Knuckles-Team Volatility Forecasting Project Workflow
Project 3 of the quant portfolio. Forecast daily volatility with GARCH and grade
it against realized vol. See quant-career-docs/reference/ml-for-finance.md and
the Risk Analyst persona.
Steps
Step 1: fetch-price-data
Agent: quant_developer
Tools: data-science-mcp
Pull historical prices via yfinance and compute daily returns.
Expected: daily-returns
Step 2: fit-garch [depends_on: Step 1]
Agent: risk_analyst
Fit a GARCH(1,1) (or variant) model to the return series using the arch library.
Capture conditional-volatility estimates.
Expected: garch-volatility-forecast
Step 3: forecast-vs-realized [depends_on: Step 2]
Agent: risk_analyst
Produce out-of-sample volatility forecasts and align them against realized
volatility (e.g. rolling standard deviation of returns).
Expected: forecast-realized-pairs
Step 4: evaluate [depends_on: Step 3]
Agent: risk_analyst
Evaluate forecast accuracy (e.g. RMSE, QLIKE). State where the model under- or
over-predicts — honest analysis.
Expected: forecast-evaluation
Step 5: github-publish [depends_on: Step 4]
Agent: quant_developer
Publish a GitHub repo with README, forecast-vs-realized plots, evaluation metrics,
clean code, and an honest "what didn't work" section.
Expected: github-repo-url
Step 6: kg-persist [depends_on: Step 5]
Agent: risk_analyst
Tools: graph_write
Persist the project and evaluation metrics as typed nodes linked to the portfolio.
Output
- A GARCH volatility forecaster graded against realized vol
- A published GitHub portfolio repo
Execution
Run this workflow as a dependency-ordered DAG. Steps with no unmet depends_on run in parallel; dependents run after their prerequisites complete.
- Run first (in parallel): Step 1 — fetch-price-data
- After level 0: Step 2 — fit-garch
- After level 1: Step 3 — forecast-vs-realized
- After level 2: Step 4 — evaluate
- After level 3: Step 5 — github-publish
- After level 4: Step 6 — kg-persist
Execution: If graph-os is reachable, offload the whole DAG via graph_orchestrate action=execute_workflow (or the kg-delegate skill) for true parallel/swarm execution. Otherwise execute the steps natively in dependency order: run steps with no unmet depends_on in parallel, then their dependents.