Financial Deep Indicator (FinDeepIndicator)

Financial indicators are essential tools for transforming raw financial data into interpretable measures for various downstream tasks, such as valuation, risk assessment, and economic analysis. However, existing financial benchmarks largely focus on answer-level accuracy and often assume that relevant data are already provided, leaving the assessment of the intermediate process of indicator construction underexplored. FinDeepIndicator is the first benchmark dedicated to evaluating Deep Research (DR) agents in end-to-end financial indicator construction. Specifically, FinDeepIndicator evaluates DR agents across four stages in indicator construction: formula specification, data collection, indicator calculation, and answer generation, and covers fundamental, technical, and macroeconomic indicators organized into 21 fine-grained sub-categories. It contains 3,350 curated question-answer (QA) pairs derived from both U.S. and Chinese markets, 10 years of historical financial data, and 800 listed companies. Extensive experiments on search-equipped Large Language Models (LLMs) and DR agents show that, while LLMs generally perform well in formula specification, their accuracy drops substantially during data retrieval and numerical execution. DR agents consistently outperform search-equipped LLMs, yet remain unreliable in realistic financial analysis settings. These findings provide insights for developing more capable and trustworthy DR agents in finance.

🏠︎ Homepage | 📑 Paper | 🐙 GitHub

Task

The task aims to evaluate the end-to-end construction of financial indicators. Given a question, an AI agent is required to correctly identify the target indicator and its corresponding formula, retrieve the necessary data from online sources, compute the indicator, and perform logical reasoning to derive the final answer. Formally, given a question (Q), the AI agent is expected to generate the indicator formula (F), retrieve the relevant raw data (D), obtain the calculated indicator value (I), and produce the final answer (A). Thus, the task can be represented as the mapping Q → (F, D, I, A).

Dateset

It contains 3,350 question–answer pairs spanning two markets (U.S. and Chinese markets), and covers three major categories, 21 subcategories, and 234 indicators.

Submission

To submit the predictions of your method, please run the evaluation script in GitHub and send your results in CSV format to OpenFinArena@gmail.com.

Please use the following subject in your email: FinDeepIndicator-{Method Name}-{Method Type}-{Team Name}-{Organization}

The Method Name should be a unique name for your system/method, not just the name of the base model used (e.g., "MyAwesomeMethod", "FinAgent-v2", not "GPT-4").

The Method Type might be "LLM with Search", "Deep Research" or other types that can best explain the type of your method. If the method type cannot be confirmed or you choose not to disclose it, please indicate "Other".

If you have a dedicated logo for your Organization, please attach the logo in the size of 100x100, which will be presented in the leaderboard.

Contact

The OpenFinArena team is open to collaboration across development, application, and partnership. Whether you have inquiries, technical questions, or are interested in contributing to the evolution of this platform, please join the OpenFinArena Google Group or drop us an email at OpenFinArena@gmail.com.