The cryptocurrency market is notoriously volatile and unpredictable, making it challenging to forecast the future price movements of assets like Bitcoin (BTC). However, some tools and methods claim to offer better insights than others, such as artificial intelligence (AI) and human traders.
To test this claim, the GNY Range Report team, which uses a machine-learning LSTM model to generate price range predictions for cryptocurrencies based on technical analysis indicators, conducted an experiment. They invited crypto traders to participate in a competition to predict BTC’s closing price on October 27, 2023.
The competition ran from October 23 to October 25, and received 206 predictions from traders. The results showed that only 56 of them had a higher accuracy than the 3% margin of error from the AI model. This means that the majority of traders failed to predict a price within a 3% deviation from the actual outcome.
The team also compared the average predictions from both the crowd and the AI model on each day of the competition. They found that:
- On October 23, the crowd predicted $31,168 per BTC, while the AI predicted $29,861. The crowd was closer to the actual price of $33,892.02.
- On October 24, both the crowd and the AI predicted almost the same price, around $33,090 per BTC. They were both close to the actual price.
- On October 25, the AI predicted $33,976 per BTC, while the crowd predicted $34,128. The AI was slightly closer to the actual price.
The team concluded that the average prediction from the crowd was $32,795.33 per BTC, while the average prediction from the AI was $32,298.33 per BTC. The crowd was closer to the actual price by $1,103.31.