#EconTwitter
New paper with
@YuehuaTang
:
"Can ChatGPT Forecast Stock Price Movements?"
Using news headlines, we explore
#ChatGPT
's potential in predicting stock market returns.[1/7]
📖 My new book is out soon!
Everything you wanted to know about AI and stock markets.
The Predictive Edge: Generative AI and ChatGPT in Financial Forecasting
Homework, cover letters... stock forecasts?
University of Florida professor
@alejandroll10
says ChatGPT *might* just be a top stock picker... someday.
He breaks down his fascinating research paper on what could be the future of trading:
🏆Very happy to win the Blackrock best paper award with
@YuehuaTang
!
@UF
3rd prize in two weeks!
Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models
Paper link:
My second post is out!
Human-Level Referee Reports Using Claude, Not ChatGPT
How to obtain harsh, helpful referee reports to improve your research; and why large context windows enable a comprehensive understanding of texts.
Link ⬇️
Paper with
@NickRoussanov
challenging the core of arbitrage pricing theory and questioning the trade-off between systematic risk & expected returns.
Do Common Factors Really Explain the Cross-Section of Stock Returns? Not really!
#EconTwitter
#fintwit
Higher risk yields higher returns, says an old investing maxim. Research from
@alejandroll10
and
@Wharton
’s
@NickRossanov
challenges this notion, using AI to help eliminate risk that is not compensated with extra returns. Read about the model here: .
Can ChatGPT be used to predict stock moves?
University of Florida professor
@alejandroll10
is studying that question by using the AI to parse through news headlines and if they are good or bad for a stock:
Finance professors get paid more because our students earn more income, with a high possibility of exceptional income and happily pay higher tuition and donate more! 💲💵💹
📢We're hiring an Assistant or Associate Professor in Finance focusing on Fintech/ML!
📚Come join our department at the University of Florida!
@UF
@UFWarrington
🎓Top Market Salaries and No State Tax!
☀️Best weather all year!
I am on the committee this year.
Link ⬇️
ML in finance academia is not using the latest developments. Two causes:
1. It takes time to adapt the newest techniques into working papers. You need to write it down and adapt it to an interesting academic setting
2. Referees are skeptical of exotic methods
@GaryMarcus
Vital: coding in python, translating code from python to R, converting between latex writings to presentations and vice versa, solving equations (via Wolfram), rephrasing thoughts
The best reason to use Python over R for statistics is that ChatGPT can run python code in its sandbox.
Pandas has a horrible syntax but who cares if you don't have to write it explicitly anymore.
🏆 Thrilled that our paper, "What If Option Closing Prices Were Trustworthy? A Machine Learning Approach," won the Best Paper in Derivatives Award!
Link:
Thread 🧵
🔍 Options lack a robust mechanism for closing prices. How can we solve this? (1/N)
@EconTalker
No, it's one of the best econ podcasts available. The historical archive contains more than enough econ. Much easier for people interested in econ to find it with the current name. It also helps incentive you to do some econ episodes which I love.
I really look forward to the 6th Future of Financial Information Conference, taking place on 27-28 May 2024 at Stockholm Business School. And you should too! 🌟
As session chair, I am thrilled to see two wonderful papers in my session:
What Drives Trading in Financial Markets?
In finance:
- Good pay
- Good job market atm* (8/8 strong academic placements among our PhD students this year, though we are good and this year was unusually good even for us)
- Good outside options
* past performance is no guarantee of future returns etc
We gave ChatGPT $50k of our money to see if it can beat hedgefunds
1500 people joined and now another $1 million+ is copying ChatGPT via Autopilot
So lets have some fun here
No clue how it'll do, but will be fun to do it with everyone
More details on "The ChatGPT Challenge"👇
Our research has just been featured in the latest episode of
@excessreturnpod
I always enjoy talking about the foundations of asset pricing!
Does peer-reviewed theory help predict the
cross-section of stock returns?
with
@achenfinance
@TomZ_Econ
Those of us who use factors have been taught that they need a risk-based or behavioral explanation.
We have been taught to avoid data mining.
But what if both are wrong?
We are joined by
@achenfinance
and
@alejandroll10
whose new paper suggests exactly that.
🚨 Webinar alert 🚨
150+ registered participants within a few hours of the initial announcement can hardly be wrong - you will not want to miss the upcoming Future of Financial Information -
@FutFinInfo
#webinar
either!
Date: 21st June at 4pm CET (10am New York time)
1/🧵
Teaching and Research with ChatGPT Newsletter 🤖
I'll post about using ChatGPT in finance, research, and teaching.
How to use ChatGPT in class to teach coding. 🧑🏫
How (not) to generate research ideas using LLMs. 🧑💻
How (not) to use ChatGPT when writing. 🔖
Link ⬇️
💹New results in the space of open source financial LLMs:
💻We added Llama 2 to our Financial Natural Language Understanding and Prediction Evaluation Benchmark (
#FLARE
) Leaderboard
👨💻
#FinMA
looking well and we're currently improving it!
@JiminH_AI
It's crazy the number of opportunities
#ChatGPT
opens in financial economics research. If you have an idea that needs text analysis or ChatGPT but don't know how to implement it let me know and we can collaborate!
#EconTwitter
⏯️New podcast recording with
@JeremyDSchwartz
in Behind the Markets!
We had a wonderful talk about the role of artificial intelligence in finance, including ChatGPT, factor models, and the state of asset pricing.
Podcast:
Live from
@Wharton
we have
@alejandroll10
talking his research on using Chat GPT for stock selection strategies and state of finance and general asset pricing research.
