License

MIT © 2025 Maria Westrin

Portfolio Risk Report — Transaction-to-Report Pipeline (Python)

A reproducible, bank-style analytics pipeline that goes from raw transaction rows (BUY/SELL/DIVIDEND) → clean panel data → risk metrics + exposures → shareable HTML report.

Why this project (business value)

This project demonstrates an end-to-end workflow a data analyst/quant analyst would deliver:

• Data quality checks & validation (sanity checks, duplicates, missingness)
• Position building from transactions (signed volume/value, cumulative holdings)
• Daily portfolio panel construction (price/position/market value/exposure)
• Risk reporting (VaR, volatility, drawdown) + segment analyses
• Professional artifacts: HTML report + figures + CSV tables

Key outputs (what recruiters should open first)

• HTML report: report/portfolio_risk_report.html
• Figures: report/figures/*.png
• Tables: report/*.csv

Tip: add 1–2 screenshots from the HTML report into assets/ and show them here.

What the notebook does (high level)

1) Data ingestion & schema checks

• Reads financial_raw_data.csv
• Type casting for date and timestamp
• Overview table of dtypes, missingness, and basic stats

2) Data quality & consistency validation

• Duplicate checks (transaction_id)
• Consistency check: price_per_share * volume ≈ transaction_value
• Flags relative error and inspects worst rows

3) Build positions from transactions

• Signed volume/value (BUY = +, SELL = −)
• Instrument key primarily from ticker_symbol (fallbacks exist)
• Cumulative position_units per instrument over time

4) Build daily panel

• Daily snapshot per instrument
• Daily price series from last trade per day + forward-fill
• Returns computed from price (and optional comparison vs daily_return if present)

5) Portfolio return (robust definition)

• Computes portfolio return via P&L / Gross Exposure: [ r_t = \frac{\sum_i pos_{t-1,i}(P_{t,i}-P_{t-1,i})}{\sum_i |pos_{t-1,i}P_{t-1,i}|} ]
• Stable even if net exposure is near zero (gross-based denominator)

6) Risk reporting

• Historical VaR 95% (daily)
• Daily + annualized volatility
• Max drawdown on equity curve
• Exposures by sector and risk rating
• Optional signal check: analyst_rating(t) → next-day return(t+1) and robust regression outputs (HC3)

Data requirements

This repo does not need to publish your private dataset. The notebook expects a CSV with transaction rows.

Required columns (minimum)

• transaction_id
• date, timestamp
• ticker_symbol (or equivalent identifier)
• transaction_type (BUY/SELL/DIVIDEND)
• price_per_share, volume, transaction_value
• currency

Optional (used for richer analyses/plots/tables)

• company_name, sector
• risk_rating, liquidity_score, spread_pct
• analyst_rating
• daily_return (if present, used for QC comparison)

Reproducibility (how to run)

Option A — Notebook (recommended)

1. Open Financial_Analysis.ipynb (or financial_analysis_maria_westrin.ipynb)
2. Restart Kernel → Run All
3. Open report/portfolio_risk_report.html

Option B — From terminal (WSL/macOS/Linux)

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
jupyter notebook