What a Whale Alert Actually Tells You
For the Bitcoin–Ethereum comparison, the crucial object is not a “whale” in the abstract but a sequence: a large transfer is recorded, an alert makes it salient, and researchers observe what happens in a defined window afterward. This chapter explains that sequence and the study's measurement rules so the later participation and volatility results can be read as bounded evidence. The payoff is practical: an address movement can be visible enough to influence behaviour without revealing who acted, why they acted or how much of the surrounding market activity occurred somewhere else.
A transfer becomes a signal in stages
A public blockchain gives observers a detailed record of a transaction. They can see an address, the amount transferred and the time at which the transaction was included in the ledger. They can often trace the movement through subsequent records. That visibility is a major difference from a conventional private account book, but it does not turn the address into a named person. The owner may be an individual, a fund, an exchange or another service, and the transfer's economic purpose may be unclear.
The record is also not the same thing as a trading instruction. A transfer from one wallet to another could accompany a purchase or sale, but it could also be a custody move, a rebalancing operation or a movement between related addresses. On Ethereum, a transfer may sit within a programmable environment that includes smart-contract activity. Even when the asset and amount are known, intent remains an inference unless other information supplies it.
An alert service adds a second layer to this record. Whale Alert collects selected large movements and broadcasts them to followers. The transfer may already be on-chain, but the broadcast reduces the work required to find it. That change in monitoring cost matters because information has an attention dimension. A wallet movement that most observers would never discover can become a prominent event once an alert account identifies it, labels it and distributes it in real time.
This makes the alert timestamp analytically useful. The Philadelphia Fed study uses the time at which the alert was publicly disseminated to identify when information about whale activity became salient, rather than assuming that the underlying transaction's ledger time alone captures the moment when other participants could react. The distinction is subtle but important. A response after the broadcast may reflect the transfer, the new attention around the transfer or both. It may also overlap with a price move, news item or other market event that arrived at nearly the same time.
The study therefore treats the alert as an event around which behaviour can be measured, not as a complete explanation of behaviour. A simplified sequence looks like this:
The event study begins with a 15-minute control window immediately before the alert. It then examines several post-alert windows, including the first 15 minutes and later intervals out to an hour for participation measures. For volatility, the study compares minute-return variation over horizons ranging from 15 minutes to 24 hours. This design asks whether activity after the alert differs from activity just before it, while also accounting for factors such as the preceding one-hour return, preceding 24-hour volatility and regular time patterns.
A matched pre-alert window is useful because crypto markets trade continuously. There is no daily opening bell that supplies a natural before-and-after boundary, and activity can differ by hour or year. Comparing the alert window with a nearby period gives the analysis a local baseline. It does not guarantee that the two periods are identical. A market shock can begin before the alert; a different alert or news event can occur during the window; and traders who saw the underlying transfer before the broadcast may already have acted.
The timing also explains why a result should not be described too broadly. If a measured change appears in the first 15 minutes and fades by the next hour, the evidence concerns a short-lived response. It does not show that the event changed long-term ownership or created a durable price trend. If volatility differs over 24 hours, that still says nothing by itself about whether a particular trader earned a return. The unit and horizon are part of the claim.
The study's whale is a measurement rule
Because complete historical wallet balances are not available, the study defines a whale wallet through observed behaviour: a wallet that conducted a transfer exceeding $50 million at least once during the sample period. The threshold is a way to identify unusually large on-chain events. It is not a universal boundary between sophisticated and unsophisticated people, and it is not evidence that the wallet's owner held more than that amount continuously.
That distinction prevents several common errors. A wallet can be one of several addresses controlled by the same entity, so a single-address rule can split one holder into multiple observations. A wallet can also belong to an exchange or an institution rather than to an individual investor. Conversely, an entity with substantial influence might distribute activity across addresses in a way that the rule does not capture. The category is therefore best read as “wallets with a recorded transfer above the study's threshold,” not as a census of wealthy people.
The data combine transaction-level records for native BTC, ETH and WBTC with Whale Alert notifications. The underlying on-chain histories cover long periods beginning in 2015 for BTC and ETH and in 2019 for WBTC; the alert history begins later, in December 2017, and runs through the end of 2025. Those different starting points matter. The analysis cannot treat every year of on-chain data as a year in which the same alert-based observation was available.
