An in-depth analysis into the realm of hidden emotions obtained from plain-sight news platforms reveals an unrivaled link for predicting cryptocurrency prices, new research from FIU Business finds.
The insight: using artificial intelligence (AI) and algorithms to outperform the market.
Published in the March 2024 issue of Decision Support Systems, the research incorporates signals delivered by news outlets and social media posts and combines them with trading data - including daily open, close, highest and lowest prices, and daily trading volume - to predict future bitcoin prices.
“Our analysis transforms this data into actionable insights through predictive analytics focusing on investor utility, while identifying the most powerful signals by comparing their performance in prediction,” said Ziyang Zhang, a doctoral student at FIU Business and one of the researchers. “We can increase the prediction accuracy and provide a good suggestion for value progression.”
The researchers’ analysis used 55 signals generated by MarketPsych Analytics, a real-time aggregator of financial sentiment data from 4000-plus news and social media outlets, and classified them into five categories - emotional, price, factual, slang and general. Although the focus of the research was on bitcoin, the methodology and strategy are relevant and adaptable for a broader spectrum of cryptocurrencies, the researchers said.
Florent Rouxelin, assistant professor of finance at FIU Business, and Patricia Angle, at the University of Oklahoma, collaborated on the research.
“It was not just finding the most performing signals of these 55,” said Hemang Subramanian, FIU Business associate professor of information systems and business analytics, who led the research. “Using combinations of the top-performing signals gives us more power - fear is more predictive than FOMO, which is more predictive than relevance.”
The researchers then developed an app and algorithms that combines market data with benchmark metrics. This software can also evaluate various portfolios and compare their performances.
Subramanian described it as a 24-hour trading system guided by the 55 signals and combinations of seven more returning signals. In their tests, various signal-driven portfolios outperformed markets by up to 39.6% on a risk-adjusted basis.
“You chart when you have indication that the market is going to go down, buy when you have indication that it’s going to go up,” he said. “Basically, you make a decision based on the algorithms’ showing the highest returns.”
The project began in 2020 when Subramanian received a private data grant from Refinitiv (now MarketPsych) to analyze the bitcoin market. The data gathering took place from May 2022 through May 2023, during the growing decline in bitcoin’s market valuation, from $60,000 to $20,000, when many people and businesses declared bankruptcy or exited the market.
The focus in 2025 will be to commercialize the app model and conduct market discovery. Once the system is deployed, users can trace the daily price and signals and predict the price of bitcoin for the next day. Companies and investors will be able to deploy the service for cryptocurrency overall, Zhang explained.
Florent Rouxelin, assistant professor of finance at FIU Business, and Patricia Angle, at the University of Oklahoma, collaborated on the research.