Cryptocurrencies as Contagions: Predicting Bubbles via Epidemic HMMs

TL;DR

Is a crypto bull run a financial phenomenon or a biological one? This paper argues it is the latter. By repurposing a Hidden Markov Model (HMM) originally designed to track influenza outbreaks, the authors successfully predicted cryptocurrency price bubbles using Reddit data. Their strategy more than doubled the returns of a "Buy and Hold" approach by identifying when the "contagion" of an investment idea begins to spread.

The "Psychological Contagion" Problem

Traditional finance struggle to define "bubbles" until after they burst. However, behavioral economists like Robert Shiller suggest bubbles are driven by psychological contagion. The problem is that most predictive models look at price alone, which is a lagging indicator. Prior attempts to use social media (mostly Twitter) often missed the "community" aspect where these infectious ideas actually breed.

The authors argue that platforms like Reddit, with dedicated subreddits for Bitcoin or Ethereum, represent isolated "populations" where the spread of enthusiasm can be measured much like a virus entering a city.

Methodology: Flu Tracking for Finance

The core of this research is a 2-state Hidden Markov Model (HMM). While a standard HMM might look at simple states, this version uses a specific Inductive Bias:

1. The Non-Epidemic State (0): Modeled as Gaussian White Noise. Social activity is random and lacks a clear trend.
2. The Epidemic State (1): Modeled as an Autoregressive process (AR(1)). This is the "Aha!" moment—during an epidemic, today's infection count is highly correlated with yesterday's. By using AR(1), the model captures the momentum and "viral" feedback loops of the boom phase.

Input Indicators from Reddit

The model monitors three specific Reddit metrics plus trading volume:

• Posts: Daily volume of new threads.
• Subscriber Growth: New interest from outsiders (new "susceptibles" being infected).
• New Authors: Increased participation within the community.

Figure 1: The model uses a moving window approach to capture the most recent social dynamics.

Trading on "Consensus"

The authors tested two ways to trigger a "Buy" signal:

• Unanimous Voting: All four HMMs (one for each indicator) must signal an "Epidemic" state.
• Averaging: Calculating the mean probability across indicators.

The results showed that Unanimous Voting was far superior. It acted as a noise filter, preventing the strategy from entering "false positive" bubbles caused by temporary news spikes.

Experimental Results: Beating the Market

The strategy was tested on Bitcoin, Litecoin, Ethereum, and Monero. The results were striking:

• Ending Portfolio: The HMM strategy turned $1,000into\ast \ast$14,804**, whereas Buy-and-Hold only reached $7,939.
• Risk Mitigation: The Maximum Drawdown was reduced from 50% to roughly 35%, and the recovery period was significantly shorter.

Figure 2: The system accurately identifies the massive 2016 Ethereum bubble and exits before the catastrophic drop.

Critical Insight & Future Outlook

The beauty of this model lies in its simplicity. By recognizing that asset bubbles share a mathematical fingerprint with viral outbreaks, we can move away from complex sentiment analysis (which is often wrong) and focus on the structural behavior of data.

Limitations: The model occasionally flags "negative bubbles"—panic events (like the DAO hack) that create epidemic-like social volume but result in price crashes. Future iterations would benefit from a "Price-Direction Filter" to ensure the epidemic is a bullish one.

Takeaway for the Industry: Reddit is a goldmine for leading indicators. While Twitter is the "town square," Reddit subreddits are the "clinical labs" where investment contagions are born.