AI transforms financial news into stock forecasts: an incredible breakthrough by US scientists

For years, the financial press has served as a source of valuable information for investors of various levels. Recent studies by scientists at Cornell University (USA) have shown that this information can be used to train algorithms for a new financial forecasting model.

The experts used approaches from machine learning, natural language processing (NLP) and finance to create a new, interpretable machine learning model. The created model is able to analyze information specific to stocks and industries and predict financial returns with greater accuracy than traditional models.

According to the authors research , one disadvantage of machine learning is that it is not interpretable. Researchers often cannot accurately interpret the results produced by complex models. The work of specialists uses text data from news to create interpretable machine learning models, where important features are clearly visible and easy to read.

Experts say financial news can help you better understand which stocks are associated with certain tradable assets, such as Exchange Traded Funds (ETFs).

To develop their models, scientists analyzed a huge amount of financial news published on the Internet from 2013 to 2019. Based on this data, the algorithm associated certain assets and words with specific stocks and industries.

Thus, the researchers created two separate models:

• the NEUSS model for predicting the returns of individual stocks;
• INSER model that identifies important words for each specific industry for more accurate revenue forecasting.

The new approach paid off: the NEUSS model outperformed the standard Fama-French 5-factor forecasting model by 50%, while the INSER model performed 10% better than the comparative benchmark without taking into account industry information. The researchers concluded that this discovery is an AI revolution in finance, and the discovery contributes to the development of this process.