Trading Strategies
Four models across every time horizon — intraday, half-day, multi-day, and long-term. All built on the same underlying edge. All graded on the same scale. Read the edge first; each model below is one way of harvesting it.
// The Core Edge
For every instrument we follow, we maintain a statistical estimate of its "normal" price range — recomputed nightly, exponentially weighted toward recent volatility. The range is calibrated so that roughly 80% of normal sessions stay inside it. When price punches outside, it's by definition a 1-in-5 tail event.
In a tail event, the higher-probability outcome is mean reversion back to the range, not continued movement in the breach direction. Across years of backtesting, recovery rates in the high-70s to low-90s show up consistently on the cleanest setups, depending on instrument and regime.
That's the edge. Everything else on this page is a different way of harvesting it.
What we grade on (and why it isn't recovery rate)
The conviction tier (A through E) on every signal doesn't measure how often the trade works. It measures the average dollar profit per trade across the model's full backtest history. One number that captures both win rate and payoff size in the same metric — so a strategy with a 90% recovery rate but tiny payoffs lands in the same place as a 65% rate with bigger payoffs, if their average dollar outcomes match. Most subscribers care about the dollar number; few care about the rate.
Three horizons, one edge
Day Trade trades the edge intraday — open and close in the same session, hundreds of instruments under continuous watch. Multi Hour holds the breach for half a session to two sessions, letting the reversion play out over hours rather than minutes. Swing Trade filters the same intraday breaches for the small subset that historically have multi-day continuation legs and holds them for 7 to 28 days. Different timeframes, different filter stacks, same underlying premise — and the same A–E grading system across all three.
// 01 · Intraday
The fastest engine, and the foundation the other two build on. Day Trade runs the band mechanic on the broadest universe — around 200 instruments across equities, futures, FX, and crypto — checked every minute of the session. Trades open and close the same day. Typical hold: minutes to a few hours.
What's specific to this strategy
Universe. ~200 instruments, the broadest of the three strategies. Equities, index futures, FX majors, crypto majors. Calibration per instrument; K typically lands around 1.5σ either side of the prior close.
Trigger. Every 1-minute bar is checked against the band. High ≥ upper edge fires SELL, targeting the lower edge. Low ≤ lower edge fires BUY, targeting the upper edge. Only the first cross per side per day — the same stock can't trigger the same direction twice in one session.
Exit window. All positions close before the session ends. No overnight risk. The strategy is calibrated for the snapback inside the same trading day, where the band statistics actually apply.
// 02 · Half-day
Same edge as Day Trade, slower. Multi Hour catches the same breaches but holds them for hours instead of closing same-session, giving the mean reversion time to play out fully. Best on dislocations driven by positioning rather than news. Typical hold: 12 hours.
What's specific to this strategy
Volatility decay tuned for the longer horizon. The EWMA uses a 21-day half-life — a slower decay than Day Trade — so the bands respond steadily to regime shifts without overreacting to a single noisy session. Calibration defaults: K_L = 1.45σ below the prior close, K_U = 1.52σ above. The asymmetry is intentional: real markets crash faster than they rally, so the downside edge is set tighter than the upside.
Conviction overlays. On top of the standard breach detection, four overlay lanes (Peak, High, Scaled, Workhorse) check for additional confirmation — multi-timeframe agreement, volume profile, cross-asset confluence, regime alignment. A breach that lights up multiple overlays carries materially higher conviction.
Why the slower window wins. By the time half a session has passed, idiosyncratic intraday wiggle has averaged out, and only the breaches with real reversion power are still in profit. The noise that scalps trade through gets filtered by time itself.
// 03 · Multi-day
Multi-day trades, on moves that have legs. Swing Trade watches the Day Trade signal stream and only fires when a short-term dislocation also shows a strong historical pattern of running for days, not hours. Holding period is 7 to 28 days — and the bar is high enough that most candidates get rejected.
How a signal gets graded
Every short-term dislocation Day Trade fires on is run past two scoring systems in parallel, each looking at the setup from a different angle.
A consensus panel — a primary read plus six overlays. The overlays look at different things: how the macro regime is set up, how the ticker has been behaving versus its sector, where institutional positioning sits, and so on. The more overlays that agree with the primary, the stronger the grade.
A set of specialists — independent of the panel. Each specialist hunts for a narrower setup (elite long opportunities, rank-one momentum, event-driven directional moves). When a specialist's threshold is met, the signal goes out — even if the broader panel was on the fence.
Both systems pull from a pool of validated parameter sets, each carrying its own conviction tier earned over years of historical testing. The grade you see is whichever validated set fired at the highest tier.
// Shared framework
Every signal we publish carries a grade from A (highest conviction) down to E (lowest). The grade reflects one thing and one thing only: how much profit the setup has made on average across years of historical testing. No win-rate gimmicks. No Sharpe theatre. Just expected outcome.
Why the thresholds differ between strategies
A Day Trade lasts hours. A Swing Trade lasts weeks. You can't measure them against the same yardstick. Grade A on Day Trade means averaging at least 0.7% per trade — small per trade, but you take a hundred a year and it compounds. Grade A on Swing Trade means averaging at least 6% per trade — much larger per trade, but you take far fewer. Both are A within their own strategy; you can't read across.
How the grade on a single signal is decided
When a setup appears, every validated parameter set in the strategy's pool looks at it independently. Most don't fire — their threshold conditions aren't met. Some do. The signal grade is whichever fired variant carried the highest tier. A Grade A signal therefore isn't "the average variant thinks this is great" — it's "at least one A-tier variant fired on this setup." A much higher bar.
The safety floor and the quarterly refresh
Any variant with fewer than 10 trades in its backtest history is automatically E-tier, regardless of average — ten trades is the minimum sample needed to tell edge apart from luck. Tiers get recomputed every quarter as backtests refresh. A variant that loses its edge over the last quarter gets downgraded automatically, no human intervention.
| Strategy | A | B | C | D | E |
|---|---|---|---|---|---|
| Day Trade | ≥ 0.70% | ≥ 0.50% | ≥ 0.30% | ≥ 0.15% | < 0.15% |
| Multi Hour | ≥ 4.50% | ≥ 3.50% | ≥ 2.50% | ≥ 2.00% | < 2.00% |
| Swing Trade | ≥ 6.00% | ≥ 5.00% | ≥ 4.50% | ≥ 4.00% | < 4.00% |
Numbers are the historical average return per trade required for the corresponding tier. Any variant with fewer than 10 trades in its backtest history is automatically E-tier.
// 04 · Risk framework
Every signal we publish includes three numbers up front: where to enter, where to take profit, and where to cut the trade if it goes against you. There are no open-ended directional bets, no "wait and see" on the downside. From the moment a signal goes out, the math of the trade is fully defined.
Entry, target, stop
Entry is the price at the moment the setup triggered — the level where the strategy says the edge appears. Target is the take-profit level: for a band-breach trade, it's the opposite edge of the band (the mean-reversion anchor). Stop is a defined offset on the opposite side of entry, sized to fail fast when the setup is wrong rather than bleed through a slow drawdown.
Sizing
We publish the signal geometry — entry, target, stop, conviction grade. We don't tell you what size to trade. That stays with you and your account. The general principle most subscribers run with: bigger size on the higher-grade signals, smaller on the lower ones, and scale down on the more volatile instruments so each trade puts roughly the same amount of dollar risk on the table.
Why static stops, not trailing
Trailing stops sound appealing — lock in profit as the trade moves your way — but they're noise traps. A normal pullback inside a winning trade can knock you out of a position that would have hit its target. The trade's edge was modelled assuming you hold to the target or get stopped at the defined level, not somewhere in between. Static stops respect that.
// 05 · Macro framework
What is the economy actually doing right now, and which markets tend to win in this kind of environment? Macro Model takes the two big macro questions — is growth speeding up or slowing, is inflation rising or falling — and uses them to place the current environment on a four-quadrant grid. Each quadrant has a track record of which asset classes lead. Rather than picking one quadrant outright, the model publishes the full probability that we're in each one, plus a forward projection at 1, 3, and 6 months.
The four regimes
Q1 Goldilocks — growth up, inflation down. High-beta equities and tech historically lead. Q2 Reflation — both up. Commodities, energy, materials and financials outperform. Q3 Stagflation — growth down, inflation up. Gold and defensive sectors win. Q4 Deflation — both down. Long-duration bonds and defensive staples lead; the worst regime for risk assets.
