Bitcoin and Deep Neural Networks, Introducing Evolution One.

[QuantumTrading]

Hey guys,

We are looking for some feedback on a project currently under development.

Quantum Trading develops HFT software, and artificial intelligence for the forex, and equity market however we have been increasingly interested in applying this to bitcoin.
I myself have been using bitcoin since the very early days, and noticed a few things that are very specific to bitcoin. This makes it hard to trade through traditional methodology.

After a few years of research we have come up with a model that is currently being back tested, but so far has shown in upwards of 96 percent trade accuracy. The software is
using a deep neural network to be able to learn from the market at an extraordinary rate.

The network is currently being trained, so of course as the epoch count grows higher, the network will get better.

The project we have launched has been named Evolution One. This is a revolutionary new type of crypto currency portfolio management system, and trading platform.  The software will come as two parts, the stand alone trading platform, and the virtual management service.

The stand alone platform will allow traders to have better control over there trading activities, without the need to rely on being signed into multiple exchanges and there trading platforms. Evolution One allows you to manage all of your accounts in one place.

If traders decided to also take advantage of the Evolution One Virtual Management Service, you will enjoy advanced intelligent management of your whole portfolio, advanced statistics on the bitcoin market, and more. All run on our deep neural net as a SaaS offering.

I am planning on making the trading platform open source, allowing the community to interact with the neural net, and develop your own plugins. The platform as a free service, will allow you to trade and manage crypto currency trades, but will not have the advanced intelligence of the neural network backend.

The platform is going to be rolled out in a stepped service as research into this area is still being conducted. Most likely, the service will launch with a intelligent BTC management service first, as we are still developing the models for portfolio.

The neural net will be closed source, as our models are designed in house and are proprietary. Committed to the advancement of artificial intelligence however, all research will be open and we encourage submissions from the public to our gitlab repo and research portal.

Quantum Trading is currently also looking to do an ICO for this project, however we still need to figure out the details with this. We may introduce some sort of dividend product once we open our quantitative fund to the public and become registered and regulated in Canada.

Let me know what you guys think about this

[1714854756]

1714854756

[1714854756]

1714854756

[Mangiafuoco]

What are you mean "96 percent trade accuracy"? Overfitting?

[QuantumTrading]

What are you mean "96 percent trade accuracy"? Overfitting?

https://quantumtrading.ca/images/129-14-17.png

This is not final results, but as the network takes in and processes data it learns from it. It is using a combo of a deep nueral networks, and re-enforcment learning.

As the program learns from the data it digests, it theoretically can get trade results that are this high.

I will be releasing a whitepaper detailing this project in the next few weeks.

I will also be opening up research to the public in the next while, once we have more data from our tests. Training something like this with as much data as we are takes a long time. But with the new GPU cluster I have plans to build it should speed research up.

[Thekool1s]

Wow, if you guys can pull this off, it will be a game changer. I have a few questions, from where are you getting your data? From which exchange? And how many years of data is the AI is being trained on? Really excited for this project. Good luck!

[QuantumTrading]

Wow, if you guys can pull this off, it will be a game changer. I have a few questions, from where are you getting your data? From which exchange? And how many years of data is the AI is being trained on? Really excited for this project. Good luck!

Thanks for the support!

Right now testing and training of the prototype is with data from Kraken, and we will be offering kraken only support in the first prototype for release. This allows a launch within an acceptable timeline. The network is trained using data obtained from Quandl, a financial data provider who has a built in Python API. We are going back to the creation of Kraken for the duration of the data.

The final version will have many datapoints it will use. Inputs to this include current and historical price data from major exchanges(any that support websockets within there Api, and ones who have data accessible in Quandl)and other datapoints such as bitcoin mining hashrate, total google searches, transactions per day and so on.  The duration of training will depend on how far back data goes, but the system will train from the date of data creation to current.

If you guys would like updates on this project, and would like to support us, please take a min to sign up on our mailing list.
http://quantumtrading.us15.list-manage.com/subscribe?u=3d6d446d17041fea4e03b26c6&id=6f8babeb17

[ferumflex]

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.

[QuantumTrading]

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.

[Thekool1s]

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?

[ferumflex]

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.

[QuantumTrading]

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.

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.