Investing & Dividends Half-true — works only if you do the unspoken work

By the Income Reality Check Editorial Team · Updated 2026-05-25 · 7 min read

Lewis Jackson’s Claude Code quant prompt: what the Markov pitch leaves out

Verdict: Half-true — works only if you do the unspoken work. The Markov regime model in the prompt is genuine quant territory; the leap from “paste this into Claude Code” to “operate like a hedge fund quant” is not.

Lewis Jackson’s video “I Re-Created A Quant Trading Strategy With Claude Code (Insanely Cool)” has pulled north of 157,000 views, and the hook is irresistible: copy-paste a free prompt, install it as a Claude Code skill, and start trading the way quant funds do. Jackson tells viewers this is the same “hedge fund method” he found in a Twitter thread from a working quant called Rowan, and that the math behind it is “what gets college kids hired into $650,000-a-year jobs.” Is any of that wrong? Not exactly. Is it enough to make a retail trader profitable? Not even close.

What the video actually claims

Jackson spends most of the run-time teaching ten ideas from a Markov-chain regime model. Markets sit in one of three states — bull (last 20 days up 5% or more), bear (down 5% or more), or sideways (anything between). You count every historical transition between those states, build a 3-by-3 probability matrix, and use the diagonal “stickiness” plus a simple bull-minus-bear differential as your trading signal. Square the matrix for a two-day forecast. Cube it for three days. Layer a hidden Markov model on top so the regime labels stop being human-defined and start being learned from the price data itself.

The tutorial part of the video is shorter. Jackson links to his GitHub, where he hosts a one-shot prompt that installs the whole thing into Claude Code as a /markov skill, plus a Pine Script that draws the matrix on any TradingView chart. He keeps emphasising that the prompt is free, no email required, no upsell. The implicit promise sits in the framing — that retail traders run on “vibes” and “feelings about Bitcoin,” while quants run on probabilities, and that this prompt closes the gap.

He’s right about one half of that contrast. Hidden Markov models and regime-switching frameworks really are a recognised technique in quantitative finance, used to separate bull, bear and high-volatility regimes from price data (Investopedia). The half he glides past is everything that has to sit around the model before it earns money.

What the method actually requires

Start with the math itself. A 20-day return threshold of plus or minus 5% is an arbitrary cut Jackson admits is “subjective” on screen. The hidden Markov layer is meant to fix that — except HMMs introduce their own headaches. Academic work on regime detection consistently flags model specification (how many states? what observation distribution?) and overfitting as the central problems; pick the wrong state count and the model finds regimes that don’t exist out of sample. None of that is mentioned in the prompt.

Then there’s the backtest problem. Jackson briefly covers walk-forward testing — the practice of refitting the model day-by-day so it never “learns from the future” — and notes it’s “massively computationally heavy.” True. He also says you don’t need to understand it because “it’s going to be in the AI to do this for you.” That’s where reporting gets uncomfortable. Survivorship bias, look-ahead bias, and parameter overfitting are the three classic failure modes that quietly inflate strategy backtests by several percentage points a year, and a Claude Code skill that ingests a CSV and prints a probability matrix doesn’t, by itself, address any of them. You either build that discipline in or you don’t.

Now the costs you’d actually pay. Claude Code on the Pro plan runs $20 a month, with Max tiers at $100 and $200 a month for heavier usage; via the API, Sonnet 4.6 sits at roughly $3 input and $15 output per million tokens, and Opus 4.7 at $5 and $25. A walk-forward backtest over ten years of daily data, refitting every day, eats through tokens fast — heavy users routinely report $400 to $1,200 a month or more on API mode. TradingView’s Pine Script editor is free, but executing real trades requires a broker and, in most jurisdictions, paying the spread and the commission whether the signal works or not.

And then capital. The bull-minus-bear differential Jackson teaches outputs a number like “+45%,” with bigger numbers meaning bigger positions. He says “every hedge fund will have their own calculation” for how to size from that signal. They do — and the sizing rule is roughly half the game. Get it wrong with leverage and one bad regime call wipes the account.

