Skip to content

[Design] Backtest / signal-alpha harness (validate conviction & activism signals) #1401

Description

@stranske

Problem

MDB detects conviction/activism signals but does not validate them historically, so they are assertions, not evidence. WhaleWisdom/Novus ship backtests + skill scores; the literature quantifies 13F "alpha cloning" (~1.6–2.1%/qtr; treat higher claims as marketing). A backtest harness lets MDB prove its signals.

Proposed approach (for discussion)

  • Replay a strategy (e.g. "follow high-conviction new buys") over historical point-in-time holdings, entering at disclosure lag, and report return / Sharpe / hit-rate vs a benchmark.
  • Requires as-of holdings + a price source (new dependency — see open decisions).

Open decisions (owner)

  • Blocked by bitemporal holdings (T1) — cannot backtest without point-in-time data. Confirm sequencing.
  • Price/returns data source (cost + licensing): which provider, and is redistribution of results acceptable?
  • Scope: internal signal-validation harness only, or an operator-facing backtester UI (much larger)?

Definition of done (for this DESIGN issue)

Design doc: strategy spec, price-source decision, metrics, and the as-of data contract with T1. No code until T1 lands.

Notes

Part of epic #1320 (Tier-2). Blocked by the bitemporal-holdings design/impl.

Part of epic #1320 (Tier-2). Blocked by #1400 (bitemporal holdings).

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestpriority:normalNormal-priority weekly repo-review work

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions