Forecasting Returns Through Changes in the Language of News

Quantifying the novelty of news with an entropy measure helps to predict stock market trends and macroeconomic outcomes for the subsequent year.

Quantifying the novelty of news with a measure called entropy helps to predict stock market trends and macroeconomic outcomes for the subsequent year.

Going beyond prior studies that linked news sentiment with short-term asset price changes, researchers explored how variations in the distribution of news text can predict negative market returns. From a dataset comprising over 1.6 million Reuters news articles spanning 27 years, the study used advanced natural language processing (NLP) techniques to create a measure of entropy that captures the degree of unusualness or change in news language.

A key discovery was the predictive ability in forecasting stock market trends. A rise in entropy was linked to a 3% drop in S&P 500 index returns over the following 12 months, surpassing other economic predictors in forecasting accuracy. In addition, assets correlated with high entropy tended to have a negative risk premium, suggesting that investors accept lower returns on assets that offset risks linked to changes in news language.

A change in the distribution of news text was also found to be a powerful forecaster for year-ahead macroeconomic outcomes, such as unemployment rates, market volatility, industrial production, inflation, interest rates and corporate earnings. However, despite entropy’s significant forecasting ability, markets might not fully adjust to the information entropy provides due to informational constraints or slow-moving institutional capital.

The research reveals entropy’s unique informational content and its capability to capture aspects of market dynamics unexplained by existing models and traditional indicators. The findings offer a more nuanced understanding of how news affects asset prices and market dynamics. They also open new avenues for research, including exploring advancements in NLP to analyze news unusualness, as well as the potential for applying the concept of entropy to individual stock analysis.

Source: “New News is Bad News,” by Paul Glasserman, Harry Mamaysky and Jimmy Qin; SSRN, August 2023.

Discussion

BARRY J from TX posted over 2 years ago:

Eunice, thanks for bringing this article to our attention. I found it too dense with unfamiliar, non-standard terminologies, and the “findings” to be difficult to reconcile with any of the three versions of the efficient market hypothesis. So, I downloaded the abstract. Same even more densely defined terminology and unexplained conclusions. So, I downloaded the paper. It was a little easier to understand, but all its breathless “mirabile dictu” conclusions appear to be tenuous and of little value in predicting future market changes in a timely and useful manner. So, now I quote the opening summary paragraph that purports to provide an introductory summary and I welcome any comments that can decipher its meaning. “Several studies have documented that the sentiment of news text forecasts short-term changes in asset prices, with negative news forecasting negative returns. We find that a change in the distribution of news text— a measure of the novelty or unusualness of the news rather than its sentiment—forecasts negative market returns and macroeconomic outcomes over the subsequent year. Consistent with this pattern, we find that assets that positively covary with our measure carry a negative risk premium: investors accept lower compensation to hold assets that hedge the risk associated with a shift in the language of news.” For all my effort I concluded that the paper says that strange terminologies create negative sentiments that drive down interest in investments after up to a one-year delay. How that falls within the concept of something "useful" to know escapes me. I concluded that the article seems to be a self-referential tautological certainty and that someone needed to publish anything anywhere near AI to keep their jobs. I haven't had this much fun since "Godel, Esher, Bach" in the1970s. Any Indiana Jones' out there?


ROBERT A from NC posted over 2 years ago:

This from the same class of pointy-heads that developed the "volatility-is-risk" fantasy.


Don P from USA posted over 2 years ago:

Entropy is the state of change of the past to parts'n the present to da'Reality of when the Future is N-O-W . I say is not , what is done rather what they hope ; egs. Policy - Pandering - Pilfering .


You need to log in as a registered AAII user before commenting.
Create an account

Log In

Get your free copy of our special report analyzing the tech stocks most likely to outperform the market.

Download the FREE Report Here: