Chapter 6 — Quantum-inspired portfolio & sentiment

Guided time: 4–5 hours
Prerequisites: Chapter 5 — ML

Next: Chapter 7 — Hardening

1. Purpose

This chapter covers two optional differentiators often highlighted in product marketing but operationally heavy:

  1. Quantum-inspired portfolio optimization — a research-flavored layer that may run when Qiskit ecosystem imports succeed and environment flags allow it.
  2. Sentiment analysis — typically Finnhub-backed news flows surfaced over HTTP for dashboards.

You should leave knowing when to enable each feature and what happens when you disable them.

2. Objectives

  1. Explain how HybridPortfolioManager decides between quantum and classical weighting.
  2. Describe ENABLE_QUANTUM and failure modes (import errors → HAS_QUANTUM = False).
  3. Locate sentiment configuration keys and at least one HTTP route returning sentiment JSON.
  4. Articulate operational costs: dependency weight, cold start time, external API rate limits.

3. Quantum path (architecture)

Entry: engine/portfolio_manager.py

  • Reads ENABLE_QUANTUM (default on unless set to 0 / false).
  • Tries to import quantum.portfolio_optimizer.QuantumPortfolioOptimizer.
  • On import failure (missing optional deps, version skew), logs a warning and sets HAS_QUANTUM = False.

Optimizer: quantum/portfolio_optimizer.py

  • Builds a QUBO via _build_mv_qubo from expected returns and covariance proxies.
  • Current engineering note: full QAOA + MinimumEigenOptimizer wiring for modern qiskit-algorithms may still be evolving in your branch; the optimizer may log that it returns uniform weights while keeping the QUBO builder exercised. Treat quantum outputs as experimental until you verify solver behaviour in your fork.

Backends: quantum/backends.py

  • Prefers IBM if token + provider packages work.
  • Falls back to Aer simulator locally.
  • IBM provider packages may be optional; Aer path should remain available for dev machines.

4. When quantum runs vs classical fallback

Even with HAS_QUANTUM = True, the hybrid manager may require enough symbols (historically ≥3) before attempting quantum optimization. With fewer symbols, classical weights are used.

Teaching moment: quantum here is portfolio selection / weighting, not the same module as signal generation ML.

5. Sentiment path (typical)

Configuration (representative keys—verify in your tree):

  • SENTIMENT_ENABLED
  • FINNHUB_API_KEY
  • Windows / timing: SENTIMENT_WINDOW_S, polling intervals, asset lists

Code locations:

  • sentiment/ package — services and cycles.
  • render_api.py — HTTP endpoints aggregating sentiment for dashboards.

Operational caution:

  • External APIs impose rate limits; aggressive polling burns quota.
  • News sentiment is noisy and may lag spot markets.

6. Labs

Lab 6.1 — Quantum off (20 min)

Set:

ENABLE_QUANTUM=0

Restart API; confirm logs show classical optimizer messaging without quantum attempts.

Lab 6.2 — Quantum on (60 min, optional)

Set ENABLE_QUANTUM=1 on a machine with full Qiskit stack installed. Watch logs for import warnings vs successful optimizer construction.

Lab 6.3 — Sentiment route (45 min)

Call /sentiment (or your branch’s equivalent) with and without FINNHUB_API_KEY. Document status codes and JSON shapes.

7. Exercises

  1. What env var disables quantum at the portfolio manager gate?
  2. Name one practical reason to disable sentiment in development.
  3. Where is _build_mv_qubo defined, and what mathematical structure does it return?

8. Product narrative (if you sell training)

Be transparent in marketing:

  • Quantum may be experimental / fallback-heavy depending on version pins.
  • Sentiment is auxiliary data, not oracle signal.

Students respect honesty; buyers punish overselling.

9. Notebook

notebooks/01_optional_features_matrix.ipynb — table: feature on/off, deps, risks, mitigations.

10. Summary

You can now toggle headline features without mystique. Chapter 7 converts this knowledge into durable operations: secrets, tests, deployment hygiene.

Next: Chapter 7 — Hardening