📈 economics

Forecasting food price inflation in Nigeria and identifying its drivers using machine learning models

This study demonstrates that regularized regression models, particularly Elastic Net, outperform traditional time-series and complex machine learning algorithms in forecasting Nigeria's food price inflation, identifying historical prices, exchange rates, energy costs, and rainfall as key drivers while advocating for coordinated macroeconomic and agricultural policies to ensure food security.

Original authors: Oluwaseyi Samson Afolayan, Afolake Carolyn Afolami, Lawrence Olubunmi Balogun, Oriyomi Kazeem Aboaba, Adeleke Sabitu Coster

Published 2026-08-11

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Oluwaseyi Samson Afolayan, Afolake Carolyn Afolami, Lawrence Olubunmi Balogun, Oriyomi Kazeem Aboaba, Adeleke Sabitu Coster

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to predict the weather. You could look at a single thermometer from yesterday and guess it will be the same today, or you could try to build a super-complex robot that simulates every cloud, wind current, and bird in the sky to forecast the storm. This is the world of forecasting: the science of using past data to guess what will happen next. In economics, this is crucial for something called inflation, which is just a fancy word for when the price of things (like food) keeps going up, making your money buy less than it used to. When food gets expensive, families struggle to eat well, and the whole country can feel the stress. Scientists have been trying to figure out the best way to predict these price hikes for a long time. Some think simple math works best, while others believe we need giant, complicated computer brains (called machine learning) to solve the puzzle. But which one actually works better when the economy is messy and unpredictable?

This study dives into that exact question, but with a specific focus on Nigeria, a country where food prices have been jumping up and down wildly. The researchers wanted to know two things: First, can we accurately predict what food prices will do over the next two years? Second, what are the real "villains" pushing those prices up? To find out, they didn't just pick one method; they set up a massive race. They lined up a traditional, old-school math model (like a reliable, steady bicycle) against several high-tech machine learning models (like futuristic, high-speed motorcycles and even a complex robot). They fed all these models mountains of data about food prices, fuel costs, rain, and exchange rates to see which one could cross the finish line with the most accurate guess.

The results of this race were surprising. The researchers found that the "fancy" high-tech robots and complex neural networks actually stumbled and fell. They were too complicated for the job and ended up making bigger mistakes. Instead, the winners were the "simpler" models, specifically a type of math called regularized regression (think of these as smart, streamlined bicycles that know exactly when to pedal hard and when to coast). The champion of the race was a model called Elastic Net, which predicted the future with the highest accuracy, followed closely by its cousins, LASSO and Ridge Regression. The old-school bicycle (the standard time-series model) did a decent job, but the smart, streamlined bikes beat it. The complex robots (like Artificial Neural Networks) were the worst performers, getting lost in the noise.

So, what actually drives the price of food in Nigeria? The study acted like a detective to find the clues. They discovered that the single most powerful predictor of tomorrow's food prices is simply what food prices were yesterday. It's like a heavy ball rolling down a hill; once it starts moving, it keeps going. However, the study also found other important factors that push the ball. When the exchange rate (how much the local money is worth compared to foreign money) drops, food gets more expensive because importing things costs more. When fuel prices go up, it costs more to drive trucks to markets, which also raises food prices. Interestingly, rainfall acts as a brake; good rain means good crops and lower prices, while bad weather pushes prices up.

The authors suggest that while we often think we need the most complex, AI-driven tools to solve big economic problems, sometimes a simpler, well-tuned approach works better, especially when the economy is unstable. They didn't prove that complex AI is useless forever, but in this specific race with this specific data, the simpler models were the clear winners. The study concludes that to keep food prices stable, Nigeria needs to focus on stabilizing its currency, lowering fuel costs, and helping farmers deal with the weather, all while using these simpler, more reliable models to keep an eye on the prices before they spiral out of control.

Technical Summary: Forecasting Food Price Inflation in Nigeria and Identifying its Drivers Using Machine Learning Models

Problem Statement
Nigeria faces persistent and severe food price inflation, which reached 39.53% in July 2024, posing critical challenges to macroeconomic stability and food security. While the socioeconomic consequences of rising food prices are well-documented, there is a scarcity of evidence regarding the most effective forecasting approaches for Nigerian food inflation and the specific drivers behind these price dynamics. Traditional econometric models often struggle with the nonlinear relationships, structural breaks, and high-dimensional datasets characteristic of Nigeria's volatile economic environment. Furthermore, existing literature has largely focused on the welfare effects of inflation or aggregate inflation forecasting, with limited application of machine learning (ML) specifically to food price inflation or comparative evaluations of forecasting performance against classical benchmarks.