Plus we’re scooping CNBC this week with latest Siegel thoughts live from a cruise
📷 Excited to introduce PIXIU - the first open-source financial instruction dataset, benchmark, and fine-tuned LLMs! This is a big step forward for
#FinancialAI
. Check out our
#arXiv
paper () and GitHub repo ()!
#AI
#NLP
#FinTech
I found this amazing book named "Interpretable Machine Learning" by
@ChristophMolnar
while I was researching about explainability. It covers lots of core concepts from beginner to advanced. I highly suggest reading it.
Universities should consider putting together a single class on generative AI that covers the hot button issues: LLM basics, AI literacy, ethics issues, how to prompt & use AI, etc.
Otherwise a lot of this content is partially taught & is ad-hoc scattered across various classes.
Looking forward to presenting some cool new asset pricing results in the Theory-based Empirical Asset Pricing Research (TBEAR) Network!
With
@CarterDavisFin
We are happy to announce the launch of the FinNLP-AgentScen at IJCAI-2024, FinLLM, which includes three subtasks tailored to effectively and holistically address diverse financial challenges:
- Financial Text Classification: Designing LLMs to categorize financial texts as
Can LLMs really help you outsmart the market? 🤔
I explore the potential, limitations & future outlook in my latest article - a taste of the in-depth insights in "The Predictive Edge," available for pre-order!
Book out on July 11!
Article and Book links on thread⬇️
@paulnovosad
But then you reject if not all requests are satisfied. If as a referee I think clustering is causing the issue I'll ask for a specific specifications, same for the data sample. If it breaks or authors don't do it then you reject the paper
My first post is out! Based on my class using ChatGPT
How to persuade students not to use ChatGPT for essays⚡️
Why the pretraining in ChatGPT can get students into serious academic problems‼️
Link ⬇️
Heading to the
#AFA
! ✈️
Check out tomorrow our paper documenting data mining is as good as the peer reviewed process in finding asset pricing anomalies!
Testing academic ideas just became an order of magnitude easier with
#ChatGPT
If it's less costly (timewise) to write academic papers, will it result in better or worse average quality in peer reviewed journals?
#EconTwitter
This 🤯 is a very big 🤯
I have access to the new GPT Code Interpreter. I uploaded an XLS file, no context:
"Can you do visualizations & descriptive analyses to help me understand the data?
"Can you try regressions and look for patterns?"
"Can you run regression diagnostics?"
🚨Job market paper🚨
Ever thought about the differences between ETFs and index mutual funds? Heard about ETFs' advantages, incl. tax efficiency (in the US) & intraday trading, and wondered whether ETFs will eventually replace mutual funds in the index segment? My JMP suggests no.
I'm convinced consumption asset pricing models can rationalize the extremely high Sharpe ratios we observe in the cross-second of returns.
We just need to add more unobservable latent variables that affect consumption but do not materialize in-sample.
April 1st.
Why do Random Forests perform so well off-the-shelf & appear essentially immune to overfitting?!?
I’ve found the text-book answer “it’s just variance reduction 🤷🏼♀️” to be a bit too unspecific, so in our new pre-print ,
@Jeffaresalan
& I investigate..🕵🏼♀️ 1/n
This is a *lower bound* on GPT-4's predictive capacity.
Better prompts and current information will increase the performance.
See our paper for a case where it performs very well.
Can ChatGPT Forecast Stock Price Movements?
How good is GPT-4 at forecasting?
In this new paper (with
@Dr_Park_PhD
), we test this in a real-world forecasting tournament
@Metaculus
.
🔴 We find poor LLM performance compared to humans!
🔴 GPT-4 fails to outperform predicting 50% on every question!
@paulnovosad
But p-hacking means select specific results/data preparation no? You can just ask for a variety of other samples, specifications and assess the coefficients and statistical significance for each of the exercises you suggest.
Heading to the
#AFA
! ✈️
Check out tomorrow our paper documenting data mining is as good as the peer reviewed process in finding asset pricing anomalies!
Some PhD students seem to think a good way to find research ideas is to read the conclusions paragraph of good papers, and do the things authors say are natural next steps for research
This is a terrible idea! Absolutely terrible! It is hard to do worse than this!
Interesting idea! In finance, we can see the tractability trap with the pervasiveness of CARA/CRRA utility models despite their fragile empirical foundations.
Unpopular opinion: Causality is **not relevant** in the majority of
#quantfinance
modeling applications! “Successful prediction does not require correct causal identification.”
Causal relationships are important if you want to **intervene** in a system. Quant traders are not
One of my favorite conferences! Be sure to apply or attend! A perfect place to learn the latest in
#BigData
,
#MachineLearning
, and
#AI
applied to finance research!
🚨 Happy to announce that submissions are open for the
#5th
Future of Financial Information Conference, due to take place on 15/16 May 2023 at
@HECParis
- the only conference to look across all areas of finance through the lens of information. 1/🧵
I will allow (and encourage!) the use of LLMs like ChatGPT in my classes (finance and ML) this semester.
I plan on starting two newsletters with the outcomes.
One for professors/PhDs on ChatGPT in teachings (and research) and one for students.
Thoughts?
@alejandroll10
&
@NickRoussanov
's attempted murder of the APT is probably my favorite paper in the last several years. They find the more risk you hedge, the *better* your Sharpe ratio!
Highly recommend the NBER SI talk + Bryan Kelly's discussion
A simple puzzle GPTs will NEVER solve:
As a good programmer, I like isolating issues in the simplest form. So, whenever you find yourself trying to explain why GPTs will never reach AGI - just show them this prompt. It is a braindead question that most children should be able to