Signal within the Noise? An Investigative Analysis of Relationships Between Asset-Specific News and Price Moves for Various Asset Classes

Abstract

This paper explores the intricate relationship between asset-specific news sentiment and subsequent price movements across various asset classes, namely equities, cryptocurrencies, and commodities, during the particular financial landscape of 2023. This analysis is done to shed light on the mechanics of information absorption in the market, with a focus on the assets with the largest market capitalizations. Despite large capitalizations for the assets chosen, the cryptocurrency market is included to determine the efficiencies of nascent asset markets. Utilizing advanced Natural Language Processing (NLP) models, such as RoBERTa and FinBERT, sentiment analysis was conducted on a large dataset of news articles collected via the NewsCatcher API. The study aims to test whether effects on market sentiment derived from news publications can provide predictive signals for asset price movements. Through a Pearson correlation analysis, this research determines there is no significant correlation between the sentiment expressed in general news and the price movements for all asset classes studied. However, from relative comparisons using regression models and an XGBoost model with featurized datasets that include market and price features, this paper finds that the market for equities makes uninformed decisions but prices in information very quickly. In this text, "uninformed" means market participants do not have the relevant expertise to observe how a news event will affect an asset’s underlying value over time. By our definition of efficiency, markets are considered efficient even if market participants are uninformed about intrinsic value, such as the case of early 2000s dot-com bubble. Instead, they are efficient because of their ability to reflect new available information rapidly. In contrast, in cryptocurrency and commodities markets, market participants focus on salient news events and default to prevailing asset trends to make buy and sell decisions, likely due to their nature as productive assets or stores of values, respectively. In this way, contrary to popular belief, cryptocurrency and commodity markets may actually price assets more accurately than equities for large capitalization assets, with similar levels of efficiency as measured by their speed of information absorption.