Crypto Coin-Age and its Relationship with Price
Far from being human readable, blockchain data structures are not even trivially queryable by machines. However, with correct scientific application, the patterns and anomalies in the distributed data they contain can reveal a surprising amount of valuable and actionable information not previously available from traditional centralised markets.
In this series of articles, we begin to expose some of the derivative metrics available from the on-chain data, starting with the concept of ‘coin-age’, and examine its correlation with Bitcoin price, in both rising and falling markets.
By analysing each transaction executed on the Bitcoin blockchain, it is possible to calculate how long each amount of coin has been dormant. This metric gives an indication of the relative activity between long-term buy and hold investors and short-term hot money accounts.
In short, by examining the inputs of each transaction in each block and traversing their path through the transaction graph to the point at which their corresponding outputs were last mined, we are able to determine a weighted coin-age of each block in real-time and throughout the history of the Bitcoin blockchain.
We find that significant changes in weighted average coin-age correlate strongly with, and in advance of price movements in both rising and falling markets.
Anomalous spikes in mean coin-age reliably correlated with future price moves
The naive average coin-age on a given day versus average daily price throughout Bitcoin’s history is plotted below.
To calculate naive average coin-age on a given day the following equation was used,
where the sum was taken over a given day with N number of coins transacted.
The noticeable feature in this graph appears when Bitcoin experiences aggressive price movements and vice versa: rapid changes in the average age of coins being spent come before significant price movements. This feature is more apparent in the earlier years of the active Bitcoin market, such as preceding the bull runs of 2013, 2014 and 2017.
One explanation for this is that real-money market participants have significant value stored in cold wallets and preempt large trading activity by moving funds to more liquid facilities. Hence high coin-age can signify that old coins were moved, probably with an intention to imminently exchange, whereas low coin-age may signify that a lot of new coins were accumulated and retired to a cold wallet.
However this average coin-age signal conspicuously fails to account for the lack of price movement during its spike in 2015Q1 and is noticeably weaker in recent times as the market becomes less idiosyncratic, so we continue to refine this naive metric below with a weighting by value.
Weighted mean coin-age correlates better with future price
Taking the simple mean of coin-ages has a disadvantage that high value transactions with high coin-age get lost among the daily churn of low value, low age transactions. To progress further it is important to link a coin’s age to its amount. This is done by the following weighted mean formula which allows high value transactions to contribute more to the mean,
where the sums are taken over an entire day (i.e. ∑ amounts, in the denominator is the sum of all transacted coins on a given day).
In the graph above, the black line shows the weighted mean coin-age on a given day as per the given formula. To better expose the trends and smooth out noise in the black line, the green and red lines respectively show the 50d and 200d moving averages of the weighted mean coin-age.
From the graphs it can be seen that peaks in the 50 day moving average align quite well to peaks in price during bull markets in 2013–2014, and price jumps during 2017 bull run.
We also see that times when the weighted average coin-age gains or loses some momentum, where the quicker 50d indicator crosses the 200d, can be indicative of a change in price.
To quantify how well coin-age curves match the price movement, Spearman’s correlation coefficient was calculated for all three traces.
An ideal correlation of 1 would mean that whenever the price moves in a certain direction, our coin-age metric moves in the same direction. If the correlation is 0, metric behaves randomly compared to price. In our case, we see a strong correlation which fluctuates around 0.74 with lowest correlation being the black line itself.
A more insightful way to look at correlations is to look at a moving window and see how the correlation changes through time. This type of analysis is common in legacy financial markets when analysing correlations between assets.
Coin-age distributions on significant days show anomalies
An alternative way of considering transactional behaviour on the blockchain looks at the distribution of coin-ages instead of a mean.
The peaks in aged coin movement in the previous graph can also be observed using histograms of coin-ages on the blockchain. Looking at a plot of the entire history of the blockchain, one can see a smooth normal distribution with a very extended tail and a very high count of short coin-ages. We might surmise that this matches Bitcoin behaviours in 3 categories: the churn of Bitcoin day traders with holdings for less than a couple of days, medium term holdings that are held between 2 days and 30 days and long term 30 day+ holdings.
It is also possible to look more closely at individual days with significant price movements and high trading volumes. In the process of doing so a histogram otherwise similar to the above appears to contain unexpected patterns. To illustrate these let us look at the distribution of coin-ages during two particular days: 29 Nov 2013 which was the day with the highest price during 2014 bull market and 11 Nov 2013 which was a day with a peak in coin-age preceding a peak in price.
The histogram gives us a chance to get a detailed view of what is actually happening during days with high coin-age: we can see that on 11 Nov 2013 in the 8k coin-age region an anomalous number of old coins was transferred. Our analysis has shown that this anomalous high count of older coins is often present during spikes in coin-age which do precede a price move: it is present in 2017 and 2019 bull markets as well.
A possible explanation for this observation is that transfers of old coins are usually characteristic of HODLers moving their funds from a ‘cold wallet’ to an exchange or vice versa and since older coin transactions represent a more fundamental transfer of value than the daily churn of low age coins analysing these anomalies can potentially provide useful insights on price levels.
Access to this type of information can help participants to understand what type of investors are currently present in the market, and plan their actions accordingly.
A holistic view of the Bitcoin blockchain
Another way of looking at the data presented above is by using Catallact’s holistic visualization of the entire blockchain, in real-time. The high coin-age anomaly can be found in November 2013 region, highlighted by a white rectangle where it takes the form of a horizontal line. These patterns can also be seen in the other two highlighted rectangles where it can be noticed that increased numbers of old coins were moved during those periods.
It should be noted at this point that all three visualizations are different ways of viewing the same information; which is the signal of aged coins moving on-chain.
To explain, in the above diagram, each row represents a particular block at a certain block height. The points on each row represent from which historical block bitcoins have come, coloured by their contribution to that block by value. So if, on row y=5000, there is a point at x=100, that means that the block at height 5000 included a transaction that references outputs that were originally created all the way back in block 100.
Such a radar view helps users to visualize every transaction in the Bitcoin blockchain at once, it also allows users to quickly notice patterns and anomalies similar to the ones we found in the earlier histogram.
Without a team of quants and infrastructure engineers, getting timely access to this information from the raw public data is hard. Catallact distils such information into a real-time API and presentation layer for explainable and actionable intelligence.
Please follow us on medium or twitter to receive more in this series of on-chain data analyses. Or contact us directly at info@catallact.com if you have any specific questions.
Many thanks to Jack Tatar, Ron Kochman and Ruhan Wang and for their feedback on this article.
