Clone Army is a fully automated algorithmic trading system, that trades a portfolio of strategies on different markets and time frames. The strategies are mined and robustness tested with a full automatic process using SQX software. The portfolios run on live accounts of Global Prime FX brokerage. This system is for personal use only and not for sale. Hence this article is solely intended for educational purposes. It was my personal challenge to keep it as short and simple (KISS) as possible for good readability.
The mining process
The robustness testing process
Because of the fact, that strategy mining is working with historical data, we logically only have an historical view of the performance of a mined strategy. As we know the words in nearly all disclaimers of financial institutions ”past performance is no guarantee of future returns”, we have to check that this performance is not biased due to the mining process. And chances are extremely high (nearly 100%) that we have exactly this problem. That is because of the nature of strategy mining, because we are puzzling millions over millions of combinations of indicators and indicators settings to achieve a good result. To check that a strategy is not just “curve fitted”, we need to perform robustness tests. There are a variety of technics that can be used to evaluate, whether a strategy is just a best fit to a set of data or really has an edge to survive in real live trading market conditions.
figure 1: robustness test filtering
At the end of this robustness tests, we end up with some candidates, that have the potential to be used on a live portfolio. As a rule of thumb, I needed about 3‐5 days of 24/7 calculation time with an EPYC 750P2 32‐core with 2,5GHZ and 128GB RAM server to gather one single candidate. But as soon as you have this candidate: Be warned! In my experience we have to evaluate this candidates, whether they can actually proof their historical results under real live market conditions. Therefore we have to put these candidates on real live micro accounts and let them trade with the minimum available lot size (=micro). This process is what I call “micro account testing under supervision”.
Micro account testing process (“candidates under supervision”)
From Micro to Master account (“Creating a portfolio”)
After some time of mining and testing we should have a couple of candidates, that have the potential to be used to trade within a portfolio on the master account. To transition from micro to master, I personally use the following procedure. First step is to sort the strategies on the micro account (candidates) in groups of markets (see figure 2).
Figure 2: Overview micro accounts (candidates)
Therefore I group them in the following markets EURUSD (1), GBPJPY (2), GBPUSD (3), USDJPY (4), AUDNZD (5), other FX pairs (6), XAUUSD15 (7) , DAX (8), Other Indices (9) as well as Energy & Commodities (10) and BTCUSD (11). So in other words all EURUSD strategies will be sorted into the EURUSD group all DAX strategies into the DAX group and so on. All groups have an individual micro live account with the corresponding name (like f.e. EURUSD). I personally call them support portfolios, because they are the “basic pot”, where we pick our best strategies from. In figure 2 the groups (“support portfolios”) of “Clone Army” are shown. This has the advantage to keep an overview about our overall project and to evaluate intermarket performance. This gives us very interesting insights because some markets are more handy to be used with fully automated strategies than others.
Unfortunately not at all, the journey has just begun. As a next step we have to run our master portfolio and “treat it with love and caution like a raw egg”.
Excursus: Magic of uncorrelation
Before we come to the master account we have to speak about correlation. That is because this concept is very important. What will become clear with some experience is, that there is no perfect single strategy. There is no holy grail. There are only trade‐off’s. Every type of strategy like trend, mean‐reversion or bias strategies will come with their pro’s and con’s. (f.e. trend strategies work well in strong trends and bad in sideway markets). In my opinion it is not possible to change their very natural behavior, even not with the best filters or additional indicators. Performance could be improved, yes, but the very core, the nature of a strategy, is likely not possible to change. So my postulation is: We just accept this fact and live with it. But the good news is: There is a solution for this problem.
The solution is correlation. Or in other words for our specific problem: It is uncorrelation. That is the magic, that brings everything together.
Technically speaking: It is the combination of different uncorrelated strategies within a portfolio with the goal to place a winning streak of one strategy into the drawdown of another strategy. With this knowledge we can accept the fact, that we cannot eliminate but reduce overall drawdown. And this is our goal, this is the magic that lies in the application of uncorrelation. In my personal opinion, this is the closest we can get to the holy grail. And the good news is, that this is something we can actively manage. We can find the best fit for our portfolio with the usage of correlation analysis.
So let’s have a look at the correlation of strategies of “Clone Army”. You can see in figure 3 that the strategies are highly uncorrelated (green). Only two strategies are showing some moderate correlation (red).
Figure 3: Correlation of strategies of Clone Army version 1‐4 based on profit/loss by day
Running the Master account
So this thoughts bring us to the most important point running a master account: The money management (MM).
In my personal experience it is good practice to use 1% of our balance as stop loss for one single trade. For larger accounts 0,5% may be more appropriate. For smaller accounts up to 2% is realistic to stay within limits. But why do we want to use a percentage of balance and not trade with fixed lots?
The answer is: Compounded profits. What we want to achieve are strategies, that allow us to run them for years with solid profits. We definitely do not want to have a “mayfly”. This method will grow the balance of our account and consequently increase our positions, which will increase our balance and vice versa. This is a reciprocal cycle. If we use the current monthly profits of “Clone Army” which is 12,35% profit/month to make a prediction, we can see the power of compounded profits (blue line). For this graph I have chosen a hypothetical start balance of 10k EUR for demonstration.
Figure 4: Predicted profits theoretical best case versus realistic approach with base = 10k EUR
Sure the blue line (extrapolation of 12,35% into the future) is an ideal prediction and unlikely to happen in real life. So we should apply a degeneration (reality) factor to see a more realistic outcome. My personal best guess is a reality factor of 33%, that will lead us to an average monthly profit of approx. 4%, which can be considered to be reachable (orange line). Hence, with an hypothetical starting balance of 10k EUR we can think of a realistic outcome of 23k EUR in 22 months, which is quite nice (+130%) in my personal perspective.
This insight leads us to the major advantage of this kind of trading system. That is the possibility of upscaling! As already discussed while our equity is growing, our position size does this accordingly. And it is technically feasible because FX, CFD and future trading is capable of opening up to 1000 lots in most brokers’ standard capability. If we come to the point, where 1000 lots is not enough to open 1% equity stop loss of our portfolio, this will be a long journey. We would need roughly a 50 million EUR account to have this kind of issue happen. So this will be no problem for 99,99% of the users. That means in other words, if we have managed to find a good portfolio, we could let it grow without almost any restrictions.
Last but not least I would like to mention, that our master account needs some nice care. Market regimes can change. The character of markets can change. Price data feed can change. With this in mind it is a good practice to periodically evaluate the performance of our best strategies on the master account. If performance deteriorates significantly, then it could be a sign to replace a bad performing strategy with a better one. And this should not cause any major problems, because we have a big pot of support strategies! I personally check my master on a quarterly basis, that means every 3 months.
Congratulations: If everything is done correctly, you have a good chance to be successful on the financial markets
Let’s have a look at the most important point: The performance of “Clone Army” under real life conditions with real money. This analysis was started in September 2021 with beginning of version 2 of “Clone Army”, because this was the point where I had enough candidates to build a solid foundation for the master account. Therefore data from pre‐version 1 (α‐version) was skipped in this analysis.
Here you can see the performance overview in myfxbook.com.
Figure 5: Equity curve and overall stats
Figure 6: Drawdown by day
We achieve an average monthly profit of 12,35% with a maximum drawdown of 18,14% and an average drawdown of 3,49%, that I calculated manually out of the trading data. This leads us to an annual calmar22 ratio of 8,223.
What is important to notice is a smooth upsloping equity curve. This is exactly what we are aiming for. We can see also periods of drawdown. But this is ok and we have to accept this. The most important point to mention here is to have enough confidence in our portfolio to stick with it even in extending periods of drawdown. If you give it up to early, all hard work is “blown into the wind”.
- The Calmar ratio is determined by taking the investment fund’s estimated annual rate of return, usually for a three‐year term, and dividing it by its maximum drawdown.(Wikipedia)
- 12,35% profit/month x 12 months / 18,35% max DD
As additional information you can see “Clone Army’s” asset popularity.
Figure 7: Chart of asset popularity March 2022
In terms of quality score number we reach an nice 3,6.
Figure 8: Calculation und Grouping System Quality Number
If we look at the duration of trades, we see that the maximum duration is 5 days. That is logical because we only use Friday close only as trading option. So consequently we do not have to worry about weekend risk.
