The emergence of ChatGPT has catapulted artificial intelligence (AI) into the mainstream conscience.
AI’s ability to replicate functions in industries as varied as accountancy, retail and logistics has left many working in these sectors wondering when – not if – it will impact their jobs and how they can learn to work with the technology.
In the context of investing, some have begun to question whether AI will be able to predict the stock market.
The answer is far from straightforward. There have already been several advances in AI and machine learning (ML) which have changed how analysts operate and interpret markets.
Algorithms can have a hard time finding the relationship between the drivers and the outcome of a stock’s returns
These include the ability to undertake functions such as data cleaning, credit scoring and trading optimisation. The bigger question is whether AI can predict the stock market reliably.
At the moment, the answer is no, but it is not inconceivable that this will change. Geoffrey Hinton, considered the “godfather of AI”, recently stated that general purpose AI – systems which can learn any task that humans can do – could now exist in fewer than 20 years, when previous estimates put it closer to 50.
Financial markets are vastly different to other disciplines where AI performs well, though, such as the physical sciences or consumer internet domain. One major difference is that financial data has a low signal-to-noise ratio, meaning there is no single variable that determines how something will perform.
Variables including company earnings, banks’ interest rates and investor attitudes can all impact investment returns. This differs from an algorithm that offers movie recommendations to subscribers of streamed content channels based on the kind of movies the subscriber has already watched on the platform.
There are already ways AI could indirectly predict the markets by making judgements on factors that influence them
When it comes to the complex financial market, therefore, ML algorithms can have a hard time finding the relationship between the drivers and the outcome of a stock’s returns.
Another challenge for ML is the amount of available data in financial markets compared to other domains. A significant driver of an ML algorithm’s ability to accurately predict trends is the volume of data it has at its disposal.
Data traditionally used by quantitative investors is often made available only quarterly or monthly, which pales in comparison to other domains.
We might also consider the inherent fluctuation that exists within financial markets. This nonstationary characteristic differs from domains where ML has stood out, such as physical sciences which often have static systems.
While many ML algorithms can be designed to adapt to evolving systems, there remains the question of how valid and applicable the historical data used to train the algorithm is.
Although there are significant obstacles to overcome before AI might reliably forecast financial markets, there are already ways in which it could indirectly predict the markets by making judgements on factors that influence them.
One example is ML algorithms predicting company fundamentals, such as corporate earnings. Not only are there often fewer variables that can impact a company’s fundamentals, but they are also more stable than stock returns, making them better suited for ML predictions.
Finance professionals are keen to learn how to work with AI to help them gain an edge.
As AI becomes ever more entrenched into investment processes, its growing presence is making the desire for informed predictions stronger. And, as acknowledged by Hinton, AI’s capabilities are developing at a pace quicker than many expected.
By extension, its ability to make reliable predictions of stock market returns may not be too far away.
Larry Cao is senior director of industry research at CFA Institute
