BAKU, Azerbaijan, June 21. Economic forecasting has always been an imperfect science. But in recent years, the gap between forecasts and reality has become harder to ignore. Growth projections are revised more frequently, inflation estimates often miss the mark, and economic outlooks that appear reliable can become outdated within a matter of months. The problem is not simply that economists are making more mistakes. Increasingly, the challenge lies in the fact that the global economy is changing faster than the models designed to explain it.
For much of the past three decades, forecasting benefited from a relatively stable economic environment. Global trade expanded steadily, supply chains became more efficient, and major economies moved through broadly synchronized business cycles. These patterns provided economists with a large body of historical data from which reliable relationships could be inferred.
That environment has changed. Trade disputes between major economies, sanctions regimes, military conflicts, and industrial policies aimed at strengthening domestic production have altered the flow of goods, energy, and capital across borders. Companies that once optimized supply chains for efficiency are increasingly prioritizing resilience and diversification. As a result, economic relationships that appeared stable for years have become less predictable.
One example is inflation. Before the pandemic, many advanced economies experienced low unemployment and modest wage growth without significant inflationary pressure. Traditional models suggested that tighter labor markets would eventually push prices higher. Yet inflation remained subdued for years. Then, following the COVID-19 pandemic, inflation surged at a pace that many central banks and forecasting institutions failed to anticipate. Supply-chain disruptions, shifts in consumer demand, labor shortages, and energy shocks combined in ways that historical data offered little guidance on.
Economists often describe such episodes as “structural breaks” — periods when long-standing relationships between key variables no longer hold. Forecasting models rely heavily on the assumption that past patterns contain useful information about the future. When those patterns change abruptly, model accuracy deteriorates.
The pandemic exposed these weaknesses more clearly than any event in recent history. Forecasts produced in 2020 were repeatedly revised as governments imposed lockdowns, rolled out stimulus programs, and reopened economies at different speeds. Recovery patterns differed sharply across sectors and countries. In many cases, economists found themselves responding to developments rather than predicting them.
Artificial intelligence may represent another structural shift. Businesses are investing heavily in AI technologies, yet their impact on productivity, employment, and investment remains uncertain. Historically, major technological advances have often taken years to translate into measurable economic gains. Whether AI follows a similar path—or accelerates economic change more rapidly—remains an open question for forecasters.
Another obstacle is the quality and timing of economic data itself. Official indicators such as GDP, employment, and productivity are often published with delays and may undergo substantial revisions months later. By the time a forecast incorporates new information, underlying conditions may already have changed.
To address this problem, economists increasingly rely on alternative sources of data. Credit card transactions can reveal shifts in consumer spending before retail sales reports are released. Shipping and logistics data can provide early indications of trade activity. Online price tracking can offer faster insights into inflation trends. Some institutions even use satellite imagery to monitor industrial production, construction activity, and energy consumption.
These tools provide a more immediate view of economic conditions, but they introduce new challenges. High-frequency data can be noisy, incomplete, or difficult to interpret. Signals from different datasets do not always point in the same direction. Having more information does not automatically produce greater certainty.
This has contributed to the growing popularity of “nowcasting,” an approach that focuses on estimating current economic conditions rather than making long-range predictions. While nowcasting can improve situational awareness, it does not eliminate uncertainty. Models built on real-time indicators can react too strongly to short-term fluctuations and may generate false signals during periods of market stress.
Geopolitics has further complicated the forecasting landscape. Wars, sanctions, export restrictions, and strategic industrial policies can reshape global trade patterns almost overnight. Unlike interest-rate cycles or consumer spending trends, geopolitical events rarely follow predictable economic logic. Russia’s invasion of Ukraine, for example, triggered major disruptions in energy and commodity markets that affected inflation and growth far beyond the countries directly involved.
As a result, economists are adapting their methods. Machine-learning techniques are increasingly used to process large datasets and identify patterns that traditional models may overlook. Natural-language processing tools can analyze news reports, corporate earnings calls, and policy statements to detect shifts in sentiment or economic expectations. Yet few economists view these technologies as substitutes for established economic frameworks. Instead, they are becoming additional tools within a broader analytical toolkit.
Perhaps the most important change is conceptual rather than technological. Forecasting is gradually moving away from the idea that a single prediction can accurately describe the future. Instead, many institutions now emphasize scenario analysis, presenting a range of possible outcomes and the conditions under which each might occur.
In an era defined by geopolitical uncertainty, fragmented trade networks, rapid technological change, and frequent economic shocks, forecasting is unlikely to become easier. Economic models remain valuable, but their limitations are increasingly apparent. For governments, businesses, and investors, forecasts may be most useful not as precise predictions, but as structured assessments of an uncertain and rapidly evolving world.
