What does this skill do, and when should you use it?
This skill accompanies video 1 of the Quant Series, built on a framework by Roan (@RohOnChain) and refactored into a plugin by Lewis Jackson. It labels each day of any asset's history Bull / Bear / Sideways via a rolling-return rule, fits a maximum-likelihood 3×3 Markov transition matrix, and forecasts n-steps ahead via Chapman-Kolmogorov, alongside the long-run stationary distribution and a no-lookahead walk-forward backtest (Sharpe, max drawdown). Input is either any yfinance ticker (stocks, ETFs, crypto, FX, futures) or your own CSV price file. It is explicitly designed to compose — as a confirmation layer, standalone signal, or tail-risk filter — without rewriting an existing trading agent.
Runs scripts/markov_regime.py: accepts --ticker (prices fetched via yfinance) or --csv (needs only date and close columns, with common column names auto-detected); labels each bar using defaults of a 20-day window and ±5% threshold over 10 years of history; estimates a 3×3 Markov transition matrix; computes next-state probabilities and the stationary distribution; emits signal = bull_prob − bear_prob in [-1, 1]; runs a re-estimated-every-step, no-lookahead walk-forward backtest reporting Sharpe and max drawdown; optionally fits an HMM via hmmlearn that degrades gracefully if it cannot compile. With -- it prints exactly one structured object containing current_regime, signal, transition_matrix, stationary_distribution, walk_forward, and more.
- A trader with an existing momentum strategy who wants regime confirmation before entries: take longs only when the strategy and regime signal agree
- Risk management: scale position sizes down (or gate to zero) based on the asset's long-run Bear share in the stationary distribution
- An individual investor who wants a quick read on where an asset like BTC-USD or AAPL stands, how sticky that state is, and directional probabilities
- A quant with their own CSV price data who wants the same regime detection on any asset or timeframe their pipeline produces
- A TradingView user who wants the companion Pine Script v5 indicator showing the regime ribbon, live transition matrix, and stationary-distribution table on a chart
How do you install this skill?
- This is a static review; no code was executed. Script logic, backtest figures, and JSON output are independently unverified.
- Dependencies resolve dynamically from PyPI via uv with no lockfile — a supply-chain risk. Inspect the PEP 723 inline metadata and pin versions.
- yfinance may be unreachable from mainland-China networks; the ticker path may be unusable there — fall back to --csv with your own data.
- Output is a trading signal for educational purposes; validate via the walk-forward Sharpe and max drawdown before any live use.
- Repo license metadata is NOASSERTION and the publisher is unverified; SKILL.md lacks versioning and a changelog — diff carefully on updates.
- Shell / CLI
- Network access
- Local filesystem
uv (PEP 723)Pythonyfinancehmmlearn (optional, degrades gracefully)
Two commands in Claude Code: /plugin marketplace add jackson-video-resources/markov-hedge-fund-method then /plugin install markov-hedge-fund-method@markov-hedge-fund-method. No API keys, no accounts, no sudo; dependencies resolve on first run via uv (PEP 723 inline metadata, roughly 10–20 seconds). If you prefer not to install from a marketplace, markov-hedge-fund-method.md in the repo is the original one-shot onboarding prompt — paste it into Claude Code (agent mode) and it builds the identical skill from scratch locally, as a zero-trust fallback.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/jackson-video-resources/markov-hedge-fund-method.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/regime" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- detect the regime on BTC-USD
- add a regime confirmation filter to my SPY momentum strategy
- what's the long-run regime mix of AAPL
Invoke with /markov-hedge-fund-method:regime or plain English prompts like "detect the regime on BTC-USD", "add a regime confirmation filter to my SPY momentum strategy", or "what's the long-run regime mix of AAPL". Programmatic use: uv run ${CLAUDE_PLUGIN_ROOT}/scripts/markov_regime.py --ticker BTC-USD --, or swap in --csv ./my_prices.csv. Drop -- for the pretty terminal output (matrix, persistence diagonal, walk-forward metrics, HMM line). All defaults (--window 20, --threshold 0.05, --years 10, --min-train 252) are overridable.
