Anelia D. Yordanova

Finance student at NTNU. I build AI tools for the problems finance teams actually have, test them properly, and write about what's moving the markets.

See my work Read the blog
My work as an efficient frontierRisk and return, loosely interpreted

Selected work

AI applied to finance documents, asset pricing with real market data, and what happens when a company actually adopts AI.

Applied AI

Can an AI find the right number in an annual report?

Retrieval-augmented generation over financial documents. I test how chunk size and splitting strategy change retrieval quality across filings, tables and earnings text, using FinanceBench, FinQA and TAT-QA.

A document split into chunks; highlighted chunks are the ones retrieved for a question.

  • Python
  • RAG
  • Embeddings
  • Evaluation
 Asset pricing

Does carbon intensity explain European stock returns beyond beta?

Cross-sectional tests on European equities, 2015–2025, with ESG and emissions data.

  • Python
  • SQL
  • WRDS
  • Refinitiv
 AI in practice

What drives AI adoption at an accounting firm?

Seven interviews across five departments at SpareBank 1 SMN Regnskapshuset, read through UTAUT and TOE.

  • Qualitative research
  • Strategy
 Founder

Kretz: finance and operations for sports clubs

My startup. Two pilot clubs live, with accounting integrations to Tripletex, Fiken and Uni Micro underway.

  • Product
  • Frontend
  • Integrations

Market notes

Short takes on financial news: what happened, why it matters, and what I'd watch next.

All posts

    Finance lab

    Small tools I've built in plain JavaScript to make textbook models tangible. Move the sliders: this is the CAPM, live.

    Open the lab
    Required return 10.0%

    Tools I work with

    Python for analysis and AI, SQL for getting data out of databases like WRDS, JavaScript for anything people need to click on, and Excel and Power BI for models and reporting.

    Python

    pandas, statsmodels, scikit-learn and LLM tooling for retrieval and evaluation.

    # Share of evidence found in top k
def recall_at_k(retrieved, gold, k=5):
    top = set(retrieved[:k])
    return len(top & gold) / len(gold)

    SQL

    Joins, window functions and monthly panels from return and fundamentals tables.

    SELECT permno, month,
  AVG(ret) OVER (
    PARTITION BY permno
    ORDER BY month
    ROWS 11 PRECEDING
  ) AS avg_ret_12m
FROM monthly_returns;

    JavaScript

    Interactive calculators and this site, written without frameworks.

    // Present value of cash flows
const pv = (flows, r) =>
  flows.reduce((sum, cf, t) =>
    sum + cf / (1 + r) ** (t + 1), 0);

    Excel

    Financial models, sensitivity tables and lookups, built so someone else can follow and audit them.

    Enterprise value, 5-year DCF:
=NPV(WACC, FCF_Y1:FCF_Y5)
  + TerminalValue / (1 + WACC)^5

    Power BI

    Dashboards and DAX measures that turn raw data into reporting a board can read.

    Fee collection rate =
DIVIDE(
    SUM( Payments[Amount] ),
    SUM( Invoices[Amount] )
)

    About

    I study finance at NTNU in Trondheim and spend most of my time where finance meets AI.

    I have ten years in sports behind me: as a player, as a coach and as deputy leader of a club board, responsible for its finances. That last role is where Kretz started. I like problems with real data and a real person on the other end of the answer.

    Get in touch