Welcome to the world of machine investment. We look at two South African companies that offer fully computerised investment. Is this the future? From Brainstorm
For the better part of two decades, there’s been raucous debate in the investment community as to whether machines can beat humans at investing.
And machines are certainly making their case. High-frequency traders (HFTs) using algorithms that look for pre-determined trading signals now account for the bulk of market trades around the world. The holy grail for HFTs is to detect opportunities and execute trades at the speed of light, which explains why so many of them are located within metres of the stock exchange trading engines. An advantage of a millisecond or two in speed over your competition means the difference between profit and loss.
Not all algorithmic traders are looking for this kind of speed, however. An entirely new breed of machine-based trading has sprung up in recent years that is more concerned with filtering data from the tens of thousands of tradeable instruments around the world in search of opportunities. No human analyst – or team of analysts – can cover this ground the way machines can.
One such company is Rimar Capital, formed in 2014 by a group of traders, quantitative and financial analysts, and coders. “There’s simply too much data to possibly process. You aren’t just looking at the internal financial data related to tens of thousands of companies and other tradeable instruments; you also have external factors, such as Covid-19,” says Rimar Capital’s CEO Itai Liptz.
So, how did Rimar Capital’s algorithms deal with that event?
This is where AI becomes crucial. Liptz says at the first news of the outbreak of Covid-19, Rimar Capital’s AI systems were able to go back in history to study the effects of previous pandemics on world markets. Within minutes, they were able to pick up the likely extent of the drop in market prices. The AI systems issued sell orders early in the Covid-19 pandemic cycle and so protected investors from much of the carnage that followed.
Historical perspective
The ability to go back in history to assess the likely impact from black swan events such as Covid-19 would keep teams of human researchers busy for weeks. The speed with which AI can accomplish this, and act on it in the financial markets, makes a massive difference to overall financial returns.
It turns out that Covid-19 is a once-in-a-century event (although smaller pandemics such as SARS, Ebola and Swine Flu have all occurred in the last two decades). A study by Ishan Shah at Towards Data Science analysed pandemics going back as far as recorded history allows. “There are two trends emerging. One is that there is a significant fall in the number of deaths over the years. Another trend is the period has also significantly reduced. The reason for these is an improved healthcare system available to us and also a lot of effort that goes into finding and deploying vaccinations to the general public.”
If Rimar’s AI systems can detect and act at the very onset of Covid-19, how does this compare with human response times?
Rimar Capital offers a number of different investment strategies catering to different risk appetites:
• Its Short-only strategy (which attempts to profit from falling prices) is up 21% for the year to end April 2020 (and up 909% since inception).
• Rimar Fund A (a hedge fund that invests in securities based on both technical and fundamental data) achieved growth of 3.38% for the year to end April 2020, and is up 136% since inception.
• The Options Fund (which uses algorithms to trade in equity options forming part of the S&P 500) is up 6.6% for the year to end April 2020, and 1 400% since inception.
• The Options Fund+ (invested in options linked to US Treasuries and derivatives) is up 4.35% for the year to end April, and 2 400% since inception.
• The Global Macro 1 fund (using technical indicators to indicate buy and sell opportunities on the global markets) is down 37% for the year to date, but up 38 795% since inception.
These are stunning results, which in all instances have outperformed their benchmarks.
Based on these performances – bearing in mind the company was only formed in 2014 – it can be said we with some confidence that machines can beat humans at investing.
Dividend discount model
Machines can process data faster than humans, but NMRQL was formed to overcome the problem of in-built human bias in most investment decisions, by using unbiased algorithmic trading. Tom Schlebusch, CEO of NMRQL, founded the company in 2017 with former FNB CEO Michael Jordaan to offer machine-learning investment opportunities.
Millions of data points are run through the machine-based learning model before an investment decision is made, says Schlebusch.
Asked whether NMRQL is looking for inefficiencies in the markets (such as deviations from norms), he says: “We try to exploit efficiencies in three areas: data, models (algorithms) and processing power. This is best illustrated with an example. A manager is using earnings reports or the financial reports from companies as his data. He then uses a dividend discount model to build some idea of future value for his stock and does this processing using an Excel spreadsheet and a few analysts. One could argue that for this manager, the market is quite efficient as the data, the model and the process used are fairly generalised and available to almost all players.
“Contrast this with the very best guys out there who use petabytes of data that consist of unstructured data sources like sentiment data, satellite photos and news feeds, together with tick-by-tick data on almost any asset and economic and fundamental dataset.
This gets processed with deep learning quant models that can find relationships and inefficiencies in price between a multitude of assets, and then process this by using massive cloud processing and storage. I imagine the second manager would have the best edge and be able to exploit the opportunities, which others don’t have the tools to exploit.
Improving your information ratio is not only about getting calls right, but also about how many calls you make, which is called market breadth. The second manager has much larger breadth and much more info with which to improve performance.”
So do the algorithms trade automatically or is there still human judgment involved? “For many in this space, they do trade automatically as managers exploit shorter-term opportunities. At NMRQL, we’re not there yet; we still apply some human oversight,” says Schlebusch.
So how is the system performing?
“If you asked me six months ago, I would have said, ‘A little disappointing’; however, over the past six months, we made several large improvements. Most significantly was to include several regime models that try to distinguish good from bad environments. While we have underperformed the benchmark (inflation plus five percent) since inception, our numbers over the past year have significantly improved. For the year to date, we’ve been outperforming our benchmark handsomely.
We are at the infancy stage of these developments here in SA, even though we were possibly the first entrants to this environment locally.
The guys who do this well don’t publish what they do, so you have to find your own way. We’re confident that our models are improving all the time and that the long-term targets will be achieved, even more so now,” he says.
NMRQL’s systems avoided Steinhoff when others were still buying. At that time, the forecast scores on Steinhoff were just simply much poorer than other opportunities. Many other stocks offered better opportunities.
And, like Rimar Capital, its models started flashing on 25 February 2020 when the Covid- 19 crisis first started to seriously impact global m arkets. “ We sold quite a bit of our share holdings at that point and slowly sold a bit more into the drawdown. The models performed really well at the start of Covid-19,” says Schlebusch. “The bigger question is how they will perform from here onwards. We have not lost any money now since the start of the year. The question is if the crisis is over or if we are in for another move down? Time will tell.”
Schlebusch is confident that his machine learning systems will indicate when it’s time to start moving back into the market.
Despite the impressive performance of machine-based trading systems, the jury is out as to whether algorithms will completely displace humans. Markets are wild frontiers that aren’t prone to prediction. Liptz believes algorithmic trading will capture an ever larger slice of the investment universe but, like Schlebusch, he doubts it will ever fully replace them.
Schlebusch says machines will encroach on those trading models geared for shorter time periods, and that certain human functions can be entirely replaced by machines. “Will machines replace humans? This depends on your timeframe. In five years, possibly not.
The first steps will be to augment and assist, but in time, I definitely think certain human functions could be entirely replaced. The bigger question is how do we adjust to this?”
Ultimately, machines have two crucial advantages over humans: their lack of bias, and their ability to process vast volumes of data. Machines will not hold on to shares for some misplaced nostalgia. They are brutal, clinical and extremely fast. And that will only improve over time.
