Bitcoin Amplifies the Signal; Ethereum Often Filters It
This chapter focuses on the people behind the post-alert statistics: which non-whale wallets become active after a large transfer, and whether they trade in the whale's apparent direction. The comparison is native BTC on the Bitcoin network versus ETH and WBTC on Ethereum. Reading the results correctly requires one practical distinction: a change in the active-wallet mix is not the same as a change in price, and a same-direction response is not proof that followers understood the whale's motive.
Native Bitcoin changes the crowd around the event
The study's first behavioural measure asks how much of the pre-alert participant mix carries into the post-alert window. It predicts the post-alert share of active wallets in each non-whale size group from that group's share in the 15 minutes before the alert. A coefficient of one would mean persistence: the wallets active just before the alert remain proportionally represented afterward. A coefficient near zero means that the earlier mix has little power to predict the later one.
For native BTC, the reported coefficients are 0.06 for small non-whale wallets and approximately 0.00 for both medium and large groups. Each is far below the no-change benchmark of one. That does not mean that the same wallets disappear, nor that no one was trading before the alert. It means that the post-alert composition is largely independent of the immediate pre-alert composition. In conjunction with the raw activity plots, which show a rise in the share of active non-whale BTC wallets after an alert, the result points to a broadening of the crowd around the event.
This is a more specific finding than “Bitcoin becomes busy.” Total activity could rise because a few already-active wallets trade repeatedly. The composition result points instead to a larger set of wallets becoming active relative to the baseline. The three size groups also make the result less dependent on one presumed retail type. In the full sample, the largest raw increase occurs in the first 15-minute window, and activity then falls in the later windows. The alert seems to concentrate attention quickly rather than permanently redraw the participant base.
There is a useful statistical caution inside that result. The coefficient 0.06 is not a statement that 6 percent of Bitcoin wallets were new arrivals. It is the estimated slope relating the pre-alert share to the post-alert share. Its meaning is relational: knowing which share of small non-whale wallets was active just before the alert gives very little predictive information about the share active afterward. The raw activity means supply the second part of the interpretation by showing that participation rises. Neither number should be read alone as a count of people who entered the market.
The same care applies to the directional estimates. A 23.72-point increase for medium wallets does not mean that 23.72 percent of every Bitcoin holder bought, or that the whale caused a 23.72 percent return. It is a change in the measured share of the medium non-whale category relative to its matched pre-alert activity during the first 15 minutes. That denominator makes the result useful for comparing the alert window with its local baseline, but it also limits the claim. The estimate describes composition within the observed category, not the size of the entire Bitcoin market or the wealth transferred.
Taken together, the measures answer a sequence of increasingly specific questions. First, the active mix changes. Second, the change has a directional sign when the alert is labelled a buy or sell. Third, the response fades across later windows. The evidence is strongest when these pieces line up, but they remain different pieces: a broadening crowd is not automatically a crowd that trades profitably, and directional alignment is not proof of a single shared motive. This separation will matter when the comparison moves from wallet participation to price volatility.
The second measure asks whether those active participants choose the same side as the whale. For a whale buy, the paper estimates the abnormal change in non-whale buy participation relative to the matched 15-minute pre-alert window. In the first 15 minutes, the estimates are 14.81 percentage points for small wallets, 23.72 points for medium wallets and 3.50 points for large wallets. These are participation-share changes, not dollar returns and not the percentage increase in Bitcoin's price.
The sell results are broadly symmetric. After a whale sell alert, sell participation rises by 12.95 percentage points among small non-whale wallets, 29.52 points among medium wallets and 2.95 points among large wallets in the same initial window. The strongest reported response is among medium wallets in both directions. The later window coefficients are negative relative to the first 15-minute period, which is consistent with the burst of activity weakening and returning toward normal within an hour.
The pattern fits the paper's use of the term “leader-follower”: a visible whale buy is followed by more non-whale buying, and a visible whale sell by more non-whale selling. But the phrase should be read narrowly. The data show a same-direction change in participation after an alert; they do not show that every follower copied the same wallet, knew the whale's private information or earned a positive return. A participant may react to the alert as a public signal, to the recent price trend, or to both. The study reports that recent returns amplify the response in important parts of the sell model, which is consistent with momentum and the alert working together rather than with a pure imitation mechanism.
The distinction between composition and direction is useful here. A low persistence coefficient tells us that the active mix changed. The buy and sell intercepts tell us that the change was not direction-neutral in Bitcoin. Together they describe an alert that both widens the set of visible participants and aligns more of their immediate activity with the whale's apparent side. Either result on its own would be less informative. A broader crowd might buy and sell in equal measure; same-direction activity might come only from a small, already-active group. The combination is what makes the Bitcoin response distinctive.
Ethereum keeps more of its measured composition
ETH shows a different pattern. Its composition-persistence coefficients are 0.71 for small wallets, 0.69 for medium wallets and 0.63 for large wallets. These values remain below one, so the post-alert mix is not perfectly unchanged. But they are much closer to one than the native-BTC estimates. The immediately preceding participant mix carries considerably more information about who is active after an ETH alert.
