDP – and least funds went to Latin America & Caribbean
• It’s worth mentioning that South Asia's majority of Aid came for the economic support
• While East Asia & Pacific's accrued the most governmental aid in comparison with other regions

Part 3:

---

Now let's test the theory

First, generate a new variable, 'inst' that is the average of the five institutional variables. Draw the histogram of your variable

df['inst_mean'] = ((df['polity'] + df['xconst'] + df['ptss'] + df['indjud'] + df['checks']) / 5)
df['inst_mean'].head()

0    -93.438004
1   -106.349243
2     31.778065
3    -78.425232
4    -81.296829
Name: inst_mean, dtype: float32

3.a) We run first regressions with an average of institutional indices on aid and its components

First, we regress average score of indices on every type of financial aid individually and analyze the findings

import statsmodels.formula.api as smf
reg1 = smf.ols('inst_mean ~ cmt_recip_GDP', data=df).fit()
reg1.summary()

OLS Regression Results

Dep. Variable:	inst_mean	R-squared:	0.003
Model:	OLS	Adj. R-squared:	0.002
Method:	Least Squares	F-statistic:	7.456
Date:	Fri, 20 Dec 2019	Prob (F-statistic):	0.00637
Time:	10:29:11	Log-Likelihood:	-14881.
No. Observations:	2619	AIC:	2.977e+04
Df Residuals:	2617	BIC:	2.978e+04
Df Model:	1
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
Intercept	3.0137	1.661	1.814	0.070	-0.244	6.272
cmt_recip_GDP	-0.4903	0.180	-2.731	0.006	-0.842	-0.138

Omnibus:	217.842	Durbin-Watson:	1.915
Prob(Omnibus):	0.000	Jarque-Bera (JB):	87.658
Skew:	0.223	Prob(JB):	9.23e-20
Kurtosis:	2.223	Cond. No.	11.1

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

reg2 = smf.ols('inst_mean ~ governAid_GDP', data=df).fit()
reg2.summary()

OLS Regression Results

Dep. Variable:	inst_mean	R-squared:	0.005
Model:	OLS	Adj. R-squared:	0.005
Method:	Least Squares	F-statistic:	12.85
Date:	Fri, 20 Dec 2019	Prob (F-statistic):	0.000343
Time:	10:29:11	Log-Likelihood:	-14878.
No. Observations:	2619	AIC:	2.976e+04
Df Residuals:	2617	BIC:	2.977e+04
Df Model:	1
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
Intercept	-0.7533	1.431	-0.526	0.599	-3.560	2.054
governAid_GDP	2.8724	0.801	3.585	0.000	1.301	4.443

Omnibus:	230.526	Durbin-Watson:	1.936
Prob(Omnibus):	0.000	Jarque-Bera (JB):	91.131
Skew:	0.230	Prob(JB):	1.63e-20
Kurtosis:	2.211	Cond. No.	1.89

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

reg3 = smf.ols('inst_mean ~ econAid_GDP', data=df).fit()
reg3.summary()

OLS Regression Results

Dep. Variable:	inst_mean	R-squared:	0.005
Model:	OLS	Adj. R-squared:	0.005
Method:	Least Squares	F-statistic:	14.18
Date:	Fri, 20 Dec 2019	Prob (F-statistic):	0.000170
Time:	10:29:11	Log-Likelihood:	-14877.
No. Observations:	2619	AIC:	2.976e+04
Df Residuals:	2617	BIC:	2.977e+04
Df Model:	1
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
Intercept	3.8508	1.644	2.342	0.019	0.626	7.075
econAid_GDP	-2.4804	0.659	-3.765	0.000	-3.772	-1.189

Omnibus:	206.672	Durbin-Watson:	1.912
Prob(Omnibus):	0.000	Jarque-Bera (JB):	84.526
Skew:	0.216	Prob(JB):	4.42e-19
Kurtosis:	2.234	Cond. No.	3.11

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

reg4 = smf.ols('inst_mean ~ otherAid_GDP', data=df).fit()
reg4.summary()

