【Hacker News搬运】LLM在金融市场的应用
-
Title: Financial market applications of LLMs
LLM在金融市场的应用
Text:
Url: https://thegradient.pub/financial-market-applications-of-llms/
这篇文章讨论了大型语言模型(LLMs)在金融市场中的应用。文章首先指出,LLMs在处理代表单词或单词部分的标记序列方面非常出色,这使得它们能够完成诸如翻译、问答和根据简单的用户提示生成类似人类的文本等任务。 文章进一步探讨了LLMs在量化交易中的应用潜力,特别是在统计套利等股票策略中,识别自回归结构的研究。然而,由于可用于训练模型的数据量和信息含量有限,这一应用面临着挑战。例如,文章提到,在2023年的NeurIPS会议上,高频交易公司Hudson River Trading比较了用于训练GPT-3的输入标记数量与股票市场数据中可训练标记的数量。他们估计,在3000只交易股票、每只股票每天10个数据点、每年252个交易日、每个交易日23400秒的情况下,每年可用的市场数据中有1770亿个股票市场标记。而GPT-3是在5000亿个标记上训练的,所以差距并不大。 文章还指出,尽管LLMs在预测金融时间序列方面的应用存在挑战,但AI在金融市场中的应用前景仍然值得期待。例如,多模态学习是一个新兴的AI研究领域,它使用不同模态的数据(例如图像和文本输入)来构建统一的模型。在金融领域,多模态努力可能有助于将传统数据源(如技术时间序列数据)与替代数据(如推特上的情感或图形交互、新闻文章和公司报告的自然语言,或商品中心港口的货运活动卫星图像)结合起来。 此外,文章还提到了LLMs在合成数据创建方面的应用,这可能包括模拟股票价格轨迹,这些轨迹模仿市场观察到的特征。这种合成数据对于相对稀缺的金融市场数据来说可能非常有用。 总的来说,尽管LLMs在量化交易中的直接应用存在挑战,但它们在金融市场分析中的潜力仍然值得关注。随着AI模型的不断改进,LLMs可能会帮助分析师完善投资论点,揭示管理评论中的矛盾,或发现不同行业和企业之间的潜在关系。
Post by: andreyk
Comments:
jsemrau: A lot of words for not bringing much new content to the discussion.
I think the most interesting application of LLMs in Finance are<p>(1) synthetic data models for data cleansing,
(2) journal management,
(3) anomaly tracking,
(4) critiquing investments<p>All of this should be done by professionals and nothing is "retail" ready.jsemrau: 很多话没有给讨论带来太多新内容。我认为LLM在金融中最有趣的应用是<p>(1)用于数据清理的合成数据模型,(2) 期刊管理,(3) 异常跟踪,(4) 批评投资<p>所有这些都应该由专业人士完成;零售;准备好的
conorh: We are working on a project for a client which functions as an analysis tool for stocks using LLMs. Ingesting 10ks, presentations, news, etc. and doing comparative analysis and other reports. It works great, but one of the things we have learned (and it makes sense) is that traceability of the information for financial professionals is very important - where did the facts and information come from in what the AI is producing. A hard problem to solve completely.
conorh: 我们正在为一个客户进行一个项目,该项目可以作为使用LLM的股票分析工具。拍摄10公里、演示、新闻等,并进行比较分析和其他报道。它很有效,但我们学到的一件事(也是有道理的)是,金融专业人员的信息可追溯性非常重要——人工智能产生的事实和信息来自哪里。一个很难完全解决的问题。
steveBK123: LLMs labor savings will only help financial market participants if they manage to do it without hallucinations / can maintain ground truth.<p>Sure its great if your analysts save 10 hours because they don't need to read 10Ks / earnings / management call transcripts .. but not if it spits out incorrect/made up numbers.<p>With code you can run it and see if it works, rinse & repeat.<p>With combing financial documents to then make decisions, you'll realize it made up some financial stat after you've lost money. So the iteration loop is quite different.
