Research

A PRISMA-compliant review screening 847 records down to 60 qualifying studies finds the academic literature still cannot agree on whether AI-driven algorithmic trading improves or degrades market liquidity, price discovery and stability, only that it changes all three

Fuli Yang, Shiyu Lin and Jing Huang's systematic literature review, published on SSRN 15 July, is less a verdict than a map of how unsettled the underlying research still is, which matters more than another single study's headline finding for anyone trying to reason about what widespread AI trading actually does to market structure.

· Source: SSRN


Systematic literature reviews are a different kind of research output than the individual studies they synthesise, and this one is useful for exactly that reason. Screening 847 initial records across Web of Science, Scopus, SSRN and Google Scholar down to 60 studies meeting pre-specified inclusion criteria, the authors are not testing a new hypothesis about AI trading, they are auditing what the existing body of published research actually supports once weaker or non-comparable studies are filtered out. That is a more useful signal for anyone trying to separate genuine findings from the noise of individual papers.

The review's organising frame is market microstructure's three foundational dimensions, liquidity provision, price discovery and market stability, and it tracks how large language models and machine learning classifiers, not just older-generation algorithmic and high-frequency trading, are reshaping each. What comes through is not a clean answer in either direction. Some of the 60 studies find AI-driven trading improves liquidity and speeds up price discovery by processing information faster than human-paced markets could. Others find the same mechanisms concentrate risk and amplify instability when models react to the same signals in correlated ways, the herding dynamic regulators have flagged repeatedly this year.

That lack of consensus is itself the finding worth sitting with. Two decades into the shift from human to algorithmic market-making, and with LLM-based trading agents now a live deployment rather than a research curiosity, the academic literature still has not converged on whether the net effect on market quality is positive or negative. That is a genuinely different situation than, say, high-frequency trading research a decade ago, where a rough consensus eventually formed around liquidity benefits offset by tail-risk costs. AI-driven trading has not reached that stage of settled understanding yet, despite being deployed at meaningfully larger scale.

For regulators and practitioners, the practical implication is that policy built on a confident model of what AI trading does to markets is currently ahead of what the evidence supports. The FCA's principles-based approach and the SEC's reliance on existing rules rather than AI-specific frameworks look, against this review, less like caution and more like an accurate reflection of how unsettled the underlying science still is. Anyone citing a single study's finding as settled fact about AI's effect on market structure is, per this review, citing one of 60 studies that do not agree with each other.


Read the original: SSRN - A PRISMA-compliant review screening 847 records down to 60 qualifying studies finds the academic literature still cannot agree on whether AI-driven algorithmic trading improves or degrades market liquidity, price discovery and stability, only that it changes all three. Commentary is the independent editorial view of Share Trading; the original article is credited to its publisher.