Frequently asked questions

MarketGPT model details:

MarketGPT is designed to identify patterns in a company's stock performance following the publication of news items or press releases about that company. Each news article is given a score that indicates the potential impact on the stock price and the anticipated direction of the price change. Articles that exceed a certain threshold are selected by the algorithm and published on the website and mobile app.

MarketGPT was trained using a database of financial news spanning from 2018 to 2022, which includes over 1.3 million individual items. Each news item was pre-processed using the GPT-4 model with custom prompts to standardize the format of each article, facilitating easier comparison. The processed database was then merged with the performance of the mentioned stock the day after the news was published, defined as the percentage change in the stock price between the opening and closing value. This data was then used to train the model to predict expected stock performance based on the presented news item.

Sentiment analysis is a basic tool used to gauge the overall sentiment of a news item, without conducting a detailed analysis of the text. Research indicates that relying solely on sentiment is inadequate for predicting stock movements. In contrast, MarketGPT conducts a comprehensive textual analysis of the article and can accurately predict negative stock performance following seemingly "positive" news, and vice versa. For instance, an article reporting several days of positive stock returns may be classified as having a "positive" sentiment. However, MarketGPT understands that without a fundamental cause for the price increase, the market is likely to correct itself over time, leading to a price drop.

An example of a simple pattern that MarketGPT can identify is that a company's stock is likely to decrease in value after the company announces a reverse stock split. However, MarketGPT can also identify more complex patterns and combinations thereof to reach a final conclusion. This means that an article with seemingly positive news may be classified as likely to indicate a negative price change.

The specific architecture of MarketGPT is proprietary information. At this time, we are not ready to release the model as open source. However, you can download our mobile app to access daily stock picks from the model.
Stock prediction performance:

The average return of stocks selected by MarketGPT in the first three quarters of 2023 is 3.18% per day. This result is relatively consistent on a month-to-month basis, as depicted on our performance graph. However, it's crucial to remember that the stock market is inherently unpredictable, and past performance may not necessarily predict future results.

MarketGPT is designed to predict market reactions to news within the upcoming trading session. We've found this to be the most effective time horizon for predictions, as stock prices in longer timeframes may be influenced by other events or news that the market reacts to. Therefore, MarketGPT's stock performance predictions are limited to a one-day period.

MarketGPT analyzes news items related to publicly traded companies and attempts to predict the market's reaction to this information. As such, it can only provide predictions for stocks that are mentioned in the news. It cannot provide insights into the stock performance of any arbitrary company.

Currently, MarketGPT only considers companies traded on the NYSE and NASDAQ stock exchanges. It can also make predictions for most American over-the-counter stocks. However, we do not currently publish these predictions as these stocks are more challenging to short, making this information less useful to investors.

MarketGPT analyzes news items related to individual companies, not specific industries or the overall market. While these articles may reference general market trends, the predictions are primarily based on the analysis of the specific company's circumstances.

MarketGPT has been trained on market data from 2018 to 2022 and began providing market predictions in January 2023. All market predictions from January 2023 onwards are archived on our website.

You can access all historical predictions by scrolling through our home page or by downloading the mobile app and selecting the "History" tab. The mobile app groups all historical predictions by day for easier analysis.
Mobile app features:

MarketGPT utilizes multiple news aggregators to evaluate news items in real time, as soon as they are published. The resulting stock picks are updated in the mobile app every minute.

The app is free to use, and you can access the daily stock picks without any charges. However, to access certain additional features of the app, such as real-time updates on the performance of recommended stocks, you may need a subscription. This does not affect the basic function of displaying stock picks, which remains free of charge.

Yes, you can enable notifications that alert you about all the stocks that the model has selected for a given day. The app sends you daily notifications at a specific time before the market opens (from 5 to 30 minutes), allowing you to place your orders with your stock broker in a timely manner.

Once the market opens, the app displays live performance of all the stocks that the model has selected for the day. You can use this data as an indication of the performance of each stock. After the market closes, the final performance (calculated as the difference between the closing and opening prices) is recorded and made available in the "History" tab of the app.
Trading with MarketGPT:

Yes, the stock picks provided by MarketGPT are specifically tailored to the needs of stock investors. If you wan to incorporate them in your trading strategy, visit our trading help to learn about appropriate trading strategies. However, you are strongly encouraged to always do your own due diligence on the stocks that you trade, and not just blindly rely on stock picks made by aby app.

Because all the predictions of MarketGPT follow the same format and cover the same timeframe, trading stocks using its information is straightforward as it requires only two trades per day pr each stock. However, investors are strongly encouraged to perform their own research on proposed stocks before trading them. If you are a beginner in stock trading, we recommend that you first practice “paper trading” with virtual money to get yourself acquainted with the process.

If you're unable to place your market orders before the market opens, we recommend waiting until the next day's predictions rather than opening trades during the market session.

As MarketGPT's predictions have a one-day horizon, we advise against keeping your market positions open for longer. Other events may influence stock prices over a longer timeframe. Additionally, your stock broker may charge overnight fees, which could reduce your profits.

Our simulations show that using these options can negatively impact investment performance. Market volatility may cause your positions to close prematurely before profits are fully realized.

Our training data shows that the market reacts more strongly and predictably to negative news. Therefore, the current version of MarketGPT only provides "Sell" recommendations, which are easier to make and thus more reliable. We are working on an improved version of the model that will also issue "Buy" recommendations.