Table of Contents¶
Kiva, loans changing lives?¶
Data Set: https://www.kaggle.com/kiva/data-science-for-good-kiva-crowdfunding
I have made loans with Kiva in the past in the hope of helping locals secure the working capital needed to generate a cash flow. But it usually takes years before I even see a repayment of cents. There is a lot of controversy surrounding Kiva and the benefit/harm it does to small communities.
Kiva has released a data set of their loans, and so I thought it would be interesting to take a look to see how the money is being used/distributed.
Lets start by seeing what kind of data we have.
The analysis¶
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import random
pd.options.mode.chained_assignment = None
# Load the data
df = pd.read_csv(r'C:\Users\Sunny\python_workbook\kiva_loans.csv')
#Preview data
df.head()
#Check the missing data
missing_count = df.isnull().sum()
percent_missing = (missing_count/df.count()*100)
num_unique = df.nunique()
missing_data = pd.concat([missing_count, percent_missing, num_unique], axis=1, keys=['Count', 'Percent', 'Unique'])
missing_data.sort_values(by=['Count'], ascending=False)
# Drop irrelevant data
df = df.drop(['id', 'country_code', 'region', 'tags'], axis = 1)
There are no major gaps in the data, so not much work needed here (Tags isn't something I'm interested in so I dropped it).
It wasn't clear if the loans were reported in USD so need to dig into that and clean up some of the data
# See what currencies we are dealing with
print('Unique Currencies:\n', df['currency'].unique(), '\n')
print('Currency summary: \n', df['loan_amount'].describe(), '\n', sep='')
# Plot the distribution of select currencies
df_currency_test = df[df['currency'].isin(['INR', 'USD', 'PHP', 'KES'])]
df_currency_test.boxplot(column = 'loan_amount', by='currency')
plt.title('Distribution of currencies in which loans are given')
plt.suptitle('')
plt.ylabel('Curency Unit')
plt.xlabel('Currency')
axes = plt.gca()
axes.set_ylim([0,5000])
plt.show()
# Clean data
df['use'] = df['use'].str.lower().str.strip('.').str.replace(' his ', ' their ').str.replace(' her ', ' their ')
df['borrower_genders'] = df['borrower_genders'].str.split(',').str[0]
df['borrower_genders'] = df['borrower_genders'].fillna(method='ffill')
Using a sample of four currencies which have substantially different exchange rates, it appears that the distributions of loans amounts are approximately the same. Therefore its reasonable to conclude all values in the data set are reported in USD.
We're already seing some interesting stuff. Looks like there are some larger loans up to \$100,000 which would not be microfinancing. Looking at the Kiva site, they apparently do engage in some larger loans but 75% are under $1000. It will be interesting to look at these larger loans in more detail later.
#Chart funded amount and loan amounts
fig = plt.figure(figsize=(20,5))
df_truncated = df[df['loan_amount']<1000]
plt.subplot(1,2,1)
ax1 = df_truncated['loan_amount'].plot(kind='hist', alpha=0.3, bins=20, edgecolor='k')
plt.title('Distribution of loan amounts')
plt.xlabel('Loan amount ($)')
plt.ylabel('Number of loans')
plt.subplot(1,2,2)
ax2 = df_truncated['funded_amount'].plot(kind='hist', alpha=0.3, bins=20, edgecolor='k')
plt.title('Distribution of funded loans')
plt.xlabel('Funded amount ($)')
plt.ylabel('Number of loans')
#Calculate and chart funded percentage in each bin
df_funding = pd.DataFrame()
df_funding['loan'] = df_truncated['loan_amount'].groupby(pd.cut(df_truncated['loan_amount'], np.arange(0,1000,50))).count()
df_funding['number_funded'] = df_truncated[df_truncated.percent_funded == 1.00]['loan_amount'].groupby(pd.cut(df_truncated[df_truncated.percent_funded == 1.00]['loan_amount'], np.arange(0,1000,50))).count()
df_funding['funded_percent'] = df_funding['number_funded']/df_funding['loan']
df_funding.reset_index().plot.bar(x='loan_amount',y='funded_percent', figsize=(15,5))
plt.title('Percent of loans funded')
plt.xlabel('Loan size')
plt.ylabel('Percent (%)')
Over 90% of loans under under \$1000 in each bin are funded to completion. A higher proportion of smaller loans are funded to completion than large loans. This data is encouraging to see if you want to fund a larger loan to someone as it will likely be funded to completion at some point and distbursed to the borrower.
# Plot top 10 in activity, sector, and use
fig = plt.figure(figsize=(15,10))
plt.subplots_adjust(hspace=0.3, wspace=0.5)
cols = df.loc[:, ['activity', 'sector', 'use']]
for i, col in enumerate(cols):
plt.subplot(2, 2, i+1)
value = cols[col].value_counts().head(10)
value.plot(kind='barh')
plt.title(col, fontsize=18)
plt.xlabel('Count', fontsize=14)
plt.gca().invert_yaxis()
Agriculture and farming rank the most popular types ('activity', 'sector') of loans. Communities in lesser developed countries usually rely on sustenance farming, which strongly links agriculture to their survivability. The actual use of loans is predominantly for drinking water. It's unclear whether that use would also fall under an 'Agriculture' type of loan, which would make sense if it was sustenance farming. If not, I would have expected a different distribution in the activity and sector distributions.
Loan repayments are very important to look at when evaluating any type of debt system. Without repayments, the ability for investors to achieve returns and reinvest is impaired.
# Plot frequency of repayment by term of loan
fig = plt.figure(figsize=(15,5))
repayment_intervals = df['repayment_interval'].unique()
for intv in repayment_intervals:
sns.kdeplot(df['term_in_months'][df['repayment_interval'] == intv] ,shade=True, label=intv)
plt.title('Frequency of repayment across term length of loans')
plt.xlabel('Term (Months)')
plt.ylabel('Frequency')
# Plot Genders
df_truncated["gender_split"] = ""
sns.violinplot(y='loan_amount', x='gender_split', hue='borrower_genders', split=True, data = df_truncated)
print(df_truncated['borrower_genders'].value_counts())
There are almost 4X the amount of loans made to females than males. Behaviouraly, studies have shown that females are more risk adverse and have better repayment habbits. Perhaps theis makes them better loan targets. Hoever, the violin plot seems to show denser loan distribution for males in the higher end, while females are getting most of the loans in the \$200 range.
fig = plt.figure(figsize=(15,50))
# Loan count by country
value = df_truncated['country'].value_counts()
value.plot(kind='barh')
plt.title('Loan count per country')
plt.xlabel('Loan count')
plt.ylabel('Country')
plt.gca().invert_yaxis()
# Loan Distribution by country
fig = plt.figure(figsize=(15,50))
ax = fig.gca()
df_truncated.boxplot(column='loan_amount', by='country', vert=False, ax=ax)
plt.suptitle('')
plt.title('Loan distribution by country')
plt.xlabel('Loan amount')
plt.ylabel('Country')
Initial analysis is encouraging¶
The analysis above is encouraging that loans being made on Kiva are being disbursed.
+ Over 70% of loans amounts are under $1000, proving that Kiva is a microfinancing service
+ Of those loans, over 95% are funded to completion, encouraging confidence that individual contributions will be disbursed
Also seen is that the loans are being directed to fundamental needs
+ Most popular use of loans is agriculture and water