If I were to make a neural network that predicts the value of e.g. Bitcoin tomorrow based on the chart of the last month, would that work? Of course, 100% accuracy cannot be reached, but a success rate over 50% on determining if I should buy or sell Bitcoin could be very profitable. Have there been any attempts to create such neural networks so far?
1$\begingroup$ Autoregression problem assumes that there is some relationship between the present and the past values. But in fact, there is too much randomness and unpredictability. I cryptocurrency cannot be predicted with a reasonable accuracy like the stocks market in principle $\endgroup$– spiridon_the_sun_rotatorJan 24, 2021 at 20:32
There are some academic works on that, but basically what you're asking for is a time-series estimation.
Here's a paper that tries to estimate Bitcoin price, for example: PAKDD: Forecasting Bitcoin Price with Graph Chainlets.
$\begingroup$ To make this answer more useful, it may be a good idea to at least briefly describe their approach in the paper. $\endgroup$– nbroJan 25, 2021 at 12:59
It's a very complex task. The price of bitcoin could be affected by NEWs(political desicions, network upgrades, other cryptocurrencies), whales buying and selling crypto, network hashrate. You can try to search for predictions after lets say after 5 minutes, based on the data inside the orderbook on Exchanges. The market is very volatile, you can make profits of shorter period of time.