Earnings calls — and specifically the questions asked by analysts on the calls — are a better predictor of a stock's performance than management's remarks, research by investment bank Nomura revealed. Its analysis of call transcripts using artificial intelligence showed that analysts' sentiment matters more than management's "positive spin" when it comes to stock price moves. "It's striking that companies with the worst analyst sentiment in earnings calls underperform the market in the days leading up to the call, and even more dramatically underperform following the call, and vice versa," said Joseph Mezrich, head of quantitative strategy at Nomura. "This suggests a connection between analyst sentiment and price momentum around the earnings call." Analysis of S & P 500 companies' third-quarter earnings call, conducted for CNBC Pro by Nomura , showed that companies in the financial, technology and materials sectors had the largest increase in negative analyst sentiment, compared to the same period last year. Whereas companies operating in the industrials, staples and utility sectors of the S & P 500 had a smaller proportion of analysts with negative sentiment on earnings calls, according to the analysis. How will stocks perform? Nomura created two portfolios of stocks using a long-short strategy, buying stocks with the best sentiment and shorting the ones with the worst. One was based on analysts' sentiment from the earnings calls, and the other was based on management's sentiment. As the chart below shows, both strategies outperformed the broader market. But the portfolio measuring analyst sentiment beat the management sentiment portfolio by about 25 percentage points. "A simple interpretation of this result is that analysts are probably less biased. The sentiment of their language reflects reality more closely than the sentiment of management remarks," Nomura strategists Lai Wei and Thelonious Jensen said in a note to clients. The quant team at the Japanese investment bank used "natural-language processing" — a technique that uses the AI platform ProntoNLP — to perform the analysis. Third-quarter earnings calls for S & P 500 companies that took place until Nov. 4 were compared against earnings calls of the same period last year.