Can you see a recession coming in the news?
bachelor thesis · University of Cologne · Bayesian + frequentist · NLP
Recessions are declared long after they start — official committees take months to
confirm what markets and headlines already suspect. My bachelor thesis asked whether
you can nowcast US recession risk in real time by combining three kinds of
signal: classic macroeconomic indicators, financial-market data, and text-based
predictors extracted from news.
Approach
The same question, answered from two statistical worldviews: a frequentist
specification and a Bayesian one, compared head-to-head on out-of-sample performance.
That contrast — what each framework buys you, where they disagree — became the heart
of the thesis.
- Macro, financial and news-text predictors in one nowcasting framework
- Bayesian and frequentist model variants, evaluated out of sample
- Graded 1.7, supervised by Prof. Dr. R. Liesenfeld
Results
[headline finding — which signals mattered, how early the model
flagged risk, one number if possible]
Read it
The full write-up is available on request —
email me.
A public version is in the works.
What is a footballer actually worth?
end-to-end ML · web scraping · XGBoost / CatBoost
Transfer fees look irrational from the outside — €80m for one striker, €8m for
another with similar stats. This project asked how much of that price a model can
actually explain from performance data alone.
Approach
- Scraped player performance data (fbref) and transfer histories (transfermarkt)
- Cleaned and joined genuinely messy real-world data — the honest 80% of the work
- Gradient-boosted regression (XGBoost, CatBoost) with randomized hyperparameter search
- Feature-importance analysis: which stats move the market, which don't
- Full written report alongside the code
Results
[headline metric — e.g. test-set error, R², and one surprising
feature-importance finding]
Code
github.com/paulniggenaber/projects →
Predicting-Football-Transfer-Fees
Teaching a network to imagine digits
variational autoencoder · TensorFlow · graded A
A variational autoencoder that learns to generate handwritten digits —
both black-and-white and color variants — by learning a compressed probabilistic
representation of what "a digit" is.
Why it matters to me
VAEs sit exactly where my interests overlap: they're deep learning, but the core is
variational inference — KL divergence, the reparameterization trick, a proper
probabilistic model with priors. Bayesian thinking wearing a neural network.
Engineering
Deliberately built as a real piece of software, not a notebook: separate modules for
data loading, losses, network architectures and training, wired together behind a CLI
with an argument parser. Run it, configure it, extend it.
Results
Graded A. [add 1–2 sample-output images or a one-line result]
Code
github.com/paulniggenaber/projects →
Image Generation
What do patients say between the lines?
NLP · sentiment analysis · in progress
Patient reviews of medications carry signal that star ratings flatten out: side
effects mentioned in passing, hedged satisfaction, strong words about mild issues.
This project runs sentiment analysis over a large corpus of drug reviews (drugs.com
dataset) and connects the text signal to reported outcomes.
Status
Honest label: in progress. Preprocessing and sentiment augmentation are done; the
analysis and write-up are being finished.
[update when polished — planned finding / angle]
Code
github.com/paulniggenaber/projects →
Drug_Review
Product intern @ Two
2026 — now · B2B fintech (buy-now-pay-later for businesses) · Oslo
Two builds payment infrastructure that lets businesses buy from each other on
flexible terms — B2B buy-now-pay-later. I'm interning in the product department,
which means sitting where data, engineering and commercial decisions meet.
What I'm doing
[2–3 bullets on what you actually work on — projects, analyses,
tools. Keep it concrete but respect confidentiality.]
What it's teaching me
How data decisions get made when real money moves: what "good enough" evidence looks
like under deadline, how product metrics get defined and argued about, and how much of
data science in a company is really about asking the right question.
Working student, R&D data analysis @ Kuraray
Jul 2024 — Jul 2025 · Kuraray Europe · advanced interlayer solutions R&D · Germany
A year inside an industrial R&D department, doing hands-on data work on real
measurement data — and automating away the department's most tedious workflows.
Final rating: "consistently exceeded expectations".
What I built
- Python tooling (pandas, NumPy, Seaborn, Matplotlib) to analyze R&D measurement data
- An Excel/VBA automation for hazardous-material labeling with precise data filtering —
highlighted by my employer as a standout achievement
- Two Python automation tools (a file splitter and a multi-file worksheet appender)
for the department's digitalization initiative
- Interfaces between IT systems, plus documentation and colleague training
- Physical measurement data used to train AI models; photometric calculation methods
implemented in Excel and Python
Reference
A written recommendation and formal work reference are available on request.