Figure 9: Duration of trades
A very important insight is this graph is that the losers are cut off at stop loss of 1% or 2% (beginning of version 2, then changed back to 1%) respectively and winners are ride out. This asymmetrical behavior is the very root cause, why this portfolio works. It is a funny thing the algo’s come to the same conclusion of the old wisdom “cut losers and ride out winners”. Is this just coincidence?
It is hard to find something to benchmark with. So I was evaluating the portfolios of myfx autotrade.com to have any kind of comparison. I made a graph, where maximum drawdown is depicted on the y‐axis and monthly profits on the x‐axis. You can see that “Clone Army” is doing pretty well. Notably the max drawdown with respect to the achieved profit is quite low in comparison to the other portfolios. One possible explanation is that on myfxautotrade.com only forex portfolios are used. Due to the nature of “Clone Army” that is also using Indices, Commodities and Crypto, the variety and therefore uncorrelation is greater, which brings us to the postulation that more uncorrelated strategies consequently lead to less drawdown (and higher profits) compared to the average of forex only portfolios. This graph could be a first indication, that this postulation could be true.
Figure 10: Benchmark “Clone Army” compared to strategies on myfxautotrade.com
There is a lot of skepticism when it comes to full automated strategies. So one counter argument could be, that the shown profits of “Clone Army” are due to luck and have no sustainability in the future.
Although it is true, that we only can evaluate the performance until now and profits of “Clone Army” could theoretically turn dramatically worse one day after this article was written, I do personally not think it is likely to happen. Regrettably we only can evaluate what we see as performance facts here and now.
If we want to see what’s happening in the future, we have to make a reevaluation after some time. Unfortunately I do not own a crystal ball to predict the future, so that’s the only way to do it. Hence that could be a good point for an update of this article.
Regarding the second argument, that the results are only due to luck: This is highly unlikely. To achieve a profit factor of 1.4 with 271 trades only due to luck has a probability of 0,17% if we calculate it with a binomial‐distribution with a possibility of 50% to have a winner or losing trade. In reality this is much worse, so in my personal opinion it is almost impossible to reach this performance only due to luck especially if we take into account that 65%‐80% of retail clients lose money in trading with derivate products.
And finally, according to the scientific method a thesis is as long valid as long as their contrary is not proven, so we can assume that we have found a good way to make risk adjusted profits so far.
If you have questions, comments, suggestions or criticism, you can reach me on the SQX or Global Prime Discord server under ChrisWhite ID: 5390.
You can request a personal link to watch and follow the performance of “Clone Army”. Just give me a direct message on discord.
 Available markets are: Forex, CFD, futures, energies, commodities and crypto
 Builder is a strategy mining tool within SQX software
 Please refer to Strategy Quant User Guide, ebook “How to trade profitable in forex using Strategy Quant
Software” and documentation & tutorials on SQX website
 Evidence Based Technical Analysis, David Aronson, 2007, page 107ff
 Please refer to chapter cross checks – robustness tests on SQX website
 Out‐of‐sampling (OOS) testing, other market testing, testing different time frames, Monte‐Carlo (MC) analysis, walk‐forward analysis, what‐if analysis, SPP (system parameter permutation), graphical analysis on local parameter highs, manual curve evaluations, optimization analysis and micro account testing 9 Please refer to the video tutorials on SQX website
 Personal first choice are Hetzner servers https://www.hetzner.com
 P/F = profit factor
 Error type I: Bad strategy on Live versus Error type 2: Missed opportunity
 Instructions for Life – Dalai Lama
 P/F = profit factor
 Major and Minor FX pairs with very low number of candidates 16 Gold
 Dow Jones (US30), Nasdaq100 (NQ100)
 WTI (crude), NGAS (natural gas), soft commodities (soybean, wheat, corn,…)
 Quant Analyser 4.0: Tool from Strategy Quant to analyse and manage portfolios
 If you do not believe it, then think about the end of the movie “Monthy Python and the Holy Grail”
 Reduction of minus 66%
 https://en.wikipedia.org/wiki/Binomial_distribution with N=271, k=160, p=0,5
 Due to transaction costs
 Simplified, we can reject the null hypothesis H(0): “Clone Army has no edge in trading on real markets and profits are due to luck” with p=0.0017 as very statistically significant (p<0,01)
 See disclaimers on various online brokers ranging from 65%‐80%
 Evidence based Technical Analysis, The Scientific Method and Technical Analysis Chapter 3, page 94