Method and backtest transparency
The four stages (paraphrasing the book)
- Stage 1 · Basing: the MA flattens and price oscillates around it.
- Stage 2 · Advancing: price moves above the MA and the MA turns up.
- Stage 3 · Top area: the MA flattens and price swings widely around it.
- Stage 4 · Declining: price is below a falling MA.
How this site implements it (logic_version = v1w210s, spot long-only)
- MA: 210-day simple moving average. Crypto trades 7×24, so 30 weeks = 210 daily bars; the book uses weekly bars, this site decides daily, so the two are not equivalent.
- Slope: 20-day change of the MA; threshold = 0.2% × a volatility factor (60-day close volatility / 2%, clipped to 0.5–2.5), ×1.5 for altcoins.
- Stage 2: close > MA and slope > threshold; Stage 4: close < MA and slope < −threshold; when price hugs a flat MA the stage is 1 (previous clear stage was 4) or 3 (previous clear stage was 2).
- A stage change needs 3 consecutive days. 'Undetermined' is a category added by this site, not in the book, and has two kinds: 'Insufficient data' = not enough valid daily bars yet for the 210-day MA or its 20-day slope (mostly coins listed less than about 230 days ago); 'Unclear' = enough data, but no stage condition is met, or a stage was met but not yet for 3 consecutive days. Totals and history count both as 'Undetermined'.
- Universe: Binance spot USDT pairs (excluding stablecoins, wrapped and leveraged tokens), top 200 by 30-day volume each day. The historical universe is rebuilt day by day, including coins delisted later.
Signal rules
- Buy (long entry): close crosses above the 210-day MA, the MA's 20-day slope > 0, and volume ≥ 1.2 × the 14-day average; altcoins additionally need BTC in Stage 1 or 2 that day (the BTC market gate kept from the original altcoin strategy). Exit: close crosses below the MA, or closes below MA × 0.90.
- In a spot context Stage 4 (declining) means "avoid / holders exit": the stage label is still computed and downloadable, but no short signal is attached to it.
- Entries only when the coin is in the universe that day; exits regardless. If a held coin has a missing day (no exchange bar, e.g. a trading halt or token swap) it is closed at the last close before the gap (reason: data gap) — prices across a token swap are not comparable.
- Returns assume a fill at the close the day after the signal (the same convention as the research backtests), net of 0.1% fees per side and volume-tiered slippage (≥$100M 5 bp, $20–100M 15 bp, $5–20M 40 bp; below $5M is treated as untradeable and gets no entry signal; exits there use 40 bp). Equal size, each signal independent — not the return of any portfolio.
Relative strength vs BTC
Carried over from the original code: RS = close ÷ BTC's 364-day MA × 100, then normalized against its own 364-day history (×10, capped at ±100). The book defines relative strength as price ÷ market index, computed once a week, and reads it against a "zero line" separating long-term outperformance from underperformance, but it gives no formula for the zero line (Ch. 1 "Definitions", Ch. 4 "Relative Strength"). Our formula is the original code's implementation, not the book's; it is for ranking only and does not drive signals.
Why the spot version is long-only
- Spot markets cannot be shorted directly (you must hold a coin to sell it). The short signals we used to publish were not executable on spot, so from the 2026-09-27 (UTC) daily bar onward the spot version no longer issues shorts. They were discontinued because they could not be executed, not because they lost money.
- At the cut-off close every open short was closed by rule (close price + tiered slippage + 0.1% fees per side, reason "discontinued") and short signals due to fill the next day were cancelled. The former long/short series v1w210 is sealed; its history stays read-only in the "Discontinued · archive" section of the Signals page, losses and liquidations included.
- The long rules did not change by a single byte: before the cut-off the long ledger of the spot series v1w210s matches v1w210's long ledger trade for trade, so the long side's out-of-sample start is still 2026-09-23.
- Shorts have moved to the futures version (Binance USDⓈ-M perpetuals, long + short, separate ledger).
Futures version (Binance USDⓈ-M perpetuals, long + short)
Differences from the book (B1 fidelity audit)
- The book uses weekly bars and a 30-week MA; we use daily bars and a 210-day MA, classified every day. The book's stated MA formula is a simple average (Ch. 1 note), while the 30-week line on the book's charts is weighted (Ch. 1, "Reading the Weekly Chart"); we use a simple MA.
