Intro
Macroeconomic forecasting plays an important role in government policy, financial markets and business planning. Economists use predictions of gross domestic product (GDP), inflation and employment to assess economic conditions and anticipate future risks. However, traditional forecasting methods can struggle when economies experience sudden shocks, changing consumer behaviour or unexpected shifts in financial markets. Artificial intelligence (AI) is introducing new approaches by allowing researchers to process larger datasets, identify complex patterns and update predictions as fresh information becomes available.
In 2026, machine learning, time-series foundation models and AI-powered data analysis are becoming increasingly important parts of the economic forecasting toolkit. These technologies can help economists produce faster estimates, examine alternative outcomes and test how changes in interest rates, energy prices or employment might affect the wider economy. However, better technology does not guarantee more accurate predictions. The most promising approaches combine AI with established economic models, reliable data and rigorous testing. Understanding these developments is increasingly valuable for economists, financial analysts and professionals developing skills in economic data science.
Lets Dive In
Why AI Is Changing Macroeconomic Forecasting
Traditional macroeconomic forecasting relies on statistical models, historical relationships and expert judgement. Economists examine indicators such as consumer spending, industrial production, employment, interest rates and inflation to estimate the future direction of an economy.
These methods remain important, but they face several challenges. Economic data arrives at different times, is frequently revised and can change rapidly during periods of uncertainty. GDP figures, for example, are generally published quarterly. This means policymakers may need to make decisions before a complete picture of economic activity becomes available.
AI offers a different way to approach this problem. Machine learning algorithms can process large volumes of information from multiple sources. These can include government statistics, financial markets, business surveys, online prices, satellite imagery and employment data.
Rather than relying exclusively on a fixed set of historical relationships, some algorithms can identify patterns across many variables. They can also update estimates as new information becomes available. This is particularly useful when economists need to assess economic conditions between official data releases.
One important application is nowcasting. Nowcasting estimates current economic conditions before the full set of official statistics has been published. A model can combine monthly employment figures, retail activity, industrial production and financial indicators to estimate quarterly GDP growth.
The approach is already used by major research institutions. The New York Federal Reserve’s Staff Nowcast processes incoming economic information to produce an estimate of current GDP growth. Its methodology uses a Bayesian dynamic factor model and Kalman filtering, demonstrating how advanced statistical techniques can provide timely economic assessments.
AI is extending these capabilities through machine learning, alternative data and increasingly flexible forecasting models. The result is a more responsive forecasting environment, although economists must still assess whether new methods improve accuracy rather than simply increase processing speed.
AI Models Improving GDP Forecasting
GDP forecasting is one of the most promising applications of machine learning in economics. Quarterly GDP figures provide an important measure of economic activity, but the delay between the period being measured and the publication of official estimates creates a forecasting challenge.
AI models can help address this gap by combining traditional indicators with higher-frequency information. Retail transactions, energy consumption, business activity, shipping movements and satellite imagery can all provide additional signals about economic conditions.
One recent example comes from an IMF working paper published in January 2026. The research, titled Nowcasting Economic Growth with Machine Learning and Satellite Data, examined how machine learning and satellite-based nightlight information could improve estimates of real GDP.
The researchers found that incorporating nightlight data into a Random Forest model significantly improved quarterly GDP growth estimates compared with models using traditional indicators alone in their application. This is particularly relevant for countries where reliable economic statistics are limited or published less frequently.
Random Forest models work by combining predictions from multiple decision trees. This approach can capture complex relationships between variables and reduce the risk of relying on a single statistical decision structure.
Other machine learning methods are also being tested. Gradient boosting algorithms, including XGBoost and LightGBM, can identify nonlinear relationships and interactions between economic indicators. These techniques are useful when economists need to analyse large datasets containing many different variables.
However, performance depends on the quality of the data and the economic environment. A model trained during stable conditions may struggle when an economy experiences a financial crisis, pandemic or major geopolitical shock.
The most effective GDP forecasting systems therefore use a combination of machine learning, established statistical methods and expert interpretation. This allows economists to benefit from the flexibility of AI without abandoning the economic principles that support reliable forecasting.
Explore the IMF’s research on machine learning and satellite data for economic growth nowcasting.
BISTRO: A New Approach to Macroeconomic Time-Series Forecasting
One of the most interesting developments in AI forecasting is BISTRO, the Bank for International Settlements’ Time-series Regression Oracle.
