⚡️ ROI improvements

[Rikj000]

The MGM community has concluded that ROI in it's current state brings no improvements at all!
It reduced profits & increases drawdown in HyperOpt Results reported on Discord.

In it's current state I recommend leaving it completely off, however I believe we should not give up on it yet as our means to support the sell signals (that aim for selling off with profit) while reducing risk and drawdown!

Thus I propose following improvements done to ROI generation:

☑️ #85: Make it configurable in which trends ROI is allowed to trigger.

☑️ Generate the ROI on a slightly bigger candle size (perhaps configurable, perhaps auto-set) then the weighted buy/sell signals, because now ROI tends to seek to reap profits too fast, resulting in being the main sell reason, but also in decreasing the total profits made when it's enabled. Increasing the candle size for ROI only should lead to it decreasing over a longer timespan, which will hopefully balance things out.

☑️ Introduce a configurable/hyperoptable gap (roi_delay) before ROI would start counting down. This would be another way of giving the sell signals more of a chance to act as the main reason for selling off profitable trades

[Rikj000]

Since my last update to this issue I implemented the configurable roi_time_interval_scaling & roi_value_step_scaling settings.
These allow to fine-tune ROI more, and make it less bad.

However it still didn't prove to bring any benefits compared to optimizing MGM without ROI.
I would still keep away from it until above improvements have been implemented.

[Rikj000]

Slowly but surely ROI is changing into something decent with these new changes! 🔥
I haven't played with much of the config possibilities of the new timeframes section yet so feel free to do so!
Perhaps we can still improve further upon those! 🚀

(All of these tests where done with roi_when_trend already implemented and a timeframe of 30m)

Show/Hide Test Results

HyperOpt - ROI Timeframe = 30m

HyperOpt - ROI Timeframe = 4h

BackTest - ROI Timeframe = 30m

BackTest - ROI Timeframe = 4h

[Rikj000]

Happy to announce that the new roi_delay setting (Not yet committed) is another improvement! 🥳

Following test's where done using:

• HO1 spaces buy sell
• HO2 spaces buy sell roi+ roi_delay 120

Show/Hide Test Results

HyperOpt - Results

BackTest - Sell Reasons

BackTest - Summary

[Rikj000]

Test 1

For this test both HO 1 & HO 2 have been done on new random states using spaces = buy sell roi + roi_delay = 300

If we compare these to my previous set of tests then we can notice that it doesn't seem ideal to already HyperOpt ROI on the initial HyperOpt 1 Run.

Show/Hide Test 1 Results

HyperOpt - Results

BackTest - Sell Reasons

BackTest - Summary

Test 2

For this test I used the same random states as in my previously posted tests.

• HO 1: Has been done using spaces = buy sell + unclogger
• HO 2: Has been done using spaces = buy sell roi + unclogger + roi_delay = 300

These results are better then Test 1, however I think I still prefer the results of my previous post a little better
(Basically same test with roi_delay = 120)

Show/Hide Test 2 Results

HyperOpt - Results

BackTest - Sell Reasons

BackTest - Summary

[Rikj000]

roi_delay implemented in: 7c08e91...5b52cd9

Please test out various possibilities with the new ROI settings!
I so far just tested a handful of configuration possibilities so there's a good chance that we can still improve upon this section! 🚀