Modeling and Forecasting Somalia’s Economic Growth: A Comparison of Traditional Time Series and Modern Neural Network Approaches
This study compares ARIMA, ETS, and NNAR models for forecasting Somalia's annual GDP growth using 1961–2024 data, finding that the simple ETS(A,N,N) model outperformed both the ARIMA and neural network approaches in out-of-sample accuracy for the 2015–2024 period.
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
Predicting the future of a nation's economy is one of the most difficult tasks in science. Governments need to know how much money they will have to build schools, pay workers, and feed their people, but the economy is rarely a straight line. It is a jagged path of sudden booms and sharp crashes, often driven by events that have never happened before, like severe droughts or the sudden stop of foreign aid. To make sense of this, economists use mathematical tools to look at past numbers and guess what comes next. Some of these tools are old and rely on simple patterns, assuming that what happened recently will likely happen again. Others are modern and complex, using computer programs that try to learn intricate, hidden rules from the data, much like a student trying to memorize a textbook by finding every possible connection between the words. The big question for anyone trying to plan for a country's future is whether these fancy, modern tools are actually better than the simpler, older ones, or if the complexity just adds confusion.
A team of researchers in Somalia set out to answer this question for their own country's economic growth. Somalia's economy has been through a turbulent history, with years of rapid expansion followed by deep contractions, making it a perfect test case for these forecasting methods. The researchers gathered sixty-four years of data, from 1961 to 2024, recording the percentage by which the country's total economic output grew or shrank each year. They then split this history into two parts. They used the first fifty-four years, up to 2014, to teach three different forecasting models how the economy had behaved in the past. The first model was a traditional statistical approach that looks for linear connections between past years. The second was a method that gives more weight to recent events, fading out the influence of older history. The third was a modern neural network, a type of artificial intelligence designed to find complex, non-linear patterns that the human eye might miss. Once the models were trained, the researchers asked them to predict the next ten years, from 2015 to 2024, without showing them the actual answers. They then compared the predictions to the real numbers that actually happened.
The results were surprising to those who assume that newer technology always wins. The most accurate predictions did not come from the complex neural network. Instead, the simplest method, which focused only on the current level of economic activity without trying to predict a specific trend or pattern, turned out to be the most reliable. This model, known as exponential smoothing, predicted the average growth rate with an error of about 3.8 percentage points. The traditional statistical model came in a close second, with an error of roughly 4.1 percentage points. The modern neural network, despite its ability to learn complex relationships, performed the worst, missing the mark by an average of 4.7 percentage points and making some very large errors when the economy swung sharply. The researchers found that the neural network struggled to handle the extreme volatility of Somalia's economy, often smoothing out the sharp peaks and valleys that actually occurred, while the simpler models were better at capturing the general direction of the growth.
This study suggests that for a country with a short and volatile economic history, adding more complexity does not necessarily lead to better answers. The neural network, which had more "patterns" to learn, likely tried to memorize the noise in the data rather than the signal, a problem that becomes worse when there is not enough historical data to work with. The simpler models, by focusing on the most recent level of growth, managed to avoid these pitfalls and provided a more stable baseline for the future. The authors emphasize that while the simple model was the best of the three, it is not a crystal ball. The economy of Somalia remains highly sensitive to shocks like drought and conflict, which no mathematical model can predict on its own. However, having a reliable tool to estimate the general direction of growth allows planners to make more informed decisions about where to allocate resources. The study concludes that for this specific task, the quiet, steady approach of the simpler method outperformed the high-tech alternative, offering a practical tool for economic planning in a challenging environment.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.