- The book's buy point is an upside breakout from a Stage 1 base; our long signal is an MA crossover (a different event).
- The book wants breakout-week volume of at least 2× the average weekly volume of the past several months (on daily charts: breakout-day volume ≥ 2× the prior week's average daily volume) (Ch. 4, "The Importance of Volume"); we use 1.2× the 14-day average, a clearly lower bar.
- The book suggests long-term investors buy in two halves: half on the breakout, half on a pullback toward the breakout point (Ch. 3, "How Investors Buy"); our signals open in one go, which matches the book's trader approach.
- The book raises stops as the price rises (Ch. 6); we use fixed rules (close below MA × 0.90, or a cross below the MA).
- The book uses pullback entries and overhead-resistance checks; we do not implement them.
- The book works top-down: market, then sector, then stock (Ch. 3, "What to Buy: The Top-Down Approach"); we implement only the first layer (alt longs gated by BTC's stage). No sector layer; breadth is displayed but does not drive signals.
- The book's breadth measure is the share of stocks in Stages 1 and 2 (Ch. 3); the "% above the 210-day MA" on our breadth page is our own supplementary measure, not the book's.
- The book's diversification rules: roughly equal dollar amounts per position, at most 6 names for a small account and 10–20 at a time for a large one, spread across sectors (Ch. 4, "Don't Put All Your Eggs in One Basket"); we record each trade at equal size and independently, without position-count limits or sector spreading.
- Shorting: the book covers short selling in Ch. 7 and says breakdowns need no volume confirmation. Spot markets cannot be shorted directly, so our spot version is long-only; shorts are in the futures version.
Checked against the book (2026-09-26)
Source: the 2021 Chinese edition (China Renmin University Press, ISBN 978-7-300-28898-7), e-book. The e-book has no page numbers, so citations are by chapter and section.
- 30-week MA: the book's stated formula is a simple average; the lines on its charts are weighted (Ch. 1).
- Breakout volume: the author says there is no fixed number; his rule of thumb is at least 2× the average weekly volume of the past several months (Ch. 4).
- Buying in two halves: for long-term investors; traders buy the full position at the first breakout (Ch. 3, 4).
- Relative strength: price ÷ market index, weekly, read against a zero line; no formula for the zero line is given (Ch. 1, 4).
- Shorting needs no volume: stocks can fall of their own weight (Ch. 7, "When to Sell Short").
- Market breadth: the book's measure is the share of stocks in Stages 1 and 2 (Ch. 3); Ch. 8 lists eight long-term indicators, none of which is "% above the MA".
- Position sizing and diversification: roughly equal dollar amounts, number of names scaled to account size, spread across sectors (Ch. 4).
Known deviations
- The universe exclusion list treats WLFI (a governance token) as a stablecoin and leaves it out. The spot rules are frozen until 2027-03-23, so this is not changed here; the futures version has put WLFI back into its universe (2026-09-27).
Backtest transparency (research results, 2026-09-22, static)
These numbers come from a separate research process (A1 baseline replication, B2 pre-registered backtest verdicts) and do not update daily. This site's track record is not directly comparable with the B2 numbers below: both use the same execution convention (fill at the next day's close, 0.1% fees per side, tiered slippage), but three things differ — (1) the BTC market gate uses this site's own 210-day MA stages, while B2 uses the original 160-day MA stages; (2) B2's altcoin long legs also carry the original RS gate, this site does not; (3) the universe is the top 200 by 30-day volume with average daily volume ≥ $5M here, versus every coin with average daily volume ≥ $5M in B2. B2's short legs also include funding; this site's spot version does not short (the archived short ledger excludes funding). The site's own rules are measured by the backfilled track record on the Signals page.
A1 frozen baseline: original logic (BTC 160-day / altcoins 105-day, 46-coin universe) and controls
| Leg | Total return | Sharpe | Max drawdown | Trades | Window |
|---|---|---|---|---|---|
| Original BTC strategy (MA160) | -2.32% | -0.030 | -10.57% | 5 | 2020-02-01~2025-12-31 |
| Control: BTC buy & hold | +833.12% | 0.926 | -76.63% | 1 | 2020-02-01~2025-12-31 |
| Control: single MA160 filter | +298.68% | 1.095 | -31.78% | 27 | 2020-02-01~2025-12-31 |
| Control: same-structure random entry (median) | +9.69% | 0.223 | — | 5 | 同 BTC_native_bt_futures |
| Original altcoin strategy (46 coins, equal weight) | -5.98% | -0.125 | -17.41% | 276 | 2020-01-01~2025-12-31 |
| Control: 46-coin buy & hold | +516.39% | 0.785 | -84.87% | — | 2020-01-01~2025-12-31 |
| Control: single MA105 filter | +612.91% | 0.875 | -76.79% | — | 2020-01-01~2025-12-31 |
The A1 baseline is the original logic before the restart, not this site's v1w210. Its altcoin leg was later corrected by B2: with data extended back to 2017, the same rule on the same 46 coins returns +57.2% (Sharpe 0.81; 45 equal-weight accounts, denominator convention per B2 r1); the A1 result depended on where the data started.
B2 verdicts on the pre-registered upgrade hypotheses (0 of 8 passed)
| Hypothesis | Δ Sharpe | Random pct. | Verdict |
|---|---|---|---|
| H1a 210-day MA (altcoins) | -0.01 ± 0.39 | 21.8 | Underpowered |
| H1a_btc 210-day MA (BTC) | +0.24 ± 0.52 | 45.2 | Underpowered |
| H1b 30-week MA (altcoins) | -0.29 ± 0.39 | 13.8 | Underpowered |
| H2 Exact Stage 1→2 breakout | -0.04 ± 0.40 | 1.9 | Rejected (significantly worse in a window) |
| H3 Volume multiple | -0.23 ± 0.29 | 8.7 | Underpowered |
| H4_mans Book RS zero-line gate | -0.42 ± 0.37 | 3.9 | Rejected |
| H5_breadth Breadth gate | -0.08 ± 0.26 | 23.4 | Rejected (significantly worse in a window) |
| H5_ma MA-breadth gate | -0.17 ± 0.25 | 16.3 | Rejected |
| H6 Trailing stop | +0.04 ± 0.21 | 42.3 | Underpowered |
| H7 Stage 4 short | -0.23 ± 0.56 | 5.9 | Rejected (significantly worse in a window) |
Δ Sharpe = upgrade leg − control leg (full window 2020-07-29 to 2025-12-31), ± standard error; random percentile = the upgrade leg's percentile among 2,000 same-structure random entries.
Short rule (H7) by window (B2, full market, 1× notional, with liquidation rule)
| Leg | Window | Total return | Sharpe | Max drawdown | Trades | Win rate |
|---|---|---|---|---|---|---|
| H7_short | Full window | -1.26% | 0.016 | -18.18% | 2811 | 28.5% |
| H7_short | Bull 2020-07~2021 | -4.44% | -1.682 | -5.03% | 264 | 16.9% |
| H7_short | Bear 2022 | +7.51% | 1.003 | -3.91% | 479 | 44.1% |
| H7_short | 2023~2025 | -3.57% | -0.060 | -18.18% | 2068 | 26.0% |
210-day long (H1a, B2 structure with BTC gate and RS gate), full window
| Leg | Window | Total return | Sharpe | Max drawdown | Trades | Win rate |
|---|---|---|---|---|---|---|
| H1a_alt | Full window | +1.28% | 0.270 | -2.35% | 295 | 24.2% |
| H1a_btc | Full window | +35.91% | 0.432 | -27.24% | 7 | 42.9% |
Stage label as a coin-selection factor (H8): 20-day rank IC = -0.015 ± 0.017 (labels paired with returns from t+1), negative and not distinguishable from zero; the research does not recommend it as a universe pre-filter.
research.json asof 2026-09-22T15:22:40Z · B2-回测裁决/r1-修订/results/verdicts.csv、leg_metrics.csv(r1 修订版) · A1-复现基线/baseline_metrics.csv