steveBK123: LLM的劳动力储蓄只有在金融市场参与者设法做到这一点而没有幻觉的情况下才能帮助他们;能够维持基本事实<p> 当然,如果你的分析师节省了10个小时,那就太好了,因为他们不会浪费时间;t需要读取10Ks;收益;管理通话记录。。但如果它吐出不正确的;虚构的数字<p> 有了代码,你可以运行它,看看它是否有效,冲洗&;重复p> 通过梳理财务文档来做出决策,您可以;我意识到它在你之后编造了一些财务数据;I’我赔钱了。所以迭代循环是完全不同的。
hydershykh: I think some of the financial applications around LLMs right now are better suited for things like summarization, aggregation, etc.<p>We at Tradytics recently built two tools on top of LLMs and they've been super popular with our usercase.<p>Earnings transcript summary: Users want a simple and easy to understand summary of what happened in an earnings call and report. LLMs are a nice fit for that - <a href="https://tradytics.com/earnings" rel="nofollow">https://tradytics.com/earnings</a><p>News aggregation & summarization: Given how many articles get written everyday in financial markets, there is need for a better ingestion pipelines. Users want to understand what's going on but don't want to spend several hours reading through news - <a href="https://tradytics.com/news" rel="nofollow">https://tradytics.com/news</a>
hydershykh: 我认为目前围绕LLM的一些金融应用程序更适合于总结、聚合等。<p>我们Tradytics最近在LLM之上构建了两个工具,它们;We在我们的用户案例中非常受欢迎<p> 收益记录摘要:用户希望对电话财报会议和报告中发生的事情进行简单易懂的总结。LLM非常适合这一点-<a href=“https://;/;tradytics.com/:盈余”rel=“nofollow”>https:///;tradytics.com;收益</a><p>新闻聚合&;总结:考虑到金融市场上每天都有多少文章被写出来,我们需要一个更好的吸收渠道。用户想要了解;正在进行,但不要;我不想花几个小时阅读新闻-<a href=“https://;/;tradytics.com/!news”rel=“nofollow”>https:///;tradytics.com;新闻</a>
btbuildem: There were some developments using LLMs in the timeseries domain which caught my attention.<p>I toyed with the Chronos forecasting toolkit [1], and the results were predictably off by wild margins [2]<p>What really caught my eye though was the "feel" of the predicted timeseries -- this is the first time I've seen synthetic timeseries that look like the real thing. Stock charts have a certain quality to them, once you've been looking at them long enough, you can tell more often than not whether some unlabeled data is a stock price timeseries or not. It seems the chronos LLM was able to pick up on that "nature" of the price movement, and replicate it in its forecasts. Impressive!<p>1: <a href="https://github.com/amazon-science/chronos-forecasting">https://github.com/amazon-science/chronos-forecasting</a><p>2: <a href="https://imgur.com/a/hTRQ38d" rel="nofollow">https://imgur.com/a/hTRQ38d</a>
btbuildem: 在时间序列领域中使用LLM的一些发展引起了我的注意<p> 我摆弄了Chronos预测工具包[1],结果不出所料地相差悬殊[2]<p>然而,真正吸引我眼球的是“;感觉“;在预测的时间序列中——这是第一次I;我看过一些合成的时间序列,看起来像真实的东西。股票图表有一定的质量,一旦你;我看了足够长的时间,你经常可以判断一些未标记的数据是否是股价时间序列。看来chronos LLM能够抓住这一点;自然”;并在其预测中重复这一点。给人印象深刻的p> 1:<a href=“https://;/;github.com/:亚马逊科学/!时间预测”>https:///;github.com/;亚马逊科学;时间预测</a><p>2:<a href=“https://;/;imgur.com/:a/!hTRQ38d”rel=“nofollow”>https:///;imgur.com/;a;hTRQ38d</a>