BuySideFinBench

A single-contributor, bilingual Chinese/English OpenCompass benchmark of 180 multiple-choice questions testing buy-side equity-research skills such as DCF valuation and three-statement linkage.

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categorydomain
Subcategorybuy-side equity-research and valuation reasoning (bilingual multiple-choice)
Page statusactive
Metricaccuracy (exact match after answer normalization)
Directionhigher_is_better
Unit%
Dataset size180
Dataset licenceApache License 2.0, per both the Hugging Face dataset card and the OpenCompass documentation page

What it measures

BuySideFinBench targets the analytical skills of a buy-side equity-research analyst rather than the sell-side, news-driven or surface-level knowledge tasks its own documentation says dominate existing finance benchmarks. It covers six subjects: tracing cash impact across the income statement, balance sheet and cash-flow statement; discounted-cash-flow valuation mechanics; comparable-company analysis; interpreting financial ratios in context; IFRS-versus-US-GAAP accounting-standard distinctions; and sensitivity/scenario analysis. Every subject is offered in both Chinese and English, so a score can also be read as a same-skill language comparison. Questions are original scenarios or paraphrased from public materials -- CFA/CICPA preparatory texts, SEC/HKEXnews/CSRC filing examples, and IFRS Foundation/FASB publications -- rather than reproduced exam content.

Task format

Four-option multiple-choice question (A-D) in one of six buy-side analysis subjects and one of two languages; the model returns a single letter answer, prompted five-shot from each subset's own five-question dev split.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub