Introduction-to-Quantitative-Finance: A Chinese-Language Quant Curriculum Plus an arXiv Radar
AI+金融（量化）：1.多因子股票量化框架开源教程 2.学界和业界的经典资料收录 3.AI + 金融的相关工作，包括LLM, Agent, benchmark(evaluation), etc.
At a glance
- What is it?
- This repository from Barca0412 bundles a multi-factor equity research tutorial, a curated link map of quant tooling, and an automated AI + Finance arXiv Radar. It is a knowledge base and a data pipeline, not a trading library, and the two halves have very different adoption costs.
- Who is it for?
- Adopt it as a reading list and as a paper-tracking pipeline if you read Chinese and want a pre-clustered AI + Finance feed; the arXiv Radar half is the part with real machinery behind it. Do not adopt it as a backtesting engine, a factor library, or a source of runnable strategy code, because the repository does not claim to be any of those.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 4 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What this repository is, and what it is not
The README describes the project as a Chinese-language open knowledge base aimed at quantitative researchers and learners. It has three stated deliverables: a multi-factor equity research tutorial, a curated map of tools and courses, and the AI + Finance arXiv Radar. The tutorial section says it plans to open-source research from the Quant Group of the Hunan University Fintech Association, and includes a framework diagram. Everything else in the repository is a link collection: data sources and alternative data, backtesting frameworks, factor mining and evaluation, portfolio optimization and risk control, fund research and FoF, high-frequency trading, and a long list of external quant projects.
That distinction matters before you clone anything. There is no strategy engine here, no order simulation, no broker adapter. The README never claims there is. If you arrive expecting a runnable multi-factor backtest, you will find a syllabus and a bibliography instead. If you arrive expecting a maintained index of AI + Finance papers with a working pipeline behind it, that is the half of the repository that has actual code and generated data files.
The arXiv Radar is the part with a real pipeline
The README states that the site integrates an /arxiv/ sub-site built on VitePress, replacing an older arXiv documentation experience. The listed capabilities are a paper list, trend charts, semantic search, institution filtering, and tag aggregation. The data lives in three files: data/papers.json, data/stats.json, and data/embeddings_index.json. The pipeline code sits in scripts/arxiv_radar/. Refresh is triggered with npm run arxiv:update, and the README notes that this command also refreshes the machine-updated status block embedded in the README itself, which is delimited by ARXIV_RADAR_STATUS markers.
The status block is the most concrete evidence of scale in the whole repository. It reports a latest update of 2026-09-09, 18503 indexed papers, 9056 focus papers, a latest publication date of 2026-09-08, and 10 monitored categories. Those numbers come from the README's own status block, not from any independent measurement, and they say nothing about quality. What they do establish is that the pipeline runs on a schedule and that the focus subset is roughly half the index, which implies a filtering rule exists somewhere in scripts/arxiv_radar/ that decides which papers are worth surfacing. The README does not document that rule, and that omission is the first thing a serious user should chase down.
Semantic search without a documented embedding model
The presence of data/embeddings_index.json tells you the Radar does vector search rather than keyword matching. What it does not tell you is which embedding model produced those vectors, how chunking was handled for long abstracts, or how often the index is rebuilt relative to the paper list. Those are not nitpicks. If papers.json refreshes daily but the embedding index lags, newly indexed papers become invisible to semantic search while remaining visible in the list view, and a user has no way to notice the gap from the interface.
The README gives no schema for the three data files either. A contributor who wants to add a tag or fix a misclassification has to read the pipeline source to learn the record shape. For a project whose stated contribution path includes improving Radar tags, that is a real friction point. The contributing section asks people to discuss ideas in GitHub Discussions first and to submit small fixes as pull requests with source and use-case notes, which is a reasonable process, but the missing data documentation makes the Radar half harder to contribute to than the tutorial half.
Getting it running: two commands and a lot of reading
The README does not include a full installation walkthrough. What it does give is npm run arxiv:update for refreshing the Radar data and the README status block, and a table of entry points: the documentation site for systematic study of the multi-factor framework, the /arxiv/ page for paper tracking, the resource map for tools and courses, and GitHub Discussions for suggestions. The site itself is at barca0412.github.io/Introduction-to-Quantitative-Finance.