MSc Data Science @ BI Norwegian Business School
Aug 2025 — Jun 2027 (expected) · Oslo · average grade: A
A data science master's built on a business school foundation — the point, for me,
is learning ML and AI in a way that stays connected to decisions someone actually
has to make.
Coursework so far (all A)
- Advanced Statistics
- (Big) Data Curation
- Object-Oriented Programming
- Microeconomics
- Accounting, Valuation & Financial Economics
Focus
Deep understanding and application of machine learning and AI on a strong
business-domain foundation — with my personal thread of Bayesian and causal methods
running through everything.
Exchange semester @ Bocconi, Milano
fall 2026 · Università Bocconi · Milano, Italy
My third master's semester happens at Bocconi — one of Europe's strongest schools
for quantitative finance and economics, in a city that's very good at reminding you
there's life outside the library.
Plan
[courses / focus — fill in once course selection is confirmed]
Goals beyond the classroom: [e.g. basic Italian, one trip a month,
a Milan running route worth bragging about]
BSc Economics @ University of Cologne
Sep 2022 — Jul 2025 · grade 1.8 ("good") · 180 ECTS
Economics gave me the question-asking habits; I used the elective space to build a
de-facto quantitative-methods specialization on top.
Chosen focus
- Statistics & Econometrics
- Empirical Methods & Data Analysis
- Analysis of Multivariate Data
- Market Design: Auctions & Matching
Thesis
Nowcasting Recession Risk: A Bayesian and Frequentist Approach Using
Macroeconomic, Financial and Text-based Predictors — graded 1.7.
Full project page →
How I actually use these
stack & methods, honestly labeled
Skill lists are cheap, so here's the honest version: what I reach for daily, what
I've shipped with, and what I'm actively leveling up.
Reach for daily
Python (pandas, NumPy) for everything data; SQL for anything that lives in a
database; Matplotlib/Seaborn when I need to see it; LaTeX when it needs to look
serious.
Shipped real things with
scikit-learn and XGBoost/CatBoost (transfer-fee project), TensorFlow (VAE),
web scraping (fbref/transfermarkt pipelines), Excel/VBA automation (Kuraray,
where it was genuinely the right tool).
Actively leveling up
[e.g. causal inference project in progress, SQL window-function
reps, MLOps basics]
The differentiator
The econometrics/Bayesian background: I don't just fit models, I care about
identification, uncertainty and whether the effect is real. That's the lens I bring
that a pure CS path usually doesn't.
The longer story
Kerpen → Köln → Oslo → Milano → ?
I grew up in Kerpen, just outside Cologne. I started out studying civil engineering
during the COVID lockdowns, realized it wasn't my path, and switched to economics —
the best decision of my degree-choosing career, because it's where I met
econometrics and discovered that statistics is actually about arguments, not
formulas.
During the bachelor I worked the whole way through — landscaping, arena logistics,
fitness coaching, and finally a year of real data work in industrial R&D at
Kuraray. Then Oslo: an MSc in data science at BI, straight A's so far, and an
internship in fintech on the side.
[add a personal paragraph — what drives you, what kind of team
you want, what you're like to work with]
Elsewhere
GitHub ·
[LinkedIn URL] ·
email
Ten years of youth football coaching
2012 — 2022 · ERFA 09 Gymnich & Deutz 05 · DFB C + B licenses
I started coaching young and kept at it for a decade, eventually earning the DFB C
and B coaching licenses. Youth football is the best communication training there is:
you learn to explain complicated things simply, to people who'd rather be doing
something else, and to adjust the message per kid.
What transferred
- Explaining > knowing — an insight nobody understands doesn't count
- Reading a room and adjusting on the fly
- Patience with the long game: development over instant results
[a favorite coaching story or proudest moment]
Ultra running
long distances · questionable decisions
Running long is my favorite way to think — and Oslo's forests are absurdly good
for it.
The numbers
[longest run / race results / weekly volume — whatever you're
happy to share]
Currently training for
[race, distance, date, goal]
Fitness instructor
Feb — Aug 2022 · XtraFit, Cologne · trainer B + A licenses, personal trainer cert
Before the data career got serious I coached people in the gym — licensed fitness
trainer (B and A) with a personal-trainer certification, working the floor and the
occasional 6 a.m. shift at XtraFit Cologne.
Coaching adults one-on-one is a different skill from coaching kids in a team: more
listening, more habit design, less shouting across a pitch.
