Key highlights

• The peak in market uncertainty, influenced by US policy and economic conditions.
• The decline in US exceptionalism in equity returns and the growing case for stronger European stock returns.
• The efficient implementation and execution in quantitative strategies due to lower returns in recent decades.
• The use of AI and ML to enhance traditional quantitative factors, while addressing challenges related to model complexity and interpretability.
• The need for ongoing improvement and adaptation of quantitative models, even in smaller markets like South Africa.

What a time! Uncertainty in the market has peaked, and it goes without saying that what is currently unfolding in the US has a lot to do with it. At this year’s UBS Quant Conference, there were a couple of discussions on the idea of US exceptionalism vs the case for Europe using different data. United States (US) exceptionalism seems to be on the decline in terms of equity returns, and the case for stronger returns from European stocks is gaining traction. Policy uncertainty in the US, the war of tariffs and its effects on the US and its trading partners and Germany’s fiscal stimulus plan are all seen as evidence for this. The forecasted gross domestic product (GDP) growth for the European Union region is improving, and in contrast, the US is declining. Despite all the evidence, the concentration in global stock indices like the MSCI All Country World Index means that global stock returns will depend on US stock returns.

From generating alpha to defending alpha

As we know, doing these four things well is important to building successful quantitative products: data, alpha signal generation, portfolio construction (this includes risk management) and implementation (also called execution). The returns from quantitative strategies have been lower in the last two decades, compared to the two decades before that. Because of this, efficient implementation and execution is more important now than ever. However, there are methods for how to improve fund rebalancing and using transaction cost analysis and market impact to improve trading. Better rebalancing techniques include trigger-based rebalancing or rules-based rebalancing. The idea is to reduce turnover and lower trading costs, while maintaining as much of the alpha signal as possible. The market impact reduction methods are aimed at executing large trades in such a way that the market impact on price movements and liquidity is minimised. We should see more research in this area in the coming years as defending the alpha becomes more important.

Artificial Intelligence (AI) and Machine Learning (ML) find their place 

With quantitative strategies, data is critical and that is why even today there still is a big push from quantitative funds to use alternate data (like credit card payments, satellite imaging, shipping estimates) to eke out alpha. Despite the efforts, this has proven to be very difficult.

The focus has shifted from using alternate data to using ML and AI techniques to enhance traditional quantitative factors like value, momentum and quality. Non-linear techniques like gradient boosting, neural networks and random forests are proving to be better at predicting stock returns than traditional linear methods. The discussion around parsimony versus complexity of these alpha models is always interesting. Traditionally, quantitative managers wanted to keep their models simple, using only the data that was deemed necessary. There was a push to keep models simple because increasing complexity usually meant increasing the noise and decreasing the signal. But with non-linear methods, traditional thinking is challenged, and slightly more data and variables can be tested because these methods tend to be better at modelling the relationships between signals and alpha. Even if traditional thinking is challenged, the importance of balancing the number of signals and variables to maintain a good signal-to-noise ratio is as important as ever.

In the past, there have been some questions around the interpretability of these advanced machine learning techniques and rightly so. The workings and output from these methods are usually difficult to explain and quantitative managers are hesitant to invest client funds because of this. A lot of new work has focussed on improving the interpretability of complex methods and finding the right balance between improving existing methods, while trying to explain and understand how they work. This area of research is particularly interesting because it can break down barriers and make fund managers more open to using machine learning and artificial intelligence.

Overall, there is a need for constantly improving and updating quantitative ideas and models, more so in the current investment environment.

Although South Africa is a tiny part of the world and not on the investment radar of these developed market economies, we can translate their research into a South African context and use their ideas to improve our client offerings.

As quantitative investors, we have always believed that adding complexity for the sake of it can be detrimental to fund performance. Given the complexity of financial markets, simpler methods implemented efficiently usually work better. It is important to model liquidity in concentrated and illiquid markets like South Africa, and that is why we have done work on more efficient ways of handling turnover through better rebalancing. While we have also been concerned about the interpretability of machine learning and artificial intelligence models, within our team, we have started experimenting with them and trying to find the right balance between better alpha signals and understanding what is driving them.