Nigeria Federal Budget Observatory

Why Nigeria's Budget Deserves a Dashboard

Nigeria's federal budget has grown from ₦677 billion in the year 2000 to ₦54.99 trillion in 2025, a nominal increase of 81 times over. Yet in real purchasing power terms, adjusted for decades of inflation, that growth looks far more modest. In dollar terms, it barely keeps pace with 2014. This tension between what the numbers say and what they mean is exactly why this project exists.

Public finance data in Nigeria is plentiful but scattered. The Budget Office of the Federation publishes appropriation acts. The Central Bank of Nigeria (CBN) maintains monetary and exchange rate records. NEITI releases quarterly FAAC breakdowns. BudgIT translates budget documents into accessible formats for civil society. The challenge is not finding the data. It is putting it all in one place and making it speak.

This project pulls 25 years of approved budgets, actual expenditures, sector allocations, state FAAC distributions, inflation figures, exchange rates, oil prices, population data, and public debt records into a single interactive Streamlit dashboard. Eight modules allow users to explore the data from different angles: nominal figures, inflation-adjusted values, dollar equivalents, per-capita allocations, state-level distributions, budget implementation rates, machine learning forecasts, and debt sustainability simulations.

25 Years of Nigerian Budgets, in Seven Acts

Understanding the dashboard requires understanding the story it tells. Nigeria's fiscal history from 2000 to 2025 is not a linear growth story. It is a narrative shaped by oil booms, debt relief, recessions, a global pandemic, and a currency that lost nine-tenths of its value against the dollar in a single decade.

Where the Numbers Come From

Every figure in this dashboard traces back to a primary or credible secondary source. The table below documents each dataset, its origin, and the range it covers. No data is fabricated. Where figures for 2025 are labelled as estimates, they are based on signed appropriation acts or official government projections at the time of publication.

How the Project is Built

The dashboard follows a deliberate two-file architecture. All data logic, constants, and loading functions live in data.py, which acts as the single source of truth for every number in the project. The application interface lives entirely in app.py, which imports from the data module and renders eight navigable pages. This separation means the data can be audited, tested, or extended without touching any UI code, and the UI can be redesigned without risk of breaking data integrity.

# File structure
nigeria_budget_v2/
├── app.py            # 1,216 lines: 8-page Streamlit UI
├── data.py           #   299 lines: all datasets + loaders
└── requirements.txt  #     7 dependencies

# data.py exports 5 loader functions
load_budget_df()    # 26-row master table (2000–2025)
load_sector_df()    # 16-row sector allocations (2010–2025)
load_faac_df()      # 37-row state FAAC data (2023–2024)
load_impl_df()      # 10-row implementation tracker (2016–2025)
build_ml_features() # 24-row feature matrix for XGBoost

Data Pipeline

Raw data from the sources listed above was compiled, cross-referenced, and encoded directly into data.py as Python dictionaries. This choice was deliberate. Rather than relying on CSV files that could become separated from the codebase or corrupted, embedding the data as code means the entire project is self-contained in two Python files. Each loader function applies transformations on top of the raw figures.

The master budget table begins with 12 raw columns per year and produces 20 derived columns through calculations applied at load time. These include inflation-adjusted values using a CPI index rebased to 2015, dollar-equivalent budgets at annual average exchange rates, per-capita figures divided by population in millions, and debt-to-GDP ratios computed from the debt stock and GDP columns.

# Real-term deflation inside load_budget_df()
# Step 1: Build cumulative CPI index from annual rates
cpi_idx = [100.0]
for rate in df["CPI_Inflation_Pct"].iloc[1:]:
    cpi_idx.append(cpi_idx[-1] * (1 + rate / 100))

# Step 2: Rebase to 2015 = 100
base_2015 = cpi_idx[df.index[df["Year"] == 2015][0]]
df["CPI_Index_2015"] = [v / base_2015 * 100 for v in cpi_idx]

# Step 3: Deflate nominal budget to constant 2015 naira
df["Budget_Real_Bn"] = (
    df["Budget_Approved_Bn"] / df["CPI_Index_2015"] * 100
).round(1)

Navigation Architecture

Streamlit's default st.radio widget renders as a list of dot selectors, which is functional but visually weak for a portfolio project. The dashboard replaces this with a custom navigation system using st.button components styled via CSS targeting button[kind="primary"] and button[kind="secondary"]. Page state is persisted through Streamlit's st.query_params, which writes the active page to the URL. This means users can share direct links to specific modules, and the browser back button works correctly.