Because the site is built on VitePress, anyone who wants to run it locally will need Node and the project's package.json scripts, which the README does not enumerate. The tutorial section is explicitly described as planned rather than complete: the README lists what is included now (quant and machine learning reference books, technical indicator backtest code, sell-side quantitative research reports, investor sentiment and behavioral finance papers) and what is still to come (quant written tests and MFE practice problems, portfolio optimization, machine learning factor mining). Treat the tutorial as a work in progress with a published outline, not a finished course.
The curated list is the real product, and its weakness is link rot
The bulk of the repository is an annotated link map. It names external projects across categories: qlib and MultiFactor under multi-factor frameworks; backtesting.py, vectorbt, and zipline-reloaded under backtesting; AlphaAgent, alpha-gfn, and RD-Agent under factor mining and machine learning; Riskfolio-Lib and PyPortfolioOpt under portfolio optimization; FinRL under reinforcement learning trading; hftbacktest and VisualHFT under high-frequency trading. Each entry gets a one-line description in Chinese. The value is in the grouping and the short annotations, which reflect choices a practitioner made rather than an alphabetical dump.
The cost of that design is maintenance. Every entry is an outbound link to someone else's repository, and the README already contains at least one truncated link (the hummingbot entry ends mid-URL in the supplied text) and one empty section heading for fund research and FoF with no content behind it. A link map degrades quietly. Nothing in the pipeline checks whether the listed projects still build, still accept issues, or still exist. If you adopt this repository as a reading list, expect to verify each link yourself before you invest time in it.
Where this is the wrong tool
If you need to backtest a factor, this repository will not do it. It points you at backtesting.py, vectorbt, and zipline-reloaded, and it points you at MultiFactor for a Huatai-report-based multi-factor backtest framework, but it does not provide the engine. If you need a factor library with computed values, it does not provide one. If you need a paper feed in English with full-text search across publishers, the Radar covers arXiv only, across ten monitored categories that the README does not name.
The language boundary is also a hard constraint. The tutorial, the annotations, and the resource map are in Chinese. An engineer who does not read Chinese can still use the /arxiv/ sub-site and the data files if the paper metadata is in English, but the curated list, which is the largest part of the repository, becomes a list of names without the reasoning that makes it useful. That is not a defect, it is a scope decision, but it determines who can actually extract value here.
How it differs from an awesome-list or a framework like qlib
Two comparisons are worth making. Against a plain awesome-list, the difference is the Radar: this repository ships generated data (papers.json, stats.json, embeddings_index.json), a scheduled pipeline in scripts/arxiv_radar/, and a status block that updates automatically. That is infrastructure an awesome-list does not have. Against a framework like qlib, which the README itself lists under multi-factor frameworks and describes as an AI-oriented quantitative investment platform supporting automated factor mining, the difference is category. qlib executes research; this repository teaches and indexes it. Choosing between them is not a trade-off, because they do different jobs. The honest framing is that this project is a companion to tools like qlib, not a substitute.
The MIT license removes most adoption friction for the text and the pipeline code. It does not extend to the linked external projects, each of which carries its own license, and it does not cover the sell-side research reports and reference books the README mentions, which are third-party copyrighted material regardless of how they are indexed here. If you plan to redistribute anything beyond the repository's own content, check the original source.
Maintenance cost and who should adopt it
The Radar half has a low, bounded maintenance cost: one npm run arxiv:update command, three data files, and a pipeline directory. The link-map half has an unbounded one, because it depends on the health of dozens of external repositories the maintainers do not control. The tutorial half has the highest cost of all and is the least finished, which is why the README marks several topics as pending.
Adopt it if you read Chinese and want a structured entry point into multi-factor equity research, or if you want a pre-clustered AI + Finance paper feed and are willing to read the pipeline source to understand the filtering. Do not adopt it if you need executable strategy code, English-language instruction, or a guarantee that every listed tool is current. The single most useful thing to verify first is the focus-paper rule in scripts/arxiv_radar/: 9056 of 18503 indexed papers are marked as focus, and until you know how that split is computed, you cannot judge whether the Radar is surfacing what you care about.
Editorial conclusion
Adopt it as a reading list and as a paper-tracking pipeline if you read Chinese and want a pre-clustered AI + Finance feed; the arXiv Radar half is the part with real machinery behind it. Do not adopt it as a backtesting engine, a factor library, or a source of runnable strategy code, because the repository does not claim to be any of those. Before relying on the Radar, open scripts/arxiv_radar/, confirm which of the ten monitored categories it queries, and check whether data/embeddings_index.json is regenerated on the same schedule as data/papers.json, since a stale embedding index would silently degrade the semantic search the site advertises.
Community notes