Machine learning-powered unit trust launched in SA - TechCentral

Machine learning-powered unit trust launched in SA

Michael Jordaan

NMRQL Research, a new financial technology start-up co-founded by former First National Bank CEO Michael Jordaan, has launched a unit trust fund that uses machine learning to drive research, analysis and stock selection.

The NMRQL SCI Balanced Fund, administered by the Sanlam Collective Investments platform, is a collective investment scheme approved by the Financial Services Board that aims to achieve steady long-term growth of capital and income.

“This will be achieved by investing in a diversified portfolio of domestic and international assets, where the asset allocation and stock selection is systematically managed using machine-learning algorithms,” NMRQL said in a statement.

“The machine learning-powered, computational investment process … allows NMRQL to discover hidden patterns in underlying big data,” it said. “Once discovered, these patterns can be exploited to forecast returns across all asset classes and markets, resulting in steady long-term growth of capital and income.”

According to the company, the fund is suitable for institutions, fund of funds and high-net-worth individuals with a moderately aggressive risk appetite and an investment horizon of five years or longer.

It may comprise a combination of assets in liquid form, money market and interest bearing instruments, bonds, corporate debt, equity securities, property securities, preference shares and convertible equities.

Its benchmark is the multi-asset high equity category as recorded by the Association for Savings and Investment South Africa. An annual investment fee of 0.9% is inclusive of management and administration costs, with a 10% performance fee applied should the fund outperform the average performance of all funds within the category.


“This new investment philosophy essentially changes the investment management process from a biased, human-centric investment process to a non-emotive, unbiased algorithmic-driven process that is continuously learning and adapting to changing environments,” said NMRQL co-founder and CEO Tom Schlebusch.

“Machine learning equips fund managers with the tools to assess historical and present data, to help predict future risks and returns based on large volumes of data. At NMRQL, we process around two million data points each time we rebalance our portfolio. This could include quantitative, fundamental, economic or technical variables in order to discover and exploit repeatable patterns, helping us achieve our goal of delivering superior returns for clients.”

Jordaan, who co-founded NMRQL with Schlebusch, said machine learning has already disrupted the fund management industry globally. “In addition to the vast amount of data that the algorithm is able to process, the investment philosophy eliminates emotive decision making, which allows the model to remain rational at all times.

“As humans we suffer from various cognitive biases. These biases negatively impact our objectivity and reasoning skills daily, and are compounded when financial repercussions are involved,” said Jordaan.

Stuart Reid, chief engineer at NMRQL, said the algorithm the company uses is “testable” and allows the fund managers to use historical data to investigate exactly how the fund would have behaved using only information available at that point in time.

“By using more than a thousand different models and applying an algorithmic voting system, NMRQL is then able to produce portfolios with the best possible chance of outperformance,” Reid said. — (c) 2017 NewsCentral Media


  1. Greg Mahlknecht on

    I googled and found their minimum disclosure document – if this is the right one ( ) – it’s not off to a good start. Down almost 25% in just under 2 years.

    Is that 10% performance fee normal? I’ve had a number of similar investments over time, and I’ve never seen that! So if it outperforms the AVERAGE of similar fees, you take 10% of your profits, whereas the funds that beat this one, you probably won’t pay that 10%? Colour me not impressed. If they under-perform, there’s no penalty, I bet.

  2. I think the combination of a rock star executive and tech buzz words obscures the fluff in this article

    Their basic premise is that finding patterns in the past allows them to predict the future. While this has had great success in many industries, I’m not sure this is as true in financial markets.

    They claim their algorithms are completely rational. However they don’t mention that algorithms can be plagued by built in biases which effectively renders them irrational.

    Its not about running thousands of models. Its about running as many models as you need to get a good-enough outcome. This may indeed be 1000s of models. Or it may be 10. It depends.

    I’d also question the claims around ‘testability’ on historical data. What matters more is how the models perform on fresh data. No one cares bout how the models would have performed in the past. What they should care about is how the models perform in the future.

  3. Greg Mahlknecht on

    I find it funny that they point out “testability” on historic data as a feature, yet if you look at the history, those tests have failed miserably.

    In this case, I do care how it performed in the past, because the answer is “HORRIBLE”, and with absolutely no history of success, it’s hard to sell the promise of future outperformance to customers.

  4. I completed an advance course in applied data science at Harvard, in a nutshell your rational is correct. The same argument can be applied to models for weather, probability theory (lotto), This article is riding the buzz word wave, and this investor he is a clueless banker tycoon trying to throw weight around..

© 2009 – 2019 NewsCentral Media