Crypto Momentum Model

Ready. Click Refresh data to pull the latest daily candles.

No data loaded yet.

What this actually is

This is not a price forecaster. No model reliably predicts what a coin will do over the next 5–10 days — crypto returns at that horizon are dominated by noise, news and flows that nothing in a price chart can see.

What this does is cross-sectional momentum: it scores every coin in the universe on how strongly and how cleanly it is already trending, ranks them against each other, and surfaces the top few. The bet is that relative strength persists a little longer than chance — a real, documented, and small effect. It shows up across many trades and is invisible in any single one.

A realistic expectation in a healthy market is somewhere around 52–57% of picks positive over 10 days, with the winners a bit larger than the losers. Anything claiming much better than that is either overfitted or lying. The Backtest tab is there so you can check this one against its own record rather than take my word for it.

The five things it measures

Every coin gets scored on five components. Each is ranked against the other coins on that same day (a percentile), so the score means "strong relative to the field right now," not an absolute number that drifts with the market.

Risk-adjusted momentum (40%) — returns over roughly 1, 3 and 6 months, each divided by that coin's own volatility. Dividing by volatility matters: a 30% move in a calm coin is far more meaningful than a 30% move in one that swings 30% every week. The most recent 2 days are skipped, because very short-term moves tend to snap back.
Trend quality (25%) — a straight line fitted to the last 40 days of log price. It rewards the slope multiplied by how well the line actually fits (R²). A coin that grinds steadily upward scores high; one that doubled in a single candle and went flat scores low, even though both have the same total return.
Breakout freshness (20%) — how close the price is to its 60-day high, plus whether the daily range is expanding. This is what catches moves that are just getting going.
Volume confirmation (15%) — the last 10 days of dollar volume against the last 60. Rising participation means real money is behind the move rather than a thin drift.
Overheating penalty (subtracted) — RSI above 75, or price stretched more than ~2.5 ATR above its 20-day average, cuts the score. This is aimed squarely at your 5–10 day horizon: something already vertical is the worst thing to buy for the next week.

The gates — what gets thrown out before scoring

A high score on a coin in a downtrend is meaningless, so a few hard filters run first. A coin must be above its 50-day average, its 20-day average must be above its 50-day, it needs at least 220 days of history, at least the minimum daily dollar volume you set, and RSI must be under 92 (that last one excludes genuine blow-off tops).

The market regime gate — the part that matters most

This is the single most valuable rule here, and it is worth more than the ranking itself. When Bitcoin is below its 100-day average, altcoin momentum strategies don't merely underperform — they get taken apart, because in a crypto downturn almost everything correlates to 1 and the strongest recent movers fall the hardest.

So the model checks Bitcoin's trend and the market's breadth (how many coins are above their own 50-day average) before it will hand you anything. On risk-off days it tells you to sit out rather than serving you the best of a bad lot. The backtest applies the same rule, so what you see tested is what you'd actually have run.

How to read the backtest honestly

Three numbers matter far more than the headline return:

The decile chart. Every coin-day is bucketed into ten groups by score, and each bucket's average forward return is measured relative to the market that day. If the top bucket doesn't beat the bottom bucket, and the pattern doesn't slope reasonably smoothly in between, the score is sorting noise — and no amount of good headline return changes that. This is the test most tools quietly leave out.
Excess over the market, not raw return. In a bull market every strategy looks brilliant. The only question worth asking is whether picking the top 5 beat simply holding the whole universe over the same days. That spread is the actual edge.
The worst stretch. Max drawdown and the losing periods tell you what you'd have to sit through. Most strategies are abandoned in the drawdown, not in the backtest.

Known biases in the backtest — read these before trusting any number above.

Survivorship bias (the big one). The universe is built from coins that are liquid on Binance today. Coins that collapsed, delisted or died over the test window aren't in it. Real-world results would have been worse — possibly materially. There is no clean way to fix this without paid point-in-time data.
One market cycle. Two years is a handful of independent regime changes, not a large sample. It is enough to reject an obviously broken model, not enough to confirm a good one.
Overlapping windows. The decile statistics use overlapping forward windows, which makes the sample look larger than it is. The t-statistic shown is computed on non-overlapping samples for that reason, and is still optimistic.
Costs are an estimate. Fees and slippage are modelled as a flat round-trip you set in Settings. Real slippage on a fast-moving mid-cap is worse than on Bitcoin.
No tuning was done to make this look good. The weights and lookbacks are conventional defaults chosen before seeing any results. You can change them in Settings — but be aware that adjusting weights until the backtest sparkles is precisely how people fool themselves. If you change something, change it for a reason you could explain out loud.

Where the data comes from

Daily candles come straight from Binance's public market-data API, fetched by your browser — no server, no API key, no account. Signals are computed on completed daily bars only (UTC close), never the partial day in progress, so what you see live matches exactly what the backtest measured. Data is cached locally so a morning refresh only pulls the new day.

Not financial advice. This is a research tool: it ranks relative strength, it does not predict prices. Assume you can lose money.