deepcreditbench/DeepCreditBench

DeepCreditBench是一个用于企业信用风险评估的基准数据集，基于公开可用的真实世界企业、市场和宏观金融数据构建，而非合成或LLM生成的记录。它支持三个评估层级：第一层（基于时间跨度的企业违约预测）、第二层（在受控分布偏移下的基于场景的信用风险评估）和第三层（使用结构化LLM提示的分析性信用评估）。数据集组织在firm-month级别，每个观察对应一个公司在每月时间戳上的信息。DeepCreditBench is a benchmark dataset for corporate credit risk assessment, built upon publicly available real-world corporate, market, and macro-financial data rather than synthetic or LLM-generated records. It supports three distinct evaluation tiers: the first tier focuses on firm default prediction based on time spans, the second tier covers scenario-based credit risk assessment under controlled distribution shifts, and the third tier involves analytical credit evaluation using structured LLM prompts. The dataset is organized at the firm-month level, where each observation corresponds to the information of a company at a monthly timestamp.