
Quantitative investment research traditionally starts from structured data: prices, yields, positions, fundamentals, risk factors and portfolio constraints. Yet much of the information that shapes financial markets is not naturally tabular. News, company filings, central-bank communication, research notes, event streams and alternative data arrive as text or other non-parametric inputs. At the same time, investors operate across many different and overlapping regimes, from long-term macroeconomic and monetary cycles to political, geopolitical and shorter-term market developments. Alternative sources of data may contain information about these regimes that is not captured directly by conventional market variables, but their unstructured and heterogeneous nature makes them difficult to use systematically.
Large language models (LLMs) create a new research opportunity. Rather than treating an LLM as a stand-alone forecaster, it can be used as a research tool that interprets different sources of information, extracts structured variables and helps identify market regimes. Its contribution can then be tested using the same principles applied to any investment signal: economic intuition, reproducibility, realistic timing, out-of-sample performance and comparison with transparent baselines. This assignment is about determining whether LLM-derived information is stable, timely and useful enough to improve a clearly defined investment decision.
For institutional investors, a large amount of potentially relevant information about changing market regimes is available only in unstructured form. Political developments, central-bank communication, macroeconomic events, company information, news and alternative data may all provide evidence about the regime in which markets operate. An LLM can be used to interpret these heterogeneous sources and translate them into structured signals describing dimensions such as growth, inflation, monetary policy, political risk or market sentiment. These LLM-derived signals can then be combined into a market-regime representation that can be used as an input to a financial model, investment strategy or portfolio decision.
The central task of the assignment is therefore: use an LLM to interpret/ convolute heterogeneous macroeconomic and alternative information, translate it into structured signals, and combine those signals into a market-regime classification. A simple financial model should then be used to test whether the resulting regimes contain meaningful explanatory or predictive information. The financial model is the validation step within the regime-classification research, rather than a separate research direction.
For example, the student could classify regimes using dimensions such as growth, inflation, monetary policy, political risk or market sentiment, and then test whether these regimes improve a simple return, volatility or risk model. A three- to five-year setting may be used to investigate whether macroeconomic and alternative information helps identify developments that are not fully captured by either slow-moving economic-cycle indicators or short-lived news and sentiment signals. These examples guide the validation of the regime classification without changing the core research task.
The methodology for deriving the market regimes is open. The LLM should first translate heterogeneous information into structured signals. Reinforcement learning could then be used to learn how these signals should be combined, weighted and updated to form the final regime classification. Alternatively, the student may investigate another suitable classification method. In all cases, the core requirement is the same: raw macroeconomic and alternative information is transformed by an LLM into structured signals, which are converted into market regimes and evaluated through a simple financial model.
The broad theme of this assignment is therefore:
To what extent can LLMs transform heterogeneous macroeconomic and alternative data into meaningful market-regime classifications for a selected asset class?
The specific asset class will be selected at the start of the internship, based on the relevance and availability of macroeconomic, alternative and market data. A strong thesis will define the information set, the structured signals produced by the LLM, the regime-classification method and the simple financial model used for validation. Where a relevant internal regime model is available for the selected asset class, it may provide an additional benchmark. Otherwise, the student will define an appropriate conventional baseline for evaluating the LLM-derived regimes. The main research contribution should remain the construction and assessment of the market-regime classification. Particular attention should be paid to look-ahead bias, data leakage, prompt and model instability, model drift, out-of-sample testing and reproducibility. Experts are also available for consultation in the entire process.
The specific asset class and final dataset will be selected at the start of the internship, based on research relevance, data availability and the suitability of a simple financial model for validating the regime classification. Candidate inputs include time-stamped news, company filings, central-bank publications, macro-economic event streams, alternative data and corresponding market prices or portfolio variables. Bloomberg terminal and API access may be used where appropriate and available. The student should document the data, prompts, model settings, feature definitions, timestamps and evaluation splits sufficiently clearly for the research to be reproduced and reviewed.
This assignment is offered by PGGM as part of its Academic Excellence Program (AXP). The student will work alongside PGGM investment professionals and gain experience with institutional investing and quantitative research. During the program, the student will have opportunities to share ideas, attend presentations from investment teams and learn more about working at PGGM.
We are looking for a quantitative Master’s student with a STEM, econometrics, quantitative finance, computer science or related background who is interested in combining thesis research with an internship. The student should be comfortable with Python, empirical research and financial-market data. Experience with machine learning, natural-language processing, optimization or reinforcement learning is useful, but the ability to design a careful experiment is more important than prior experience with every method. The student is expected to translate the open research theme into a focused thesis question and to communicate both positive and negative findings clearly.
At PGGM we are proud of our core values: trust, teamwork and innovation. These values arose from the best stories of our colleagues, moments when they were at their best. We believe those moments occur when you can be 100% yourself. At PGGM, we value the unique combination of your personal and professional attributes. Because what really counts at PGGM are people, one hundred percent! At PGGM, you can count on:
Is this your new role at PGGM?
Would you like to join us and share the responsibility of ensuring a good pension for more than 5.6 million pension participants? Then apply for this vacancy.
Do you have any questions? Please contact recruiter Meryem el Ajjouri via meryem.el.ajjouri@pggm.nl.
More information about the application process can be found on our website: Our application procedure | PGGM.
At PGGM, we have created a hybrid working environment that enables us to combine flexibility with collaboration. Our employees have the opportunity to work both in the office and remotely, depending on their personal preferences and the nature of their work. By offering hybrid working, we aim to give our employees the freedom to shape their own working style, while at the same time retaining the benefits of a shared workplace.
Upon receiving your application, we will follow a number of steps to see if we are a good match for each other. During the recruitment and selection process for new employees, PGGM collects and processes personal data, which may also include reviewing social media profiles. PGGM believes it is highly important that this is done carefully, confidentially, transparently, and in accordance with the General Data Protection Regulation (GDPR) and related laws and regulations. Please read this document to understand what happens to your personal data.
Note to Agencies and Intermediaries
We are aware that many external parties would like to assist in our search or offer their services for this vacancy; however, at PGGM, we recruit our new colleagues directly. Acquisition regarding recruitment and selection assignments or unsolicited CV submissions are therefore not appreciated and will not be processed. Should we require external support, we will contact you directly.
We always communicate via official PGGM email addresses (@pggm.nl). Have you been approached by a third party and/or do you doubt whether the contact is genuinely on behalf of PGGM? Please feel free to verify if there is indeed a legitimate partnership with third parties by sending us an email or giving us a call.
We are committed to an affordable, high-quality, and sustainable pension for pension funds—our clients—and their participants. In addition, we contribute to a liveable world, healthy working environments, and vital aging.
We are a non-profit cooperative pension administration organization. We offer our clients pension administration, asset management, and policy advisory services. PGGM manages the pensions of various pension funds, their affiliated employers, and their employees.