How Investor Attention Amplifies Bitcoin and Ethereum Jumps

A bullish crypto headline can arrive during a quiet afternoon or in the middle of a market frenzy. Its wording may be identical, but the surrounding conditions can differ dramatically. Understanding that difference could improve how investors interpret sentiment and prepare for sudden price moves.

A new study accepted for publication in the International Review of Financial Analysis examines how sentiment and collective attention jointly relate to jump risk in Bitcoin (BTC ) and Ethereum (ETH ).¹ Researchers Alev Atak and Philip Hans Franses combine AI-analyzed social media posts with Wikipedia traffic and minute-level prices. Their central finding is that positive sentiment has a stronger association with upward-jump intensity when attention is elevated.

The broader insight is useful: knowing what investors are saying becomes more informative when paired with a measure of how much attention the subject is receiving.

How AI Measures Crypto Sentiment and Attention

The researchers examine Coinbase Exchange prices from January 2022 through December 2023. They classify an upward or downward jump when a one-minute return exceeds twice the standard deviation of returns over the preceding hour in the corresponding direction.

This definition identifies unusually large moves relative to recent conditions. It does not require every jump to exceed the same percentage threshold. A move that looks exceptional during a quiet period might be less unusual during an already turbulent hour.

For sentiment, the authors use FinBERT, a language model designed to analyze financial text, to score cryptocurrency-related Twitter posts. They average those scores each minute. Collective attention comes from hourly Wikipedia pageviews for Bitcoin, Ethereum, and Cryptocurrency, measured against the preceding week.

The distinction matters. Sentiment measures tone, while attention measures unusual interest. Increased pageviews do not reveal whether readers intend to buy, sell, or simply understand an unfolding event.

Why Attention Changes Bitcoin and Ethereum Jump Risk

The study uses a Hawkes model, which captures the tendency of events to arrive in clusters. In this application, it separates a changing baseline rate of jumps from the additional clustering associated with previous jumps.

Sentiment and attention enter the baseline together. Their interaction allows the same sentiment shift to have different modeled associations depending on the attention environment.

These comparisons hold attention fixed within each scenario. Neutral attention corresponds to an attention score of zero; elevated attention corresponds to two standard deviations above the recent weekly mean. The sentiment shift is measured on a scale from minus one to plus one.

The percentages describe changes in fitted baseline intensity, meaning the modeled arrival rate of upward jumps. They do not imply a 42% Bitcoin return or a 49% Ethereum return. Nor are they percentage-point increases in jump probability.

The researchers also find stronger clustering for downward jumps than upward jumps. That suggests risk monitoring should consider both the information environment and whether abrupt moves are already occurring in succession.

Crypto Sentiment Needs Market Context

A useful interpretation is that attention can indicate how widely a narrative is being absorbed. An upbeat message during low interest may reach a small audience. Similar sentiment during intense interest may coincide with broader participation and faster repricing. The study documents an association, rather than proving this transmission mechanism.

Context also includes the subject of the discussion. Securities.io’s recent coverage of Bitcoin and monetary policy expectations examines how AI-analyzed investor conversations can contain information about subsequent returns. Together, these research directions suggest sentiment dashboards could benefit from identifying what people are discussing as well as their tone and level of interest.

For example, positive discussion about an upgrade and relief over anticipated interest-rate cuts could produce similar sentiment scores while reflecting different economic drivers. A monitoring system that groups them together may overlook those differences.

The proposed opportunity is richer context, with separate measures for narrative, attention, and trading conditions. That is an extension inspired by the findings, not a system validated by this paper.

Better Forecasts Do Not Guarantee Profitable Trades

The researchers train their main forecasting models on 2022 and test them on 2023. Their predictive specifications use sentiment from the previous minute and attention from the previous hour, helping avoid reliance on information unavailable when a forecast would be made.

Text-augmented models improve out-of-sample predictive log-scores by approximately 16% to 20% relative to standard specifications. This evaluates how well the model assigns probabilities to observed events. It is not a corresponding increase in trading accuracy, portfolio returns, or the proportion of profitable transactions.

Calibration tests also find no evidence against the text-augmented forecasts in the 2023 test sample. That supports their historical probability estimates, but does not establish that performance will persist indefinitely.

Execution introduces another challenge. Even a useful warning could arrive after spreads widen or protection becomes expensive. Fees, slippage, and financing costs can absorb a statistical advantage. The paper does not demonstrate a profitable strategy after these costs.

Turning Attention Signals Into Risk Monitoring

The findings may be most useful for deciding when to scrutinize exposure. A trader, exchange, or portfolio manager could investigate whether rising attention and jump clustering warrant tighter monitoring before attempting to use them as automatic trade triggers.

• Track sentiment alongside attention and recent jump clustering.
• Check liquidity and execution costs before acting on a forecast.
• Validate the model on newer data and additional exchanges.

Data quality deserves equal attention. Securities.io’s analysis of crypto wash trading and market volatility highlights why apparent activity can be misleading. Social attention, reported volume, and genuinely available liquidity measure different things and should be checked independently.

The study also treats qualifying jumps as binary events. A move barely crossing the threshold and a much larger move receive equal event status. Estimating jump frequency therefore leaves an essential portfolio question unresolved: how severe might the next move be?

Coinbase Connects Crypto Trading With Market Infrastructure

Better monitoring becomes commercially relevant as platforms expand trading and hedging services. For investors interested in that infrastructure, Coinbase Global (COIN ) provides a public company connection. Its exchange supplied the study’s price data, although that does not imply Coinbase uses the researchers’ framework.

Coinbase’s June 2026 commentary on regulated crypto derivatives access describes its expansion of U.S. institutional access to perpetual futures and options. Such products create more ways to express views and manage exposure, increasing the relevance of understanding changing volatility conditions.

In its second-quarter 2026 results, Coinbase reported a 10.3% crypto trading volume market share and said subscriptions and services represented 48% of net revenue. Those figures provide context for a business extending beyond transaction fees into broader financial infrastructure.

The investment question is whether that expansion builds durable customer relationships and revenue across market conditions. Increased volatility can stimulate activity, but severe stress can also weaken confidence. This study does not test Coinbase shares or establish an earnings benefit from attention spikes.

Latest Coinbase Global (COIN) News and Developments

What Crypto Investors Should Take From the Research

The findings support treating sentiment as a conditional signal. Its association with sudden price moves varies with attention, while existing jump clusters add another layer of risk.

The evidence remains limited to historical Bitcoin and Ethereum data from one exchange. Conversations and prices may respond to the same underlying news, preventing a clean causal conclusion. Newer market structures and changing social platforms also require fresh validation.

Still, the framework offers a constructive direction for crypto analytics: assess tone, attention, and turbulence together. Understanding those conditions could help investors distinguish a popular narrative from a measurable change in short-term risk.

References:

¹ Atak, A., & Franses, P. H. (2026). Attention-amplified sentiment and jump risk in Bitcoin and Ethereum markets: A text-augmented Hawkes framework. International Review of Financial Analysis. Advance online publication. https://doi.org/10.1016/j.irfa.2026.105431