None of this means the underlying ideas are fake. It means a 20-minute YouTube video is the wrong place to learn them well enough to bet money.

Is this really how hedge funds trade?

Mostly no, and partly yes. Quant funds — Renaissance, Two Sigma, D.E. Shaw, Jane Street’s quant trading arm — run dozens to thousands of signals at once, with regime detection as one feature among many, embedded in factor models, market-microstructure models, statistical-arbitrage books, options-pricing surfaces, and aggressive transaction-cost modelling. Their edge often lives in execution and infrastructure as much as in the signal itself. A standalone 3-by-3 Markov matrix on daily closes is closer to a teaching example than to a live institutional book.

The $650,000-out-of-college number Jackson cites isn’t fabricated either. Levels.fyi shows Citadel’s L1 quantitative-researcher band starting around $336,000 and reaching about $642,000 at L3; Jane Street’s published base alone is around $300,000 plus discretionary bonus. Those packages exist. They go to PhD-tier mathematicians and computer scientists hired into desks with billion-dollar capital pools, eight-figure data and compute budgets, and risk teams that vet every signal before it gets a dollar. None of that ports through a copy-paste prompt.

What you’d realistically earn

Here’s the part the video can’t quote, and you should sit with it before installing anything.

The Securities and Exchange Board of India studied 1.13 crore unique individual derivatives traders across FY22-FY24 and found 93% lost money, with aggregate net losses of ₹1.81 lakh crore (~$21.7 billion) over three years — an average loss of about ₹2 lakh per trader. The figures from the regulator’s later FY25 update were no better (SEBI). In Europe, ESMA’s product-intervention work cites National Competent Authority data showing 74% to 89% of retail CFD accounts lose money, with average per-client losses between €1,600 and €29,000 (ESMA). The UK’s FCA made its retail CFD restrictions permanent in 2019 and estimates they save UK consumers between £267 million and £451 million a year in avoided trading losses (FCA). The U.S. SEC’s own day-trading warning is blunter: “Day traders typically suffer severe financial losses in their first months of trading, and many never graduate to profit-making status” (SEC).

Those numbers don’t change because you added a regime filter. A realistic outcome for a retail trader who paper-trades this method for six months, papers a positive expectancy, and then trades small live is somewhere between “small consistent loss after fees” and “modest, lumpy gains that disappear in the next regime break.” The realistic outcome for someone who skips paper-trading, leverages up, and trusts the prompt? Read the four paragraphs above again.

Who actually wins this game

A method like this has two real audiences. The first is people who already have the prerequisites — they read Hamilton’s regime-switching paper, they know what a hidden state is, and they want a faster way to prototype something they were going to build anyway. For them, the Claude Code skill is a productivity tool, not a strategy. They’ll catch the overfit before it eats them.

The second audience is people who think the prompt is the strategy. They are, statistically, the 93% in the SEBI sample. The video doesn’t tell them that, because the channel’s business — and Jackson’s “01 systems” community linked in the description — runs on the first impression, not the six-month outcome.

A useful comparison: we’ve covered the same income-promise pattern in a Claude Code YouTube-monetisation pitch, and a different version of it in a “great market flip” prediction video. The shape is consistent: a real technique, framed as a shortcut, with the unspoken work cropped out of frame.

Who this is (and isn’t) for

If you’ve already done quantitative work, have a brokerage account you can afford to lose 5% of without panicking, and you treat the prompt as a sketch to extend — go for it. Build a proper walk-forward harness around it. Add transaction costs. Test on out-of-sample data the model hasn’t seen.

If you’re newer than that — if you don’t already know what “look-ahead bias” means, if you’ve never sat through a 30% drawdown on paper, if the $20/month Claude Code subscription would be your first trading-related cost — this prompt is not the on-ramp Jackson’s framing suggests. Read the SEBI study. Open a paper-trading account first. Treat the next 12 months as tuition, not income.

What to remember

The math in the video is genuine. The prompt is genuinely free. The unspoken middle — backtest discipline, risk sizing, regulator-grade base rates — is genuinely where retail traders lose, and a Markov matrix doesn’t fix it.

View counts and channel details reflect the moment of review and may have changed since publication.