OLS Regression Results

Dep. Variable:	inst_mean	R-squared:	0.004
Model:	OLS	Adj. R-squared:	0.004
Method:	Least Squares	F-statistic:	11.56
Date:	Fri, 20 Dec 2019	Prob (F-statistic):	0.000683
Time:	10:29:11	Log-Likelihood:	-14878.
No. Observations:	2619	AIC:	2.976e+04
Df Residuals:	2617	BIC:	2.977e+04
Df Model:	1
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
Intercept	3.2461	1.602	2.026	0.043	0.105	6.387
otherAid_GDP	-0.8263	0.243	-3.400	0.001	-1.303	-0.350

Omnibus:	234.080	Durbin-Watson:	1.915
Prob(Omnibus):	0.000	Jarque-Bera (JB):	89.354
Skew:	0.216	Prob(JB):	3.95e-20
Kurtosis:	2.204	Cond. No.	7.66

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

Summary: We have an interesting finding displayed here. It shows that governmental financial aid is the only type of aid which helps to improve institutions in the recipient country. Its P-value is close to zero which speaks of its statistical significance. However, other types of financial aid (the economic aid, the other aid, and the total aid) have a negative effect on the institutions. All of those variables have a P-value close to zero which means they also are statistically significant. There’ve have been many critics of financial aid, which has not resulted in the improvement of institutional qualities in receiving countries. Usually, those funds are never put to expected employment and are consumed by the bank accounts of the small number of people. Of course, we cannot explain (and I never intended) the effect of the financial aid by one variable as there are other variables that we have not controlled for, but it gives some idea if those aids are working towards a positive direction or towards a negative direction.

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.cbook as cbook

figsize=(15,5)
plt.style.use('seaborn-whitegrid')
df.plot(kind='scatter', x='cmt_recip_GDP', y='inst_mean', color='purple')
plt.xlabel('Total Aid')
plt.ylabel('Institutional Quality')
plt.axis('tight')
plt.xlim(xmin=0)
plt.xlim(xmax=140)

(0.0, 140)

• The scatter graph shows that overall Total Financial Aid is close to being equally distributed with good and bad institutional quality, but it's still leaning towards the negative side of the institutional quality score

3.a) Next we'll run the regression with all three types of Aid:

reg4 = smf.ols('inst_mean ~ governAid_GDP + econAid_GDP + otherAid_GDP', data=df).fit()
reg4.summary()

OLS Regression Results

Dep. Variable:	inst_mean	R-squared:	0.017
Model:	OLS	Adj. R-squared:	0.016
Method:	Least Squares	F-statistic:	15.31
Date:	Fri, 20 Dec 2019	Prob (F-statistic):	7.11e-10
Time:	10:29:12	Log-Likelihood:	-14861.
No. Observations:	2619	AIC:	2.973e+04
Df Residuals:	2615	BIC:	2.975e+04
Df Model:	3
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
Intercept	4.4092	1.694	2.602	0.009	1.087	7.731
governAid_GDP	4.5392	0.857	5.294	0.000	2.858	6.220
econAid_GDP	-1.9184	0.739	-2.597	0.009	-3.367	-0.470
otherAid_GDP	-1.0085	0.288	-3.504	0.000	-1.573	-0.444

Omnibus:	215.725	Durbin-Watson:	1.920
Prob(Omnibus):	0.000	Jarque-Bera (JB):	83.847
Skew:	0.201	Prob(JB):	6.21e-19
Kurtosis:	2.221	Cond. No.	8.53

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

• Once again, all the Aid variables are statistically significant (P-value ≈ 0) but the effect of the Governmental Aid has a higher positive coefficient then before, while economic and other aid remains with a negative coefficient.

Part 4:

---

Institutional variables are slow-changing variables. Changes from one year to another may not be considerable. To overcome this issue, we can use the mean values of measured over time.

