Data science in finance

Data science is the work of answering questions with data: collecting it, cleaning it, testing a statistical or machine learning model on it and explaining the result to the people who decide. In finance the data are loan files, card and account transactions, securities prices, customer records and the reports a bank sends to its supervisor. AI in finance runs on the same data, and many AI projects in a bank start as a data science project.

Common data science use cases in finance are credit scoring, fraud alerts on card payments, screening for money laundering, cash and liquidity forecasts, pricing and customer churn. In asset management, data scientists build signals from market data and from alternative data such as news text. Data analytics in banking is the older, descriptive half of the job: reports and dashboards that show what happened. Data science adds models that estimate what happens next, for example the probability that a borrower defaults, which is the core number of credit risk.

In a bank, data quality takes a large share of a data scientist's time, because supervisors check it. The Basel Committee published its Principles for effective risk data aggregation and risk reporting, known as BCBS 239, after many banks in the financial crisis could not add up their risk exposures fast enough. The ECB followed with its Guide on effective risk data aggregation and risk reporting, which states what it expects from the banks it supervises on top of BCBS 239.

Central banks also collect loan-level data. AnaCredit, the ECB's analytical credit dataset, holds details on single loans of more than €25,000 to companies and other legal entities in the euro area, with 94 data attributes and 7 identifiers per loan. Loans to private households are not in it.

For models that make or prepare decisions, BaFin published principles for the use of algorithms in decision-making processes. BaFin splits a model's life into development, where it is selected, calibrated and validated, and application, where people interpret its output, and it asks for a clear owner of each model. The EU AI Act adds duties for high-risk systems, among them credit scoring of consumers.

Frankfurt is the seat of the European Central Bank, the Deutsche Bundesbank and Deutsche Börse Group, and Deutsche Bank and Commerzbank have their head offices there. The Bundesbank's research data center, the RDSC, gives researchers access to anonymized microdata on banks, companies, securities, investment funds and households, for independent, non-commercial projects and under strict conditions.

For a degree, Frankfurt School of Finance & Management runs a Master of Artificial Intelligence & Data Science: four semesters and 120 ECTS credits, with classes on two weekdays and Saturday so students can work alongside. Company projects run with firms such as Deutsche Börse Group and PwC, and the school reports that 30 percent of a recent class started in banking. Practitioners meet at Frankfurt Data Science, a Frankfurt group for applied data science and AI whose meetups mix short talks with campfire sessions, several of them at TechQuartier.

The programs of recent AI events for the financial industry show where data teams spend their time. At fAInance, the half-day conference of Sopra Steria and Fraunhofer IAIS for staff of financial institutions, the demo areas covered AI in software testing, in governance, risk and compliance, and against financial crime. The AI & Financial Market Data meetup in Paris brought quantitative developers from Deutsche Bank together with people from FactSet, Amundi and QuestDB to compare how they store and serve market data for models. In Hamburg, KI Exchange put sovereign AI infrastructure, AI governance and DORA compliance on one program.

The last point reaches every data platform. A bank that runs its data and models in the cloud falls under the Digital Operational Resilience Act, which requires a register of all ICT service contracts. The choice of a data vendor is therefore also a question for the risk and outsourcing teams.

Finance Loop is the meeting place for people who build data and AI models at banks, asset managers, insurers and fintechs. It connects the finance, IT and AI communities, with events in Frankfurt and also in Munich, Berlin and Hamburg. Finance Loop has a cooperation with Frankfurt Data Science for joint events and knowledge exchange, with AI governance and applied use cases as main topics. Together with Finteda and Frankfurt Data Science it supports Claude Hacker House, a hands-on series on building with Claude for business and finance.

Finance Loop was a partner of AI in Finance Paris, organized by Finteda, and supported the AI & Financial Market Data meetup in Paris. It organizes the Frankfurt Quantum Finance Forum with Frankfurt School, Deutsche Bundesbank, IBM and Finteda. All dates are on the events page.

They build and maintain the models behind credit decisions, fraud alerts, forecasts and prices. Much of the work is finding and cleaning data, testing a model against past cases and writing down how it works, because a model validation team and later the supervisor read that documentation. The rest is explaining results to the people in risk, sales or treasury who act on them.

In Frankfurt, data science finance jobs are at the banks, the Bundesbank, the ECB, Deutsche Börse and the consultancies that work for them. Munich has the head offices of Allianz and Munich Re, where data scientists work on insurance pricing and claims. Of the Frankfurt School data science graduates, 30 percent went into banking and 14 percent into consulting, most of them in Germany.

In Germany the topic appears mainly at AI conferences for the financial industry, such as fAInance in Frankfurt, which is held in German and open only to people who work at financial institutions, and KI Exchange in Hamburg. Smaller meetups of Frankfurt Data Science and Finance Loop appear in the calendar above.

For a full degree next to a job, the Frankfurt School master with classes on two weekdays and Saturday is made for that case. For a first step, a hands-on evening such as Claude Hacker House or a meetup shows what the work looks like, and the Finance Loop knowledge hub explains the finance side, from market data to credit risk, in plain words.