WBTC falls between the two cases, with coefficients of 0.27 for small, 0.31 for medium and 0.51 for large non-whale wallets. That intermediate pattern is not a minor technical detail. WBTC is linked to Bitcoin exposure, but its measured transfers take place on Ethereum. Its position helps separate two explanations that are easy to confuse. If only the asset label mattered, WBTC might be expected to look like native BTC. If the network and venue environment matter, WBTC should show some of the Ethereum pattern. In these composition results, it is much closer to the latter: less persistent than ETH but far more persistent than native BTC.
The directional estimates reinforce the contrast. For ETH, the full-sample results show no broad immediate same-direction response among non-whale groups. The reported exception is the largest non-whale group after an ETH sell alert, with an estimated 0.76-point increase in sell participation. That is very different in scale and breadth from the double-digit Bitcoin responses. The correct conclusion is not that ETH participants never respond. It is that the study finds little evidence of a broad, immediate imitation pattern in the measured Ethereum-network wallet activity.
WBTC buy responses are not statistically significant in the full sample. The only marginal exception reported for WBTC is a 7.63-point increase among small investors after a WBTC sell alert. The raw plots add a more qualified detail: the largest non-whale investors on both WBTC and ETH tend to buy after whale buys and sell after whale sells. The larger groups may therefore be more responsive to direction, even where the broader population does not move together. Meanwhile, the smallest Ethereum-network retail group can reduce activity after either type of whale alert. A muted average can contain several different subgroup responses.
This is why a comparison table of “BTC versus ETH” can mislead if it hides the denominator. The BTC estimates describe changes in the share of non-whale groups active in the measured on-chain environment during a short window. The ETH and WBTC estimates describe the same kind of constructed outcome on a different network. They do not say that Bitcoin has more investors in every sense, that Ethereum has fewer retail traders in total or that one asset is intrinsically more rational. They say that the observed wallet-level response to these selected alerts is broader and more directional on native Bitcoin than on the Ethereum-network observations.
A plausible explanation follows from how activity is distributed across the two rails. Bitcoin's measured native transfers expose a market in which the alert can reach a broad set of on-chain non-whale participants. Ethereum is a programmable platform with smart contracts, tokens and more varied transaction roles. Its on-chain activity may be more dominated by institutions, exchanges and contract operations. Retail users may trade mainly inside centralised exchanges or on layer-2 venues, with those actions later appearing as large balance transfers between exchanges rather than as direct retail wallet movements.
That explanation is useful because it connects the statistics to a mechanism: the network can filter or aggregate the signal before it reaches the measurement. It is not, however, a proven mediation chain in the study. The regressions do not separately identify how much of the muted response is due to exchange settlement, smart-contract execution, layer-2 activity, different liquidity or different investor beliefs. The fact that the data do not show a broad Ethereum on-chain reaction is not evidence that no one noticed the alert or that no off-chain trading occurred.
The same point applies to WBTC. Holding the token's broad economic association with Bitcoin constant does not hold the market environment constant. On Ethereum, WBTC can be transferred among wallets, exchanges and contracts within a programmable ecosystem. That changes who is visible, what a transfer may represent and where a retail response might be recorded. WBTC therefore functions as a diagnostic bridge, not as a perfect experiment: it makes the rail difference harder to ignore while leaving other differences unresolved.
The sequence diagram describes a route, not a claim that every actor follows the same path. On Bitcoin, the study observes a broad on-chain response after the alert. On Ethereum, some activity may be recorded through a different mix of addresses, exchange balances and contract interactions, leaving a more stable measured composition. The arrows from venues to wallets are therefore a reminder of possible filtering and aggregation, not a direct estimate of their size.
What the comparison establishes
The strongest evidence is comparative and specific. Native BTC alerts are followed by a sharp change in the composition of active non-whale wallets and by same-direction buy or sell participation in the first 15 minutes. ETH retains more of its prior measured composition and shows little broad immediate directional response. WBTC, despite its Bitcoin connection, follows the Ethereum-network pattern more closely in the composition measure. The evidence therefore makes it difficult to explain the result with the asset name alone.
At the same time, the comparison does not rank the networks or identify a universal winner. A broad response can mean greater sensitivity to public information, but it can also expose smaller participants to a fast-moving signal and its risks. A stable measured composition can mean that the alert is less influential for direct on-chain wallets, but it can also mean that much of the relevant reaction is taking place in venues the study does not observe. More visible imitation is not automatically better market quality, and less visible imitation is not automatically resilience.
The next question concerns price uncertainty. Participation can broaden and align with a whale's apparent direction, yet volatility depends on how prices move through the event and over what horizon. The BTC and Ethereum-network paths may therefore diverge again when the outcome changes from “who traded?” to “how variable were returns after the alert?”