BISTRO uses a transformer-based architecture adapted to macroeconomic time-series data. Unlike traditional models designed for one particular forecasting task, it aims to provide a more general-purpose approach to predicting economic variables.
The model is trained using a large repository of macroeconomic information maintained by the BIS. This allows it to identify relationships across different economic indicators and generate forecasts without requiring every task to be developed around a completely separate model.
The BIS demonstrated BISTRO’s capabilities through an analysis of the 2021 inflation surge. Traditional statistical benchmarks often projected that inflation would return towards its historical average. BISTRO was better able to anticipate the persistence of the inflation surge in the research application.
This is important because economic relationships can change. Inflation may remain elevated when supply constraints, energy prices and wage pressures continue to influence the economy. Models that rely too heavily on historical averages can struggle to recognise these conditions.
BISTRO also supports scenario analysis. Economists can generate a baseline forecast and then examine how alternative assumptions about explanatory variables affect the results. This creates opportunities to test different economic conditions without constructing a completely separate forecasting model for each scenario.
The BIS has made supporting scripts available through its BISTRO GitHub repository. The workflow can also be used with Google Colab, making the model more accessible to researchers who want to experiment with their own datasets.
BISTRO is an important development, but it should not be treated as a universal replacement for traditional forecasting. Its results still depend on the variables, historical information and assumptions used in each application. Independent testing remains essential.
MOIRAI and the Emergence of Time-Series Foundation Models
Another important development is the emergence of time-series foundation models. These systems apply ideas from large language models to numerical sequences rather than written text.
One example is MOIRAI, developed by researchers at Salesforce AI Research. It is designed to work across different time-series datasets and forecasting tasks, offering a more general approach than models built for a single economic indicator.
Traditional forecasting systems often require specific model structures for GDP, inflation, employment or other variables. Foundation models aim to reduce this separation by learning patterns from broad collections of time-series data.
This can make experimentation faster. Researchers can test different forecasting horizons and datasets without starting every project from the beginning.
However, general-purpose time-series models must be assessed carefully before being used in macroeconomic policy. Economic data has characteristics that make it different from many commercial forecasting datasets. It is often revised, affected by policy decisions and influenced by structural changes that are difficult to predict.
MOIRAI and similar models therefore represent promising research tools rather than automatic replacements for specialist macroeconomic models.
The wider trend is significant. Economists are beginning to explore whether foundation models can provide useful baseline forecasts, identify unusual patterns and support scenario analysis across different economic variables.
How AI Is Improving Inflation Forecasting
Inflation forecasting is another area where AI can provide meaningful advantages. Consumer prices are influenced by many interconnected factors, including wages, energy costs, supply chains, exchange rates, interest rates and consumer demand.
Traditional inflation models can struggle when these relationships change quickly. The inflation surge following the pandemic demonstrated how supply shortages, energy costs and strong demand can interact in ways that challenge established forecasting assumptions.
Machine learning provides additional ways to examine these relationships. Random Forest and gradient boosting models can analyse large groups of price indicators, while quantile regression forests can estimate a distribution of possible inflation outcomes rather than producing only one central prediction.
This distinction is important for central banks. Policymakers need to understand not only the most likely inflation path but also the risks surrounding it. A forecast that shows a wider range of possible outcomes can support more informed decisions about monetary policy.
The European Central Bank has discussed the use of quantile regression forests to produce inflation density forecasts. These methods help economists assess the likelihood of different inflation outcomes and understand the uncertainty surrounding a central estimate.
Another development involves large language models and automated price classification. Central banks increasingly have access to web-scraped prices and online product information. However, these datasets can contain millions of products with inconsistent descriptions and categories.
The BIS Innovation Hub’s Project Spectrum explored how generative AI could help classify products using large-scale price information. The project used the European Central Bank’s Daily Price Dataset, which contains billions of daily observations covering millions of unique products.
This type of classification can help economists turn large volumes of online pricing information into more useful signals for inflation analysis. It can also support faster monitoring of changing prices between official statistical releases.
However, generative AI introduces its own challenges. Processing costs, inconsistent classifications and model errors can affect the usefulness of the results. Human review and statistical validation remain important.
Read the BIS report on Project Spectrum and generative AI for inflation nowcasting.
Can Generative AI Models Predict Inflation Reliably?
The growing use of general-purpose AI models has created interest in whether tools such as ChatGPT can forecast macroeconomic indicators directly.
