The Machine Learner

Hi Folks,

I am starting a new C2 strategy that trades equities using an artificial intelligence model. The plan is to do end of week reviews which will help me identify trades that could have been avoided if the model had access to some information which was present before trades were entered into.
Hopefully, having other traders chip in with their insights will help me iterate faster on what information sets will be useful to incorporate into my model.

  • Currently, the strategy enters 20 long and 20 short equity positions on Mondays at the market open and exists on Friday close to market close.
  • Entries are structured to have equal weights as much as possible.
  • All trades have a stop loss that is 10% away from the entry price representing a 10% max loss per trade.

In backtests, the strategy could go into prolonged drawdowns for months (1-3) and some drawdowns were as large as 40%.

I have been trading the long leg in a trading212 account for the past 3 weeks. The walk-forward results match what the model predicted so I am confident the model should do well going forward but time will tell.

I am looking forward to getting some useful comments with time.

40% drawdowns? The big question is are you yourself willing to invest with your own real money in a strategy/system that has a 40% drawdown? If the answer is no then I think you had best try and figure out how to reduce the drawdowns.

A 40% drawdown means potentially you can lose up to almost half your account value.


Seeing as it is an equity-based strategy that is always fully invested in the market (except weekends), this drawdown looks normal. The benchmark I have in mind is the S&P 500 and seeing as that index goes into much deeper drawdowns, a 40% drawdown with respect to this system is not surprising.

One way around this drawdown is to trade a fraction of the strategy. If one follows all trades but holds 50% of the proposed positions, then the drawdown expectation gets cut in half.

I am actually trading full positions but only the long leg (my broker does not allow shorting) and I expect to lose more than half my account at some point.
I see drawdowns as part of every profitable system. At the end of the day, profit factor and probability of winning trades will determine the survivability of the system given a large number of trades and time.

@AlgoSystems Thanks a lot for the comment and I look forward to learning more from you.

No problem @MachineLearner, just trying to give some advice. The community in C2 usually prefer lower drawdowns but that also depends on the return percentage as well. ie. If you have a 40% drawdown and the yearly returns are 300% or higher it may be suitable for some investors.

Good trading!


Truth be told, 300% a year is way out of what this model is capable of. But let’s see.
Good trading to you too!

Notes on 30 August 2020

The Machine Learner is a performance-oriented specialist investment manager currently managing a single equity strategy. The main benchmark of this strategy is the S&P 500.

The AI behind Machine_Learner_01 employs a highly selective and systematic process in selecting only stocks that have a high probability of ending the week in green or red. Since the portfolio is equally long and short, its exposure to the entire market is very low.


Going forward I am reducing the number of stocks I hold in a week from 40 to 10; 5 longs and 5 shorts. The reduction is coming from further analysis that shows that a reduction in the stocks improves the annual return of the strategy, but this does not increase in the expected drawdown.

Good trading to you all!

Would be nice if you post link to you system here.

@JITF I will include the link in my posts going forward. Thanks for the suggestion.
@LiveForexSignals the system is actually one week old on C2. I have included the link in this post.

That’s odd, I couldn’t find it yesterday. :thinking:
But anyway, thanks for the link.

Notes on September 2020

Changes in pricing: All three of my systems are now priced at $0 for October 2020. Pricing for a system increases by $20 if we end the month in green and reduces by $20 if we end the month in the red.

System Update: The initial iteration of the AI was holding positions for only a week but the steep and quick drawdown in September called for a change. Not many people can stomach that rollercoaster in real life. The AI now looks for more medium-term swings anywhere from a week to four weeks. This means expected returns are going to be lower but drawdowns are going to be more manageable as evident in Machine_Learner_02 and Machine_Learner_03. All open positions have a 10% stop-loss.

September was a very difficult month for Machine_Learner_01.
Machine_Learner_01 went into the first of its projected drawdown of 19%. The system has since clawed back to -13% for the year.
Return:         -14.8%
YTD:             -13.0%
Drawdown:   18.9%

Very good first month for this strategy as it managed to make a beautiful 2.4% return from the last two weeks of trading 2 weeks in September.
Return:        2.4%
YTD:            3.5%
Drawdown: 0.9%

Similar performance to Machine_Learner_02 over the same two-week period but it’s expected to underperform the first two systems over the long run.
Return:        2.4%
YTD:            3.2%
Drawdown: 0.9%

Hope you all simulate the system for the rest of October to see if it can help smooth out the performance of your portfolios.

Good trading to you all!

Just a quick note that changing the monthly fee is always ‘backcalculated’ to the initial start affecting all monthly’s performance.

That’s also going to be a problem when at some point you want to trade bigger and charge a higher monthly fee. This will possibly push past results into the red. I’m running into that point now myself :-\

I have been thinking about that myself. But I guess it keeps us honest, no? If I want to charge a fee that will put my performance in the red, then I have not earned that fee, in a sense.

Now that I am thinking about it, I should update the terms. It might be reasonable to add that the fees only increase as long as the increase in fees means the system out-performs the S&P500 over the life of the system. Definitely, something to ponder. Thanks a lot for the information @RaoulSuurmeijer.

1 Like

Mid-month update October 2020

Changes in pricing: All three of my systems will stay priced at $0 for the rest of October through November and December. Pricing adjustments will start on January 1st, 2020. Pricing for a system increases by $20 if we end the month in green and reduces by $20 if we end the month in the red.

System Update: The new iteration of the AI has been doing great, and has also passed a number of robustness checks I have been running over the past few weeks. I will continue to test out different scenarios to make sure the system is as fat-finger proof as possible. The underlying strategy will not change for the foreseeable future, but at some point, I plan to expand the information set I am subscribing to so that the AI has a larger information set to learn from.

TOS: I recently got two subscribers and like everyone else, they are wondering why I am not trading my own system. The truth is that I am currently trading a more risky version of the strategy in my IBKR account.

You can match the attached trades with the trades in any of the systems, and conclude the picks overlap 100%. I plan to link the account to my C2 account once I hit $35,000. Any amount less and I risk getting a PDT strike on my account giving the stop losses the system employs.

Return Expectations: My system description does not include expected returns because I do not trust the absolute numbers that come out in backtests. I trust that drawdowns will be at least twice as large but compounded returns are always too optimistic because we cannot account for all the costs inherent in trading a long-short equity strategy. What I can say is that the systems are robust in their ability to make money over time. A version trained on data up to 2004, still makes money when brought to the 2017-2019 market. So a reasonable prior is that we will make money over time. How much money we can only find out by adjusting this prior with the information we are getting from the “live” trading.

All three systems are in the green so far for the month.
Still on its way to recovering from the drawdown.
YTD:             -8.6%
Drawdown:   18.9%

YTD:            6.8%
Drawdown: 0.9%

YTD:            6.4%
Drawdown: 1.3%

Good trading to you all!