# A Financial Brain Scan of the LLM

## TL;DR

This paper introduces a transparent and lightweight method to analyze and steer large language models in economic forecasting, enabling understanding and manipulation of their reasoning related to sentiment, technical analysis, and bias correction.

## Contribution

It presents a novel approach to map LLM reasoning to plain-English concepts and steer their biases without performance loss, facilitating social science research.

## Key findings

- Mapped economic forecast concepts to sentiment, technical analysis, and timing.
- Steered models to vary risk aversion, optimism, and pessimism.
- Achieved interpretability and bias correction without performance degradation.

## Abstract

Emerging techniques in computer science make it possible to "brain scan" large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance. We also show that models can be steered to be more or less risk-averse, optimistic, or pessimistic, which allows researchers to correct or simulate biases. The method is transparent, lightweight, and replicable for empirical research in the social sciences.