Marcelas is an algorithmic trader on Darwinex Zero who runs AMCG and has just launched TRDF. In episode three of Behind the DARWIN he tells Blayn how he built a seven-figure crypto portfolio as a teenager, lost almost all of it, and rebuilt with a single mean reversion algorithm. He optimises it on one or two years of data and tests it on twenty.
Key takeaways
- Marcelas switched to algorithmic trading in 2022 because his ideas were sound but his discipline was not.
- His main algorithm runs one mean reversion strategy across nine forex pairs. It is around 8,000 lines of code, mostly mathematical formulas.
- He optimises on one or two years of data and treats the remaining 20 years as out of sample. He optimises five parameters at most.
- He leaves out EUR/USD and the yen. Central bank surprises push them too far for mean reversion to work.
- His academic survey of traders found almost everyone trades on emotion at least sometimes, including those who rate themselves as flawless.
Watch the full interview:
Who is the trader behind AMCG and TRDF?
Marcelas started his first business at 13, reselling trading cards, and was making around €1,000 a month by 14. That money went into his first broker account. He lost his first €2,000 within two weeks.
Early positions in tech stocks and then crypto grew his account to around 3.6 million. Trading on margin through the 2022 bear market took nearly all of it. His final Bitcoin long, at 1.5 times, was liquidated too.
"But losing the millions of dollars, it was the best thing that ever happened to me." 11:41
Today he is finishing a master's in international business, teaches at his university and is writing research papers on trader behaviour.
Why did he move to algorithmic trading?
He decided the problem was execution. He had good ideas, then held losing trades and refused to give up on them.
"I need a machine that executes my ideas flawlessly without me being in the equation to mess things up again." 7:25
He taught himself Pine Script to test ideas, then MQL4 to turn the ones that worked into Expert Advisors on MetaTrader. His first algorithm aimed for steady monthly returns over the highest total return. He still designs every strategy that way.
How does he avoid overfitting a trading strategy?
He optimises on a small slice of history and tests on everything else. Early on, nearly everything he built was overfitted.
"So in the beginning I was always going in circles, I think I found the golden grail and then it breaks and then I think I found the next golden grail and then it breaks again." 24:00
His process now has four parts. He optimises on one or two years and checks the result against around 20 years of out of sample data. Where the algorithm breaks, he looks for a general formula that handles that type of market, never a patch for the single moment. He caps optimisation at around five parameters, following Jim Simons' idea that a model can have too few variables or too many. Then he runs the finished strategy on other currency pairs.
"Obviously you can't overfit for another symbol while testing it on a different one." 30:32
What does his mean reversion strategy trade?
It trades forex pairs where daily behaviour stays in a tight range. He looks for stretches where price moves away from where it should sit and trades back towards it. Position size increases as the move stretches further.
He avoids stocks, indices and metals. Too many days fall well outside the normal range. EUR/USD is out for the same reason, and the yen because of the Bank of Japan's long one-way moves. NZD/CAD takes around a third of the allocation. The strategy works on bigger moves, roughly the one to four hour charts. He does not believe in scalping.
The jump from one pair to nine came from a tool he built with ChatGPT. It logs equity during MetaTrader backtests and runs millions of correlation scenarios to check how the pairs could draw down together.
What did his research find about trader psychology?
His paper, "Inside the Trader's Mind: Evidence on the Perception Behaviour Gap in Individual Trading", surveyed traders, including some from Darwinex. He hid the real questions among unrelated ones so people would not realise what was being tested.
"We still had a very strong finding. It was almost everybody trades on emotion, at least sometimes." 18:12
The bigger finding was the gap between how traders see themselves and what they actually do. Traders who rated themselves as flawless were among the most likely to admit emotional decisions.
What is his advice for new traders?
Test properly first.
"Even five years of data are absolutely meaningless if you don't test it properly without overfitting." 57:47
He is happy for a strategy to take a long time and grow complex in small pieces. His other lesson is personal: money is not everything.
"Don't let it eat your life." 59:05
He chose Darwinex Zero because it rewards a long-term track record over quick profits. Blayn and Marcelas plan a follow-up episode in eight months to see how TRDF has developed.
Watch the full interview
The full episode includes Marcelas sharing his screen to walk through the pairs in TRDF and his scenario testing. Watch Behind the DARWIN with Marcelas on YouTube. His paper will be shared in the Darwinex Zero Discord, where you can put your questions to him.
Building your own track record? Start on Darwinex Zero.
FAQ
What is mean reversion trading?
Mean reversion trading assumes price that moves too far from its usual level will tend to move back. Marcelas trades forex pairs where daily behaviour stays in a tight range. He avoids markets where central bank surprises can push price in one direction for weeks.
How do you test a trading strategy for overfitting?
Optimise on a small part of the data and test on the rest. Marcelas optimises on one or two years, then checks the result against around 20 years of out of sample data. He also runs it on other assets. A strategy cannot be overfitted to a market it was never optimised on.
How many parameters should a trading strategy have?
There is no fixed number. Too few and optimisation finds nothing useful. Too many and it fits the noise in your data. Marcelas works with five at most and accepts that the right number depends on how much each parameter changes the strategy.
Is Darwinex Zero a prop firm?
No. Darwinex Zero lets traders build a verified track record that can attract investor capital through Darwinex, an FCA-regulated asset manager. There is no challenge to pass.
Thanks for reading,
Darwinex Zero.
*Darwinex Zero and the domain www.darwinexzero.com are trade names used by Tradeslide Technologies, a company registered in the United Kingdom under number 14398381.
The contents of this blog post and video are for educational purposes only and should not be construed as financial and/or investment advice.