Statistics, econometrics and programming

At the ECB's Macroprudential Policy Division, I worked with granular loan-level and supervisory data at very large scale to study the effect of capital requirements on bank lending. The data warehouse combined roughly 100 million individual AnaCredit loan observations with bank-level supervisory time series; the broader research environment exceeded 400 GB.

I programmed SQL infrastructure for joining and extracting datasets, recurring table updates and distributed processing. I connected the database layer to econometric work in R so that large-scale queries could feed regression and panel-data analysis.

The research examined whether banks close to regulatory capital constraints reduced lending during the pandemic and whether actual capital-requirement relief supported credit supply. I am explicitly acknowledged in the related ECB Working Papers 2644 and 2720.

In the ECB's Bond and International Markets Division I developed analytical tools for financial markets and portfolio management. My work covered bond, foreign-exchange and broader market data, portfolio monitoring and risk indicators.

In Python I analysed more than 100 GB of tick-by-tick FX trading data, including episodes of heightened volatility. In Tableau I built interactive tools for parametrised Value-at-Risk calculations and monitors for equities, volatility, commodities and portfolio performance.

The work also included database architecture for managed-portfolio and foreign-exchange-operation data, supported by Bloomberg, Thomson Reuters and other professional market-data systems.

In ECB Supervisory Banking Statistics I worked with very large supervisory banking datasets and developed technical infrastructure for financial, credit-risk, liquidity, stress-test and data-quality indicators.

In SAS I programmed analytical SQL queries, dataset creation, institution-level indicator extraction, validation rules and reporting processes. A redesigned code base reduced one daily supervisory-report extraction from approximately six hours to two.

I used Tableau for visualisation and Python/scikit-learn for machine-learning processing and response labelling of supervisory comments.

Across the ECB and ESRB I worked on statistical modelling of financial stability, including credit cycles, bank capital, portfolio risks, real estate and other systemic vulnerabilities.

In MATLAB I conducted time-series analysis and forecasting and developed large VAR models to identify drivers of non-performing loans, including macroeconomic and house-price dynamics.

In R/RStudio I worked with time series and panel data, built infrastructure for loading, processing and data-quality analysis, and mined characteristics of banking groups and loan portfolios. I also used Stata for advanced regressions and panel-data processing.

During my first DG ECFIN assignment I developed econometric models for macroeconomic-imbalance analysis; professional records explicitly identify EViews and VAR modelling.

A major topic was Sweden's housing market and household indebtedness. The Commission's public Country Report used an econometric shock-simulation framework to quantify housing-price and interest-rate scenarios, including effects on GDP, consumption and residential investment.

Public documents confirm my work on econometric modelling in this policy area but do not attribute each individual specification to specific team members, so the description remains deliberately conservative.

At DG ECFIN I prepared the European Commission's official macroeconomic and fiscal forecasts. Country forecast datasets contained roughly 159 variables covering GDP and its components, inflation, employment and unemployment, wages, external balances and public finances.

The quantitative workflow included national accounts and high-frequency indicators, data-quality and consistency checks, revisions and forecast-error analysis, alternative scenarios and maintenance of forecasting tools. Forecasts had to be internally consistent within each economy and with the common assumptions of the EU-wide exercise.

I am explicitly named as a contributor in ten published European Economic Forecasts from Summer 2022 to Spring 2025.

While working on Slovakia's Recovery and Resilience Plan I addressed a quantitative problem that could not be handled satisfactorily with a standard headline price index. Rapid inflation and construction-cost increases required an objective way to reassess investment targets.

I developed several variants of a quantitative methodology combining comparable statistical data with cost indicators relevant to particular construction and infrastructure investments.

The project is an example of statistical modelling used directly for policy design and implementation rather than only for forecasting or academic research.

I also used Python for machine-learning processing. A documented use case involved automated processing and response labelling of supervisory comments using scikit-learn.

This combined textual and structured information with classification workflows and complemented my broader use of Python for data analysis, visualisation and large financial datasets.

At the ESRB Secretariat I worked on the Systemic Risk Survey, which collected assessments of financial and systemic risks from participating institutions. I contributed to the survey's data infrastructure, compilation and visualisation as well as econometric analysis.

I programmed a new interface for more automated collection of responses, replacing a process that had relied more heavily on individual Excel submissions. The important outcome was a more robust statistical infrastructure for a recurring European risk survey.

I used MATLAB for the econometric side of ESRB work; the surviving material does not uniquely identify the programming language used to build the collection interface.

My Master's thesis at the University of St. Gallen focused on forecasting GDP across European economies using advanced time-series and machine-learning approaches.

I compared Bayesian Vector Autoregression with neural networks using Bayesian regularisation. The work covered model specification, parameter optimisation, variable selection and systematic comparison of forecast accuracy.

The project formed part of the methodological foundation for my later forecasting, econometric and large-data work in European institutions.

Alongside specialised statistical languages I have also programmed tools for day-to-day analytical work. In Excel and VBA I built advanced functions, add-ins, data tools and automated updates for chart packages combining external data sources.

I also created add-ins for automatic chart formatting and recurring analytical requests. In Microsoft Access I built databases of financial and economic data.

Automation was a by-product rather than the objective: the aim was to make recurring analytical work faster, more consistent and less prone to manual error.