Stop Blaming Miners for Falling BTC Prices

👷 Understanding supply-side dynamics through the lens of coinbase guardians.

Felipe Gaúcho Pereira
Nov 28 · 10 min read
Illustrations belong to Maggie Appleton.

The bulk of the data here was cordially provided by Token Analyst 🙏

Bitcoin's primary economic activity is stupid simple. One plugs hash rate into a network, for the chance to get coins that it spits.

If Bitcoin had miners, alone, it'd still survive. With miners and coin buyers, it already has a market — and value. From a first-principles economic perspective, all else is complementary.

Despite, mining remains little understood by the general public. Much of its modus operandi is perceived through truisms, rather than actual data:

  • "Hash follows price" (or "price follows hash");
  • "Miners basis-cost of production for a bitcoin is a natural price bottom";
  • "Miner capitulation is a prerequisite for a bull run";

And so on. In fact, a prevalent preconception is that miners have a somehow disproportionate stronghold of Bitcoin's prices.

The red line indicates the (virtual) 50% security threshold, while the black line is the Gini coefficient as a measure of market share distribution. From "A Deep Dive into Bitcoin Mining Pools ", 2018.

⏱️ 1. Why Miners Matter

In 1931, Harold Hotelling formalised the notion of Scarcity Rent, which bases much of the economic theory around the supply of nonrenewable resources:

If net prices are not rising at the rate of interest, as a condition of equilibrium, the present value that could be received from selling in some periods would be higher than in other periods.

Apart from the fact bitcoin has appreciated much faster than any rate of interest, this proposition captures the useful notion that miners actively care about when to produce and sell resources — in other words, we can generally assume they sell reserves when they believe prices will fall, and hold when they believe the opposite. It’s a primitive but fundamental dynamic: the rate at which fresh new coins are released into the market is a function of the inter-temporal arbitrage miners practice.

Boltzmann's analysis of net inflows/outflows of major BTC mining pools indicated F2Pool and SlushPool led a sell-off prior to the BSV drama past year, with numbers "17.5 standard deviations below [the] 3-month trailing average".

Worth remembering, supply is inelastic on Bitcoin. No matter how many miners show up to compete, there will always be a preset number of BTC being issued.

But this shouldn't hide the fact that miners hold the keys to Bitcoin fresh supply — quite literally.

Every time a coinbase becomes spendable (100 blocks after it's appended to the blockchain), its miner makes a decision, conscious or not, whether to move its reward or to keep it still.

Hence, miners's collective "hive mind" control a tap that increases / decreases the available supply of new bitcoin in the market.

💦 2. A Tap… or a Drip?

Bitcoin's asymptotic coinbase rewards imply that miners' "economic firepower" diminish over time. Invariably, movements in the outstanding stock of bitcoin hold greater potential implications for prices than variations within the newly introduced supply.

In early 2009, 100% of coins transacted on-chain were BTC that had just been mined. In mid 2012, this proportion had already fallen by 3 orders of magnitude.

Miner’s Share of (on-chain) Volume (MSV) is a simple metric to gauge the relevance of mining output negotiation in relation to all on-chain volume.

MSV = volume of txs. including coinbase-originated input(s) / total on-chain volume

Since almost all (nowadays) mining rewards are spent from the receiving address on the same day they arrive, the MSV will be, in practice, an inverse measure of on chain volume (gradually becoming smaller, and adjusting abruptly every halving).

An MSV of 0.01% means that a monopolistic miner needs to dump 3-months-worth of coinbases (ignoring the rate of change) in order to match just 1/100th of the daily on chain volume.

If the chart above is not compelling enough to illustrate miner's littleness amidst the broader BTC-verse, below is a visualisation that includes off-chain negotiation volume, in spot and derivative markets.

On Chain Volume = raw data multiplied by 4/9 to approximately minimize the effect of change outputs. Spot volume = Bitcoinity's data multiplied by 10% (again, an approximation to get closer to "real volumes").

🎭 3. Miners, Who Art Thou?

In just a couple of years, the hobbyists that forged the first hundred thousand blocks were almost fully replaced by pools. In 2011/12 it became commonplace to self-identify, but some pools may have as well been operating before, unbeknownst to us.

The well documented transition from amateur to professional mining becomes evident on this chart recently plotted by Clain:

Most of the economic analysis on miner activity so far has tackled two key questions: (1) how to model the game theoretic competition that is mining?; and (2) how to map pools in order to track their power/behaviour?

A breakdown of 81 papers related to "Bitcoin Mining" (out of 1032 with the keyword "Bitcoin"), on SSRN, between Jan-2013 and July-2019.

Those in the latter category rely on heuristics that can be generalised as follows:

  • clusterize addresses that receive coinbase outputs;
  • categorize these clusters as pools, individuals or "unknowns";
  • track bitcoin inflow/outflow to and from these wallets over time (and alternatively, between these wallets and other specific clusters, such as exchanges).

Boltzmann, BitMex Research, CoinMetrics and TokenAnalyst have explored data produced with variations of this method in the past.

🏷️ 4. (Labelled) Miner Outflows

There's 2 popular lists of labelled bitcoin addresses accessible online: and Many custom clustering tools have obviously been devised too — some using these lists as starting points.

Below, we explore data from TokenAnalyst - an extract of the dataset that goes from October 2014 to October 2019.

The flows mapped cover transactions from Antpool, BTCTOP, BitClubNetwork, F2Pool, Slush and ViaBTC to the exchanges Bitstamp, Bittrex, Binance, Bitfinex, BitMex, Huobi, Kraken and Poloniex.

