COHI: Coinbase Output Herfindahl Index
Recently I’ve been working on a metric to quantify how quickly coinbase outputs move off miner balance sheets and get integrated into the Bitcoin floating supply. My goal is to see if the halving is putting increased sell pressure on miners.
Below is a chart showing my best version of the metric. It provides useful signal, but I’m still getting comfortable with it. Low values represent increased miner selling, and high values miner hodling.
Full chart here.
Some general observations from this work:
Contrary to my expectation, COHI multiple usually spikes a bit right after a halving event, then draws down into the cycle as miners sell into an appreciating market. This seems to be consistent with other, orthogonal metrics such as Miner Outflow Multiple from Glassnode. My best guess is that this is because weak miners drop out of the market immediately on a halving, leaving only the strong.
I expect miners to take advantage of the recent price runup and start selling into the market soon. It will be interesting to see what impact this has on the price. Some have speculated that miners no longer have enough firepower to damper BTC price in the new halving epoch.
In general miners are very good at timing (or moving) the market, so these miner activity metrics should be closely watched even if they don’t have predictive power outside of sudden spikes.
When the Bitcoin block subsidy halved from 12.5 to 6.25 BTC on 05/11/20 miners were suddenly earning half the revenue they were previously. Miners typically need to sell their Bitcoin revenue to pay mining costs denominated in fiat: chips, electricity, etc.
Strong miners (high margin) can afford to hold some BTC on their balance sheets, waiting to sell until the price appreciates to earn more. Weak miners (low margin) don’t have that luxury and would need to sell immediately to cover costs and stay afloat.
The downward shock to the block subsidy should cut margins for miners across the board. So the halving should have two effects on miners bringing fresh BTC to the market:
Miners have less BTC to sell overall, so supply of new BTC is decreased
Miner margins are cut substantially, forcing previously strong miners to sell Bitcoin before they would have otherwise, bringing more BTC to the market in the short term (this was my hypothesis at least, the metrics say the opposite!)
These two effects can offset each other in the short term but effect #2 should dissipate over time as miners adjust to the new normal. When effect #2 has run its course I expect the Bitcoin price to increase based on simple supply and demand.
To track effect #2 I developed the Coinbase Output Herfindahl Index (“COHI”). A Hefindahl index measures the concentration of a quantity in a distribution. For COHI I use a Herfindahl index to measure the concentration of TXO originating from a coinbase output.
At creation, a coinbase output is fully concentrated in a single TXO and cannot be moved for 100 blocks. After 100 blocks the miner can move the TXO and take actions that will reduce the concentration of that TXO and reduce the herfindahl index for that coinbase output:
Pay out participants in the mining pool, splitting the TXO many ways.
Send the TXO to an exchange to sell, where it can enter circulation and continue getting split and mixed
Mix the TXO with outputs from other sources, diluting it
All of these actions are consistent with the coinbase output getting increasingly assimilated into Bitcoin’s free float supply stock.
Let’s see what the raw COHI time series looks like:
I plot COHI using a 28 day moving average to reduce volatility. The vertical dotted white lines are the halvings. Full interactive chart here.
The COHI value has been trending down over time, which seems healthy as that represents miners increasingly operating at break-even (healthy competition) and more participants mining in general.
The COHI value seems to increase after a halving (contrary to my expectation) but then falls over time. This could be due to a lag effect where miners don’t feel the pressure of decreased margins for a bit. This pattern is also seen in the Miner Outflow Multiple chart below from Glassnode.
COHI local minima tend to occur at price peaks (miners selling at maximum profit) or sometimes at price bottoms (weak miners capitulation).
Compare COHI to the Miner Outflow Multiple from Glassnode:
COHI bottoms in 2014 and 2016 correspond to peaks in miner outflow multiple, but the 2019 and 2020 COHI bottoms do not line up with high miner outflows, they lag them a bit. COHI aligns with the date the coins were mined, whereas the miner outflow multiple aligns with the date the coins were moved, so it makes sense that it aligns with price movements better.
Since the COHI time series trends down over time I created a “COHI Multiple” metric by dividing the 28-day moving average of COHI by the 1-year moving average of COHI. This normalizes the metric and makes miner selling (low multiple) and miner hodling (high multiple) clearer as it compares it against recent history.
You can see clearly from this chart that the COHI multiple usually spikes a bit right after a halving event, then draws down into the cycle as miners sell into an appreciating market. Miners seem to be pretty good at timing the market. Or they move the market, IDK.
Since COHI can be tracked for each coinbase output over time we can generate curves that show COHI decreasing over time as the coinbase output gets more and more integrated into Bitcoin’s free float supply.
The chart below shows the mean COHI curve for the periods before and after the Bitcoin Halvings. At day=0 the COHI value is 1, representing full concentration of the TXO. Over time the TXO gets less and less concentrated and COHI falls.
COHI barely changed before/after the first halving (blue/red), but the baseline level of concentration was higher
The COHI curve shifted up after the second halving (green), compared to before (purple). This is consistent with what we see in the time series charts.
The COHI curve again sifted up after the recent third halving (orange) compared to before (light blue). The shift looks a little stronger in magnitude compared to that in the second halving.
Here are curves for each month in 2020:
These curves could be interesting to compare across mining pools as well to get a sense of how those pools operate.
There are some clear limitations to this metric
Can be gamed by miners splitting TXO or mixing
Can only look 4-6 transactions deep before it's computationally infeasible (for me)
Have to make assumptions about TXO attribution when mixed with coins from other sources
If more participants join a mining pool, the coinbase TXO will get split more ways and contribute to a lower COHI
Still doesn’t seem to provide a crystal clear signal, but can be useful in concert with other on-chain metrics
Let me know if you have any questions or comments about COHI. Feel free to DM me on twitter (@typerbole) or email at firstname.lastname@example.org.