AI won’t pick your stocks

Generative AI has moved very quickly from research labs into everyday life and tools that can draft an email or answer a question have sometimes been presented as universal problem solvers, including for investing.

Every adviser is currently being pitched some version of 'AI-powered investing', but the investment professionals running HSBC Asset Management’s Active Quant Equity strategy caution that AI’s role today should stay well clear of portfolio decisions.

AI doesn't create alpha

Daniela Hamoui, global head of the quant product specialist team at HSBC Asset Management, said: "I don't believe that AI creates alpha. AI can help you organise information, capture inputs that help differentiate and can help identify signals, but not necessarily generate new alpha altogether.

"We're a fully bottom-up manager. We know why we hold stocks and why we buy stocks; we will not buy something that we can't trace or explain. We have to have an economic intuition of why we're buying a stock."

There is a specific reason large language models (LLMs) make an uneasy fit for portfolio decisions. They work by sampling from a probability distribution over likely next words, generating an answer that looks statistically 'typical' of their training data rather than one that has been reasoned through and checked.

That process can produce fluent, plausible-sounding answers that are nonetheless wrong and it comes with no reliable measure of how confident the model actually is in what it has just said. For a task like picking stocks with client money behind it, that combination of fluency and unpredictability is an obvious problem.

Where AI genuinely saves time

None of which means HSBC Asset Management's quant team has no use for the technology, so long as a qualified human being stays in control and checks any AI-generated input against other evidence rather than acting on it directly.

The clearest, least controversial benefit so far has simply been speed. Hamoui said: "In terms of where AI has genuinely changed what we're doing, using it for internal research and productivity helps us enhance efficiencies from a research perspective, but also, for example, how we back-test or run simulations of new strategies. This is much faster now.

"Everything in our process becomes more efficient using AI."

AI, therefore, does not touch the HSBC Active Quant Equity models that decide which stocks are bought or sold. It simply means the research process that supports those models, testing ideas, running historical simulations and refining signals now happens faster than it used to.

An interesting frontier use is having AI convert unstructured text into something a systematic process can actually use. LLMs represent words and documents as points in a numerical space, where texts with similar meaning sit close together and dissimilar ones sit far apart.

Once language has been converted into numbers in this way, it becomes possible to measure how the tone of a piece of text is shifting or how closely current commentary resembles patterns seen before past profit warnings or upgrades. This measurement can be fed into a model alongside more familiar data such as prices and fundamentals.

Hamoui gave a concrete example of where HSBC Asset Management is exploring this. "We take very high -quality business news and what we do is quantify the topics that are dominating," she said. "Think about inflation risk or financial stress, for example. You see these topics coming through the news sources and then we understand how much coverage they're getting and how the tone is shifting over time."

One aim, she said, is to build what the team calls a recession narrative index, an intensity measure that could help explain stress emerging in markets by linking it to the themes dominating the news.

How a signal earns its place

For an AI-based process to hold any place in HSBC Asset Management’s Active Quant Equity approach, it has to clear a specific bar.

"We only use it if there's an economic intuition, a reason why this would hold up, and if it improves the results within our existing framework," Hamoui explained. "We're not just going to take it on its own. We always ask: does it add value to our existing framework?"

Across the strategy, any new input, whatever its source, is constructed clearly, tested against historical data, combined with existing indicators and monitored over time. Signals that prove unstable or unreliable are constrained or dropped; only those that add robust, explainable information are incorporated more fully.

That caution is deliberate, Hamoui said. "We don't want to just market ourselves as the next -gen AI and say 'we have new alpha because of AI', because that is not serious,” she concluded.

When we're talking about portfolio decisions or investing people's money, we're very cautious on that front. It's really important that we keep that discipline and governance around it."

Source: HSBC Asset Management, data as at July 2026.

The value of investments and any income from them can go down as well as up and investors may not get back the amount originally invested.

The value of investments and any income from them can go down as well as up and investors may not get back the amount originally invested. Characteristics are for illustrative purposes only, as at the date indicated, and subject to change without prior notice. Any forecast, projection or target where provided is indicative only and is not guaranteed in any way. HSBC Asset Management accepts no liability for any failure to meet such forecast, projection or target.

This product is based overseas and is not subject to UK sustainable investment labelling and disclosure requirements.