🚰 Total Outflows

At first glance, a straightforward depiction of monthly miner outflows to exchanges may look very suggestive:

Although the same plot seems less telling in log scale:

Since we're not interested in evaluating the possibility that miners are impacting prices a month after it happens, we'll be looking at granular, daily data, hereon.

🔦 On Coverage / Representativeness

Ceteris Paribus has calculated these pools “currently make up ~40% of hashrate. With ~657,000 BTC mined per year, (ex. tx. fees) these pools are left with ~270,000 BTC”.

Of those ~270,000 yearly mined BTC, the dataset accounts for close to ~8,500 sold through exchanges in 2019 (~3% of all mined BTC).

We can infer a similar degree of underrepresentation by comparing the total mapped outflow of Slush to exchanges (the most constantly active pool in the dataset) to alternative estimates of what it has historically mined.

Slush: ~800k BTC mined so far vs. 6k cumulatively sent to exchanges since late 2014.

Although Slush’s BTC payouts are likely concentrated in the early days (before 2014), it seems safe to assume we’ll be only covering something between 1–5% of the pool’s actual flows, hereon. This seems like a good approximation for our coverage for the sum of tracked pools, too.

The question we are left with is: do these miner outflows tell us anything meaningful about price returns?

🎎 Correlations

The answer is: no (at least that we can tell with the data at hand).

Correlogram and ADF tests for stationarised Total BTC Outflows and Prices.

After (1st-order) differencing the series, and checking for stationarity, we sought after meaningful relationships between any of the mapped pools' moves and bitcoin's prices.

In correlation scatter plots like the ones below, we typically want to be able to spot that "in occasions when BTC outflows increase, prices decrease" (or the opposite; or anything distinguishable, really).

Correlation between diff(Slush) and diff(BTCPrice); then diff(TOPBTC) and diff(BTCPrice), along the years.

Distributions that resemble a shapeless amoeba or a cross are suggestive of weak relationships (changes in one variable don’t consistently describe changes in the other).

We picked Slush and TOPBTC here because (1) they are the most active pools in the dataset and (2) the same plot for Total BTC Outflows appears to contain even less signal.

Correlation between diff(Total BTC Outflows) and diff(BTCPrice).

The correlation coefficient for both series studied above reveals no meaningful relationship with price returns.

A p-value of 0.062 indicates a ~94% probability that the very slightly negative correlation between Slush outflows vs. price returns is not a product of chance. The even smaller correlation coefficient for TOPBTC cannot be taken seriously due to its outrageous p-value.

(vs. Price Returns) Slush: r=-0.044; p-val=0.0626 | TOPBTC: r=0.004; p-val=0.8834.

Rigorously speaking, none of these relationships can be deemed statistically significant.

Distinct aggregations on the data sample (weekly, monthly, …) don’t seem to unearth special insight, as the following plot demonstrates. We correlated changes in the sum of outflows to exchanges to price returns, for different timeframes, only to find a discrete down trend as the window gets wider.

The spikes (left chart) may be indicative of some form of regimented behaviour on the part of miners. But the observed correlations are way too small nevertheless.

Finally, we tested the differences in the sum of BTC outflows for cross correlation with price returns, but failed to notice meaningful signals, independently of the chosen lag.

There's a myriad of methods that could be applied to more fully characterise these relationships— information coefficient tests, conditional distributions, etc. But the near absence of apparent correlation makes it much more unlikely that any causation exists — so that’s where we stop for now, sparing the dataset from further digging.

👣 5. Next Steps

So far, we concluded:

  • Miners actually command a very tiny portion of all BTC “selling-power” in the markets.
  • The data sample relied upon is not fully representative of the observed phenomena. This analysis probably accounted for 1–5% of all actual miner outflows (we can't tell how much ultimately goes to exchanges, neither how much never does).
  • We didn’t identify meaningful bivariate relationships between miner outflows to exchanges and BTC prices. This doesn’t mean we’ve falsified the hypothesis that they exist (nor that miners have some degree of influence over prices).

A hunch of ours is that distinct pool idiosyncrasies may distract one from global patterns in miner behaviour.

Address-labelling and pool-mapping surely provide insight on Bitcoin's supply side. But “unknowns” have been eating market share steadily since 2017. Relying on the permanent identification of addresses is ultimately an arms race against anonymising techniques.

On the continuation of this article, we'll explore alternative ways of assessing the raw on-chain fingerprint left by miners.

📝 Notes on heuristics:

Pools approach payouts in different manners. The simplest conceivable way is to send payouts to all pool miners in one transaction. Few, if any, large pools do this. Some use an iterative approach: pay one miner, transfer the remaining balance to a new address, repeat. Some randomly choose a number of miners to pay in one transaction, then transfer the remaining balance to a new address, further distributing the remainders in subsequent transactions.

Even coinbase transactions (which can theoretically have >1 output addresses) can carry ordinary pool payouts (F2Pool and Eligius have done this in the past)! One must be careful when trying to map coinbases directly to one single entity.

Some popular pool payout schemes (Romiti, Judmayer, Zamyatin, & Haslhofer, 2019)

The bottom line is: following funds belonging to miners and child entities may be more complex than it seems.

In this article, for example, we described the MSV (Miner’s Share of On Chain Volume). The metric takes into account the 1st transaction out of the address that received the coinbase. Now that you have in mind the payout schemes just described, you may realize the indicator does not distinguish between "paying out", “reshuffling” or “spending”.

We’ll address that in more detail on the continuation of this series.

📚 Bibliography

The block hashing algorithm, revisited. Source.

Paradigma Capital

Home is where the node is.

Felipe Gaúcho Pereira

Written by

Partner @ Paradigma Capital; co-founder @ Paratii

Paradigma Capital

Home is where the node is.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade