AI inflation forecasting has been deemed “absolutely useless” by a leading investment strategist, with a new study revealing that generative AI models like ChatGPT are significantly inferior to established, low-tech alternatives. The findings, highlighted in a MarketWatch report on Tuesday, May 12, 2026, underscore a critical limitation in the much-hyped capabilities of artificial intelligence within macroeconomic analysis.
Investors’ enthusiasm for AI’s predictive power often outstrips its actual performance, especially when it comes to complex economic indicators such as inflation. Joachim Klement, head of strategy at Panmure Liberum, a prominent U.K. investment bank, minced no words in his assessment, stating that ChatGPT’s record in macroeconomic forecasting is so poor it renders the technology “absolutely useless.” This strong claim is backed by rigorous academic research, challenging the prevailing narrative around AI’s readiness for critical financial applications.
Cleveland Fed Model Outperforms Generative AI
The study, titled “ChatMacro: Evaluating Inflation Forecasts of Generative AI,” directly compared ChatGPT’s inflation predictions against those of the Federal Reserve Bank of Cleveland’s inflation nowcasting model. The results were stark: ChatGPT’s error rate was up to 12 times greater than that of the Cleveland model. This substantial discrepancy suggests that while AI excels in many domains, AI inflation forecasting remains a significant hurdle it has yet to overcome effectively.
“ChatGPT’s error rate was as much as 12 times greater than that of the Cleveland model.”
The Cleveland Fed’s model, a ‘low-tech’ tool that has been in use for years, consistently delivers more reliable forecasts. This proven accuracy provides a critical benchmark against which newer, more sophisticated technologies like generative AI are measured. The study’s findings are particularly pertinent for financial institutions, policymakers, and investors who rely on precise inflation projections for strategic planning and risk management.
The Pitfalls of Hindsight Bias in AI Training
One of the key challenges identified in evaluating AI’s macroeconomic forecasting abilities is the issue of hindsight bias. The study’s authors meticulously highlighted the importance of ensuring that AI models are not inadvertently trained on data that includes future information, which would artificially inflate their perceived accuracy. Even when explicitly instructed to use only publicly available data prior to a specific date, researchers found evidence that ChatGPT appeared to be “using forwarding looking information despite being asked not to.”
This critical observation underscores the need for true out-of-sample testing to accurately gauge an AI model’s predictive power. Many past analyses claiming out-of-sample results may have inadvertently overlooked this subtle but significant bias, leading to an overestimation of AI’s capabilities. For those seeking reliable economic insights, understanding these methodological nuances is paramount.
As of its latest prediction, the Cleveland Fed model forecasts the CPI’s 12-month rate of change for May to be 4.2%. This provides a concrete, data-driven figure for market participants, contrasting sharply with the unreliable output of generative AI. For further insights into financial market trends, explore our related Finance news.
Implications for Financial Professionals and Investors
The implications of these findings are substantial for financial professionals and investors. Relying on AI for critical economic forecasts, particularly for something as impactful as inflation, could lead to significant misjudgments and suboptimal investment decisions. The study serves as a strong reminder that while AI offers revolutionary potential, its application in highly sensitive and complex areas like macroeconomic forecasting requires rigorous validation and a healthy dose of skepticism.
Ultimately, the message is clear: when it comes to predicting inflation, traditional, proven models currently offer superior accuracy compared to the current generation of AI. Investors and analysts should proceed with caution and prioritize models with demonstrated out-of-sample predictive power over the allure of cutting-edge, yet unproven, artificial intelligence for critical financial decisions.
