Quote from: ferumflex on September 19, 2017, 12:48:53 PM

Quote from: QuantumTrading on September 18, 2017, 07:34:40 PM

Quote from: ferumflex on September 18, 2017, 06:02:42 PM

Have few questions:
You said that you have accuracy of 96%, this is on training or validation data?

Do you analyze only market data or you also parse news?

I think technical analysis does not work. You probably should combine it with some kind of fundamental analysis.

I'm interested into why you don't think technical analysis works as our findings have been quite different.

Bitcoin seems to follow a very set pattern of supply and demand, more then other assets. This is why we have seen news come out and even though the price drops, it does not go into free fall like it used to(remember silk road closure). If you study the order book during news events, and then also when it is just normal trading, you can see examples of this. Now that the market is more evolved and more money is at play, news does not affect the market nearly as much as it did in the past.

These kinds of bots also do not perform as per say traditional technical analysis such as with MACD, ATR, RSI etc. They more often employ much more advanced statistics such as GARCH volatility models, random walk theory, Alpha and Beta values, etc.

This still does not mean news should not be looked at. One other thing we have noticed in our research is the bitcoin market is extremely emotional. People sell on news that in reality does not matter much, and contently churn the market. We are still figuring out the exact thing to look at and are running tests on a tweets/google searches per day to trigger a news event. The cool thing about the Neural Network is you can feed just about anything through it and it will process it into usable data.

Also that statistic is based on training data so far. It is not trading live yet as the network is currently being trained.

Because technical analysis can not predict news about prohibition of the ICO in China.

Another question. If your model is so good. Why do not you trade by yourself and take profit? Why do you need ICO?
You can invest your money and get really rich.

Quote from: Thekool1s on September 19, 2017, 10:19:26 AM

You should focus as much as on the news as you focus on the data. Many of people who buy bitcoins are here just for quick profits, the moment they say things going south, they cashout. I agree with ferumflix.

News data will be essential for the AI here is an example, on a pump day, there will be lots of tweets, same is true on a dump date, how will the neural network differentiate. The future trends? The answer is in understanding the context. Will it be able to do that?

To answer both your questions.

Yes, the final version will have a natural language processing system to quantify twitter tweets, and news articles from coindesk and coin telegraph. I have been thinking of a type of "emotions scale" to measure fear vs greed. This would be done on a separate neural net running side by side and feed its output into the main networks inputs.

I would argue the point however that news plays such a large factor. I would even go far ahead as to be able to spot most news events within technical supply and demand zones. We were on the trading floor when the news broke about China, and we followed it very carefully as our fund has a few large positions within the market. We already saw with a high probability there would be a large correction after the market hit the 5000 area. This is a large physiological zone and many traders had large sell orders in the books to take profit off the table. It tested resistance at 4700 again and failed, reaching more supply. Market then followed a head and shoulders pattern and dumped to hard demand at the 3000 area.

Even though yes, the news was significant, I would not have doubted a drop even if China did not ban exchanges. This system of supply and demand plays a massive part on bitcoin trading. This is why we have seen cases of news breaking and the opposite of general consensus or nothing at all has happened.

Even though this system will not be high frequency, most HFT systems completely ignore news. This is because in the millisecond time frame they work on, its all done with statistics and volatility models such as GARCH and news does not really apply on this time frame.

The Bitcoin market does not trade like it used to, there is way more money involved, major firms are starting to take part, and institutional traders are starting to bite.

As far as your question about the ICO goes, right now business activities are being funded through our privately managed fund. We are reinvesting capital from the fund into this, however this space is uncharted ground. No firm I have seen has yet to do research into the kind of statistical analysis we are doing within the crypto currency market. Deep Learning is also a relatively new field of data science, and has not been dabbled with to much when applied to bitcoin trading. To perform the research we would like to at scale, we would need a much larger budget to hire the required researchers, build multi GPU clusters for training, and hire the required programing team.