AI outperforms humans in predicting company profits, study finds
Posted on Wednesday 1 July 2026
Now, a new study by the University of York has revealed that AI can successfully rewrite the rules of financial forecasting by spotting complex, hidden patterns in balance sheets that the human brain simply cannot process.
Machine learning
By deploying advanced machine learning techniques to answer the simplest yet most vital question—will a company's earnings be higher or lower than the previous year?—researchers achieved a 70% accuracy rate.
Unlike traditional forecasting methods that are heavily influenced by investor psychology, stock price speculation, or biased analyst forecasts, this study relied entirely on raw accounting data. The models analysed fundamental financial statements - such as cash flow quality, leverage, and operating profitability - across a dataset of more than 21,000 U.S. firms spanning 55 years (1967–2022).
The researchers tested various machine learning methods, discovering that artificial neural networks performed best. Inspired by the structure of the human brain, these networks excel at evaluating thousands of interconnected variables simultaneously to uncover non-linear relationships that human analysts collectively miss.
Spot signals
Co-author Dr Keith Anderson, from the University’s School for Business and Society, said: “We have shown that machine learning does have the power to spot signals hidden within the many variables in company accounts that humans are collectively missing at the moment.”
Co-author Dr Dimitrios Stafylas, from the School for Business and Society added: “Machine learning can uncover hidden drivers of profitability and improve forecasting accuracy beyond traditional models.”
One of the more revealing findings from the study concerned the difference between short-term and long-term profitability. Over one-year, earnings were driven heavily by recent momentum.
Quarterly earnings trends proved especially powerful: companies whose profits had been improving quarter by quarter were significantly more likely to continue growing.
Over five years, the picture shifted. Longer-term profitability depended far less on recent trends and far more on the underlying quality of the business - return on invested capital, operating profitability, and financial stability.
Earnings growth
To prove the real-world value of this AI "weighing machine," the researchers constructed investment portfolios based on the neural network's long-term predictions.
The results were stark: a portfolio consisting of companies the AI predicted were most likely to show earnings growth outperformed those least likely to grow by 7% to 10% per annum.
The algorithm proved most dominant when analysing value stocks and small-to-medium-sized firms, which are generally less heavily followed by Wall Street analysts.
Financial truths
Furthermore, the study revealed that algorithms trained on the entire cross-industry dataset outperformed sector-specific models, proving that broad financial truths cut across all industries.
While financial markets will always retain an element of unpredictability shaped by politics and human behaviour, this study proves that machine learning can fundamentally improve portfolio construction, risk management, and corporate decision-making.
The future of financial forecasting no longer belongs to the human eye alone, but to the algorithms capable of seeing through the noise.