# Page state via URL query params
params = st.query_params
active = params.get("page", "overview")

for pid, icon, label, color in PAGE_META:
    clicked = st.button(
        f"{icon}  {label}",
        type="primary" if pid == active else "secondary",
        use_container_width=True,
    )
    if clicked:
        st.query_params["page"] = pid
        st.rerun()

Caching Strategy

All five data loader functions are wrapped in Streamlit's @st.cache_data decorator, which serialises the returned DataFrames to memory after the first call. On subsequent navigation between pages, no recomputation occurs. The World Bank API calls in the live data module use a one-hour TTL, and the exchange rate API uses a 30-minute TTL, balancing freshness against unnecessary network calls.

Building the Budget Forecast Model

The central question the ML module tries to answer is: given what we know about oil prices, exchange rates, population, economic growth, and debt levels, what should we expect Nigeria's federal budget to look like over the next five years? A linear regression could answer part of this. But Nigeria's budget does not grow linearly. It is driven by non-linear interactions between macroeconomic variables, making tree-based ensemble methods a stronger fit.

Why XGBoost

XGBoost (Extreme Gradient Boosting) is an ensemble method that builds decision trees sequentially, with each tree correcting the residuals of the previous one. It handles non-linear relationships without requiring the data to be normalised, manages interactions between variables automatically, and provides feature importance scores that make the model interpretable. For a dataset with 24 training rows and 9 features, a shallow XGBoost (maximum depth of 3) provides enough flexibility without overfitting.

Feature Engineering

The nine features fed into the model are a mix of direct economic indicators and engineered terms. The two most important design choices are the lag terms and the interaction term.

# Feature engineering in build_ml_features()
df["Lag1_Budget"] = df["Budget_Approved_Bn"].shift(1)   # prev year budget
df["Lag2_Budget"] = df["Budget_Approved_Bn"].shift(2)   # two years prior

# Oil price × exchange rate captures naira oil revenue
# (when oil is $80 and rate is ₦1,500/$, naira revenue = 80 × 1500 = 120,000)
df["Oil_USD_x_FX"] = df["Oil_Price_USD"] * df["USDNGN"]

df["Debt_to_GDP"]  = df["Debt_Stock_Bn"] / df["GDP_Bn"]

The oil-times-FX interaction term deserves particular attention. When oil is priced in dollars and Nigeria's revenues are reported in naira, the relevant quantity for budget planning is not the dollar oil price alone but the product of that price and the exchange rate. A higher oil price with a weaker naira still produces more naira revenue even if the dollar value is unchanged. Encoding this interaction explicitly gives the model information that neither variable carries alone.

The Forecasting Challenge for Tree Models

Tree-based models have a well-known limitation for time series forecasting: they cannot extrapolate beyond the range of values seen during training. When asked to predict a budget for 2027 given features that all lie within the training distribution, the model returns a value anchored to the training data range rather than projecting the underlying trend forward. This is not a bug in XGBoost. It is a fundamental property of decision trees.

The solution implemented in this dashboard is a blended forecast. The XGBoost prediction is combined with a drift estimate derived from the user-set macroeconomic assumptions, specifically the implied nominal budget growth from the inflation and GDP growth sliders. Early forecast years lean more on the model output. Later years lean more on the drift term, reflecting the reality that the model's signal degrades rapidly with distance from the training data.

# Blended forecast: GBM signal + growth drift
implied_growth = np.clip((inf_fc + gdp_fc) / 100 * 0.6, -0.1, 0.40)

for i, yr in enumerate(fc_years):
    gbm_pred   = model.predict(scaler.transform(feat_row))[0]
    drift_pred = last_b1 * (1 + implied_growth)

    # Alpha increases over horizon: more drift, less GBM signal
    alpha    = min(0.5 + i * 0.12, 0.65)
    pred_val = (1 - alpha) * gbm_pred + alpha * drift_pred

Model Performance

The model is validated using TimeSeriesSplit cross-validation, which respects the temporal ordering of the data. Unlike standard k-fold cross-validation, TimeSeriesSplit ensures that future data is never used to predict past values. With four folds, the cross-validated mean absolute percentage error sits around 34 percent, reflecting the high volatility of Nigeria's budget trajectory across periods as different as the 2016 recession and the 2025 post-subsidy-removal surge. In-sample, the model fits to within 0.2 percent.

