Tracking the Gulf Economy in Real Time: Machine Learning and Non-Oil GDP Nowcasting

The IMF develops a machine-learning nowcasting framework to estimate quarterly non-oil GDP in GCC countries in real time, addressing long data lags and oil-driven distortions in headline GDP. By combining high-frequency indicators with transparent model selection, the approach delivers timely, accurate insights to support faster and better-informed economic policy decisions.

The IMF working paper Nowcasting GCC GDP: A Machine Learning Solution for Enhanced Non-Oil GDP Real-time Prediction, produced by researchers at the International Monetary Fund (IMF) under its Middle East and Central Asia Department, tackles a practical problem facing policymakers in the Gulf Cooperation Council (GCC). In these oil-exporting economies, official GDP data arrive with long delays and are frequently revised, while headline growth figures are often distorted by oil price swings. This makes it difficult for governments, central banks, investors, and analysts to understand what is happening in the economy right now. The study argues that focusing on non-oil GDP, rather than total GDP, provides a clearer and more policy-relevant picture of domestic economic activity, especially as GCC countries pursue diversification and structural reforms.

From Forecasting to “Nowcasting”

Instead of forecasting growth far into the future, the paper focuses on nowcasting, estimating current or very near-term GDP using high-frequency data. The authors show that nowcasting is especially valuable in the GCC, where quarterly GDP is often published three to six months late. Non-oil GDP is a better indicator of consumption, investment, and private sector activity than total GDP, which is heavily influenced by oil production decisions and global energy prices. By nowcasting non-oil GDP, policymakers can better track underlying economic momentum and respond more quickly to emerging risks or opportunities.

Using Machine Learning, Not Just One Model

A key innovation of the paper is its model-agnostic approach. Rather than relying on a single statistical model, the IMF team tests 22 different models, including traditional regressions and modern machine learning methods such as random forests, gradient boosting trees, elastic net regressions, and support vector machines. Each model has different strengths, and testing many of them reduces the risk that results depend on one specific technique. This flexibility is particularly important in the GCC, where economies are changing rapidly, and historical relationships between variables can shift over time.

Turning Big Data into Clear Signals

The framework draws on a wide range of high-frequency indicators, including consumer prices, retail transactions, trade flows, industrial production, purchasing managers’ indices, financial market data, credit conditions, interest rates, and oil prices. The authors also include regional and global indicators from major trading partners, reflecting the openness of GCC economies. To keep the models transparent and useful for policy, the paper uses Shapley value decompositions, a tool that shows how much each indicator contributes to the final nowcast. This allows the researchers to drop weak indicators, keep strong ones, and explain which economic signals are driving changes in estimated growth, rather than treating machine learning as a “black box.”

What the Results Show

The results highlight both similarities and differences across GCC countries. Oil prices remain important everywhere, even for non-oil GDP, because they affect government spending, liquidity, and confidence. Consumer prices and point-of-sale data capture domestic demand, while business surveys signal production and services activity. Financial indicators are especially important in countries with deeper markets, such as Saudi Arabia and the UAE. At the same time, each country’s model reflects its own structure and data availability. Saudi Arabia’s nowcasts benefit from detailed domestic indicators linked to construction, exports, and financial markets, while Oman and Kuwait rely more on external and regional signals. Overall, the models track historical non-oil GDP closely and respond well to economic shocks, including the COVID-19 downturn and recovery. In Saudi Arabia, the IMF’s nowcasts even perform slightly better on average than official early GDP estimates.

Why This Matters for Policy

The paper concludes that this is the first region-wide, systematic application of machine learning nowcasting tailored to the GCC. By combining timely data, multiple models, and clear interpretation tools, the framework gives policymakers a practical way to monitor economic conditions in real time. As data quality improves and new indicators become available, the approach can be expanded further. In a region facing volatile oil markets, ambitious reforms, and global uncertainty, the study makes a strong case that real-time, non-oil GDP nowcasting is no longer optional; it is essential for effective economic decision-making.