The use of Artificial Intelligence (AI) for price predictions is growing in popularity in the cryptocurrency market. With that in mind, the GNY Range Report team decided to test whether its model could make better predictions than crypto traders.
GNY Range Report uses technical analysis indicators to generate price range predictions for cryptocurrencies with a machine-learning LSTM model. Results were shared on November 2 on the project’s website.
For the mentioned test, the team launched a Bitcoin (BTC) price prediction competition to forecast BTC’s price closure on October 27. In this contest, crypto traders made different guesses daily from October 23 to October 25.
Picks for you
There was a total of 206 predictions, of which 56 had a higher accuracy than the 3% observed from the artificial intelligence. This means that 56 traders predicted a price within a 3% variation from the final result.
“Our first observation was that the average accuracy of the group was higher than we anticipated.”
— GNY Range Report Team
Breaking the results from the price prediction contest
At the same time, the team collected data from its GNY Range Report on each one of the three days. The result was notable. Bitcoin price ended up being at $33,892.02 by the final day, and the predictions were as follows:
On October 23, the average prediction for crypto traders was $31,168, beating the GNY Range Report prediction of $29,861 by a huge margin.
Interestingly, the price predictions of both the crowd and the AI were almost the same on October 24, at $33,090 per BTC from the former group and $33,058 from the latter.
On October 25, the artificial intelligence beat traders by a thin margin. Predicting Bitcoin to close at $33,976 versus the crowd prediction of a $34,128 closing price on October 27.
Notably, by getting the average prediction from each day, crypto traders predicted a closing price for Bitcoin of $32,795.33. Meanwhile, the GNY Range Report AI registered a 3-day average of $32,298.33. The former was a closer average than the final result at $33,892.02, missing by $1,103.31.
This shows an interesting limitation of the machine-learning LSTM model in dealing with price forecasts on larger time frames based purely on technical analysis indicators. Nevertheless, still close enough to identify an uptrend and adjust accordingly through the days.
