AI in systematic equities: Why discipline matters more than speed

Dr Maximilian Stroh, CFA, Head of Research, discusses how AI can support equity investing, where its limits lie, and why robustness, diversification, and human judgement remain central in an increasingly AI-enabled investment environment.

Max, many investors use the term AI very broadly. What does AI mean in your investment process?

We separate different use cases. In forecasting, AI usually means machine-learning methods aggregating signals in a way that can capture non-linear relationships or interactions that traditional linear models miss. Natural language processing for signal generation is another area, especially when we analyse news, reports, or other unstructured information. In research and operations, agentic AI is a productivity and knowledge-management tool, for example, helping researchers navigate complex codebases, migrate code, or screen large bodies of academic articles.

Where does AI sit in your equity investment pipeline today?

At the core, machine learning is used in security selection, but not in isolation. We start with a linear forecasting framework based on established concepts such as value, quality, and sentiment. Machine learning can then help us identify non-linearities and interactions that a more traditional model does not fully capture.

The important point is that AI refines the signal rather than replacing the established investment process. We evaluate whether the combined approach improves the overall forecasting system and, ultimately, whether it improves portfolio outcomes after accounting for risk, diversification, liquidity, and transaction costs.

And does it improve results?

Yes, although the impact is measured. Machine learning tends to deliver a modest but positive contribution. In most cases, the more noticeable improvement is in risk-adjusted returns rather than a dramatic increase in overall alpha.

Many investors worry that AI is fundamentally changing market behaviour. How do you think about the risk of AI-driven market regime shifts?

AI can influence markets through at least two channels. First, it can change company fundamentals by affecting business models, margins, competitive dynamics, and capital allocation. Second, it can change how quickly investors process information.

From an investment perspective of a systematic manager, the objective is not to predict every AI-driven trend in advance. The more realistic objective is to build a framework that can cope with changing leadership, changing correlations, and shorter signal half-lives. Diversification across factors, regions, sectors, and individual positions remains essential.

Another topic investors often raise is alpha decay – the gradual erosion of returns as successful signals become crowded. Does AI make this problem worse?

Not necessarily. Agentic AI lowers the cost of searching through data, which means more investors can discover similar patterns at the same time. Some of those patterns may be economically meaningful, but many will simply be artefacts of the sample period. The danger is that AI can produce many appealing signals that may look more promising than they really are.

This is why research discipline becomes more important, not less. If AI is treated as a magic signal generator, the risk of overfitting increases. If it is treated as one tool inside a controlled research process, it can be valuable.

What do those controls look like in practice?

We do not need every model to be simple, but we do need every model to be reliable and robust. That means proven data quality, the usage of holdout data, multiple-testing controls, sensitivity analysis, implementation-cost assumptions, liquidity analysis, and ongoing live monitoring.

We also look at whether a model improves the portfolio, not just whether it improves a statistical metric in isolation. Forecast accuracy is useful, but it is not the final objective. The final objective is a better investment outcome.

Beyond stock selection, are you using AI elsewhere?

Very much so. One of the most immediate benefits is in speeding up workflows. AI-driven agents can support tasks such as migrating legacy signals to new platforms, searching internal documentation, summarising research material, or helping researchers review code and text more efficiently.

Some workflows can be automated technically, but we deliberately keep them inside controlled environments. Outputs are reviewed, code changes are version-controlled, and responsibility remains with the researcher or quant engineer.

What are the main areas you are focused on now?

Three areas stand out. First, we are extending AI support across more of the research process, not only idea generation but also evaluation. That raises important quality-control questions, because AI-generated outputs can sound convincing while still being wrong.

Second, we are revisiting how we analyse news and other unstructured data. LLMs have already shown value in measuring macro sentiment, and these approaches may also improve equity selection.

Third, we are working on machine learning models that are better aligned with portfolio objectives. Instead of optimising purely for statistical accuracy, the goal is to optimise how a model contributes to risk-adjusted returns at the portfolio level, after realistic implementation assumptions.

Finally, if you had to sum up your philosophy on AI in investing for professional investors, what would it be?

AI is a powerful addition to the systematic investor’s toolkit, but it does not suspend the basic rules of quantitative research. Input data needs to be clean. Signals still need economic rationale. Models still need statistical validation. Portfolios still need diversification and risk control. And accountability still needs to sit with people.

In that sense, the firms best positioned to benefit from AI will be those that combine speed of adoption with disciplined integration.

Conclusion

AI is transforming asset management, but the transformation is more disciplined integration than wholesale replacement. For professional investors, the relevant question is not whether a manager uses AI, but how AI is governed, measured, and embedded in the investment process.

Used well, AI can improve forecasting, research productivity, and the analysis of unstructured information. Used poorly, it can accelerate overfitting, crowding, and operational risk.