To keep one whale event from being blended with another, the main sample removes events with another whale transaction within plus or minus 120 minutes. After that isolation rule, the paper reports 6,645 BTC whale transactions and 5,075 ETH whale transactions. The buy-versus-sell analysis is narrower because some alerts do not identify a direction clearly. It retains 3,471 BTC buys and 2,413 BTC sells, and 2,473 ETH buys and 2,370 ETH sells.
These filters improve interpretability while also narrowing the population to which the results apply. An isolated event is easier to study than a cluster of transfers, but clusters may be precisely what happens during a genuine market shock. Dropping alerts without a clear buy or sell label avoids inventing direction, but it removes observations that may have contained useful information. The resulting estimates describe selected, unusually large and relatively isolated events, not all large-holder activity.
The comparison also divides non-whale activity into small, medium and large wallet groups. These are observed activity categories used to ask whether the alert reaches different kinds of participants. They should not be mistaken for fixed social classes or a universal definition of retail. The group attached to a wallet can depend on the study's measurement period and transaction history. A label such as “large non-whale” means large relative to the non-whale categories in the analysis, while still below the whale rule; it does not establish the owner's total wealth or expertise.
This matters for interpreting what changes after an alert. Suppose the share of active small wallets rises. That could mean more wallets entered the event, that existing wallets traded more often, or both, depending on how the share is constructed. The study's composition measure focuses on the ratio of active wallets in each size category. Its persistence coefficient uses the pre-alert share as a predictor of the post-alert share. A coefficient of one would mean that the earlier composition carries through one-for-one. A value well below one indicates that the post-alert mix is not simply a continuation of the immediately preceding mix.
The separate change model asks a different question. It subtracts the pre-alert share from the post-alert share and estimates the abnormal change after a whale buy or sell. The first 15-minute intercept is therefore not a price return and not the whale's own trade size. It is the estimated change in the share of a non-whale category relative to its matched pre-alert activity. The later window terms show whether the initial change persists, weakens or disappears.
The volatility model changes the unit again. It uses annualized variation in minute-level returns, including realized volatility calculated from the square root of summed squared returns. A positive estimate means more return variation after the alert than in the comparison periods; a negative estimate means less. Neither sign tells us whether the asset went up or down. A price can rise steadily with relatively low minute-to-minute variation, or move sharply in both directions while ending near its starting level.
The study estimates these relationships separately for BTC and ETH price volatility and also examines cross-market pairs. That allows a WBTC alert, for example, to be evaluated against BTC volatility even though WBTC activity is recorded on Ethereum. It also means that “the whale effect” is incomplete shorthand. A full statement needs at least the alert asset, the price asset, the time horizon and the baseline.
What the design can—and cannot—identify
The controls make the comparison more informative than a simple chart of what happened after each alert. Preceding returns help account for momentum; preceding volatility helps account for the market's existing turbulence; hour and year effects absorb recurring time patterns; and the volatility analysis also uses day-of-week controls. These choices reduce some obvious alternative explanations.
They do not create a randomized experiment. Large transfers are chosen by whoever controls the wallets, and alerts are selected and disseminated by a monitoring service. The same forces that prompt a whale transfer may also affect smaller traders. A whale may move assets because a price change, policy announcement or liquidity problem is already under way. In that case, the alert could coincide with the market response rather than initiate it.
The data also do not observe everything that matters. Activity inside a centralised exchange can settle on private internal books. Ethereum's ecosystem includes smart contracts and layer-2 venues, so retail behaviour may not appear as a direct on-chain wallet transfer during the measured window. The study's on-chain exclusions and wallet proxy leave beneficial ownership, private intent and some venue activity unresolved. Absence from the measured record is not proof that no one reacted.
The appropriate interpretation is consequently layered. The study can tell us how identified non-whale wallet activity and minute-return volatility are associated with selected, publicly broadcast, isolated large-transfer events. It can show that the associations differ between native BTC on Bitcoin and ETH or WBTC on Ethereum. It cannot, from these data alone, identify every causal step from a whale's private decision to a follower's trade, establish that followers made money, or infer a permanent difference in the two ecosystems.
That boundary is not a weakness to hide; it is what makes the later comparison intelligible. Once the event is defined as a broadcast signal and the outcomes are kept separate, the central empirical question becomes sharper: did the alert bring a different crowd into the market, did that crowd trade in the whale's apparent direction, and did price variability move with it? The next comparison addresses those questions directly on the Bitcoin and Ethereum-network rails.