How the read is built
The two axes. Each axis is built from a weighted basket of macro factors. Growth uses GDP (30%), Industrial Production (25%), Non-Farm Payrolls (25%), and Retail Sales (20%). Inflation uses Core CPI (30%), Core PCE (30%), Sticky CPI (25%), and 1-year inflation expectations (15%). Each factor is checked against its anchor (2.0% year-over-year) and the direction it's moving in. The two pieces are blended 60/40, then normalised against history.
The probability output. Instead of declaring "we're in Q3," the model measures how close the current macro position sits to each of the four quadrant centres and turns those distances into a probability. So you get a read like "47% Q3, 26% Q2, 16% Q4, 11% Q1" — and an explicit measure of how uncertain that read is, instead of the vague "confidence score" most macro tools publish.
Forward projection. A model trained on historical regime transitions projects the current probabilities forward 1, 3, and 6 months. In parallel, the system surfaces the five most-similar historical months and reports what actually happened next in those analog cases — empirical base rates, not extrapolation.
Supporting models
Recession probability. Blends two well-known specifications: Estrella-Mishkin's yield-curve probit (1998) and the Chauvet-Piger / Hamilton coincident-indicator regime-switching probability published monthly by the St. Louis Fed.
Curve inversion tracker. Watches the 3-month / 10-year Treasury spread (T10Y3M). Reports current inversion state, days since last sign change, and the historical pattern of how many months elapsed between un-inversion and recession in each prior episode back to 1982.
Net Fed liquidity. Fed balance sheet minus reverse repo minus the Treasury General Account. Reports current level, 30 / 90 / 180 / 365-day changes, and a 90-day z-score against a 5-year window.
Cross-asset anomaly engine. Daily scan across asset-class relationships against trailing 5-year history. Severities flagged at |z|>1 (notable) and |z|>2 (extreme).
Print nowcasts. Daily-refreshed forecasts for the six tier-1 macro releases (Payrolls, CPI, ISM Services, Jobless Claims, Retail Sales, PPI), with confidence bands derived from the last 24 prints' residuals.
Data feed — FRED (Federal Reserve Bank of St. Louis), with vintage reconstruction so backtests never see revisions that hadn't been published at the historical decision point.
// 06 · Long horizon
The founder's long-term book, published live. Where the three trading strategies fire short-to-medium-term signals, Investing tracks the positions he holds with his own capital — every entry, every rationale, every exit trigger, in real time. Audited multi-year track record: +178% (2023), +94% (2024), Sharpe 2.10.
How the book is built
Core (60-65%). The anchor positions, held against multi-quarter theses. Low turnover. These don't move unless the thesis breaks. Thematic tilts (15-20%). Discretionary overlays driven by the prevailing macro regime. Rotated when the Macro Model quadrant shifts. Cash reserve (~20%). Dry powder for adding to existing positions on dispersion events, or starting new positions when conviction firms up.
When positions move
The book gets a weekly review. Positions only change on one of three triggers: the thesis breaks (news or sector damage that invalidates the original case), the macro regime shifts in a way that changes which asset classes have a tailwind, or the position has reached its long-term target and it's time to take chips off the table. Average holding period is around 11 months. Some positions have been in the book for years.
// 07 · Access
Four subscription tiers, four models, one ATLAS macro framework bundled with every tier. Pick the smallest tier that gives you what you need.
Single Model
$20 / month
Pick any one of the four models. ATLAS macro framework included. Full dashboard + push/email alerts.
Choose a Model →All Models
$50 / month
All four models (Swing Trade + Multi Hour + Day Trade + Investing) + ATLAS. 7-day free trial.
Start 7-Day Trial →Pro Access
$5,000 / quarter
All Models + read-only REST API + ATLAS deep-feed + full backtest archive + daily macro PDF. 1 seat.
Apply →Pro Access · Team
$10,000 / quarter
Pro Access + 5 seats + private Slack + quarterly strategy call + compliance logs. 3 seats remaining.
Apply →ATLAS macro framework is included with every tier. Single Model and All Models bill monthly; Pro Access and Pro Access Team bill quarterly. Cancel any time from your account dashboard. Past performance does not guarantee future results.