H-030: Live replication of H-028: daily forecasts of bigger-than-usual moves, published before 12:00 UTC
— Family: forecast · Status: registered
Do daily forecasts of big crypto moves work live? Data is being collected. The pre-registered test runs on 2027-11-01, once a year of data has accumulated.
Forward data collected: 0 of 368 days (0.0%). First test on 2027-11-01.
Result
No engine results yet: this study is kept for the record and is not re-run.
Why it might work
Repeats H-028 live from 2026-10-29 with the same frozen monthly models. Every morning the workflow computes the forecasts from yesterday's close with that month's model (fitted on the month's first day, stored once in datasets/forecast_models/ and never refitted) and appends them to datasets/forecast_log.csv; the commit to GitHub proves they existed before the outcome. Only forecasts published before 12:00 UTC count; the outcome runs from the open of the 12:00 UTC hourly candle to the next day's (datasets/noon_prices.csv: Binance hourly files, Gate.io API), so the real delay between data and trade is in the test. The risk threshold of each day is the coin's median absolute daily move up to the decision day, stored with the forecast. First tested after about a year of live forecasts. Shares alpha with H-029 (0.05 / 2).
Verdict history
Pre-registered specification
kind: forecast_live
target: risk
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
execution:
time_utc: '12:00'
periods:
- name: forward
start: '2026-10-29'
significance:
method: circular_shift
n: 2000
alpha: 0.025
criterion:
metric: brier_skill
pooled: true