Exchange on-chain metrics: good, bad, ugly

Glassnode analyst talks about the importance of analyzing the on-chain activity of bitcoin exchange addresses, the processes ofaccurate tracking and the limitations of such analysis.

Cryptocurrency exchanges are important drivers of the digital asset market. It is on them that most of the trade takes place and prices are determined.

Therefore, it is important to track on-chain activity.exchange bitcoin addresses - for example, how much of the bitcoin supply is held on them at any given time - as well as the amount of BTC entering or withdrawing from exchange addresses. This data can provide invaluable information about market liquidity, investor behavior and the supply side of the BTC market.

At the same time, most users lack an understanding of the processes used to track exchange wallets and obtain high-quality on-chain data about crypto exchanges.

In today's article, I will talk about theseprocesses, about the problems of creating exchange metrics and their inherent limitations, as well as provide some general recommendations on how to correctly read such metrics and what to look for.

In a nutshell:

Tracking stock data is a flawed process as each exchange uses unique wallet management methods and wallet addresses are dynamic and constantly changing.
Individual data points such as large inflowor capital outflows should be considered preliminary until they are verified and confirmed. Professional analysts - including Glassnode - are more likely to take a conservative approach to such data in order to limit false positives and present the most accurate data in the output.
The reliability of data points in stock metrics can beconsidered increasing over time, since exchange wallets make transactions and interact with each other, on the basis of which the heuristic rules and clustering algorithms used by analysts over time increase the level of marking of these wallets and, consequently, the accuracy of the data obtained.
Historical data in exchange metrics can also change over time due to the addition of new addresses to exchange clusters - either by an algorithm based on heuristic rules or manually.

Good

The Bitcoin blockchain is an open ledger that allows an observer to analyze every transaction ever committed and estimate the number of coins held or moved by any address on the network.

To track the activities of exchanges, you need to knowwhat addresses they control. These addresses can then be tracked and their activity data aggregated to create metrics that reveal invaluable market insights.

Typical exchange metrics include:

balance of exchange addresses;
the volume of coins entering and withdrawing from exchange addresses;
the number of deposits / withdrawals from these addresses.

</p>

Why are stock metrics important?

Here are some examples of exchange behavior that might be of interest:

Exchange reservesThe depletion of exchange liquidity reserves (as we have already observed) is indicative ofinvestor sentiment and provides us with valuable insights into their current behavior.
Withdrawing capital from exchangesmay mean that investors are moving their coins to self-storage and/or long-term cold storage.This can be regarded as a bullish signal, given that storing BTC in your own wallets canbe seen as presumably more long-term, which means that investors are confident in the growth of the BTC price.Other possible explanations include increased activity by OTC brokers and custodial services as a result of high institutional demand, or the use of capital in other financial services (e.g., as collateral for loans).
Capital inflows to exchangesmay indicate increased trading activity or investor interest in profit-taking and/or rebalancing portfolios for risk management.

How are exchange bitcoin addresses tracked?

Various mechanisms are used to track addresses, which can be divided into three main categories: verified addresses, external sources, and clustering.

Verified addresses

This is a simple and straightforward step.This category includes addresses officially confirmed as controlled by a particular exchange. For example, when the exchange officially announced (publicly or privately) that a particular address really belongs to them. This also includes addresses confirmed in the process of direct interaction with the exchange (for example, through a deposit of funds).

External sources

Since we are talking about a public blockchain and exchanges interact withWith millions of users, tags for stock addresses can be found scattered around the Internet.This is the crowdsourced part of putting and tracking stock tags.For many others, this is not the case, especially when different sources contradictto each other — for example, the source ofAAssociates the address with the exchangeX, and the sourceBlinks the same address to the exchangeY.

Professional analysts beforeuse address labels obtained from open sources, subject them to verification, which ultimately determines whether the address really belongs to a particular exchange. Such a quality control process can include, among other things, an analysis of the activity of addresses, their type, interaction with other network actors, the structure of the balance and the number of external sources of confirmation of this label. Processes can be streamlined and include both automated and manual checks.

Only if the address label can be verifiedwith a very high probability and without conflicting data points, it is recognized as verified and added to the pool of addresses belonging to a particular exchange.

Clustering

Heuristic algorithms and clustering arepowerful tools to automatically detect addresses belonging to exchanges. With the help of heuristic rules and clustering, a huge number of addresses can be identified based on just a few initially verified labels. This is possible through the use of powerful data analysis techniques for statistical inference based on patterns and characteristics inherent in Bitcoin's UTXO-based design. In fact, based on just a dozen initial bitcoin addresses, hundreds of thousands of new addresses can often be identified. This step is essential in order to properly track stock tags and create metrics that paint the big picture. Without these measures, meaningful exchange metrics are nearly impossible. With an increase in the amount of data and improvements in methodology, this approach becomes more accurate over time.

But, as is the case with labels from externalsources, the data obtained in this way needs to be optimized to reduce the number of false positives. If the likelihood of a tag's validity is not very high, we at Glassnode will not mark such an address as a stock address. We would rather miss an address than mark an address with a low confidence level.

In general, the combination of these methodologies providesa powerful framework that allows you to get a fairly complete picture of on-chain activity of exchanges and provide transparency in relation to these pillars of the market through highly informative metrics.

