Source of truth: Coinbase BTC-USD hourly data.
Candle used: The 1-minute candle at 11:59 PM ET on December 28, 2025.
Equivalent in UTC: Dec 29, 2025, 04:59 UTC
Price used: The “Close” value of that candle
Rounding math:
Math.round(105999.49) → 105999
Math.round(105999.50) → 106000
Resolution Date Time: Dec 28, 2025, 11:59 PM ET
1,000🏅 Top traders
| # | Trader | Total profit |
|---|---|---|
| 1 | Ṁ746 | |
| 2 | Ṁ295 | |
| 3 | Ṁ143 | |
| 4 | Ṁ64 | |
| 5 | Ṁ61 |
@uair01 confidence ranges would be nice instead of 10^-17 precision. Also any justification (or repudiation) on the applicability of ARIMA to predict noisy short-term market prices, or any other caveats to keep in mind. Thanks!
@deagol You're correct of course. I'll try to vibe code that.
I guess no prediction algorithm can handle Brownian processes. This is just fun to play with.
We'll see if this works:
https://manifold.markets/predyx_markets/bitcoin-price-on-december-31?r=dWFpcjAx
This looks better:
Date,Close
2025-12-26,86811.97555555556
2025-12-27,86389.21111111112
2025-12-28,85966.44666666667 <<< I expect to lose with this prediction :-)
2025-12-29,85543.68222222223
2025-12-30,85120.91777777778
2025-12-31,84698.15333333334
Test prediction and real prediction are the same?
I think there's a bug in my code :-)