What are this skill's strengths and limitations?
- Zero-trust friendly: the original build prompt is kept in-repo so you can read every line before installing
- No-lookahead walk-forward backtest that re-estimates the matrix at every step
- Asset-agnostic: any yfinance ticker or your own CSV, with minimal input requirements (one date and one close column)
- Graceful HMM degradation and a clean, well-documented JSON contract for programmatic integration
- Includes a TradingView Pine Script indicator for visual verification of the framework
- Backtests are historical, not forward-looking — the repo itself states this repeatedly
- GitHub license metadata is NOASSERTION; only the README claims MIT via an umbrella LICENSE in a parent directory, which you should verify
- No evidence of a test suite or third-party audit; the repo has no topics and no formal releases
- Walk-forward Sharpe/max drawdown can be NaN when history is too short; HMM states are labelled by ascending mean return, so a 'Bear' state can still be net-positive, requiring careful interpretation
- The Pine indicator (v5) mirrors the Python defaults but is a separate implementation with no stated cross-verification
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Markov Hedge Fund Method (Regime Detection Skill) this page | 45 · Not recommended | ★ 487 | 5mo ago | NOASSERTION |
| Daisy Financial Research | 55 · Use with care | ★ 29 | 5mo ago | — |
| Financial Analyst Skill | 52 · Use with care | ★ 62 | 7mo ago | MIT |
| Kronos Technical Analysis Skill | 44 · Not recommended | ★ 21 | 5mo ago | — |
| Monetary Sentinel | 57 · Use with care | ★ 2.1k | 2mo ago | AGPL-3.0 |
The repo names no competitors. Compared with quant backtesting frameworks that embed regime features internally, this skill is deliberately a lightweight, composable layer — it does not require replacing your existing strategy or backtesting stack.
How did FollowSkills review this skill?
Read-only data analysis; no API keys/accounts/sudo; data flow (yfinance or user CSV) explicitly disclosed; graceful HMM degrade; clear disclaimers; MIT license with framework attribution to Roan. Deducted: the actual script is not provided for review, so least-privilege behavior and input validation cannot be verified; uv resolves dependencies dynamically from PyPI with no lockfile or dependency audit; no rollback mechanism.
Documentation is self-consistent: field-by-field JSON contract, non-zero exit with {"error} on failure, HMM graceful-degrade path, no-lookahead walk-forward claim all align. Deducted: the script source is not in scope-provided evidence, so key paths (matrix estimation, NaN handling, CSV column auto-detection) cannot be statically confirmed; no tests or edge cases are visible.
Trigger semantics are precise: named scenarios (regime detection, confirmation layer, risk gate, sizing filter), clear inputs (ticker or CSV), stated defaults. Deducted: no Chinese-language support mentioned; yfinance may be unreachable from mainland-China networks, and core function depends on it; behavior on short histories only partially disclosed via walk_forward NaN note.
Well-layered docs (README → SKILL.md → JSON contract → composition patterns), complete MIT LICENSE, clear attribution, and a zero-trust manual install path. Deducted: repo license metadata is NOASSERTION; SKILL.md has no version or changelog; maintenance responsibility and update path are unclear (a personal YouTube link is not a maintenance commitment).
Claimed capability is complete with copy-ready composition snippets and a concrete marginal value story (signal/risk layer). Deducted: static review cannot verify outputs are directly usable; walk-forward backtest correctness and reported Sharpe are unchecked; financial-benefit claims per anchors require executed evidence, which is absent.
Auditable primary material exists: full JSON contract, documented defaults, repro params matching the on-camera demo (--threshold 0.02), and reference to the original article. Deducted: no test suite, no committed third-party execution evidence, no sample outputs to cross-check results; fact/inference separation rests on author claims.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →