Boost Your Wealth with the Combined BinH and Cluc V8 Hyper Trading Algorithm! (Patreon)

Introduction

Hi there, and welcome to this post where I test yet another high performing trading algorithm.

This time it is the Combined BinH and Cluc V8 Hyper trading algorithm. I can already tell you that 90 percent of the 50 pairs I test on, respond very well to this algo. 

Where to find it

As with many algorithms I currently test out on their merits, this one also came from Github.

If you search for the name of the strategy, you will find multiple repositories that contain the original code. Code search results (github.com)

There are also many variations to find of this algorithm. 

And in this video I will concentrate only on the one that has the best results on my specific backtesting setup. And as you already know, that is the Hyper version.

So lets open the code and see what’s inside…

According to the text in the code, this algorithm is inspired by the CombinedBinHAndCluc strategy made by Iterativ.  So if you like this algorithm, the original author deserves all the credits. I’m just the guy that test it out and show you my results on Youtube.

Now, I tested that original code a while ago and currently it sits just above the middle in my current Strategy League. Strategy League - Dutch Algotrading

And since I already discussed the inner workings of the main algo in that video, I will only do a high level overview of the strategy in this video. 

raw.githubusercontent.com/PeetCrypto/freqtrade-stuff/f6c38def0b2fe01d8d42e47740cdeab53511527b/CombinedBinHAndClucV8Hyper.py

The strategy

Let’s first talk about something that I thought was a coding error at first.

At the start of the code, after the SSL indicator has been defined, you will see strategy class. At the very start of this class, the author has defined the following settings for ROI and Stoploss.

Also the trailing stop loss has been set to False here.

However, if you scroll a little bit further down, there is a second set of settings that is used in this algorithm.

These settings effectively overwrite the initial settings. And therefore might confuse you if you use this algorithm on your own setup.

But yYou can also see which exact settings are used if you look in the log of your bot if you backtest this strategy.

Now here you see that the second ROI is used at the minimal ROI line. And also that the second stop loss and trailing stop has been set.

So in this case the strategy uses the multi timeframe ROI, A stop loss of -27.4 percent and a configured trailing stop.

So the reason I think this has been coded by the author like this is that he or she initially used no ROI, Stop loss and Trailing stop loss. And later decided to do hyperoptimization. And these settings were then added to the original strategy file to keep everything in one place. Hence the added Hyper in the name. 

So besides the hyperopted results that I initially tested, I will also show some of the results with the initial settings. So no TSL, No ROI and no Stop loss. Just the bare trading rules that are in this file.

Now speaking about hyperoptimization, these are the settings you can alter to determine of a setting should be taken into consideration when optimizing and strategy execution.

And proceeding further down the code. The algo also makes use of a custom stop loss and a custom sell method.

I think it is highly probable that the optimized settings for ROI and stop loss settings get in the way of these defined methods here. But since it is performing well I don’t complain about it.

There is use of informative pairs on the 1 hour timeframe and informative indicators on the 1 hour timeframe.

There is also a method where the normal timeframe indicators are defined.

And the usual indicators method makes use of the dataframe that is created by that normal timeframe indicator method. 

From that moment on the Buy trend signals are defined, and you can see that multiple sets of conditions create multiple buy signals. Each one focused on specific circumstances where the author thinks that a buy should occur.

For example, this buy condition in the strategy combines multiple checks to determine an optimal buying point based on trends and volatility:

It checks that the current price and the shorter EMA (50-period) are above the longer EMA (200-period) on both the 5-minute and 1-hour timeframes, confirming a strong uptrend across both intervals.

The condition ensures the price is within a defined threshold from recent highs, indicating buying on dips within an upward trend. It checks this with two rolling maxima over 2 and 12 periods, comparing them to the current closing price adjusted by specific threshold values.

It also ensures the price is below the lower Bollinger Band from the previous period and has decreased compared to the previous close, indicating a potential rebound opportunity. 

