Home Creators Posts Import Register Favorites Logout
haven't archived this post yet. have a subscription? use the importer!

Downloads

  • averagestrategy-berlinguyinca.7z

Missing 1 file, as this post has not yet been imported.

Content

In this post I will give you all the output I got from backtesting the "Average strategy" which is posted on my blog page and Youtube video. 

This output is in the file which you can find in this post.

For you I also have additional information about the way I tried to optimize this strategy.

The first time I optimized this with the Freqtrade Hyperopt function I used the SharpeHyperoptLossDaily. But this gave pretty bad results (which means here that it actually still produced a winning trade strategy but for some reasons optimizing the parameters made this strategy not perform much better than the default values).

Because I studied more about the difference between Sharpe and Sortino ratios, my preferred way of determining if a strategy is good or not slowly moved towards the Sortino ratio. 

Luckily the Freqtrade bot also has the possility to optimize with the SortinoHyperoptLossDaily setting, I also did a complete optimization run with this setting.

So in total I had to let my PC run for about two days to do two 1000 epochs optimization runs. 

These two optimization runs produced the backtest results you see below:

You can see that the Sharpe optimization settings totally did not produce good results. All the ratios and factors produced worse results (however the Sharpe ratio stayed the same though). 

This is why my preferred hyperopt setting for optimizing strategies like these is the Sortino (daily) setting.

Again, all backtest results are in the file as an attachment to this post to be as transparant as possible.

If you have any feedback, please give it to me through this channel as I am starting to react less and less to youtube comments. It is too much to respond to and I have to make choices to wich public I will respond to.

THANKS FOR YOUR SUPPORT AND GOOD LUCK FINDING THE BEST ALGO STRATEGY FOR TRADING BOTS!


Comments

Malcolm

Thanks, good points. I have moved to using the MaxDrawDownRelativeHyperOptLoss loss function. I find drawdown is what kills my runs - if there is 50% profit, but 75% drawdown it feels kind of pointless since it's dangerous and difficult to sit through watching the bot "lose" a pile in the short term in live trading. So I'm more than happy to trade a bit of profit for less drawdown. It feels a lot better live trading with a low drawdown strategy.

Patrice Savard

How do you define the spaces for hyperopt? I cant figure out how on this strategy

Dutch Algotrading

This specific strategy was created by mr Wohlgemuth. These specific spaces are chosen by him. To define spaces yourself you can start with the 'usual' value that is normally used with the indicator and create a space around that. IE.g. buy_range_short = IntParameter(5, 20, default=8) from this strategy uses the 8EMA as the default value and the space around it is 5 to 20. I hope This makes it a little bit more clear.

Marc Forn

I've tried to reproduce the 1d results with the data provided and I'm a little deviated from 1299% (getting 1000%). Here you can find the config: run freqtrade backtesting --strategy AverageStrategy --timerange 20170906-20230101 and the coins used are the ones extracted from your files. What I'm not doing well?

Dutch Algotrading

Command looks good. Only I use --timerange=-20230101. Also using another exchange can differ in fees. Sometimes looking at the pair and exit reasons can give a clue about what's different too. Maybe the dataset is different... I added all my results in export files as well as a final suggestion to check the differences.