Large language models can analyse economic reports, summarise financial information and interpret written data. They can also generate numerical predictions when prompted with historical information. However, their ability to produce a plausible forecast should not be confused with proven forecasting accuracy.
A February 2026 working paper from the Federal Reserve Bank of San Francisco examined ChatGPT’s ability to forecast US inflation. The research, titled ChatMacro: Evaluating Inflation Forecasts of Generative AI, found that real-time out-of-sample predictions were largely inaccurate and stale, despite results that appeared more competitive in simulated historical forecasting exercises.
This finding highlights a crucial distinction. A model may perform well when researchers recreate historical conditions using information available in a controlled test. However, it may perform less effectively when asked to produce forecasts using genuinely current information.
Economists therefore need to distinguish between generative AI tools that assist with research and specialist forecasting models designed to predict numerical economic indicators.
Generative AI can still provide value. It can help researchers review economic reports, prepare code, compare assumptions and explore possible explanations for changes in inflation. Specialist statistical models should remain responsible for producing and validating the numerical forecasts.
The Federal Reserve research is a useful reminder that forecasting performance must be demonstrated through independent testing rather than assumed from the apparent intelligence of an AI system.
Read the San Francisco Fed’s ChatMacro research.
AI and Employment Forecasting
Employment forecasting presents a different set of challenges. Labour markets respond to economic growth, business confidence, wage pressures, interest rates and changes in technology. Employment data can also be revised, while official unemployment statistics may not capture every change in hiring demand.
Machine learning can help economists combine conventional labour market indicators with alternative datasets. These may include online job postings, recruitment activity, business surveys, payroll information and changes in vacancies across different industries.
Natural language processing can also help analyse job descriptions and identify changing demand for particular skills. This can provide early signals about which occupations are expanding or contracting.
For example, an increase in vacancies requiring specialist technology skills may indicate stronger demand in a particular sector. When combined with wage information and employment statistics, these signals can help economists assess labour market conditions.
Classification algorithms can group job postings by occupation, experience level and skill requirements. Forecasting models can then examine how these patterns relate to employment growth and wage movements.
However, online vacancies are not a complete measure of labour demand. Some employers recruit through internal networks, agencies or direct applications. Others leave advertisements online after positions have been filled. These limitations can introduce bias into employment forecasts.
AI models also need to distinguish between a change in job advertisements and a genuine change in employment. A business may advertise more positions without completing the recruitment process.
For this reason, machine learning employment forecasts are most useful when combined with official labour market statistics and other reliable indicators.
Platforms Allowing Economists to Test Multiple Economic Scenarios
The growing availability of AI forecasting models is changing how economists examine alternative economic outcomes. Rather than relying on a single forecast, researchers can test several assumptions and compare their potential consequences.
Scenario analysis is particularly useful when economic conditions are uncertain. An economist might want to understand how GDP, inflation and employment could respond to higher interest rates, rising energy costs, weaker consumer spending or stronger productivity growth.
Several platforms already provide advanced economic modelling and scenario analysis.
Oxford Economics Global Economic Model
Oxford Economics provides a comprehensive global forecasting and scenario platform through its Global Economic Model.
The system links 87 economies and supports forecasts extending to 2060. It includes more than 70,000 variables and allows users to construct custom scenarios by adjusting assumptions and simulating economic shocks.
Economists can examine how a change in one country affects trading partners, commodity markets and wider financial conditions. For example, a scenario involving higher US tariffs could be used to examine possible effects on trade, inflation and economic growth across different regions.
The platform also supports workflow integration through APIs and an Excel plug-in. This can help research teams incorporate economic forecasts into existing analytical processes.
Oxford Economics’ model is not simply an AI chatbot. It is a structured economic modelling environment that combines established macroeconomic relationships with scenario tools and regularly updated forecasts.
Its value comes from the ability to test alternative assumptions consistently across interconnected economies.
Explore the Oxford Economics Global Economic Model.
Moody’s Analytics Global Macro Forecast
Moody’s Analytics provides global macroeconomic forecasts and alternative scenarios through its economic data platforms.
Its Global Macro Forecast covers more than 130 countries, jurisdictions and regions, with a 30-year forecasting horizon. The underlying structural model includes more than 12,000 variables covering GDP, employment, inflation, financial markets, demographics and other economic indicators.
Economists can examine baseline forecasts alongside alternative scenarios. These include probabilistic scenarios and narrative scenarios designed around emerging economic risks.