Bad

The logic described above sounds quite simple: define exchange addresses and track changes in their balances, that is, how many coins are received and withdrawn from them.

Yes, in theory everything is so, but, as usual, in practice everything turns out to be somewhat more complicated. Let's take a closer look at some of the problems associated with address labeling.

Exchange addresses are not static

By initially defining a list of stock addresses and simply tracking their balances, you won't get very far. The set of addresses belonging to a particular exchange is constantly changing, and changes greatly.

For example, the figure below shows the quantityaddresses associated with a specific exchange. Consistent growth of the corresponding address cluster is evident, currently numbering almost 25 million units. It is worth noting that the lion's share of these addresses have a zero balance; the number of addresses with non-zero balances remains below 100K for the most part. This is just one example of the dynamic nature of the ever-changing exchange wallets.

Therefore, the biggest challenge is building a reliable system that can track these changes and keep the current set of exchange addresses up to date.

Exchanges can constantly create new addresseswallets and, of course, take advantage of this opportunity. This can be the creation of new cold wallets, to which large quantities of coins are transferred. The exchange redistributes balances to new wallets on a regular basis. In addition, many exchanges do not reuse blockchain addresses, instead constantly creating new ones (for example, to receive change or to move funds to another address).

In addition, the exchanges adhere to highsecurity standards, and often use mechanisms that include complex movement of funds on-chain according to templates that are unique for each specific exchange. These internal mechanisms are very different from exchange to exchange and need to be identified and tracked on a per-exchange basis.

Finally, on-chain activity just gets complicated withtime. Exchanges today are multi-billion dollar companies that provide financial services that go beyond simple spot trading. Many offer their clients futures trading and / or have created an infrastructure for the safe custody of institutional capital. While in theory the underlying settlement level is independent of this development, this is reflected in the number and complexity of on-chain transfers of funds. For example, from a network point of view, it is not always possible to immediately determine whether all custodial wallets are included in the set of flagged exchange addresses. It depends on how these addresses are used by the exchanges, which must be properly studied before drawing conclusions.

Considering all of the above, it is clear that the correcttracking exchange addresses is not a trivial task. Analyst companies use the best available mechanisms to get values that are as close to reality as possible. But given the nature of exchange addresses, it should be clear that:

On-chain data about exchanges can sometimes be imperfect, at least at the level of individual transactions.Some room for uncertainty remains in any case, as data gaps or false positives are possible in relation to specific deposit or withdrawal transactions. Despite the use of advanced technologies, it is sometimes impossible to immediately detect the creation of a new cold wallet (address for the first time on the network), which receives internal funds from a known exchange address. Many heuristic rules are triggered only for certain actions and patterns appear over time. And note that this is especially true for sudden large transactions - the average size of internal exchange transactions often significantly exceeds user transactions (see graph below).

However, significant inaccuracies appear only occasionally - in general, stock exchange metrics are quite accurate, especially if we take into account mid-term and long-term trends.

Evil

What are the implications of the above? Simply put, the values of stock metricsmay change... New information may appear thatadds (or in rare cases removes) the stock address tag. Information can come from the channels mentioned above: for example, a new address can be officially verified by the exchange itself and / or by an algorithm that works and attaches an initially unconfirmed address to an existing cluster.

Ultimately, this means that historical values in stock metrics can change over time. This is always worth keeping in mind.

Conclusion

Does this mean that on-chain metrics are useless? Not at all, on the contrary!

Even if every single exchange streamcapital is not always possible to immediately verify, it is very important to have a general understanding of the activity of exchanges. Stock metrics are for the most part complete and have served as a source of invaluable information for researchers, investors and traders for many years.

Transparency in relation to the actions of exchangesextremely important, especially given the number of reports of fake trading and fake trading volumes in the past. Analysis of the on-chain activity of exchanges gives access to a completely new verifiable, objective and incorruptible data source, and can be useful to any investor.

At the same time, for a more accurate interpretation of the data obtained, it is useful to understand how these metrics are calculated, at least in general terms.

Main conclusions

Collection and analysis of on-chain exchange data is complextask, and individual capital flows can sometimes be verified only after some time. This is because exchanges use complex on-chain processes with a constant change of blockchain addresses.
Historical values of exchange metrics canadjust over time. This is due to a) clustering algorithms that automatically update the set of exchange addresses as new statistical information is received, and b) manually adding tags to new verified exchange addresses. Although the algorithmic addition of exchange clusters occurs on a daily basis, this only slightly affects recent data. Adding new labels can have more impact on historical data, but this happens quite rarely.
Analytical companies optimize the receiveddata to reduce false positives. This means that the likelihood of removing a tag from an address assigned to an exchange is much lower than adding a new address to the cluster. If the address is associated with a particular exchange, it almost certainly will not change.
Exercise due diligence regardingshort-term data on exchanges. This is especially true for large (single transaction) capital outflows. Such movements always require careful study. A sudden outflow of ₿10 thousand from an exchange address may well turn out to be just an internal transfer. Although many of them are immediately detected by algorithms, some cannot be quickly identified and are only checked manually over a period of several hours.
All of the above is especially important to keep in mind.when using stock metrics in intraday trading. First, the initial capital inflow / outflow transactions can be reversed if they are ultimately identified as an internal transfer. Second, as historical data changes, your models and backtesting can be affected. Always keep this in mind if you are training a model based on stock metrics.