Six Key Insights on Investing with Artificial Intelligence

What’s Quant 2.0?

Quantitative investing is a systematised approach to selecting investments that’s been around since the 1980s. It tends to be run by automated algorithms. Early versions focused on single investment factors, like the ‘small cap effect’ or ‘value’.

Over the years, these algorithms have become increasingly complex. Ever more computing capacity has allowed for vast amounts of data to be incorporated into complex algorithms. These algorithms are trained with historical data series in what is known as machine learning to generate increasingly accurate forecasts – forecasts that can be checked against actual outcomes – before going live as tools with predictive potential. These machine learning algorithms have developed to the point where their capacity to find connections within the data merits being called artificial intelligence.

At this point, AI has helped quantitative investing evolve into the next generation, what we like to call Quant 2.0.

Why use AI?

The computational power involved in AI and the vast reams of data it can digest means that ever more complex associations can be drawn out, making the models even more effective at spotting what may happen in the future given the facts of the day.

Many early quantitative models were limited to finding linear relationships within the data – direct connections between the inputs – to make predictions. Eventually, they were able to also identify more complex, non-linear relationships. AI allows the algorithm to unearth not just relationships between the data, but how different data series interact and influence each other in given situations. Together, these three – linear and non-linear relationships as well as interactions – give a clearer signal about how a given stock may perform in a given period than traditional quantitative approaches.

What makes our approach to AI different from other AI investment strategies?

Tremendous interest in AI has, unsurprisingly, resulted in a great many AI-labelled products. We believe many are spurious. We think our approach to AI is more comprehensive.

Our proprietary AI models were developed over many years by our experts, who are typically PhDs in physics and mathematics. The models have been trained with some 400 characteristics from multiple data series, incorporating periods of around 15 years and then repeatedly tested under different economic backdrops. The rigorous process is designed not only to maximise forecasting power but also to ensure that the models aren’t overfitted – that’s to say that they don’t only produce accurate results under a narrow set of circumstances but that they work under changing economic environments.

Despite the long development process, our AI models continue to evolve. New data series are constantly being evaluated. Those that meet our strict criteria are then tested within the models over many months and under rigorous conditions before they’re adopted, improving forecasting accuracy potential even further.

Do we really know how these black box models work?

AI and earlier machine learning models were often called ‘black box’ because, even for their developers, it was difficult to understand how they generated their results.

That’s another difference our proprietary AI has from other AI models. Our quantitative team spends considerable time and effort understanding how results are produced – knowing which aspects of the forecast are derived from linear relationships in the model, which come from non-linear relationships and which from the interactions of the data series. Being able to trace the path from input to output within our model gives us a near unparalleled transparency in how our model works, making it far from a black box. This transparency, in turn, means that not only can we explain the sources of our model’s investment performance, but also allows us to identify when something might be going wrong with the model.

Is this fully automated investing?

Yes and no. Yes in the sense that the model’s forecasts tell our investment managers whether to sell or buy a given stock. Because of our rigorous testing, we have faith in our model’s results to help guide decision making.

But humans – intelligent and experienced humans – are still necessary for the continued development and refinement of the model. They’re still necessary to ensure that the model is working correctly. And they’re necessary to specifying the portfolio’s constraints. For instance, our AI is a factor neutral, sector neutral, geography neutral strategy. Any stock recommendations made by the model for Quest AI-Driven have to be implemented with those constraints in mind. And should clients wish for bespoke portfolios – say having a particular geographic or stylistic preference – then those constraints also have to be taken into account.

What’s in it for clients?

We created Quest AI-Driven with a view to producing a strategy that has the same profile and level of riskiness as the wider market but that produces an additional return potential. Even an additional modest return annually can, over time, compound dramatically.

But unlike actively managed strategies – which also aim to outperform the market – the fact that our AI approach requires relatively fewer humans means that it has lower management costs. Typically our approach to AI costs little more than the passive strategy notwithstanding the expertise needed to build and maintain the model.