AI and Quant Investing - Powerful New Tools, Not a Magic Wand

Generative artificial intelligence (AI), and in particular large language models (LLMs), has moved very quickly from research labs into everyday life. Tools that can draft emails, answer questions or create images have made AI highly visible and have sometimes been presented as universal problem solvers, including for investing. Experience from quantitative investing suggests a more balanced view: different problems require different tools, all methods have strengths and weaknesses, and human judgement remains essential. This note explains in simple terms what LLMs are, where they are strong and weak, and how their most useful features can be harnessed inside a disciplined quant equity process.

What is new in AI – and why it matters for investors

Recent progress in AI has been most spectacular in image and language processing, where large artificial neural network models can detect patterns in very rich, stable data sets. LLMs are language models trained on huge text corpora using a statistical method called maximum likelihood estimation, so that the texts they generate reproduce the conditional word frequencies found in the corpus. When you ask a question, the model aims to produce an answer such that the combined question and answer look “typical” of what appears in its training data. This simple objective, applied at scale, is enough to produce answers that are often syntactically correct, semantically plausible and apparently well-reasoned.

LLMs are also generative: they do not just classify existing data; they create new content by sampling from a probability distribution over possible next words at each step. This built in randomisation allows variety and creativity, but it also means that the same question can receive different answers at different times. For investors, this combination of fluency and randomness is crucial: it demonstrates the power of quantitative techniques, but also shows why no single method, however sophisticated, should be treated as a magic wand for portfolios.

Responsibility: why humans must stay in charge

Because LLMs rely on random generation, can hallucinate and lack reliable introspection, their answers are inherently somewhat unpredictable and do not carry a trustworthy measure of confidence. Some assertions may reflect fake knowledge or inadequate reasoning, and the tool itself cannot be held responsible for the consequences. In high stakes domains such as medicine, law or finance, this raises obvious responsibility concerns: if an AI generated suggestion leads to a bad decision, liability rests with the human professionals and organisations that chose to rely on it.

For investment management, this means LLM outputs should be treated as inputs into a broader process, not as final decisions. A qualified human being with a critical mind needs to remain in control, checking AI generated material, integrating it with other evidence and bearing responsibility for the outcome. As AI adoption widens, robust supervision and governance remain key.

The real opportunity for quant investors: turning text into data

The most useful competence of LLMs for quantitative investing is not that of a modern oracle delivering buy or sell calls, but their ability to provide a numerical approximation of meaning and a measurable notion of semantic similarity. To operate on language, LLMs represent words, sentences and documents as points in a very high dimensional vector space, in which texts with similar meanings are close and dissimilar texts are far apart. This representation is learned empirically, using the same likelihood method that trains the model to match corpus statistics, and its effectiveness is demonstrated by the level of syntactic and semantic competence LLMs achieve.

Once text is embedded numerically, investors can compute distances between documents, identify clusters and extract quantitative features from language. This opens many possibilities: scoring changes in the tone of company communications, tracking themes in earnings calls, measuring how similar current news flow is to past periods that preceded profit warnings or upgrades, or extracting information from regulatory and sustainability reports. Quantitative experts can inject these text-based signals into their models alongside traditional data such as prices and fundamentals, to help with return prediction or risk measurement. This extension of the information set is probably the most tangible contribution of the LLM revolution to quantitative investment and is already driving a very active area of empirical research in finance.

Narrow AI, generative AI and disciplined quant equity

More broadly, many of the most robust benefits of AI – for example in medical imaging – are likely to come from standard, narrow AI systems designed for specific tasks, rather than from general purpose generative chatbots. In finance, this suggests the greatest value will come from AI modules that do particular jobs well, such as extracting signals from text or improving risk measurement, rather than from handing portfolio decisions to a generative model whose outputs are randomised and sometimes hallucinatory.

Quant equity strategies provide a natural framework for using AI in this way. They are built on systematic rules, diversification and explicit risk controls, rather than on single big bets or purely discretionary stories. Within this framework, AI derived text measures can be treated like any other signal: they are constructed clearly, tested on historical data, combined with other indicators and monitored through time. Signals that prove unstable or unreliable can be constrained or discarded; those that add robust information can be incorporated more fully. Human portfolio managers and researchers remain responsible for the design of the models, the challenge of the signals and the final investment decisions.

For investors, the practical message is that AI – and LLMs in particular – are best viewed as advanced tools that can broaden and deepen the data used in a disciplined, risk-controlled quant process, not as stand-alone decision engines that remove uncertainty or guarantee better returns.

The views expressed above were held at the time of preparation and are subject to change without 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. Source: HSBC Asset Management, June 2026