Quote from: Wesellfiles on August 10, 2018, 11:30:32 AM

Quote from: KingScorpio on August 10, 2018, 10:52:40 AM

Quote from: megalinux on August 10, 2018, 10:48:11 AM

Quote from: KingScorpio on August 10, 2018, 10:30:56 AM

Quote from: megalinux on August 10, 2018, 10:24:46 AM

In fact, I have highlighted how the project is an experiment. However, machine learning algorithms can help in short-term trading decisions due to the ability to analyze large amounts of data in fractional seconds. In the long term it is obvious that this can not happen.

the market isnt systematic or periodic, like natural life and death cycles....., bird winter refuge routes, or salmon breeding return wanderings

it cant be described with algortihms...

there are trends, fashions that are temporarily.

Yes, in principle I agree with you, but I would like to try. The project as I pointed out is experimental. It's in an embryonic phase, and right now I'm analyzing the results. The goal is to understand how accurate the proposed value can be.

i think coding an interexchange arbitrage bot is more secure, and more successful than trying to prognose bitcoin

but there are many doing that already

I've done both - I've built an arbitrage bot and an AI that finds trading patterns (genetic approach).
The arbitrage bot isn't efficient because the markets are moving so fast that you don't get the prices that you want - even if your sever is directly connected to the exchange server. It doesn't matter how fast your algorithm is - it's physically impossible (talking about bitfinex). You can maybe earn something on slower markets but it's less lucrative and it takes a lot of work to do and event after that it isn't guarateed to work.
It can also be that you have calculated that you will earn a lot of money but after a few months, when the project is finished it's 1/10 of that because numbers like trading volume have changed significantly. What, maybe, could work are HFT strategies implementing a mix of insurance Mathematics and Ergodic theory which is a bit like trying to beat roulette at the casino but this takes a LOT of work and time.
Cross exchange arbitrage can end up being a huge problem because sometimes coins are locked for deposit and withdrawal and then people will pump them usually. Unexpected exchange behaviour in general is a problem. And I haven't even started to talk about all the numbers like fake volume that a lot of exchanges are displaying (like 95% fake volume).

The AI is funny - the more complex your trading strategy becomes, based on the data you feed the bot with, the less it will work in the future. It's because you have the too opposites "General Rule" vs "Specific rule". The more specific a rule is the less it will work in general and the more general a rule is the less it will work in a specific case. So what the Bot wants to find is a (relativly) simple, general rule. A rule like that does not exist. By saying "does not exist" I mean that there is no simple, general rule that would have made you more money than just buying and holding bitcoin (if you analyze all bitcoin trading data on Bitstamp from 2015 until now)

I have used some genetic algorithms to search for patterns, but now I'm trying to extract a good result (with extreme difficulty) from the prediction algorithms.