Umberto Collodel

Senior Economist, Monetary Policy, Central Bank of Malta  |  PhD, Paris School of Economics

I extract forward-looking macro signals from unconventional data — central bank language, web-scraped prices, news and capital flows — to anticipate macro-financial stress before it shows up in official statistics.

Central bank communication LLMs & alternative data Nowcasting & early warning
Umberto Collodel
Experience
Central Bank of Malta
IMF Independent Evaluation Office
Education
PhD, Paris School of Economics

About

My research asks how much we can learn about the macroeconomy before official data arrive. I use large language models, market data and high-frequency alternative data to measure how central bank communication moves market uncertainty, how political pressure on central banks shapes rate expectations, how online retail prices can nowcast inflation in real time, and how accurate international organisations' growth forecasts really are.

At the Central Bank of Malta, I work in the Monetary Policy and Eurosystem Relations Department, preparing the analyses and briefing the Governor ahead of ECB Governing Council meetings.

Before the CBM, I worked with the IMF Independent Evaluation Office on the evaluations of the Fund's emergency response to the COVID-19 pandemic and of Fund advice on capital flows. I hold a PhD from the Paris School of Economics (2022), supervised by Prof. Agnès Bénassy-Quéré.

Research

Journal publications1
Journal of Macroeconomics, 2026
Market-Based Monetary Policy Uncertainty Shocks in the Euro Area

with Vanessa Kunzmann (Deutsche Bundesbank)

We introduce a market-based measure of monetary policy uncertainty: the change in the standard deviation of OIS rates in a three-day window around ECB announcements. Announcements generally raise uncertainty about future rates regardless of the sign of the policy surprise, and this uncertainty moves asset prices — raising nominal yields, lowering stock returns and increasing euro exchange-rate volatility.

Working papers3
arXiv 2610.32300, 2026
Jawboning the Fed: Do Markets Listen When Administrations Talk?

with Michela Baretti (Universidad de Alicante) and Jacopo Zacchè (Central Bank of Malta)

Using Reuters newswire from 1988 to 2025, classified by an open-weight LLM ensemble, we provide the first long-run evidence on whether markets price political pressure on the Federal Reserve. Dovish pressure lowers near-term rate expectations by about one basis point per event — roughly a four-percent rise in the implied probability of a cut at the next FOMC — fading beyond three months. The effect comes entirely from presidential commentary, holds outside the Trump presidencies, and leaves inflation compensation unchanged: markets price anticipated accommodation, not a loss of credibility. Hawkish pressure has no detectable effect.

Impact of dovish and hawkish pressure on federal funds futures
Dovish vs. hawkish pressure on fed funds futures across maturities
CBM Working Paper 4/2026
Interpreting the Interpreter: Can We Model post-ECB Conferences Volatility with LLM Agents?

Central banks cannot observe how markets will interpret their communication before release. I use large language models to simulate 30 heterogeneous traders reading ECB press conference transcripts and measure their cross-sectional disagreement. Across 293 Governing Council meetings, this measure correlates at about 0.5 with realised OIS volatility, outperforms standard text-based alternatives, and keeps its predictive power after controlling for the policy surprise and liquidity — offering an ex-ante tool for assessing how communication will land.

Simulated vs. realised OIS disagreement
Simulated disagreement after ECB press conferences
RBA Research Discussion Paper 2026-05
Real GDP Forecasts by International Organisations

with Martin McCarthy (Reserve Bank of Australia) and Prakash Loungani (Johns Hopkins University)

We study real GDP growth forecasts by international organisations — forecasts that shape policy at those institutions and in national governments. We compare their accuracy with other forecasters, test whether they are rational, and examine why rationality tests are rejected.

Figure from Real GDP Forecasts by International Organisations
Real GDP forecasts by international organisations
Policy publications2
Central Bank of Malta, 2025
Crumbs Make a Loaf: Using Online Product Prices to Nowcast Inflation

with Massimo Giovannini and Sarah Spiteri (Central Bank of Malta)

Food is about 16% of Malta's HICP. We build a dataset of over 2,700 products and more than two million daily web-scraped supermarket prices, classified with string matching and LLMs, and compare a naïve benchmark, a minimum-distance method and a mixed-frequency machine-learning model. Out of sample, online prices improve food inflation nowcasts.

IMF Independent Evaluation Office, 2023
The IMF's Forecasting and Policy Advice Formulation Processes

Background Paper 4, evaluation of the IMF's emergency response to the COVID-19 pandemic

Replication materials — data, code and documentation — are on GitHub.

PhD Thesis

The Interplay between Macroeconomic Forecasting, Financial Crises and Complexity

Paris School of Economics, 2022. Supervisor: Prof. Agnès Bénassy-Quéré

Three empirical chapters. The first shows that early warning systems for external crises forecast relatively well, comparing global and domestic indicators and the output costs of missed versus predicted crises. The second evaluates short-term growth forecasts by major institutions and the private sector, and how their errors relate to the business cycle and politics. The third builds a new textual dataset from IMF archival documents to track how crisis discourse has evolved across the Fund's membership — released as the Complex Crises Database.

Early warning system performance
Early warning system performance

Data

PSE Working Paper 2021-47
The Complex Crises Database: 70 Years of Macroeconomic Crises

co-developed with Manuel Bétin (OECD)

A text-mining database built from roughly 23,000 IMF country reports, using the Fund's archival documents to detect crises. It covers 20 types of economic, financial and non-economic events for 181 countries from 1950 to 2019.

Contact

Profiles GitHub LinkedIn
Location Valletta, Malta