Portfolio Risk Dashboard (Python → Power BI Web)

A reproducible, end-to-end mini-project that builds a portfolio risk dataset in Python (prices → positions → portfolio returns → risk metrics → stress scenarios) and publishes it as a Power BI (web) dashboard using an Excel upload (OneDrive/SharePoint).

This repo is designed to demonstrate “employer-relevant” skills:

• clean Python pipeline + file structure
• metrics common in risk analysis (drawdown, rolling vol, beta, VaR/CVaR)
• Power BI modeling + visuals (slicers, relationships, KPI cards, line + bar charts)
• GitHub-ready documentation + artifacts

---

What’s inside

Outputs (generated by the pipeline)

The pipeline produces these tables (CSV) used by Power BI:

• portfolio_daily.csv — daily portfolio value + portfolio return
• risk_daily.csv — drawdown, rolling vol (30/60/252), beta, VaR/CVaR
• risk_contrib.csv — (example) risk contribution shares by ticker
• stress_daily.csv — scenario paths (baseline + historical stress windows)
• dim_date.csv — date dimension for slicers
• dim_asset.csv — asset dimension (ticker, asset class, sector)

Power BI is loaded from an Excel workbook generated by Python:

• data/processed/powerbi_upload.xlsx

---

Dashboard (Power BI Web)

Power BI files created in the web service cannot always be committed as .pbix (org settings often block downloads). Instead, this repo stores dashboard artifacts (screenshots/PDF export) in:

• docs/powerbi/

Recommended report pages:

• Overview: Portfolio Value, Drawdown, Daily Return & 30D Vol, KPI cards, Risk Contribution (Top 10)
• Stress Tests: Baseline vs scenarios + scenario-only view

---

Quickstart

1) Create a virtual environment and install dependencies

python -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt

2) Configure the universe (tickers)

Edit: data/sample/universe.csv Example format:

ticker,asset_class,sector SPY.US,Equity,US Equity TLT.US,Bonds,US Treasury GLD.US,Commodity,Gold

3) Run the pipeline (CSV outputs for Power BI)

python src/01_fetch_prices.py --start 2018-01-01 --end 2025-12-26

python src/02_make_positions.py

python src/03_build_portfolio.py

python src/04_compute_risk.py

python src/05_stress_tests.py

Expected outputs (in data/processed/):

prices.parquet

positions.csv

portfolio_daily.csv

returns_asset.csv

risk_daily.csv

risk_contrib.csv

stress_daily.csv

dim_date.csv

dim_asset.csv

4) Build the Power BI upload workbook (ensures real Excel date types)

python src/07_export_powerbi_excel.py

This creates: data/processed/powerbi_upload.xlsx

Power BI Web (exact import path)

Upload workbook to OneDrive / SharePoint

Upload: data/processed/powerbi_upload.xlsx

Import in Power BI (web) Power BI → My workspace

Import → OneDrive/SharePoint

Select powerbi_upload.xlsx

Select the 6 tables/sheets and Transform data

Confirm:

first row is headers

date columns are type Date

numeric columns are type Decimal number

Close & Apply

Relationships to create (Model view)

Create these relationships (1:* , single direction, active):

dim_date[date] → portfolio_daily[date]

dim_date[date] → risk_daily[date]

dim_date[date] → risk_contrib[date]

dim_date[date] → stress_daily[date]

dim_asset[ticker] → risk_contrib[ticker]

Visuals (recommended)

Overview page

Slicer: dim_date[date] (Between)

Line: portfolio_daily[portfolio_value] by dim_date[date]

Line/Area: risk_daily[drawdown] by dim_date[date]

Combo: daily return vs 30D vol (portfolio_return columns, vol_30 line)

Cards: latest value / max drawdown / vol / beta / VaR

Bar: risk contribution (Top 10) using risk_contrib[rc_share] by ticker

Stress Tests page

Baseline chart (scenario = Baseline)

Scenario chart (exclude Baseline) with scenario slicer

Notes on the metrics

Rolling vol (30/60/252) is null for early dates because the window needs enough history.

VaR/CVaR uses a rolling historical window (e.g., 252 trading days).

Beta is computed relative to a benchmark series (defined in the risk script).

Troubleshooting

Power BI slicer doesn’t filter charts Ensure charts use dim_date[date] on X-axis (not Year-only fields).

Verify relationships are active and date columns are type Date.

Check “Edit interactions” to ensure slicer filters each visual.

Excel dates imported as text Always create the workbook via:

python src/07_export_powerbi_excel.py

This forces real Excel date types.

Figures (Power BI)

Author

Arina Veprikova