For example, users can explore how a change in interest rates, financial conditions or energy prices could affect economic growth and employment. The platform also provides API and Data Buffet access, supporting integration with wider analytical workflows.
Moody’s approach combines economic theory, statistical forecasting and expert analysis. Its value lies in the breadth of its data and the ability to compare different economic outcomes within a consistent framework.
For organisations that need detailed country-level analysis, this can provide a useful foundation for strategic planning and risk assessment.
Explore Moody’s Global Macro Forecast.
New York Federal Reserve Staff Nowcast
The New York Federal Reserve Staff Nowcast provides another example of how advanced statistical models can improve the speed of macroeconomic analysis.
The platform estimates current US GDP growth by processing incoming economic information as it becomes available. It uses a Bayesian dynamic factor model and Kalman filtering to handle different data frequencies and changing information flows.
This approach allows economists to update their assessment as new indicators are released. It is particularly useful when official quarterly GDP estimates are not yet available.
The model also provides a useful comparison point for AI-based forecasting. It demonstrates that sophisticated statistical methods can remain highly competitive without relying on generative AI.
Economists can use this type of nowcast alongside machine learning forecasts to identify differences in assumptions and assess whether alternative models provide additional information.
Explore the New York Fed Staff Nowcast.
Comparing AI Forecasting Models and Economic Platforms
The range of available forecasting methods means economists need to understand what each tool is designed to achieve. Some models focus on identifying patterns in numerical datasets, while others provide complete economic frameworks for testing policy and market scenarios.
BISTRO offers a flexible approach to macroeconomic time-series forecasting. Its transformer architecture allows researchers to generate baseline and conditional forecasts, making it particularly interesting for economists exploring alternative assumptions.
MOIRAI represents the wider development of time-series foundation models. Its general-purpose approach may reduce the need to build separate forecasting models for every dataset, although economic applications still require careful validation.
Random Forest and gradient boosting models are useful for analysing large datasets and identifying nonlinear relationships. They can support GDP, inflation and employment nowcasting, particularly when researchers have access to alternative indicators.
Quantile regression forests provide another important capability by estimating a range of possible outcomes. This helps economists understand uncertainty rather than relying exclusively on a single forecast.
Oxford Economics and Moody’s Analytics provide broader economic modelling environments. Their platforms allow users to test multiple scenarios across countries and variables, combining structural economic relationships with updated forecasts.
The New York Fed Staff Nowcast provides a specialist example of real-time GDP monitoring using established statistical methods.
These approaches are complementary rather than direct substitutes. An economist may use a machine learning model to identify an emerging signal, compare it with a traditional nowcast and then explore the wider implications through a global scenario platform.
The strongest forecasting workflow is therefore likely to involve several tools rather than one universal AI solution.
Why Forecast Accuracy Still Depends on Data Quality
Despite the progress of AI forecasting, data quality remains one of the biggest limitations.
Macroeconomic datasets can contain missing observations, inconsistent definitions and historical revisions. GDP figures are often revised as additional information becomes available. Employment statistics can also change following updated surveys and administrative records.
This creates a challenge for machine learning. A model trained on revised historical information may appear more accurate than it would have been using only the information available at the original forecast date.
Economists therefore need to use real-time data where possible. This allows them to evaluate whether a model could genuinely have produced the same forecast under the conditions that existed at the time.
Overfitting is another concern. A complex algorithm may identify patterns that explain historical data but fail when economic conditions change. This is particularly important in macroeconomics, where the number of reliable historical observations can be relatively limited.
A model may also perform well in one country but poorly in another. Differences in economic structure, labour markets, financial systems and data availability can affect results.
Reliable forecasting requires consistent back-testing, comparison against traditional benchmarks and clear reporting of uncertainty. Economists should also examine how forecasts perform during periods of disruption rather than relying solely on average historical accuracy.
AI can improve the speed and flexibility of economic analysis, but it cannot remove the uncertainty inherent in predicting human behaviour and complex economic systems.
The Skills Economists Need for AI-Powered Forecasting
The expansion of AI in macroeconomic forecasting is creating demand for professionals who understand both economics and data science.
Economists increasingly benefit from knowledge of Python, statistical programming, machine learning and time-series analysis. These skills make it easier to prepare economic datasets, test forecasting models and evaluate prediction errors.
Understanding traditional econometrics remains equally important. Professionals need to recognise the assumptions behind regression models, dynamic factor models, vector autoregressions and other forecasting techniques.
