using Gekko in the real world
Hi everyone!

I am so amazed by Gekko and everything it can do, and (like many of you, I'm sure) I've also been frustrated sometimes trying to get it to behave as I want it to when actually running in the real world. I just wanted to query the community to see if anyone has insights/ideas on how to use Gekko once you've done all your backtesting and paper trading and want to run it live.

One big thing: I run Gekko on a raspberry pi, and I use pm2 to keep it running all the time (restarting automatically after program crashing, power outage, etc).

npm install pm2 -g
cd gekko
pm2 start gekko.js
pm2 start gekko.js --ui
pm2 startup systemd

which is great, as far as it goes. But if I'm running a live strategy, that strategy does not restart automatically... and that's kind of 95% of the point for me of keeping Gekko running, so I'm not sure what to do there.

Another big thing I've noticed, this related to strategy design: in smaller markets (where, it seems, it's generally easier to find opportunities for big gains), the default value for slippage is way too low. I generally adjust it to 0.3 for the markets I'm interested in, but that's more of just a feel-it-out number based on looking at the general size of spreads in the order book and guessing. I've searched and have not found a scientific method for calculating slippage. What do you guys do?

Related to that, in backtesting it's not too hard to use 1 minute candles and come up with a high-frequency strategy with insane percentages, but try to execute that irl and you will get badly burnt. Most of my strategies use 15 minute candles, and when I look at each roundtrip I try to see if the trades could start and end 15 minutes later without losing anything in the wash. Generally that seems to be a good sign that the strat could be executed for real, as when you place a limit order on the books in a small market it often takes that long to be filled. I'd be curious again if anyone has similar/different ideas on how to tell if a strategy could actually work.

I recently found @SirTificate 's neuralnet strategy and I'm obsessed with it now. If you want to see something crazy, run it on ETH-ARK for the last seven months with 15 minute candles, default values except change learning rate to 0.1. Has anyone used this live? I'd love to know how it worked out.

And in general, if you use Gekko live with real money and have any tips, please share! Thanks all!

Attached Files
.png   Screen Shot 2018-06-04 at 8.52.00 AM.png (Size: 55.55 KB / Downloads: 54)
pm2 startup systemd

Forum Jump:

Users browsing this thread: