One Size Fits None: Heuristic Collapse in LLM Investment Advice

TL;DR

This paper reveals that large language models tend to simplify complex investment decisions by overly relying on risk tolerance, highlighting the need for careful input sensitivity auditing in high-stakes advisory roles.

Contribution

It introduces the concept of heuristic collapse in LLMs, demonstrating that they often reduce multi-factor decisions to a few dominant inputs, especially in investment advice.

Findings

Investment decisions are mainly driven by self-reported risk tolerance.

Web search reduces but does not eliminate heuristic collapse.

Model scale and web augmentation alone do not resolve heuristic collapse.

Abstract

Large language models are increasingly deployed as advisors in high-stakes domains -- answering medical questions, interpreting legal documents, recommending financial products -- where good advice requires integrating a user's full context rather than responding to salient surface features. We investigate whether frontier LLMs actually do this, or whether they instead exhibit heuristic collapse: a systematic reduction of complex, multi-factor decisions to a small number of dominant inputs. We study the phenomenon in investment advice, where legal standards explicitly require individualized reasoning over a client's full circumstances. Applying interpretable surrogate models to LLM outputs, we find systematic heuristic collapse: investment allocation decisions are largely determined by self-reported risk tolerance, while other relevant factors contribute minimally. We further find that web search partially attenuates heuristic collapse but does not resolve it. These findings suggest that heuristic collapse is not resolved by web search augmentation or model scale alone, and that deploying LLMs as advisors requires auditing input sensitivity, not just output quality.