4.a) In this part we'll calculate the mean values of variables for each country

df_col = df.groupby('country')['polity','ptss','indjud', 'checks', 'inst_mean', 'cmt_recip_GDP', 'governAid_GDP', 'econAid_GDP', 'otherAid_GDP'].agg(['mean'])
df_col.head()

	polity	ptss	indjud	checks	inst_mean	cmt_recip_GDP	governAid_GDP	econAid_GDP	otherAid_GDP
	mean	mean	mean	mean	mean	mean	mean	mean	mean
country
AGO	-92.505363	-130.665817	-98.048538	-72.002342	-88.511253	3.884730	0.299521	0.679730	2.905479
ALB	4.513433	33.264633	-35.186028	-35.144299	-4.479980	3.540002	0.497637	0.885168	2.157197
ARG	117.838768	66.719826	18.773989	72.824478	73.179199	0.066264	0.002378	0.028331	0.035554
AZE	-94.313599	-23.433115	-116.255142	-64.243759	-74.473793	1.632052	0.125706	0.740080	0.766265
BDI	-64.302223	-100.556137	-79.531349	-76.140251	-77.051071	14.166568	1.647794	2.287014	10.231760

4.b) Now, we'll run a regression with the mean scores of those indices and compare the results

reg = smf.ols('inst_mean ~ governAid_GDP + econAid_GDP + otherAid_GDP', data=df_col).fit()
reg.summary()

OLS Regression Results

Dep. Variable:	inst_mean	R-squared:	0.097
Model:	OLS	Adj. R-squared:	0.070
Method:	Least Squares	F-statistic:	3.597
Date:	Fri, 20 Dec 2019	Prob (F-statistic):	0.0162
Time:	10:29:12	Log-Likelihood:	-567.39
No. Observations:	104	AIC:	1143.
Df Residuals:	100	BIC:	1153.
Df Model:	3
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
Intercept	6.9604	8.169	0.852	0.396	-9.247	23.168
governAid_GDP	17.1017	7.333	2.332	0.022	2.553	31.650
econAid_GDP	6.5243	6.836	0.954	0.342	-7.037	20.086
otherAid_GDP	-7.1370	2.548	-2.801	0.006	-12.192	-2.082

Omnibus:	3.032	Durbin-Watson:	1.953
Prob(Omnibus):	0.220	Jarque-Bera (JB):	2.015
Skew:	0.127	Prob(JB):	0.365
Kurtosis:	2.367	Cond. No.	8.56

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

• We can observe that overtime gov. Aid has a strong influence on the improvement of the institutions. Its P-value is less than 2% thus variable is statistically significant with the 95% confidence interval.
• Also, now we see that the financial aid targeted at the economic development translates in positive results over time. However, P-value is extremely high so I'll leave judgment to the audience if this indicator should be used to deduct any strong conclusions.
• Other aid remains a negative contributor to the development of the institutions and we'll leave it there

Part 5:

---

5.a) We will add the income variable 'recipientGdp_log' to the model, to control for income and explain if it's statistically significant to include and/or if it's a reasonable candidate to be in our model

we convert 'recipientGdp_log' to float type and fill NaN results with the mean values of the _GDPlog

df["recipientGdp_log"] = pd.to_numeric(df["recipientGdp_log"])
df["recipientGdp_log"].fillna((df["recipientGdp_log"].mean()), inplace=True)
df["recipientGdp_log"].isnull

<bound method Series.isnull of 0       7.813625
1       8.390363
2       9.100701
3       6.449562
4       7.024415
          ...   
2838    7.963833
2839    8.276392
2840    7.771903
2841    9.158718
2842    7.245142
Name: recipientGdp_log, Length: 2843, dtype: float32>

We regress our model with the log value of the income, which represents percentage change in income:

reg = smf.ols('inst_mean ~ governAid_GDP + econAid_GDP + otherAid_GDP + recipientGdp_log', data=df).fit()
reg.summary()

OLS Regression Results

Dep. Variable:	inst_mean	R-squared:	0.158
Model:	OLS	Adj. R-squared:	0.156
Method:	Least Squares	F-statistic:	122.2
Date:	Fri, 20 Dec 2019	Prob (F-statistic):	9.94e-96
Time:	10:29:12	Log-Likelihood:	-14660.
No. Observations:	2619	AIC:	2.933e+04
Df Residuals:	2614	BIC:	2.936e+04
Df Model:	4
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
Intercept	-273.8100	13.427	-20.393	0.000	-300.138	-247.482
governAid_GDP	4.6791	0.794	5.893	0.000	3.122	6.236
econAid_GDP	2.0174	0.709	2.843	0.004	0.626	3.409
otherAid_GDP	0.8537	0.281	3.037	0.002	0.303	1.405
recipientGdp_log	34.0350	1.631	20.864	0.000	30.836	37.234

