Reasoning Beyond the Obvious: Evaluating Divergent and Convergent Thinking in LLMs for Financial Scenarios

TL;DR

This paper introduces ConDiFi, a benchmark for evaluating both divergent and convergent reasoning in large language models within financial scenarios, highlighting differences in model performance on creative and decision-making tasks.

Contribution

The paper presents ConDiFi, a novel benchmark that assesses reasoning beyond factual accuracy, specifically targeting financial decision-making and creative thinking in LLMs.

Findings

GPT-4o underperforms on novelty and actionability

DeepSeek-R1 and Cohere Command R+ excel in actionable insights

Models show significant variation in reasoning capabilities

Abstract

Most reasoning benchmarks for LLMs emphasize factual accuracy or step-by-step logic. In finance, however, professionals must not only converge on optimal decisions but also generate creative, plausible futures under uncertainty. We introduce ConDiFi, a benchmark that jointly evaluates divergent and convergent thinking in LLMs for financial tasks. ConDiFi features 607 macro-financial prompts for divergent reasoning and 990 multi-hop adversarial MCQs for convergent reasoning. Using this benchmark, we evaluated 14 leading models and uncovered striking differences. Despite high fluency, GPT-4o underperforms on Novelty and Actionability. In contrast, models like DeepSeek-R1 and Cohere Command R+ rank among the top for generating actionable, insights suitable for investment decisions. ConDiFi provides a new perspective to assess reasoning capabilities essential to safe and strategic deployment of LLMs in finance.