Hi,
Today I saw that gekko now supports GDAX under the branch develop. I'm switching to gekko to see how it goes. There's more active community there.
Zenbot is a great project but, it seems that anyone is maintaining it anymore.
Good luck!
albertolive
All 14 comments
I have made working branch completely supporting Bitfinex.
nedievas
on 4 Mar 2017
I see your branch, does it still work on GDAX or did you have to remove that logic to get it to run on Bitfinex? I could spin up a second docker instance, but it'd be handy if the same bot could trade on both (maybe perform arbitrage in the future).
praeluceo
on 5 Mar 2017
I had to rewrite default_logic. Still, you may have logics separately, just use separate configs for each one.
nedievas
on 5 Mar 2017
@albertolive is this profitable than gekko?
arruah
on 31 Mar 2017
@nedievas If I just add the configuration files from your brunch here? Will this work?
arruah
on 31 Mar 2017
@arruah You may try.
nedievas
on 1 Apr 2017
@albertolive It might interest you that I'm now working on zenbot 4.x, see #134 , will be a complete rewrite of zenbot.
I tried gekko recently and was a bit let down.
First, I couldn't make a simulated profit using the built-in strategies that came close to buy/hold. I think it's asking a lot for average users to come up with and implement a winning strategy just to make any amount of profit, or else you have the silly scenario where it would be more profitable to NOT use the bot, haha.
Second, it appeared that when it places an order, it uses the exact ask/bid price, increasing chances that it will be filled immediately, and thus subject to taker fees (and indeed, the paper trader assumes every trade will have a 0.25% fee). Ideally a bot tries to avoid fees by marking up/down the price slightly and re-adjusting the bid periodically. I made the mistake of not doing that with zenbot 3.x, and it's a big difference. Consider that on a $1200 trade, 0.25% is $3. With 90 days of trading, 1 trade per day, is $270 of fees, or 22% of your capital! So it's clear that a bot that accumulates fees is going to barely break even, and defeats the "passive profit" goal of bot trading.
Third, the backtester iterates candles, not trades, making it impossible to model a limit order strategy. You have to assume that the trade will go through at the candle's close price, which assumes market-type orders, which assumes fees (see above point for why a bot that assumes fees is bad). Another mistake I made in zenbot 3.x, and fixing in 4.x. In 4.x the simulator iterates trades, enabling real modeling of limit orders and slippage.
I'm curious what you guys might find better in gekko, so I can try to incorporate that into zenbot 4.x.
carlos8f
on 1 May 2017
It's good. In what time frame can we expect a new version? What about implementing the capabilities of a neural network? As I understand it does not work now.
arruah
on 1 May 2017
@arruah see #134 and #128
carlos8f
on 1 May 2017
Can you offer paid support?
arruah
on 1 May 2017
No, zenbot is for now a hobby-level project, and I don't offer support
carlos8f
on 1 May 2017
Hi @carlos8f, it's good to know you are back :) I tried gekko but I quit for things you commented so I'm exited to see the new zenbot. For your info, with RSI and default logic I won 700$ but I lost all after that.
albertolive
on 1 May 2017
@carlos8f Anyway, I made some donation. :) If you are interested in the future in providing paid support. arruah @ gmail.com
arruah
on 1 May 2017
@arruah thanks for the donation! Zenbot 4 is out now, please give it a try.
carlos8f
on 22 May 2017
Related issues
MCrypto
路
5Comments
dymex0
路
3Comments
mrzobot
路
3Comments
michaelr524
路
3Comments
ituhin
路
3Comments
Most helpful comment
@albertolive It might interest you that I'm now working on zenbot 4.x, see #134 , will be a complete rewrite of zenbot.
I tried gekko recently and was a bit let down.
First, I couldn't make a simulated profit using the built-in strategies that came close to buy/hold. I think it's asking a lot for average users to come up with and implement a winning strategy just to make any amount of profit, or else you have the silly scenario where it would be more profitable to NOT use the bot, haha.
Second, it appeared that when it places an order, it uses the exact ask/bid price, increasing chances that it will be filled immediately, and thus subject to taker fees (and indeed, the paper trader assumes every trade will have a 0.25% fee). Ideally a bot tries to avoid fees by marking up/down the price slightly and re-adjusting the bid periodically. I made the mistake of not doing that with zenbot 3.x, and it's a big difference. Consider that on a $1200 trade, 0.25% is $3. With 90 days of trading, 1 trade per day, is $270 of fees, or 22% of your capital! So it's clear that a bot that accumulates fees is going to barely break even, and defeats the "passive profit" goal of bot trading.
Third, the backtester iterates candles, not trades, making it impossible to model a limit order strategy. You have to assume that the trade will go through at the candle's close price, which assumes market-type orders, which assumes fees (see above point for why a bot that assumes fees is bad). Another mistake I made in zenbot 3.x, and fixing in 4.x. In 4.x the simulator iterates trades, enabling real modeling of limit orders and slippage.
I'm curious what you guys might find better in gekko, so I can try to incorporate that into zenbot 4.x.