Methodology
The study employs a comprehensive analytical framework combining conventional time-series techniques with regularized econometric and machine learning models.

Data: The analysis utilizes monthly food price data from the World Food Programme (WFP) covering 66 markets, 38 commodities, and 14 Nigerian states (February 2015–May 2026). This was integrated with exogenous macroeconomic and climatic variables, including exchange rates, Premium Motor Spirit (PMS) pump prices, the FAO Global Food Price Index, and rainfall data.
Preprocessing: The dataset was cleaned to remove temporal inconsistencies, and food price inflation was constructed as a percentage change (Month-on-Month and Year-on-Year) to stabilize variance. Augmented Dickey-Fuller (ADF) tests confirmed the stationarity of the inflation series in levels.
Models: The study compares a benchmark Autoregressive (AR) model, selected via the Akaike Information Criterion (AIC) as AR(12), against several advanced models:
- Regularized Regression: Ridge Regression, Least Absolute Shrinkage and Selection Operator (LASSO), and Elastic Net.
- Dimensionality Reduction: Partial Least Squares (PLS) Regression.
- Nonlinear Machine Learning: Artificial Neural Network (ANN).
Evaluation: Models were assessed using out-of-sample forecasting metrics: Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Median Absolute Deviation (MAD). Feature importance was analyzed through correlation analysis and coefficient examination in regularized models.

Key Results

Forecasting Performance: Contrary to the expectation that complex nonlinear models would outperform linear ones in volatile environments, the study found that parsimonious regularized regression models provided the highest forecasting accuracy.
- Elastic Net achieved the best performance (RMSE = 9.94), followed closely by LASSO (RMSE = 9.95) and Ridge Regression (RMSE = 10.02).
- These models outperformed the benchmark AR(12) model (RMSE = 10.31) and significantly surpassed the PLS (RMSE = 16.51) and ANN (RMSE = 17.92) models, which exhibited poor predictive performance and negative out-of-sample $R^2$ values.
Drivers of Inflation:
- Historical Food Prices: Identified as the single most robust predictor across all models (particularly by LASSO), confirming strong inflation persistence.
- Macroeconomic Factors: Ridge Regression highlighted exchange rate movements and pump prices (fuel costs) as critical complementary predictors, reflecting the transmission of currency depreciation and energy costs to food prices.
- Climatic Factors: Rainfall was identified as a negative predictor, indicating that favorable weather conditions help stabilize prices by improving supply.
- Sectoral Trends: Prices for animal-source foods and edible oils showed the most rapid increases, driven by their dependence on imported inputs and energy-intensive production, while staple foods also experienced significant structural price shifts post-2022.

Significance and Contributions
The paper claims to contribute to the literature in four primary ways:

Forecasting Horizon: It provides out-of-sample forecasts for Nigerian food price inflation over a two-year horizon using a comprehensive market price dataset.
Model Comparison: It offers a systematic comparison of benchmark autoregressive models against regularized regression techniques and nonlinear machine learning algorithms, demonstrating that simpler, interpretable linear models often outperform complex nonlinear algorithms in structurally volatile, moderate-sample environments.
Driver Identification: Beyond prediction, the study identifies and ranks the principal macroeconomic, market, and climatic drivers of food inflation, offering empirical evidence on the mechanisms (e.g., cost-push inflation, exchange rate pass-through) shaping price dynamics.
Policy Relevance: The findings suggest that integrating regularized forecasting models into national monitoring systems can serve as an effective early warning tool. The study advocates for coordinated policies addressing exchange rate stability, energy costs, agricultural productivity, and climate resilience to mitigate food price volatility and enhance food security in Nigeria.

The authors conclude that while machine learning offers flexible predictive capabilities, the specific nature of Nigeria's food inflation—characterized by strong temporal persistence and linear transmission of macroeconomic shocks—is best captured by regularized regression models that balance predictive accuracy with interpretability.

Technical Summary: Forecasting Food Price Inflation in Nigeria and Identifying its Drivers Using Machine Learning Models

More like this