BTC / USDT · live from Binance · connecting

A small experiment in guessing where Bitcoin goes next.

I built this to learn how forecasting works. Every 15 minutes it guesses whether the price will be higher in 1, 3 and 24 hours, and every guess is checked later. Please don't trade on it.

“Invest in preparedness, not in prediction.”

Nassim Nicholas Taleb, The Black Swan
— — in 24 h

Loading the model…

 

Chart

Guess vs. reality

The line is the real price. The band shows where the model expected the price to be an hour later. The arrows are its hourly 1-hour calls: green for up, red for down, filled when it was right.

“Don't cross a river if it is four feet deep on average.”

Nassim Nicholas Taleb
Price Expected range (80%) ▲said up, was right △said up, was wrong ▼said down, was right ▽said down, was wrong

“History and societies do not crawl. They make jumps.”

Nassim Nicholas Taleb, The Black Swan

Record · last 7 days

How often it was right

50% is what a coin flip would get. Even at 53%, trading fees would likely eat the difference. How I tested it is in the experiment report.

“The problem with experts is that they do not know what they do not know.”

Nassim Nicholas Taleb, The Black Swan
HorizonLive

Latest checked predictions

MadeForCallPrice thenPrice afterResult

Every prediction is saved as a CSV file on GitHub.

“A Black Swan for the turkey is not a Black Swan for the butcher.”

Nassim Nicholas Taleb, The Black Swan

Method

How it works

The model runs in your browser. A GitHub job runs it too, every 15 minutes, and saves the record.

“Wind extinguishes a candle and energizes fire.”

Nassim Nicholas Taleb, Antifragile
  1. 01

    Reads the market

    A week of minute prices for Bitcoin, Ethereum and Solana, plus the Fear & Greed index.

  2. 02

    Guesses up or down

    It only learns the direction, not how far the price moves.

  3. 03

    Says how sure it is

    55% should be right about 55% of the time. Below 52% it makes no call, because those were coin flips. It can't predict the size of a move, so the price estimate is just today's price. At 24 hours it was no better than a coin flip in testing, so take that one lightly.

  4. 04

    Gets checked

    Every forecast is compared with the real price later. Forecasts made late after an outage don't count.

  5. 05

    Retrains every night

    New settings are only kept if they do better on days the model hasn't seen.

“Don't tell me what you think, tell me what you have in your portfolio.”

Nassim Nicholas Taleb, Skin in the Game