Introduction
Since the inception of Bitcoin, researchers have traditionally modeled block arrivals as a Homogeneous Poisson Process (HPP). This model assumes that blocks are discovered at a constant average rate, independent of historical events. However, this simplification fails to account for the dynamic nature of the network, where mining intensity is influenced by hardware evolution, fluctuating electricity costs, and market volatility.
This research challenges the HPP assumption by applying a multivariate Hawkes process to Bitcoin block arrivals. By accounting for history-dependent "excitement," the authors demonstrate that block mining is not a steady stream but a bursty process influenced by market price jumps and internal network feedback loops.
Methodology and Data
The study utilizes a trivariate Hawkes process model to capture the interactions between three distinct event types:
- Block Arrivals ($B$): The timestamps of successfully mined blocks.
- Positive Price Jumps ($P_u$): Significant upward movements in Bitcoin's price.
- Negative Price Jumps ($P_d$): Significant downward movements in Bitcoin's price.
To define price jumps, the researchers employed a rolling 10%/90% quantile window (3-hour duration) on 5-minute Volume-Weighted Average Price (VWAP) data from Coinbase. This approach ensures that "jumps" are identified relative to current market volatility rather than arbitrary fixed percentages. The model parameters were estimated using Maximum Likelihood Estimation (MLE) with a sum-of-exponentials kernel to capture multiple time scales of influence.
Analysis of Interactions and Market Reflexivity
A core finding of the research is the presence of significant cross-excitation between the market and the blockchain. The model reveals that market activity, regardless of direction, acts as a catalyst for mining intensity. Both positive and negative price jumps increase the future rate of block discovery, likely because price movements signal changes in mining profitability or trigger re-hedging activities by large mining pools.
Furthermore, the model captures "market reflexivity" or mean reversion. The estimated kernel norms indicate that a positive price jump increases the likelihood of a subsequent negative jump, and vice versa.
As shown in the interaction matrix above, block arrivals also exhibit self-excitation ($B \to B$). When a block is found, the probability of another discovery increases shortly thereafter, which may be attributed to the synchronization and information propagation lag within the global network of mining pools.
Empirical Results
The Hawkes process model significantly outperforms the standard Poisson model across all event types. To validate the fit, the researchers used the Random Time Change Theorem to transform event timestamps; if the model is accurate, the transformed data should follow a unit-rate Poisson process.
The Q-Q plots demonstrate the Hawkes model's superior performance. For block arrivals, the Hawkes model achieved a slope deviation of just 0.002, a tenfold improvement over the 0.020 deviation seen in the Poisson model. The improvement was even more stark for price jumps, where the Poisson model failed to capture the heavy-tailed nature of market volatility.
Conclusion
By moving beyond the simplistic Poisson assumption, this research provides a more nuanced understanding of the Bitcoin ecosystem. The multivariate Hawkes process successfully quantifies the feedback loops between market volatility and blockchain health. Future work may expand this model by incorporating endogenous variables such as global hash rate shifts, electricity price fluctuations, or longer-term memory kernels to better capture the multi-scale dynamics of financial markets.
Summary updated