Three real analyses on Kenyan data

The previous 11 modules gave you the vocabulary. This one asks you to use it. Three real analyses on Kenyan data, each replicating an analysis published on this site, end-to-end in Python.

Project 1 — the bank-rates spread

Using the bankrates DataFrame (21 months of commercial bank lending and deposit rates), compute the spread (lending − deposit), plot it over time, and run a regression of the spread on a linear trend. Is the spread widening, narrowing, or stable? What does it tell you about competitive intensity in Kenyan banking?

bankrates['spread'] = bankrates['lending_rate'] - bankrates['deposit_rate']

fig, ax = plt.subplots(figsize=(10, 5))
ax.plot(bankrates['month'], bankrates['spread'])
ax.set_title('Lending-deposit spread, Kenyan banks')
plt.show()

# Regression: spread on a linear time trend
bankrates['t'] = range(len(bankrates))
import statsmodels.formula.api as smf
model = smf.ols('spread ~ t', data=bankrates).fit()
print(model.summary())

Project 2 — pension allocation shift

The pension DataFrame has 14 half-years of pension-industry asset allocation by asset class. Compute the share of government securities each half-year, plot the trend, and identify the inflection point. What was happening macroeconomically at that inflection?

# Compute share
pension['govt_share'] = pension['govt_securities'] / pension['total']

# Plot
fig, ax = plt.subplots()
ax.plot(pension['period'], pension['govt_share'])
ax.set_ylabel('Share of govt securities')
plt.show()

Project 3 — M-PESA growth rate

The mpesa DataFrame has Kenyan mobile-money volumes over time. Compute the year-over-year growth rate, plot it, and fit a linear trend. Is the growth rate decelerating? At what implied saturation level?

mpesa['yoy_growth'] = mpesa['volume_bn'].pct_change(periods=12)

fig, ax = plt.subplots()
ax.plot(mpesa['date'], mpesa['yoy_growth'])
ax.set_ylabel('YoY growth (%)')
plt.show()

What you've learned

If you can do these three projects without copying — load data, transform it, plot it, fit a model, interpret the output — you have working Python at an applied-economist level. The toolkit from here is mostly more libraries (scipy, scikit-learn, networkx) and more data, not more language. The language is mostly behind you.