Omnibus:	26.973	Durbin-Watson:	1.998
Prob(Omnibus):	0.000	Jarque-Bera (JB):	22.363
Skew:	0.157	Prob(JB):	1.39e-05
Kurtosis:	2.673	Cond. No.	95.9

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

• When we add income to the model, we see a massive reduction in standard error and all of the aids show a positive effect on the institutional quality. Income itself has strong explanatory power with a massive (34.04) coefficient.
  • This can be explained by the fact that financial aid has a strong effect on the countries with higher GDP, because in general wealthier countries already have well-established institutions.
  • Also, in general, it’s always a good idea to control for the income to eliminate the bias involved with it
• The Adjusted-R2 increases from .007 to 0.156 which means that the given variables fit in the model and explain it better

5.b) Next, I will add some of the most interesting and statistically significant variables to the model which I picked from testing several combinations.

df.dtypes

year                datetime64[ns]
recipientCode                int16
country                     object
recipientRegion             object
weight                        int8
pop_weight                 float32
calpha                     float32
polity                     float32
xconst                     float32
ptss                       float32
indjud                     float32
checks                     float32
cmt_recip_GDP              float32
governAid_GDP              float32
econAid_GDP                float32
otherAid_GDP               float32
blt_GDP                    float32
recipientGdp_log           float32
recipientPop_log           float32
rc_wdi_lifexp              float64
rc_wdi_ttr                 float64
rc_wdi_urban               float64
rc_oilgas                     int8
missing                       int8
rc_ht_region                  int8
rc_ht_colonial                int8
recipientIncL                 int8
recipientIncLM                int8
recipientIncUM                int8
recipientIncH                 int8
bilateral                  float32
devFund                       int8
devBank                       int8
inst_mean                  float32
dtype: object

reg = smf.ols('inst_mean ~ governAid_GDP + econAid_GDP + otherAid_GDP + recipientGdp_log + rc_oilgas', data=df).fit()
reg.summary()

OLS Regression Results

Dep. Variable:	inst_mean	R-squared:	0.267
Model:	OLS	Adj. R-squared:	0.265
Method:	Least Squares	F-statistic:	189.9
Date:	Fri, 20 Dec 2019	Prob (F-statistic):	6.33e-173
Time:	10:29:12	Log-Likelihood:	-14478.
No. Observations:	2619	AIC:	2.897e+04
Df Residuals:	2613	BIC:	2.900e+04
Df Model:	5
Covariance Type:	nonrobust

	coef	std err	t	P>|t|	[0.025	0.975]
Intercept	-301.9990	12.612	-23.946	0.000	-326.729	-277.269
governAid_GDP	4.4326	0.741	5.981	0.000	2.979	5.886
econAid_GDP	0.6920	0.666	1.040	0.299	-0.613	1.997
otherAid_GDP	0.7896	0.262	3.010	0.003	0.275	1.304
recipientGdp_log	39.8180	1.550	25.683	0.000	36.778	42.858
rc_oilgas	-55.2162	2.802	-19.705	0.000	-60.711	-49.722

Omnibus:	20.102	Durbin-Watson:	1.961
Prob(Omnibus):	0.000	Jarque-Bera (JB):	20.482
Skew:	0.216	Prob(JB):	3.57e-05
Kurtosis:	2.969	Cond. No.	96.6

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

• I ended up adding - if the country has natural resources like oil and gas - and the findings are astonishing!
• We see that countries who possess and are exporting oil and gas show massively negative effects of financial aid on institutional quality. The coefficient of an oil & gas variable is -55 which is nearly double the income effect. It's P-value is close to 0
• Addition of the oil & gas variable also made our model better fit with adjusted R-squared from previously 15.6% jumping to 26.5%.

Many conclusions can be done based on these results. But the one thing is evidential that countries who have good institutional quality benefit from additional funds received from other countries targeted on the further improvement of those institutions. Also, government aid works best. Giving money to countries with natural resources like oil & gas has deteriorated their institutional quality. My findings are close to what original paper suggests, even though I've not gotten into using lagged values of aid variables, which authors of the paper consider to have better explanatory power as per fact after receiving aid it takes some time until we see the results of those funds.