Data visualisation is another valuable skill. Economists must communicate forecasts clearly, explain uncertainty and show how alternative assumptions affect potential outcomes.
Online learning provides a practical route for developing these capabilities. Courses in macroeconometric forecasting, machine learning and Python-based time-series analysis can help learners build relevant technical knowledge.
For career changers and existing finance professionals, combining economic understanding with practical data skills can support opportunities in economic research, financial analysis, policy support and business intelligence.
The most effective professionals will not simply accept an AI-generated prediction. They will understand how it was produced, test its reliability and explain what its results mean for decision-makers.
Recommended Online Courses to Develop AI Forecasting Skills in 2026
Building practical forecasting skills requires a combination of economic theory, statistical knowledge and machine learning experience. The following courses provide a useful starting point across three different online learning platforms.
Macroeconometric Forecasting – International Monetary Fund
Platform: edX
Level: Intermediate
Focus: Macroeconomic forecasting, time-series analysis, VAR and VECM models, uncertainty and policy analysis
This IMF course provides a structured introduction to the methods economists use to forecast macroeconomic variables. It explores how statistical models can be developed and assessed to predict inflation, economic growth and other indicators.
Learners examine time-series properties, model specification and forecasting uncertainty. The course also introduces vector autoregression and error correction models, which are important tools for understanding relationships between economic variables.
The practical focus makes it particularly relevant to economists who want to understand how forecasting models support policy analysis. The course listing reports a 4.6 out of 5 rating from 69 reviews and more than 100,000 learners.
Explore Macroeconometric Forecasting on edX.
Machine Learning with Python – IBM
Platform: Coursera
Level: Intermediate
Focus: Predictive modelling, regression, classification, clustering, Python and model evaluation
This course introduces the practical machine learning techniques used to develop predictive models. Learners explore supervised and unsupervised learning, regression, classification and dimensionality reduction.
It also covers model evaluation and optimisation, helping learners understand how to test whether an algorithm produces reliable predictions.
These skills are relevant to economic forecasting because researchers need to prepare datasets, select suitable algorithms and measure prediction errors. Although the course is not specifically designed for macroeconomics, its practical Python training provides a useful foundation for applying machine learning to GDP, inflation and employment data.
The listing reports a 4.7 out of 5 rating from more than 18,000 reviews and over 700,000 enrolled learners.
Explore Machine Learning with Python on Coursera.
Time Series Analysis, Forecasting, and Machine Learning
Platform: Udemy
Level: Intermediate
Focus: ARIMA, SARIMA, Python, statistical forecasting, machine learning and deep learning
This course explores several forecasting techniques used to analyse time-dependent data. It covers exponential smoothing, autoregressive models, ARIMA, SARIMA and machine learning approaches.
Learners also examine deep learning methods, including artificial neural networks, recurrent neural networks and long short-term memory models.
These techniques provide a useful technical foundation for understanding how forecasting algorithms identify patterns and generate predictions from historical observations. The examples are not exclusively economic, but the methods can be adapted to suitable macroeconomic datasets.
The course listing reports a 4.7 out of 5 rating from more than 3,100 reviews and over 12,000 learners. Its content was updated in March 2026.
Explore Time Series Analysis, Forecasting, and Machine Learning on Udemy.
Final Thoughts
AI forecasting tools are reshaping macroeconomic analysis by allowing economists to process larger datasets, identify complex relationships and update predictions as new information becomes available. Machine learning methods such as Random Forest and gradient boosting are supporting GDP and employment nowcasting, while quantile regression forests and transformer-based models are creating new opportunities for inflation forecasting and scenario analysis. Developments such as BISTRO and MOIRAI also demonstrate how foundation models could make economic forecasting more flexible. Meanwhile, platforms from Oxford Economics and Moody’s Analytics allow researchers to test alternative assumptions and examine how economic shocks could affect multiple countries and indicators.
However, the future of macroeconomic forecasting is unlikely to depend on AI alone. Data quality, economic theory, independent model testing and expert judgement remain essential for producing reliable predictions. Generative AI can support research and analysis, but specialist forecasting models must demonstrate their accuracy through real-world testing. For economists and aspiring data professionals, this creates an opportunity to develop valuable skills in econometrics, Python and machine learning. As predictive algorithms become more widely adopted, professionals who can combine economic understanding with practical data science will be better equipped to interpret forecasts, test scenarios and support more informed economic decisions.
