Rafael Molinero must have driven the adults crazy.
“I always liked the ‘why’ questions,” he says. “Why is the sky blue? Why does soap work?’ Stuff like that.”
Today, those questions are along the lines of “Can the same algorithms that digital signal processing (DSP) systems use to differentiate voice from background noise also be used to differentiate true market behavior from random movement?”
The answer appears to be “yes.” Molinero used algorithms based largely on DSP to build the core of the program that has enabled Molinero Capital Management to average 18% per year since July 2005 against a maximum drawdown of just 5.9%, with a Sharpe Ratio of 1.4.
His high math aptitude helped a bit, too. “My math teacher was sending me to the French Mathematical Championships and things like that,” recalls the Paris native, who made his first trade when he was 14 and followed his passion for math to a masters in applied mathematics from the École Nationale Supérieure d’Informatique et Mathématiques Appliquées de Grenoble (ENSIMAG) in 1997.
A course on the application of physics and mathematics to stock markets struck a chord, and after graduating he landed a banking internship at Crédit Agricole, which led to a paid position in their Chicago operation. There, he had his first exposure to alternative investments — the first branch of finance that didn’t seem to bore him after a few weeks.
“France wasn’t the best environment for me,” he says. “It’s bureaucratic, slow, and seniority-based, while the United States is more of a meritocracy: work hard, and you will be awarded.”
The biggest rewards were paid tuition towards an MBA in finance and economics at the University of Chicago, where he finished in the top one percent of his class, and a transfer to Calyon, a joint venture between Crédit Agricole and Société Générale (which would become Newedge following its merger with Fimat). There, he helped manage a fund-of-funds and built their structured products department.
Through Calyon he got in contact with Rotella Capital Management, a $1 billion U.S.-based CTA that hired him to create a risk management department and to lead a team in advanced mathematics research.
“Robert Rotella also loves research and enjoyed that I would look into artificial intelligence, digital signal processing, fractals, entropy, anything from science,” he says.
It’s territory that plenty of quants have explored before, but Molinero knew he had to swim up the knowledge stream if he was going to find something others had missed.
“I always read the pure math and don’t touch any books that have the word ‘applied’ in the title,” he explains. “There’s no bias at this level.”
He quickly became intrigued by DSP. “Think about it,” he says. “On the phone, voice gets transmitted into a time series and then polluted during transmission, and DSP algorithms try to figure out which part of the transmission is voice or signal and which part is pollution or noise.”
Markets also move in waves similar to, but not identical to, the sine waves of sound, leading Molinero to ask, “If markets have a non-random component, can we extract the random one and keep the non-random one?”
That meant transforming existing algorithms, for a variety of reasons: first, markets aren’t sound waves; and second, DSP is designed to rebuild something that existed previously (sound), while traders are trying to project a bit into the future.
So Molinero, along with a colleague, spun-off from Rotella to create Molinero Capital Management in London and continue trading the algorithms he had developed.
“We decompose the price into linear (trending) and non-linear (cyclical) components, and then the algorithms identify which are signals,” Molinero explains, adding that what’s left after filtering the noise is probably the market’s real pulse. Models also generate a probability of success on each trade. “For example, if you can explain a signal with very few components, it’s probably a strong signal, which we can use to project the next directional move.”
This, he says, enables them to avoid getting whipsawed in choppy markets and also to get into trending markets earlier.
“The models adapt in frequency based on the speed at which the market is going,” he explains. “By decomposing the market, we can look at the underlying components. If these components change directions frequently, our models will trade short-term, if not our models will trade medium-term.”
The system is consistent, making money in 95% of the markets it trades despite that only 51% of all trades are winners. The winners average considerably more than the losers.
The 51% success figure reflects trades cut short by tight stops. “Risk protection is everything, and the best form of protection is knowing when to be out,” he says.
That’s one premise he doesn’t question.