Shulou(Shulou.com)11/24 Report--

The original title: "500% return?" ChatGPT, the strongest fund manager in history! "

The paper teaches you to use ChatGPT to invest in stocks, and the return can be up to 500%!

ChatGPT, is it going to replace the human fund manager?

Finder, a UK financial advisory website, said it created a portfolio of stocks chosen by ChatGPT on March 6, which rose 4.93 per cent two months later.

Over the same period, the average performance of the top 10 most popular funds in the UK was-0.78 per cent and underperformed the ChatGPT index on 87 per cent of trading days.

Similarly, the s & p 500, which includes the 500 most valuable companies in the United states, rose only 3% over the same period.

Apart from the threat to the jobs of the word workers, are the fund managers going to lose their jobs as well?

Who would have thought that foreign researchers have really written a whole paper to explore the ability of AI stock speculation in detail.

Specifically, "Hello" to ChatGPT some news headlines, let ChatGPT use emotional analysis (Sentiment Analysis) to determine the impact of these events on the stock market.

ChatGPT will determine whether an event is good or bad for the stock price, or irrelevant. After that, the researchers will score according to the results and use the real stock market returns to see if the ChatGPT is accurate.

The researchers say ChatGPT is much better than other models, such as GPT-1 and GPT-2.

This shows two things: first, the ability to predict the return of the stock market is an ability that needs to be further explored for language models. Second, more advanced language models are sure to produce more accurate predictions.

The three main data sets used by the research team are the daily return summary of the Securities Price Research Center (CRSP), the major headlines, and RavenPack.

The coverage of the data sample begins in October 2021 and ends in December 2022.

The reason for choosing this time period is also very simple. ChatGPT's own training data is up to September 2021, which means it doesn't have to predict before October. It already knows what the result will be. So, the new data after 21 years and October are just for ChatGPT to practice.

In the three data sets, CRSP's daily earnings dataset contains daily returns on the stocks of various companies listed on major U.S. stock exchanges, including stock prices, trading volumes and market capitalization.

Second, the researchers collected all the news containing the company name or stock symbol, from a variety of major news organizations, financial news sites and social media platforms.

For each company, the researchers collected all the news during the sample period.

Finally, RavenPack is worth highlighting.

Since 2003, RavenPack has been one of the leading data analysis providers in financial services.

They help companies get valuable parts out of a large amount of information quickly and easily. RavenPack's products enable customers to increase revenue, reduce risk, and improve operational efficiency.

In general, the role of RavenPack is to make sure that what is fed to ChatGPT is valid / relevant information.

In the end, the team matched 67586 headlines from 4138 companies on RavenPack.

At the same time, the researchers rated each piece of news on a scale of 0 to 100, measuring the relationship between each piece of news and the company mentioned.

A score of 0 means that a company is passively mentioned in the news and may just be picked up by hand. The researchers were looking for a 100-point correlation.

At the same time, all news only about stock market trends is excluded, which goes against the original intention of the forecast. Repeated reports of the same content were also avoided by the research team, so that the collected data could better serve the experiment.

At this point, the data to be collected is almost enough, and the next step is to analyze it.

Scoring system since the stock market to let ChatGPT output content, it is first of all to design a prompt.

The research team used the following prompt:

Suppose you are a financial expert and a financial expert with experience in stock recommendation. If this news is good news, please answer "YES". If it is bad news, please answer "NO". If you are not sure, answer "UNKNOWN" on the first line, and then explain it in a short sentence on the next line.

Is this news good or bad for the stock price of _ (company name) in _ (short-term, long-term)?

News headline: _

As you can see, the researchers asked ChatGPT Cosplay, a financial expert with experience in stock recommendations, to replace the horizontal line with specific information.

Take a chestnut:

Enter the headline: Rimini Street was fined $630,000 in a lawsuit against Oracle

Evaluation object: Oracle share price

Time span: short term

The answer to the ChatGPT output is as follows:

"YES. The fine on Rimini Street will indirectly boost investor confidence in Oracle's ability to protect its intellectual property rights and maintain increased demand for its own products and services. "

The researchers then scored according to ChatGPT's answers. YES scores 1, NO-1, and UNKNOWN 0.

If there is a lot of news about a company in a day, summarize the scores and output an average.

Finally, the predicted score is matched with the actual result.

Retail Gospel! Using headline data and generated mood scores, the researchers found a strong correlation between the results of the ChatGPT assessment and the subsequent daily returns of the stocks in the sample.

Moreover, the emotional score of ChatGPT can predict stock returns more accurately than the emotional indicators provided by existing traditional data providers.

The team believes that the reason why ChatGPT is better than the existing mood index is due to its strong language comprehension, which allows it to capture nuances in news headlines, thus making the resulting emotion scores more reliable.

Therefore, using the large language model as a tool can provide a better investment reference than the traditional emotional index.

The research team further found that ChatGPT emotional score is better than large-cap stocks in predicting the earnings of small-cap stocks. It shows that the restrictions on shareholder arbitrage may reduce the profitability of this strategy.

The research team used the emotional scores generated by ChatGPT to guide stock operations.

The specific mode of operation is to buy (positive news) or briefly sell (negative news) stocks as soon as news is released.

If the message is released within the trading time, it will be traded at the current price, and if it is released outside the trading time, it will be traded at the opening price of the next day.

(black lines represent zero cost benefits, green lines represent 5% of transaction costs, blue lines represent 10% of transaction costs, dark blue lines represent 25% of transaction costs, and red lines represent overall market revenue.)

This table shows the results of the regression analysis of the operation, which visually shows the correlation between the stock return of the next day and the emotional score generated by ChatGPT.

The rate of return is 500%? Using ChatGPT to analyze the emotion of news headlines, so as to predict the performance of stock returns is better than the traditional emotional index of major suppliers.

It shows that there is great potential to continue to develop and explore the application of large language models in the financial industry.

With the development of AI's own technology, it is reliable to design more complex models to make money in the financial market.

From a macro point of view, considering that in the future, if most financial practitioners use tools based on large language models to make decisions, it will also have a far-reaching impact on the price formation mechanism of the financial market, the dissemination of information, and market stability.

So what does the specific return look like?

The long-short strategy, that is, buying companies with good news and shorting companies with bad news, has the highest return of more than 500%.

Companies that only short strategies and focus on short selling companies with bad news have returns of nearly 400%.

The long-only strategy, buying only companies with good news, will yield a return of about 50%.

Of course, this number looks bluffing, in fact, it is an ideal situation.

But despite the cold numbers, ChatGPT does have a lot of room for this ability.

This could rewrite stock trading, and retail investors now have access to more powerful tools than corporate sentiment analysis.

In general, ChatGPT is making the work that other companies have devoted to proprietary machine learning models for many years obsolete.

It spans millions of dollars in R & D costs, and anyone can easily get this capability.

For ordinary investors, it is good to outperform the market.

For ordinary people who do not have very complex financial knowledge and experience in stock trading, they certainly cannot do such complex analysis and high-precision operations.

So based on the stock selection strategy recommended in the "ChatGPT portfolio" that outperforms the S & P 500 mentioned at the beginning:

Low debt ratio

Sustained and steady growth in history

Have assets that can create a competitive advantage.

You can also pick out good companies to help you allocate your assets efficiently.

It should be noted, however, that the response given by ChatGPT cannot be used as a factual basis for investment.

Reference:

Https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4412788

Https://www.reddit.com/r/ArtificialInteligence/comments/13dufss/a_chatgpt_trading_algorithm_delivered_500_returns/

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.