Introduction

Kenya’s 2016 interest rate regulation capped bank lending at four percentage points above the Central Bank of Kenya base rate, set at 14.5% at enactment, below prevailing market rates of around 18%. A deposit rate floor was imposed at roughly 6.2%. M-Shwari, the dominant digital platform operated by Commercial Bank of Africa, retained an exemption, continuing to charge a 7.5% monthly facilitation fee. Competitors KCB and Equity Bank saw digital rates fall from as high as 9% to 1.1% per month. The study analyses borrower and bank responses to this asymmetric reform and its welfare consequences.

Background on the digital banking sector and interest rate caps

M-Shwari launched in 2012, offering uncollateralised 30-day loans integrated with M-PESA. Within two years it had over 4.5 million active users (20% of adults). By 2017, over 80% of mobile phone owners held an M-Shwari account, with approximately 30% having borrowed. As of 2017, CBA held over 50% of loan accounts in Kenya.

Data

The analysis draws on administrative data from CBA covering a random sample of almost 10,000 customers for loans and credit limits and 5,000 for savings, opening accounts between January and March 2016. A regression discontinuity design is applied around the 14 September 2016 implementation date, with a 46-day optimal bandwidth. Within this window, approximately 2% of customers had a loan on any given day, the average loan amount was KShs 20, the average credit limit was KShs 836, and the average daily savings balance was KShs 624. Survey data from Suri, Bharadwaj, and Jack (2021) complements the administrative data for structural estimation.

Reduced form effects of the interest rate caps

After the caps, the daily probability of having an M-Shwari loan increased by roughly 10% off a 1.8% baseline, and loan amounts rose by about 1.1% (p.4). First-time access expanded by approximately 70% (p.4). M-Shwari raised credit limits for safer customers; each additional credit score point was associated with a nearly 15% increase in limits (p.11). Safer borrowers substituted toward cheaper capped alternatives, while riskier borrowers increased savings to rebuild M-Shwari credit limits. The probability of positive savings increased by 1.5 percentage points (p.12).

Theory

A two-firm model with heterogeneous repayment risk formalises these findings. The regulated competitor tightened its lending cutoff post-caps, redirecting marginal borrowers to M-Shwari. In competitive segments, M-Shwari raised limits to retain safer customers; where competitors exited, it reduced limits as competitive pressure declined. The uniform deposit rate floor reduced the cost of savings-based limit-building equally across platforms.

Structural estimation

Parameters are estimated via two-step GMM, combining administrative and survey data. Under the actual carve-out, overall average welfare fell modestly from a baseline of 1.00 to 0.97, with low credit score borrowers declining from 0.86 to 0.79, and high credit score borrowers improving slightly from 1.15 to 1.16. Under a hypothetical uniform cap, overall welfare fell sharply to 0.62: low credit score borrowers lost credit access entirely (welfare falling to 0.00), while high credit score borrowers improved to 1.30. These rankings are stable across a range of model parameter values.

Conclusion

The M-Shwari carve-out preserved access for high-risk borrowers, albeit at worse terms, yielding a modest aggregate welfare decline. A uniform cap would have generated substantially larger losses by eliminating credit entirely for low credit score borrowers. The authors suggest that policies targeting screening margins directly, such as calibrated risk-based limits or subsidies tied to first-time formal borrowing, may better protect high-risk households than uniform price ceilings.