What the Data Reveals

Running 25 years of Nigeria's fiscal data through multiple analytical lenses surfaces patterns that are not visible in the raw budget numbers alone. The five findings below represent the most significant insights the dashboard produces.

Finding One: The Inflation Illusion

Nigeria's 2025 budget of ₦54.99 trillion is 81 times larger than the 2000 budget of ₦677 billion. But when deflated to constant 2015 naira using the cumulative CPI index, the 2025 budget is equivalent to approximately ₦10.2 trillion in 2015 purchasing power. The 2015 budget itself was ₦4.49 trillion. So the real budget has grown by a factor of roughly 2.3 over ten years, not 12. Much of what looks like fiscal expansion is simply the naira becoming worth less.

Finding Two: The Dollar Decline

When measured in US dollars at the official exchange rate, Nigeria's federal budget peaked somewhere around 2013 to 2014, during the high-oil-price era before the first naira devaluation cycle. The 2025 budget of ₦54.99 trillion at an average rate of approximately ₦1,500 per dollar translates to around $36.7 billion. The 2014 budget of ₦4.64 trillion at ₦158 per dollar was worth $29.3 billion. The dollar budget has grown by only 25 percent over that decade, while the naira budget has grown by 1,085 percent. This gap is the measure of currency deterioration.

Finding Three: Debt Service Crowding Out Development

In 2022, Nigeria spent approximately ₦6 trillion on debt service against ₦2.75 trillion on capital expenditure, and only ₦2.1 trillion of that capital allocation was actually released. By 2024, debt service had risen to ₦8.27 trillion against capital releases of roughly ₦3.8 trillion. Nigeria was spending more than twice as much servicing debt as it was investing in infrastructure, education, and health combined. The dashboard's implementation tracker module places this comparison in a side-by-side chart that makes the magnitude of the imbalance immediately legible.

Finding Four: Capital Budgets That Never Get Spent

The average capital budget implementation rate from 2016 to 2025 is approximately 51 percent. In the best years, around 60 to 65 percent of approved capital spending is released. In crisis years like 2016 and 2020, it drops below 50 percent. The cumulative unspent capital from 2016 to 2024 alone exceeds ₦20 trillion, money that was approved, debated, and designated for roads, hospitals, schools, and power infrastructure but never reached the ground.

Finding Five: Nigeria's Education Budget Lags Every African Benchmark

The UNESCO recommendation is that countries allocate at least 15 to 20 percent of national budgets to education. Nigeria has never exceeded 9.26 percent in the period covered by this dashboard, and in recent years the share has fallen toward 7 percent. South Africa allocates 19.75 percent. Kenya allocates 17.5 percent. Ethiopia allocates 27 percent. Nigeria, with the largest population and one of the largest economies on the continent, ranks at the bottom of the peer comparison.

The Technical Stack

The project is deliberately lean. Every dependency earns its place either by being irreplaceable for a core feature or by reducing code complexity significantly. The entire dashboard runs from a single terminal command after a one-step install.

Core Framework

Data Layer

Machine Learning

Live Data

External APIs (Free, No Authentication Required)

Running the Project

# Clone or download the project
cd nigeria_budget_v2

# Install all dependencies (single command)
pip install -r requirements.txt

# Launch the dashboard
streamlit run app.py

# Opens automatically at http://localhost:8501

What This Project Demonstrates

The Nigeria Federal Budget Observatory is a portfolio project that sits at the intersection of data engineering, public finance analysis, and interactive visualisation. It demonstrates the ability to take scattered primary-source data, clean and enrich it through a reproducible pipeline, connect it to a live data layer, and present it through an interface designed to make complex information immediately readable.

The machine learning component demonstrates not just the ability to train a model but the ability to reason about its limitations. Knowing when to blend a tree model's output with a trend-based drift term, and being transparent about why, reflects the kind of analytical judgment that distinguishes thoughtful data work from mechanical model fitting.

The findings themselves carry genuine analytical weight. Nigeria's inflation illusion, its dollar budget decline, its debt-service crowding problem, its capital implementation gap, and its education funding shortfall are not conclusions drawn from opinion. They emerge directly from the data, sourced and cited, run through calculations that are fully reproducible in the project code.

All source code for this project is written in Python and structured as two files totalling 1,515 lines. The data is sourced from public-domain government and intergovernmental publications. Every calculation is reproducible. Every figure is traceable to its origin.