Guitar
enthusiasm currently outpacing skill
The instrument in the corner that's slowly winning. Progress is logged honestly:
slow, occasionally painful for the neighbors, deeply satisfying.
Currently learning
[song + how it's going]
The goal
[e.g. one song performed in public before 2027]
Tennis
the other racket sport
[a paragraph — how long you've played, style, favorite player,
whether you accept challenges from colleagues]
Event logistics, LANXESS Arena
Sep 2022 — Jan 2023 · A&B Events · Cologne
Service and logistics crew at Cologne's biggest arena — 15,000-person events where
the plan meets reality at 6 p.m. sharp and improvisation is a load-bearing skill.
Great early lesson in operations: most problems are coordination problems.
Football camp counselor
2020 — 2023 · Ferienfussball · Germany · volunteering
Summer football camps for kids: part coach, part entertainer, part lost-shoe
detective. Volunteering that doubled as an annual reminder of why I coached for ten
years in the first place.
Now
what I'm doing this month · updated [date]
A now page — the honest snapshot version
of a bio.
- Interning in the product department at Two (fintech, Oslo)
- Building out my public portfolio — polishing the repos behind the projects above
- Norwegian: grinding toward B1 [current status]
- Training: [current block / race]
- Reading: [book]
- Preparing for the Bocconi exchange semester this fall
How I work
principles & habits — draft, honestly still forming
I'm early-career, so this list is short and honest rather than long and borrowed.
- Uncertainty is information. An estimate without error bars is a
vibe. I'd rather say "somewhere between X and Y" than pretend precision.
- The question first. Most analysis that fails, fails at the
question, not the model.
- Explaining is part of the job. Ten years of coaching: if the
audience didn't get it, the work isn't done.
- Automate the boring thing. Kuraray habit — if you do it three
times, script it.
- [add 1–2 of your own — feedback preferences, collaboration
style]
Uses
tools, gear, setup
Software
- Python + pandas/NumPy — the workshop
- [editor — VS Code? Jupyter? something else?]
- LaTeX for anything that needs to look serious
- [other daily tools — notes app, terminal, browser]
Hardware
- [laptop / desk setup]
- [running watch / gear if you care]
This site
One hand-written HTML file. No framework, no build step, no trackers, hosted on
Cloudflare Pages. The whole site is smaller than most sites' cookie banners.
Bookshelf
reading list & favorites
Currently reading
[book + one line on why]
Formative
- [book that shaped how you think about stats/economics]
- [a non-work favorite]
- [one more]
On the list
[2–3 titles queued up]
The 2027 plan
where this is all going
I graduate in June 2027 and start my first full-time role right after. The plan,
openly:
- 2026: finish the core semesters strong, exchange at Bocconi,
ship a polished public portfolio (including a causal-inference project — my
favorite corner of the field)
- Early 2027: thesis + interviews
- Mid 2027: graduate data scientist / analyst role — Oslo first
choice, also open to Köln, Hamburg, München, Berlin, Prague, Vienna
What I'm optimizing for
A team where statistical thinking is valued, not just model-shipping — and where a
junior can learn from people who've made real decisions with data.
[refine — industry preferences, team size, values]
FAQ
the questions recruiters actually ask
When can you start?
Full-time from mid-2027 (graduating June 2027). Internships / working-student
arrangements before that: possibly — ask.
Where?
Oslo is home base and first choice. Also genuinely open to Köln, Hamburg, München,
Berlin, Prague and Vienna. EU citizen (German), so no visa needed anywhere in the
EU/EEA.
Languages?
German native, English C2 (IELTS), Norwegian around B1 and improving on purpose.
Remote?
[your stance — hybrid preference? office-first for the first
years?]
Salary expectations?
Let's talk when we both know it's a fit.
References
what others say
From my year at Kuraray Europe (R&D data analysis): a formal work reference and
a personal letter of recommendation, both rating the work as consistently exceeding
expectations. Available on request —
email me.
[optionally: one quoted line from the recommendation letter,
with permission]
Ask me about…
conversation starters that skip the small talk
- Why Bayesian methods deserve more space in business data science
- What youth football coaching teaches you about stakeholder management
- Scraping transfermarkt without losing your mind
- Moving to Norway: expectations vs. reality
- Whether a recession is visible in newspaper text before it's visible in GDP
- [add your own favorites]
Get in touch
all channels, ranked by response speed
Based in Oslo, Norway. Happy to meet for a coffee if you're local —
[favorite café] is a good default.
