Contributions Business Model

[rmendes900]

Hi @owocki ,

As I told you I was checking your code. I dint deploy yet, but I will do it. I have a few knowledge at AI and I was trying to understand. Anyhow, i think maybe you are teaching with poor info your machine, as I understood, you are doing that with the Poloniex Trading Book behavior.

• First -
  As you are trading BTC/USD and BTC/ETH pair you might get the better info for that. As I understand, the job here is to make the bot predict and take the best decision and execute. The financial market anticipate the moves, so, you have to give the bot a good source, and for this case you should work with OKCOIN, KRAKEN, BITFINEX and BTC-e, because Poloniex traders are following those companies Book, so if you have this source of learning combined with Poloniex Book, you will execute first than the others and probably will have more profit.
• Second -
  Instruments
  Spot, Futures and Swaps have different behaviors and trade mechanics.
  As your bot (by now) is working only with Spot doing arbitrage between pairs, you have to first, decide witch pair will be your net asset reference (BTC, USD, ETH.....) because doing arbitrage you only increase your asset in one.(exemple: imagine if you buy today 1 BTC for 420 and at same time you by ETH with those BTC's, and ETH goes UP 3% and you sell, but at the same time BTC goes down 12%. If your Asset in reference is USD, you are loosing money).
  For Futures and Swaps is other discussion, and we can do it further.

-Third -
Size
Your size (Amount of Trade) is very important and is part of you Risk Management.
Some Trade Books from some Exchanges doesn't have enough liquidity to your minute/hour/daily volume, and if you push too much executions at the book, you can disturb the market behavior compromising a bullish strategy (Example. You have 10 BTC to be liquidate, and the bot knows that he can start to sell, but the market is not confident yet to place big orders at that level, if you place an "anchor order" (big order) or start to execute big orders, you can clean the buy side book before it came at the level that you need)

• Fourth-
  Timeframe
  The Timeframe is extremely important specially at the Spot doing arbitrage, and might be considered according to your size.
  You have to analyze and teach that to the bot, the 12h, 6h, 4h, 2h, 1h, 30min, 15min, 5 min and 1 min. Timeframe, in order to understand the Pivots, Support, Resistance and Trade Volume. Those information are crucial, specially at the arbitrage. You cant allow the Bot to buy at the top of resistance or sell at the support (depends of the Market Trend). You have a lot of reference of this at tradingview.com and I can share some charts to you to understand better.

Well, has a lot more to include on this machine, but to make this working fine, you already did a lot and very nice job. If we start to follow those statements we can improve your code and make a money machine.

As soon as I deploy your code I let you know.
RM

[NDuma]

@rmendes900 Good call on money management = rule # 1 in trading.
BitFinex has a data portal ( bfxdata.com ) which has some nice indicators in it including short positions etc...

Market Depth to handle large trades is a consideration as pointed out re: Size and market's ability to handle particular volume. And can add strength to finding 'Support / Resistance" levels.

The more people who are reading the 'tea-leaves' the same way the more the market will behave in step with them; whether it's EMA's Fib Levels MACD or any other lagging indicators.
Leading indicators are a bit harder to come by.

[owocki]

@rmendes900 this is all very interesting, the kind of expertise I was hoping to get out of open sourcing this repo :)

Point 2 - I've been treating BTC as the net asset reference implicitly. To your point, this is not a perfect solution, I could see putting some time to make this a configurable parameter on the roadmap.

Point 4 -- Timeframe is actually in the system already. It's called granularity.

There's a lot of ideas popping up in various conversations and I need to spend some time getting organized about where to contribute my coding time next, and where to suggest potential contributors spend their time. What do you and @NDuma think are the most promising places to spend development time right now?

[rmendes900]

Nice @owocki , you can count with me to the project, I am 100% engaged.

Yes, by now as you just opened the code, will appears a lot of ideas and discussions and you gonna have to drive the ideas and discussions to the Final Project. I advice you to take as much info from everybody, organize in your mind e determine a Goal to this Project. Otherwise, you will start to develop and implement things that you not gonna need in the future, will take a lot of time and effort from everyone in here and probably unnecessary discussions.

In my point of view, you started a strong tool, but has some concept issues to the source info and decision take. I consider you prioritize this, because if you teaches right your bot will take the best decision in most of cases. Second, the analyses, trading a 12h timeframe using MACD is completely different of Trading 4h MACD, the bot has to take decisions based on what does it doing, Scalp, DayTrade, SwingTrade, LongTerm). This is crucial because the PIVOT's of support and resistance change.
Anyhow... count with me on the Project, for businness model and Code, I will deploy tonight and take a seriously and deep look inside and propose to you where can I start.

Also feel free to ad me on skype or other pm service to reach me anytime.

[rmendes900]

Hi @owocki , soory for the delay, but I am very busy with some activities here.
I was thinking here, maybe we can add some big data tool to process the data as Hadoop and delivery to machine learning. What do you think?

[owocki]

Hey @rmendes900 if it's an idea that has a high likelihood of getting us to profitability, then I'm all for it. To me, this repo is all about experimentation and figuring out what gives us an edge in the market. Lets talk about specifics on the slack channel.

[massenz]

A much "shallower" comment than @rmendes900 (awesome contribution!) but I was thinking:

My test portfolio was initialized with a 1 BTC deposit, and after 2 months and 23,413 trades, exited with 0.955 BTC. The system paid 2.486 BTC in fees to poloniex.

could this be because of the relative ratio of "volume" to fees?

Not sure how Poloniex charges (per trade or as a % of value traded?) but I can imagine that, if I invest $10 with e-trade, and they charge me $9.99, I need my shares to double in value, before I start making a profit; obviously, if I invest $100, they only need to go up 10%.

This is very interesting to me (about 6-7 years ago, when still at Google, I had a similar idea, but eventually abandoned it, as there really was not available ML software at the time) so I'm planning to contribute to the code (and the conversation) and help out.

If you are able to (ie, have the necessary data) may be a good idea to see what the outcome would have been if you had started with a 1,000BTC portfolio: would you have ended up with a 955BTC one, or a 1,955BTC one, and 2BTC in fees?

[owocki]

@massenz @rmendes900

I'm thinking at this point, it might be worth doing a deeper analysis of my results beyond just a simple count. This came up over on #5 (comment) as well, and there isn't really enough information to know exactly how promising my results were. Here is an issue I created to track that. #32

I'm hoping to put in some real cycles to this project this weekend. I'll be fleshing out exactly what direction take things on the pytrader slack channel (signup is here: #23)

[NDuma]

@massenz Most charge as a % of the trade itself; Poloniex is 0.15% maker fee & 0.25% taker fee; other exchanges are lower and/or free for maker fees.
Optimizing a strategy to prioritize making the market would be beneficial over the long run; and, in this scenario may have resulted in a net positive.