Initial position adjustment support (DCA)

[xataxxx]

Summary

Implement position adjustment / stacking / multiple buys of the same pair within a single trade.

Partly solve the issue: #2183 and #1519

Quick changelog

• Added callback adjust_trade_position to strategy that can indicate if should buy more and how much.
• Added position_adjustment_enable strategy parameter to enable this feature.
• Updated documentation to document the usage of said features.

What's new?

Implement ability for strategy to buy more of an already opened trade.
The most common scenario for this would be DCA (Dollar Cost Averaging).
The theory is that your buys signals, while good, sometimes still produce trades that first go down X% and then rise to the profit. With DCA you can buy more of the same coin when it's gone down Y% to increase your stake and lower your average open price. This results in higher probability to reach profit AND faster trade times. It can also produce bigger and more hurtful stoplosses.

Done list:

• Strategy callbacks to implement DCA logic
• Additional buy logic to allow DCA
• Trade open price averaging and total amount summing works
• Limit orders works but they follow the same timeout rules as the original buy
• Stoploss on exchange is canceled and new one is created.
• Backtesting seems to work (HyperOpting also)
• Documentation is there
• Unit tests seem to cover basic DCA functionality without breaking other tests

TODO List:

• FreqUI part is not updated yet, which means if you do buy a pair more than once with a big enough price difference, your initial buy open price is displayed way outside of the candle. Best course of action is to start showing the individual filled orders.
• Negative adjustment / Partial sell - Not planning on implementing that currently and documentation and comments in code also indicate that we support buy orders not sell orders. A warning is thrown into log when negative price adjustment is attempted.

[xmatthias]

Pretty darn good job 👍

I'm sure we'll discover some edge-cases as more people start using this - but for the moment i'd not know anything more to do on this PR.

LGTM 👍

[italodamato]

Great job to both of you! I have a question about the select_filled_orders method.

Wouldn't this condition trigger multiple orders if the first order doesn't get filled right away? I think this count should include the potential pyramidal orders created in the last cycle.

As far as I understand, we're counting just filled orders with these statuses which would not select orders we might have already placed at this level and which are still open.

[xmatthias]

Please read the documentation that was added as part of this PR.

this callback will not be called if a order is open.

[italodamato]

Sorry I missed it! It might be worth adding it to the function docstring as well.

[umutrcc]

Will we be able to do partial trading now? How to use it, can you share a detailed usage or a sample study? thank you.

[rokups]

https://www.freqtrade.io/en/latest/strategy-callbacks/#adjust-trade-position

[ghost]

This is a nice feature, but i don't think the example is very useful because profitable trades will only have a small stake, which means only small profits on these trades and losing trades will have high stakes, which means much higher losses are possible.
Maybe using buy signals from strategy to buy more coins would be better, but i don't think it is possible at the moment because pairs with open trades won't be analyzed by the bot - so maybe changing this would be a good idea?

[rokups]

Point of example is to demonstrate how to use this feature. Profitability of samples is really out of scope of technical documentation.

[ghost]

Ok, but do you think we can use additional buy signals to adjust positions?

[stash86]

Anything is possible if you know how to code it

[rokups]

Signal adjustment callback runs on each candle. You may access dataframe from this callback and see if last candle has your signal. See here https://www.freqtrade.io/en/stable/strategy-advanced/#dataframe-access

[ghost]

Thank you for your reply, this was very helpful.
With this i can access all indicators of the strategy, so i just have to put the code for my buy signal inside populate_indicators and can then use that to adjust positions.

[perkmeister]

This post on Discord may help you @incrementby1. As an aside, backtesting is broken if you remove the current_profit check @xmatthias
https://discord.com/channels/700048804539400213/920652305936842752/934926696195444756

[xmatthias]

That link kindof leads to the rules channel ... so not "quite" sure what you're trying to point me to ...

[perkmeister]

Oh sorry, its the Freqtrade Discord server, my post regarding using buy signals (or other dataframe signals) to trigger DCA buys individually. I'll fill in a github issue report regarding the backtesting.

[perkmeister]

That link doesn't open properly, yet it's the link copied directly from the post (in the DCA thread). Odd... Anyway, do a search for DCATest2_fixed.py on Discord.

[xataxxx]

A note to those who might read it later - "backtesting is broken if you remove the current_profit check" is wrong, everything works as expected. If you remove current_profit then you'll just end up buying until you run out of money. It's the responsibility of adjust_trade_position to decide if it's the right moment to buy more - if you don't check your current profit then you better check something else and do not return a stake amount if you don't want to buy.

[perkmeister]

True, but there is no difference between that and failing the current_profit check on all DCA buys. In backtesting this would fail, in live/dry it behaves correctly as it buys as much as it can with DCA and holds the trades open.

[perkmeister]

@xataxxx If you think backtesting isn't broken, run this simple strategy that just buys 3 times then sells and repeats. It should not (IMO) immdiately close trades with a profit of -0.2%, the user is in complete control with max_dca_orders and max_dca_multiplier, it should just buy the max it can and leave the trade open just like it does in live or dry. I'm concerned this is highlightng an underlying logical issue that might show itself if a current_profit check were to fail for each DCA buy, and if adjust_trade_position() is called once per candle and current_profit is the open rate of the next candle this is entirely possible. I may of course be misinterpreting things, but maybe you could explain why this behaviour shows it 'works as expected'?
DCATest2_broken.py.TXT

[xmatthias]

@perkmeister

If you think something is broken, please open an issue with all necessary details (this includes a simple strategy to reproduce the problem, but also output that's produced by it, the command used for this (if necessary some config), and maybe a description what you'd expect and what happened.

Based on the above comment, you'd expect us to first investigate what's actually happening, while then comparing it what you think would happen - which will quite frankly - simply waste our time.

[xataxxx]

Just for info for future readers, @perkmeister issue was actually caused by another tiny temporary bug in codebase that disabled adjust_trade_position completely, not in the actual backtesting functionality.

[perkmeister]

Actually as we said there were 2 problems, it seems the zero length trade issue was an issue and has been fixed in the latest update (only a few days ago it seems), and the latest update has a minor bug with the adjust_trade_position being at config level not strategy level.
@xmatthias I fully intended to raise this as a formal issue with all the outputs and the like (as I said I would in the chat), I just hadn't got round to it yet, but I take your point. I should have done so sooner, my apologies.