H-027: 每日模型对各币涨幅超过手续费的概率估计优于基础频率

未通过类别: 预测 · 状态: 已否定

日度模型能预测加密货币的涨跌方向吗？ 该想法没有跑赢简单地买入并持有，已被否定。

结果

技术细节由引擎生成，以英文显示。

pooled Brier skill -0.0089 (p = 0.012, need p < 0.025); skill above zero on 5/13 coins; paper portfolio beats holding on 4/13 and the constant share on 4/13

Sharpe：单位风险的收益（越高越好）。CAGR：年均增长率。最大回撤：从峰值的最大跌幅。p 值：结果来自运气的概率（越低越好）。“持有”列为同一资产、同一时间段的买入持有结果。术语

资产	天数	Brier skill	p 值	q (FDR)	猜中	Sharpe	持有 Sharpe	固定仓位 Sharpe	持仓时间
BTC	1760	-0.0074	0.219	0.317	52.2%	0.40	0.47	0.47	48%
ETH	1760	-0.0023	0.366	0.476	51.0%	0.16	0.20	0.20	49%
SOL	1760	-0.0117	0.505	0.597	51.2%	0.22	0.34	0.34	49%
BNB	1760	+0.0042	0.022	0.049	51.2%	0.45	0.41	0.41	48%
XRP	1760	+0.0028	0.003	0.021	54.1%	0.50	0.50	0.50	49%
ADA	1760	-0.0042	0.095	0.177	53.9%	-0.06	-0.03	-0.03	49%
LTC	1760	+0.0008	0.015	0.040	52.6%	0.13	0.10	0.10	49%
LINK	1760	-0.0119	0.680	0.680	49.8%	0.27	0.29	0.29	48%
DOGE	1760	-0.0009	0.004	0.021	53.5%	0.20	0.25	0.25	49%
XLM	1760	+0.0031	0.006	0.021	53.3%	0.29	0.29	0.29	49%
TRX	1760	+0.0034	0.005	0.021	53.8%	0.81	0.74	0.74	48%
EVER	1306	-0.1021	0.676	0.680	54.7%	-0.06	-0.11	-0.11	49%
GRAM	946	-0.0314	0.161	0.261	57.7%	-0.46	-0.35	-0.35	47%
∑	21612	-0.0089	0.012	—	52.8%	—	—	—	—

校准：预测与实际

预测	预测 ⌀	实际	天数
20%–30%	26.7%	33.3%	15
30%–40%	37.7%	44.4%	820
40%–50%	46.0%	44.4%	13847
50%–60%	53.2%	45.9%	6447
60%–70%	63.3%	58.7%	455
70%–80%	72.9%	57.1%	28

资金曲线

1 个单位的增长，对数坐标，已计入手续费。仅统计规则完全确定的交易日。

资产与区间:

为什么可能有效

Idea by F1DevBin: an honest service could publish a daily buy/sell view with a confidence level for every coin and keep score in public. Rephrased as a forecast: every day, for every coin, the probability that the next day's return beats the round-trip cost. One logistic regression shared by all 13 coins. Inputs at the close of day D: log returns over 1, 3, 7 and 30 days, distance from the 50- and 200-day averages, 7- and 30-day volatility, Bitcoin's 1- and 7-day returns, and, lagged one day for publication delay, BTC funding (7-day mean), Fear & Greed, 30-day USDT+USDC supply growth and 7-day BTC net exchange flow. Every input is a rolling 365-day z-score of its own past (min. 180 days, clipped at ±5), so coins with very different volatility share one model. The model is refitted on the first day of every month on decisions whose outcome was already known (up to two days before), L2 = 1. Costs: 0.1 % per side on Binance, 0.4 % per side on Gate.io coins (0.2 % fee + 0.2 % slippage). It passes only if, pooled over all coins from 2022 on, its probabilities beat each coin's base rate in the training window (Brier skill above zero) by more than rotating the forecasts in time does (one shift for all coins, 2000 shifts). Per-coin results are exploratory (Benjamini-Hochberg). Our earlier results (no rule beat luck on any of 156 rule x coin pairs; minute-level direction, H-001/H-002, failed) make a pass unlikely. A paper portfolio holds a share of the coin equal to the probability, trading only when the share moves by 10 points or more; it is reported against holding and against a constant share equal to its own average (timing skill would beat the constant share). Shares alpha with H-028 (0.05 / 2).

结论历史

2026-10-28: 未通过 (本引擎，数据截至 2026-10-26)

预注册规范

kind: forecast
target: direction
model:
  type: pooled_logit
  features:
  - r1
  - r3
  - r7
  - r30
  - d50
  - d200
  - v7
  - v30
  - btc_r1
  - btc_r7
  - funding
  - fear_greed
  - stable30
  - flow7
  zscore_window: 365
  zscore_min_periods: 180
  clip: 5
  l2: 1.0
  refit: monthly
  external_lag: 1
  ffill_limit: 3
universe:
- BTCUSDT
- ETHUSDT
- SOLUSDT
- BNBUSDT
- XRPUSDT
- ADAUSDT
- LTCUSDT
- LINKUSDT
- DOGEUSDT
- XLMUSDT
- TRXUSDT
- EVERUSDT
- GRAMUSDT
costs:
  default: 0.001
  EVERUSDT: 0.004
  GRAMUSDT: 0.004
portfolio:
  weight: probability
  band: 0.1
periods:
- name: oos
  start: '2022-01-01'
significance:
  method: circular_shift
  n: 2000
  alpha: 0.025
criterion:
  metric: brier_skill
  pooled: true