The Clustering Problem
A single institutional entity might control hundreds or thousands of blockchain addresses. There is no public registry mapping addresses to identities, so the challenge is figuring out which addresses belong to the same entity. This is the wallet clustering problem, and solving it, even approximately, reveals information that the entities are actively trying to hide.
The basic technique is co-spend analysis. On Bitcoin and similar UTXO-based chains, when a transaction has multiple inputs, all of those input addresses must be controlled by the same entity (since they all signed the same transaction). This simple heuristic allows you to build clusters of addresses that demonstrably belong to a single wallet. On account-based chains like Ethereum, the clustering is harder, but patterns like sequential funding from the same source, identical transaction timing, or interactions with the same smart contracts can link addresses probabilistically.
What Clusters Reveal
Once you identify a cluster of addresses as belonging to a single large entity, you can track that entity's behavior over time. You can see when they accumulate (transferring assets from exchanges to cold storage), when they distribute (moving assets to exchanges, likely preparing to sell), and when they rebalance between assets.
The timing of these movements often tells a story. A whale cluster that has been steadily accumulating BTC for three months and then suddenly stops accumulating may have reached their target position. A cluster that starts moving tokens to exchange deposit addresses is likely preparing to sell. A cluster that bridges assets from Ethereum to an L2 may be about to interact with specific DeFi protocols.
Exchange Flow Analysis Through Clustering
One of the most practical applications of wallet clustering is refined exchange flow analysis. Raw exchange inflow data (total tokens sent to exchange addresses) is useful but noisy. Clustered exchange flow, where you can identify which entity is sending tokens to an exchange, is much more informative.
If a known mining pool cluster sends 5,000 BTC to Coinbase, that has different implications than a known accumulation whale sending 5,000 BTC to Coinbase. The miner may be selling to cover operational costs, a routine and expected behavior. The accumulation whale selling is more concerning because it suggests a change in thesis from a sophisticated participant.
Similarly, large stablecoin transfers from identified institutional clusters to exchanges often precede significant buying. When you see a cluster that has historically been associated with buying activity deposit $50 million in USDC to Binance, it is reasonable to expect buying pressure in the near future.
Limitations and Countermeasures
Sophisticated entities know about clustering analysis and take steps to obscure their activity. Using fresh addresses for each transaction, mixing services, privacy-preserving bridges, and staggering transactions over time all make clustering harder. The cat-and-mouse dynamic means that the easiest-to-cluster entities tend to be less sophisticated (early holders, miners with consistent patterns), while the hardest-to-cluster are often the most sophisticated (trading firms, funds with operational security practices).
This creates a selection bias in what clustering reveals. You tend to see the behavior of mid-sophistication entities most clearly, entities large enough to be interesting but not sophisticated enough to fully obscure their on-chain footprint. The very largest and most sophisticated players may be invisible in the clustering data, which is worth keeping in mind when interpreting signals.
Tools and Data Sources
Several platforms specialize in wallet clustering and entity tracking. Arkham Intelligence has built a large database of identified entities and their associated addresses. Nansen labels wallets by category (smart money, fund, exchange) based on behavioral patterns. Chainalysis and Elliptic operate primarily for compliance but their entity resolution feeds into publicly available analytics.
For individual analysts, the approach is to start with known addresses (exchange hot wallets, identified fund addresses from blockchain explorers) and build outward using the clustering heuristics described above. Etherscan and similar explorers let you trace transaction flows manually, though this becomes impractical at scale without programmatic tools.
The practical edge from whale clustering is not in predicting specific price movements but in understanding market structure. Knowing that three large entities have been quietly accumulating for months gives you context that pure price and volume analysis cannot provide. That context makes your other analysis, whether technical, fundamental, or flow-based, more informed.