Here’s when analyst reports matter, according to Morgan Stanley’s AI

Do stock analysts truly have insights for investors about earnings or are they just riding the market?

That’s the question Morgan Stanley heard from investors after it detailed a new artificial intelligence strategy in June that studied its own analyst reports. The strategy, which relied on what the bank calls “Machine-Read Analyst Sentiment” or “MRAS,” beat the S&P 500 during a backtest.

To answer this question, researchers used the MRAS to look at times around an earnings announcement where analysts had taken a contrarian position.

They found that when analysts go against the market, their calls are often correct. Even when stocks swings go against analysts, after a few weeks, those swings often reverse themselves.

“The return trend … consistently prevails after around 20 days,” Morgan Stanley analyst Qingyi Huang said in the report.

The bank’s research team built the MRAS by training AI on more than 40,000 reports to develop sentiment scores for analyst notes. The team then used the sentiment scores in the reports with a price target change for a stock to take hypothetical positions.

The full strategy, including both the contrarian calls and those in-line with the market, had similarities to a momentum strategy, which bets that recent price moves of stocks will continue. This is part of the reason why the researchers focused on the contrarian calls to see how well the MRAS was working.

The analysis found that, for trades where the MRAS strategy was bullish and the initial market move was negative, the strategy beat the S&P 500 by an average of 1.9% over a 60-day period. The strategy saw similar success when those positions were flipped.

The report also found that the size of the first-day move was correlated with the MRAS score. In sell-offs where the analysts were positive, stocks declined just 2.6%. When the broader market and the analyst report were both bearish, the sell-off was greater than 10%.

“When a stock went down on the day of earnings, if the MRAS strategy also suggested to “sell,” it would maintain the negative trend for the next 60 days; but if the strategy indicated a “buy” signal, the stock would rally into the next earnings,” Huang said.

The report also looked at trades made outside of 20 days before or after an earnings announcement. The research found that the results of the trade were consistent with the sentiment from MRAS until 40 days after entering the trade.