Machine Learning Algorithm to Trade Bitcoin

I am wondering if you can trade Bitcoin on a trading platform like Coinbase.com or Avatrade, using your own algorithms? It looks like it is very volatile, the volatility seems to be driven by human behavior and emotions (which can be modeled and arbitraged, unlike purely random variations) making it an ideal candidate for high frequency trading.

My questions are:

Is the spread between buy and sell reasonable?
Can you short-sell it?
Is it possible to find historical daily data to design and test a trading strategy?
Can you place stop and limit orders, or is the process mostly manual?
How much you should play with? If you start too small, your gains will be eroded by trading fees.
Do you think the US government will eventually make it illegal to trade it, like what happened in Russia? More regulations could make it less volatile, and thus less interesting for a pure day trader.

I developed trading strategies in the past (see here) and I am interested to get back on the market, with the right type of commodity. Your feedback is appreciated.

DSC Resources

Services: Hire a Data Scientist | Search DSC | Classifieds | Find a Job
Contributors: Post a Blog | Ask a Question
Follow us: @DataScienceCtrl | @AnalyticBridge

Replies to This Discussion

Permalink Reply by Serzhan Akhmetov on October 26, 2017 at 9:18am

I can’t really speak on the last question but here’s my experience with bitcoin.

Most exchanges (bittrex, CEX, Kraken, Poloniex) are very similar in terms of spread. Max 5%.

Short selling (margin trading) is possible (e.g.: cex.io with 1:2 to 1:3 leverage).

Sites such as quandl.com offer historical data for a price.

You can place stop and limit orders, but on most platforms this is done through the website’s interface. Good thing is, a lot of exchanges offer access via APIs too, so you can use your own program to execute orders.

Trading fees are 0.2% - 0.5%, I think you can start trading with 10$ XD.

▶ Reply

Permalink Reply by Jonathan Armstrong on October 26, 2017 at 12:24pm

The problem is that the exchanges are not very reliable. I can't speak to all of them but I know Kraken and Coinbase have had serious delays and problems in the past.

▶ Reply

Permalink Reply by Prem Hirubalan on October 30, 2017 at 6:54am

Hi Vincent,

From experience,
- each exchange has it’s own bid/ask (no NBBO for bitcoin yet) so best bet would be to get historical data from each exchange
- BTC lends its self for HFT trading in arb, due to various prices on each exchange
- Fundamentals are driven by alt data(sentiment, transaction times) so using ML to drive those strategies could be interesting

▶ Reply

Permalink Reply by Vincent Granville on November 11, 2017 at 3:40pm

Here is an interesting answer from one of our members:

I recently joined the site and saw some really interesting article on here. I also saw that you posted some questions about bitcoin trading. I am currently algorithmically trading them so thought i could answer your questions.

1.Is the spread between buy and sell reasonable?

yes in most exchanges the minimum tick size is about 1 USD in bitmex its currently 0.1 USD but they are changing this to 0.5 USD. I think more of the issue with coins is to ask what the spread is to execute x size since this can vary quite a lot. The books can get very thin when the market is moving so where as the best prices are close, if you wanted to say sell 5 coins it could cost a lot.

2. Can you short-sell it?

Yes, there are even futures exchanges now that also let you short sell them. Bitmex and bitfinex both have shorting capability.

3. Is it possible to find historical daily data to design and test a trading strategy?

Yes but you need a bit of programming ability. You can get quotes (top of book) and trades historically from the web apis see: https://www.bitmex.com/api/explorer/ and https://docs.bitfinex.com/docs. Its possible with their apis to get historic data but it takes a while since you can get about 1000 records per query and they rate limit the number of queries you can make. It takes me 12 hours to get about 1 month of trade and quote data. You can get data for aggregated candles which wont have this issue

4. Can you place stop and limit orders, or is the process mostly manual?

Yes the apis allow for fully automated trading.

5. How much you should play with? If you start too small, your gains will be eroded by trading fees.

It can be expensive the fees are variable depending on maker/taker and volume traded. For bitfinex its about $14 for a coin ($7k). In bitmex if you don't cross the spread they pay you about $1.6 so it depends on the exchange also. There is also a new exchange that is in ICO called COBINHOOD which claims to be fee free. I am sure there is a catch somewhere, most likely in their margin fees but i would suggest if you trade on the right exchange you can make money with as little as 1000 bucks.

6. Do you think the US government will eventually make it illegal to trade it, like what happened in Russia? More regulations could make it less volatile, and thus less interesting for a pure day trader.

I don't know about this and its really unclear. So far people that have been getting into trouble are the ones exchanging cryptos for fiat not the other way around. The ICO space is also realy dodgy at the moment. This is probably the biggest event risk to the market right now.

I am an quant researcher/trader that used to work professionally in the CME and Eurex for different banks. I would be interested to get your take on different ideas for modeling techniques. I can help with your data acquisition, implementation i return.