BytePhilosopher/ethiopia-fi-forecast

Build a forecasting system that tracks Ethiopia's digital financial transformation using time series methods # Ethiopia Financial Inclusion Forecast Forecasts Ethiopia's financial-inclusion **Access** and **Usage** indicators to 2027 from a sparse, event-annotated time series — and explains *why* the mobile-money boom added 65 million wallets but almost no new account-holders. --- ## Business Problem Between 2021 and 2024 Ethiopia's operators registered **more than 65 million mobile-money accounts**. Telebirr alone passed 54 million users; M-Pesa added 10 million more. Over the same window, the share of Ethiopian adults owning *any* account — bank or wallet — moved from **46% to 49%**. Three percentage points, during the largest financial-services push in the country's history. That gap is an expensive problem for anyone allocating capital here: - **Registration counts overstate inclusion by roughly 8–10×.** A donor, regulator or investor tracking wallet sign-ups as a KPI is measuring marketing reach, not financial inclusion. - **Official targets were set against the wrong curve.** Ethiopia's NFIS-II strategy targets **70% ownership by 2025**. Nothing in the observed trajectory supports it, so programmes anchored to that number are budgeting against a plan that cannot land. - **Impact estimates borrowed from Kenya or India do not transfer.** Applied naively to Ethiopia they over-predict ownership by **13.7 percentage points**. The consortium behind this work needed a defensible answer to three questions: where will inclusion actually be in 2027, which events genuinely move it, and how much confidence should anyone place in the answer. ## Solution Overview The hard constraint is data sparsity: 87 records, and only 7 of the 30 observed indicators span two or more years. Classical per-indicator time-series modelling is not viable. So instead of forcing a model onto four Findex data points, the approach models the **events** and their transfer to Ethiopian conditions. 1. **A unified, event-annotated schema.** Observations, events, modelled `impact_link`s and pol …