Polymarket
Do Bitcoin candles have momentum?
Summary
An experiment on Polymarket's "Bitcoin up or down" markets. The hypothesis was momentum, and the data said no: candles repeat their colour less than half the time. The edges in my first version came from look-ahead leaks, so I shelved the bot before putting any money in.
Python on Binance candles, with a test that watches for look-ahead.
The idea
Polymarket's «Bitcoin up or down» markets pay out on whether a candle closes above its open. My hypothesis was momentum: a green candle tends to be followed by another green one. I built a bot with risk limits and a paper mode to trade it.
What the data said
No momentum. On 15-minute and one-hour candles, the colour repeats less than half the time, in 2024–25, in 2025–26 and out of sample. The edges my first version found in volume, volatility and wicks came from look-ahead: they used data from the very candle being predicted. A notebook now shows the leak and a test guards against it. I shelved the bot before putting any money in.
Figures