It also checks that the Bollinger Band delta and the close delta  are significant relative to the current price, while the tail is less than the Bollinger delta, suggesting a short wick under the body pointing to less selling pressure.

Finally it validates that the current volume is above zero, to ensuring there is trading activity to support the buy signal.

As for the sell signals, these are much simpler and that’s understandable because the main focus for the selling of open positions is focused on the defined custom stop loss, custom sell, roi and stop loss settings.

Now, next to the original strategy you can download from GitHub, also I did a test without the trailing stop loss set and with the ROI and Stoploss disabled. To see if this has effect on the profitability of this trading algorithm.

So let’s get into the backtest results of this algorithmic trading strategy…

Backtest results

As you can see. The performance of the strategy is pretty good on the 5 minute timeframe.

With the initial investment amount of 1000 USDT it could get a hypothetical 1304 percent gain with an average win percentage of 75 percent and a max drawdown of only 5 percent. The other thing that impresses me is the high percentage of the pairs that respond well to this strategy.

In other words, this specific algo could be used on a large range of pairs.

The equity curve of this optimized strategy also looks very promising for future use. With its nice curve that does not seem to suffer from high drawdowns.

Which we already have seen in the overview ofcourse.

On average there seems to only be a drawdown of 1.3 precent. 

As for the winrate and profit distribution, which tells a lot about the individual trades.

The boxplot on the left displays a median winrate around 0.7, with the interquartile range (IQR) primarily between approximately 0.5 to 0.8. This indicates a relatively consistent and high probability of winning trades. The outliers at the low end suggest there may be specific conditions where the strategy performs poorly.

The boxplot on the right shows a median weekly profit close to zero but remains positive, with a wide interquartile range that spans both losses and gains. This suggests variability in weekly returns. The presence of several outliers with substantial profits indicates occasional large winning trades, while the bulk of the trading results cluster closer to break-even.

Now, these plots suggest that while the strategy often wins more than it loses, the profit per trade might not be very high, leading to a profit distribution that centers near break-even with occasional large gains. But if you add all things up, then you can see that the mean profit and loss overall stays positive over the duration of the backtest. Resulting in the performance you saw a couple of slides ago.

Now how would the performance of this algorithm look like if I disabled all these optimized settings and kept it to the bare algorithm itself.

NoTSL

Well, what do you know…

Without these optimized settings the algorithm performs even better. On almost all aspects in this table. Higher end result, higher win rate, higher pairs ratio. Only the drawdown suffers a little bit from it. But not too much.

The equity curve is equally impressive, although not as smooth as the optimized version.

This results in a slightly more jagged drawdown plot. But I do think it is still very admirable to see such a good performance.

Looking at the two boxplots of the winrate distribution and profit distribution. These also show better results. 

So after all in my specific case, optimization isn’t that optimal and can even worsen the profitability of a strategy. This is actually the reason why I keep optimization of strategies to a minimum lately.

Lookahead bias

I also tested the code on possible lookahead bias. But as far as the function to detect this bias could search for, it didn’t find any.

So with this check also out of the way, where do these optimized and none optimized enter in my Strategy League…

Strategy League & conclusion

With TSL Strategy League

The original strategy enters the League on the 9th spot. 

As I said in the beginning, I tested multiple versions of this algo that I downloaded from the repositories and you can see that they all perform somewhat similar.

NoTSL Strategy League

But the surprising part is that the version that does not use the optimized parameters or trailing stop loss actually performs better on my setup.

And so it enters higher in the league then the other versions overall. Which is an unexpected, but pleasant surprise.

Ending

So with this conclusion I am at the end of this post.

I will include all the versions I found and tested to this post so you can see which differences there are. Maybe another version suits your setup and pairs preference better than mine.

if so, then please share to the community so that others can take notice of your personal experiences too.

This is it for me now, and I will see you in the next video/post.

Goodbye!