When Data Runs at Different Speeds
How can we tell if the economy is heading down a dangerous path? In recent years, economists have increasingly talked about “growth-at-risk”, i.e., the probability that future growth will slide toward extreme and undesirable scenarios. It is a simple but powerful idea: don't just look at the average GDP or inflation forecast, but question the tails of the distribution, where the greatest risks of recession, debt crisis, or inflationary spikes lie. A recent study by Massimiliano Marcellino (Bocconi) and Michael Pfarrhofer (WU Vienna) proposes new tools to tackle this task. The study, published in Economics Letters, focuses on Italy, a country with high public debt where monitoring macroeconomic risks is crucial.
Looking beyond the average: the role of tails
When it comes to economic forecasts, we are used to hearing things like “GDP will grow by 1% next year.” But that is the average. Around that number there is a distribution of possible scenarios: the central part covers the most likely outcomes, in this example slightly above or below 1%. At the tails, on the other hand, are rare but critical events: a deep recession, a sudden surge in prices, an explosion of debt.
And that is precisely where the concerns of central banks and governments lie. Studying the tails means preparing not for the “most likely” future, but for the one that can do the most damage. In other words: from “how much we will grow” to “how much we could fall (or overheat) if things go wrong (or too well)”.
Data traveling at different speeds
A practical problem is that economic indicators do not all travel at the same speed: GDP and debt are measured every three months, while inflation, industrial production, and unemployment are measured monthly. Marcellino and Pfarrhofer have therefore developed a “mixed-frequency” model that combines these heterogeneous sources to obtain real-time forecasts. To do this, the authors compared two families of tools:
- Bayesian VAR models, which are more traditional but more realistic because they take into account the fact that the volatility of economic data is not constant but can change over time;
- BART (Bayesian Additive Regression Trees) models, capable of capturing more complex and non-linear relationships, adapting better to the actual trend of the data.
More accurate forecasts
By applying these techniques to key variables such as debt-to-GDP ratio, deficit, real GDP growth, inflation, unemployment, and industrial production, the authors showed that the new measures improve the ability to predict extreme risks. In some cases, nonlinear models (BART) outperformed their more traditional competitors, especially for inflation and unemployment; in others, the updated VARs proved particularly robust.
A tool for economic policy
Monitoring risks does not, of course, mean predicting the future with certainty, but rather having more reliable maps of possible negative scenarios. For a country like Italy, this can make the difference between reacting late or intervening in time with fiscal and monetary policies.
More generally, the work shows how the frontier of econometrics is not just a theoretical exercise: new algorithms and statistical models can be translated into concrete tools for better understanding the uncertainty that accompanies our economies.
Massimiliano Marcellino, Michael Pfarrhofer, “Nonparametric mixed frequency monitoring macro-at-risk”, Economics Letters Volume 255, September 2025, DOI https://doi.org/10.1016/j.econlet.2025.112498
