why buying high and selling low?
#1
Hello all! This is my first thread here.

I've been testing a lot on the bitstamp pair USDEUR, due the the fact that it consistently trades in an ~%8 channel on the 1 minute frames. It seems this pair is very well suited for automated sniping; but the problem I'm having is that gekko almost always buys high and sells low during testing. I've tried using very short DEMA and MACD crosses, but the problem persists.

I've tried a few different strategies against this pair, including some GA via japonicus, but still seem to have this problem. Other strats on other pairs seem to work better. This morning I've been playing with the RSIBULLBEAR dynamic strat and this seems to work well on other assets.

But how to tune for something that runs in a channel like this?

Here's a shot of the chart - again it's not the swings I'm interested in, but the very predictable individual candles.

https://www.coinigy.com/s/i/5a89e6ab73a68
  Reply
#2
A little more about what I'm seeing

So I tested PPO against the same pair & timeframe, both on coinigy and gekko backtest. On coinigy these super-small (1,3) settings provide a very "blocky" PPO which should be very easy to trade, but running the same settings in a backtest gives the following:

settings:
short = 1
long = 3
signal = 5

[thresholds]
down = -0.01
up = 0.01
persistence = 2

start time 2018-02-17 01:39:00
end time 2018-02-17 19:37:00
timespan 18 hours
start price 1.24353 USD
end price 1.23352 USD
market -0.80497%
amount of trades 50
sharpe ratio -18.07
start balance 101.24353 USD
final balance 85.87304 USD
simulated profit
-15.18170%

Roundtrips
Entry at (UTC) Exit at (UTC) Exposure Entry balance Exit balance P&L Profit
2018-02-17 01:42 2018-02-17 02:02 20 minutes 100.943 100.683 0.26 -0.26%
2018-02-17 02:18 2018-02-17 02:22 4 minutes 100.381 100.080 0.3 -0.30%
2018-02-17 02:27 2018-02-17 02:38 11 minutes 99.779 99.262 0.52 -0.52%
2018-02-17 02:42 2018-02-17 03:31 an hour 98.965 98.824 0.14 -0.14%
2018-02-17 03:41 2018-02-17 03:51 10 minutes 98.528 98.398 0.13 -0.13%
2018-02-17 04:05 2018-02-17 05:16 an hour 98.103 97.645 0.46 -0.47%
2018-02-17 05:18 2018-02-17 05:44 26 minutes 97.352 96.840 0.51 -0.53%
2018-02-17 05:47 2018-02-17 06:07 20 minutes 96.550 96.418 0.13 -0.14%
2018-02-17 06:51 2018-02-17 06:53 2 minutes 96.128 95.711 0.42 -0.43%
2018-02-17 07:03 2018-02-17 07:05 2 minutes 95.424 94.964 0.46 -0.48%
2018-02-17 07:07 2018-02-17 07:31 24 minutes 94.679 94.238 0.44 -0.47%
2018-02-17 07:35 2018-02-17 07:59 24 minutes 93.956 93.833 0.12 -0.13%
2018-02-17 09:24 2018-02-17 09:29 5 minutes 93.552 93.151 0.4 -0.43%
2018-02-17 12:50 2018-02-17 12:53 3 minutes 92.871 92.398 0.47 -0.51%
2018-02-17 13:32 2018-02-17 13:35 3 minutes 92.121 91.696 0.43 -0.46%
2018-02-17 13:37 2018-02-17 13:39 2 minutes 91.420 91.000 0.42 -0.46%
2018-02-17 16:06 2018-02-17 16:08 2 minutes 90.727 90.454 0.27 -0.30%
2018-02-17 16:14 2018-02-17 16:41 27 minutes 90.182 89.914 0.27 -0.30%
2018-02-17 16:56 2018-02-17 17:00 4 minutes 89.644 89.387 0.26 -0.29%
2018-02-17 17:06 2018-02-17 17:10 4 minutes 89.119 88.848 0.27 -0.30%
2018-02-17 17:18 2018-02-17 17:20 2 minutes 88.582 88.316 0.27 -0.30%
2018-02-17 17:25 2018-02-17 17:29 4 minutes 88.051 87.787 0.26 -0.30%
2018-02-17 17:37 2018-02-17 18:23 an hour 87.524 87.236 0.29 -0.33%
2018-02-17 18:34 2018-02-17 19:16 42 minutes 86.974 86.618 0.36 -0.41%
2018-02-17 19:31 2018-02-17 19:33 2 minutes 86.358 85.873 0.48 -0.56%

..and it's always like this, NEVER one profitable trade - if the buy/sell order were reversed, it would seem near perfect.
  Reply
#3
what fees do you setup the papertrader with?
  Reply
#4
So far I've just left them stock, so:

feeMaker = 0.25
feeTaker = 0.25
feeUsing = 'maker'
slippage = 0.05

[simulationBalance]
asset = 1
currency = 100

with some more testing (RSI with very low setting but very high thresholds) I can get gekko to plot long/short "advice" very well, it looks like it should be working, but for some reason %100 of trades fail.
  Reply
#5
Buy high/sell low strats are so much easier to find.
And they would be actually brilliant if you could reverse the buy/sell signals. I too think you're on to something there.
Try the MACD default and see for yourself. The trading points are not bad, except that they should do the exact opposite. Coincidence?
  Reply
#6
if you dig into the code you can reverse it. I managed to do this with NN start to test a hedging strat (didnt work). You need to alter the long and short trigger
  Reply


Forum Jump:


Users browsing this thread: