On-chain analysis
On public blockchains, every transaction is visible. That allows building metrics, but their informational value is sharply limited by attribution problems.
Unlike stocks, a public blockchain lets you see every single movement. How many addresses are active, how much is moving, how long holdings have sat still.
That produces metrics: how many coins sit on exchanges, how long they've been untouched, how much profit or loss current holders would have at today's price.
That sounds like a real edge, and in part it is. There's no stock where you can see how many shares just got sent to an exchange.
The limit lies in attribution. An address isn't a person. One person can hold thousands of addresses, and one exchange address can belong to millions of customers. Draw behavioral conclusions from that, and you're making assumptions that are rarely disclosed.
Common metrics include active addresses, transaction volume, the age of recently moved holdings, and ratios of market cap to realized value. The latter values each unit at the price of its last move and allows statements about aggregate unrealized gains across the holder base.
The core methodological difficulty is clustering addresses into economic entities. Methods for this rely on heuristics, such as shared input usage within a transaction. These heuristics are error-prone and get deliberately undermined by privacy techniques, which is why metrics like active addresses reliably capture neither user counts nor economic activity.
A selection problem also affects validation. Available time series cover few market cycles, so thresholds for buy or sell signals get calibrated on a very small number of observations. Claims like reaching historical extremes therefore rest on a sample too small for statistical conclusions.
Summary
- Every transaction is visible, but addresses aren't people.
- Active addresses reliably capture neither user counts nor activity.
- The time series cover too few cycles for reliable thresholds.
Did you get it?
What's the core methodological problem in on-chain analysis?
Attributing addresses to economic entities. It relies on error-prone heuristics.
Why aren't active addresses a user count?
One person can hold many addresses, and one exchange address can represent millions of customers.
Why is on-chain evidence statistically weak?
Because the available time series cover only a few market cycles.
Related
- BelgiumTaxes by Country
- Chart patternsStage 3
- Active funds against the indexReality Check