How to use AlphaScreen

Protocol: run continuously from June 2019 to July 2026 across two real crashes — the COVID crash (Feb–Mar 2020) and the 2022 bear — $17k across 5 slots, next-open entries, on ~485 liquid US large-caps. Exits are the live run-the-winners rules: −8% stop, graduate at +25% then trail the 21-day EMA, 15-day deadline for anything that never graduates. GEAR-SHIFT = RS leaders gated by the breakout-timing engine.

Cumulative return over ~7 years on a $17k book. Simulated backtest, not audited live results. Reproducible from scratch/backtest_three_methods.py against a pinned price cache.

It beats the index on return, not on drawdown. +204.2% against SPY’s +188.2% — but a −36.5% worst fall against SPY’s −33.7%. You are paid for taking more risk, not less. And QQQ beat all of it (+307.4%), which is the honest benchmark most momentum strategies quietly omit.

Where it earns its keep is the shape, not the total. Broken into regimes (return / max drawdown):

It wins big in momentum tapes (2020, 2025), matches the index in the bear (−22.3 vs −22.5), and lags a broad grinding bull (2023–24: +48.3 vs +58.8). That last row is the cost of concentration — five names cannot beat five hundred when everything rises together.

Caveats, and they are not small: the universe is current large-caps, so companies that went bust are absent — survivorship bias flatters every return here, most of all in the crash windows. Trade counts are 586 over the period; the top 20 of them account for essentially all the net profit, so outcomes depend on catching a handful of outliers. Treat the strategy ranking as the robust read and the absolute returns as illustrative.

Most GEAR-SHIFT trades lose — and the book still compounds. The reason isn’t win rate, it’s payoff asymmetry. These are the actual 586 trades from the backtest above, not an illustration:

At 45.2% we clear a 36% break-even — a real edge, but a thin one, about 1.3× break-even. The −8% stop is what keeps the average loss at −7.1%; letting winners run past the old +35% target (removed in v17.6.0) is what lifts the average win to +12.6%.

But the average is a lie, and this is the part that matters. The biggest single trade returned +144%; the worst returned −15.4%. Of 586 trades, the top 20 account for essentially all of the net profit — the other 566 roughly cancel out.

That is the real shape of this strategy, and it has an uncomfortable implication: your result depends on catching a handful of outliers. Miss the top few and a year of disciplined trading nets close to nothing. It is why the exit rule matters more than the entry — every one of those 20 was a trade that was allowed to keep running, not a trade that was cleverly picked. We tested whether entry timing predicted them; it does not.

Trade economics measured over 2019-06 → 2026-07 on ~485 current large-caps. Survivorship bias applies — companies that went bust are absent — so treat these as illustrative of shape, not as a forecast.

The stock engines above hunt for the best few names. The ETF pages do something different: they deal in whole slices of the market. There are exactly two of them, and they answer different questions.

What the ladder does not do: predict returns. Next-month return is not ordered by rung — the 67% rung beat the 100% rung, and the correlation is +0.083, indistinguishable from zero. What the rungs order is the share of months that finish positive (48% / 53% / 71% / 69%). It separates fragile tape from healthy tape; the drawdown benefit is mechanical, not foresight. It also sits partly out of the market about 30% of the time, so in a decade with no crash it will lag buy-and-hold and feel like wasted money — which is what insurance feels like.

What we tested and threw away: rotating on the fastest-improving RS every 45 days (9.0% a year with a −60% fall), switching between momentum and the index on a style signal (worked in 1 of 36 configurations), and all three classical portfolio optimisers — Markowitz minimum-variance, maximum-Sharpe, and Hierarchical Risk Parity — which finished 7–9 points a year behind simply splitting money equally. None of them survived being tested on two separate windows.

RRG (Relative Rotation Graph)

A map of sectors/stocks rotating through Leading → Weakening → Lagging → Improving quadrants vs the benchmark.

Sprint

A time-boxed (3-week) basket of the highest-RS setups. Backtests show the time limit itself adds return — capital never sits in stalled names.

HMM (Hidden Markov Model)

A statistical engine that estimates the probability a stock is entering an "explosive" state; it opens up to 5 positions when strict entry gates pass.

DCR

Daily Closing Range — where price closed within the day's range (100 = at the high). Persistent high DCR signals accumulation.

Market Gate / Master Banner

A go/no-go switch computed from SPY trend + volatility. When closed, engines stop opening new positions.

Allocation ladder

How much of the market to hold, on a 0 / 33 / 67 / 100 scale, set by three trend checks at each month end. Risk control, not a forecast — see ETFs.

Breaking Down

A fund trading more than 10% below its 50-day average while still above its 200-day. The old label called any distance below the 50-day a “pullback”, which invited you to buy a dip that wasn’t one. Shown as a red badge on the leaderboard; it changes what you’re told, not what the engine buys.

Tradable core

The 45-name subset of the ETF leaderboard the books may actually buy. The leaderboard ranks ~150 so you can see the whole market; trading that whole list was measured and it was worse on both return and drawdown.

Drawdown

The worst peak-to-trough fall. It matters more than it looks: recovering a 50% fall takes a 100% gain, so a deep hole can consume most of a ten-year horizon on its own. We quote it marked daily — a month-end-only measurement misses any fall that starts and ends inside one month, and understates the real figure by several points.