BITCOIN
DIFFICULTY
HASHRATE & COMPUTATIONAL POWER
BITCOIN DIFFICULTY
AND HASHRATE
Overview
The technology at the heart of the Bitcoin network is designed to bundle
together chunks of transactions (called blocks) and validate them
approximately once every ten minutes. A reward is issued to the network
participants (miners) that contribute resources to making this
process secure. But with the huge rewards promised to the miner that
creates a new block and the rapidly increasing computing power that is
thrown at the task, how is this interval kept consistent?
To answer this, let’s take a look at the process a miner completes when
mining each block. Blocks on the network are just chunks of data: each
chunk starts with some information about itself and the block
immediately before it, continues with a list of transactions to be
broadcast to the network, and ends with a random string of letters and
numbers called a “nonce”. The miner’s job is to compile this data and
broadcast it to the rest of the network. Only then do they get to claim the
block reward - currently 6.25 BTC (~$360,000 [1]), alongside transaction
fees.
Lottery tickets
The network chooses a miner to validate a block by holding a lottery: the
first miner to find a winning ticket (by finding a nonce that, when fed
through a specific algorithm, produces a number low enough to be
accepted) gets to mine the block. By requiring the random number to be
lower than some value, the network is able to control the number of
winning tickets available. The lower this requirement is, the fewer
winning tickets there are to find, and the more guesses that must be
made before a miner wins the lottery.
Bitcoin difficulty is the mechanism by which the network balances the
number of lottery tickets against the number of miners entering the
lottery, in order to ensure that a winner is found (on average) once every
ten minutes. When the mining network makes double the number of
guesses per second, the number of available winning tickets must be
halved to ensure the same frequency of a winner being found, and so the
“difficulty” doubles. Therefore it is also a measure of the computing power
that must be expended before a block is mined, both for individual
miners and the network in aggregate.
[1] Price obtained on 4/12/2021from https://coinmarketcap.com/currencies/bitcoin/
HASHRATE & COMPUTATIONAL POWER
HASHRATE &
COMPUTATIONAL
POWER
Difficulty & Hashrate
Bitcoin’s difficulty does not respond immediately to changes in
computing power on the network. It is recalculated after every
2016th block which, at the target time between blocks (block-time)
of 10 minutes, is around once every 2 weeks. The update ensures that
the difficulty is directly proportional to the average number of
guesses per second made by the miners (in aggregate) over the
previous 2016 blocks. We call this measure of computing power the
“hashrate” - this measures how many nonces the network can guess
per second.
Figure 1.1 Average hashrate over each fortnight leading difficulty updates by around 2
weeks. Periods are approximate due to random blocktimes (average 10 minutes).
The difficulty updates are also bounded - updates can be no more
than 4 times the previous difficulty level, and no less than a quarter
of the previous difficulty level. Whilst these bounds will remain active,
they have been applied only twice - both occasions before Bitcoin
celebrated its first birthday in 2010. In fact, the difficulty has been
changed to more than 130% or less than 76% of its previous value
only 37 times in Bitcoin’s history. For these bounds to apply, the
average hashrate would also have to increase (or decrease) by a
factor of 4 over a period of 2 weeks which, with miners spread across
every continent, would likely mean disaster for much more than
Bitcoin mining.
HASHRATE & COMPUTATIONAL POWER
Expected mining rewards per second
The difficulty is only increased (decreased) following a sustained
increase (decrease) in total hashrate over the previous two weeks, in
order to maintain a blocktime of 10 minutes. During the two weeks
before difficulty is increased, miners who have not maintained their
share of that total hashrate do not see a decrease in expected profits
per second: their lower probability of mining each block is matched
by a proportional increase in the number of blocks on offer (caused
by the higher network hashrate). Their lower probability of winning
each lottery is compensated by more frequent lotteries.
This changes at the end of that two week period, when the difficulty
is raised automatically to maintain block time. This then lowers the
number of blocks to be mined in a given period (back to 6 blocks an
hour), meaning that miners who have not maintained their
proportion of global hashrate now have a lower probability of
winning each lottery and less lotteries being held.
The only way a miner can avoid this is to have scaled up their own
operation in proportion to the rest of the network, incurring higher
running costs. During the two week period before difficulty is
adjusted, these miners would see both a higher probability of
winning each block, and more blocks on offer. After the subsequent
difficulty adjustment, they keep their higher chance of winning
whilst the frequency of lotteries returns to 6 an hour. This process
locks miners into an arms race to keep up with an ever growing
global hashrate to even maintain their current revenue in the long
term. They must either increase spending and energy drain, or face
being squeezed out of the network.
Who is mining?
The costs of mining (mining equipment, land, and cooling) are much
less volatile in fiat terms when compared to the expected mining
rewards. To reduce the variance in their profits, many miners choose
to allocate their computing power to a mining pool. Rewards for
blocks mined by the pool are shared to the miners proportionally to
the hashrate they contribute, meaning that miners can lock in a
revenue stream at a fee paid to the pool operators.
Hashrate from pools like these now mines ~50% of new blocks, but
they have at times contributed around 80% of the global hashrate.
The figure below shows the disappearance of the hashrate
contributed by China-based pool “Poolin” as a result of the state's
recent ban on cyrptoasset-related activities.
HASHRATE & COMPUTATIONAL POWER
Figure 1.2 Distribution of hashrate between pools and other sources from 2019 to
today. (Blockchain.com)
Whilst many of the first blocks were mined by a small group of
miners, it is now much more difficult to mine profitably as an
individual. The global hashrate is increasingly dominated by
professional mining outfits that are able to withstand short-term
volatility in their returns and provide the initial capital for specialised
mining equipment. Despite this, the ever-rising market price of BTC
has repeatedly overcome rising mining expenditure, and could even
incentivise mine-and-hold strategies if the trend continues.
MODELLING HASHRATE
MODELLING
HASHRATE
Trend
Since the first block was mined, hashrate has grown near-
exponentially. Along with larger hashrate levels comes an increase in
volatility of that hashrate.
Figure 2.4 Volatility of hashrate changes increases with an exponentially increasing
hashrate.
Plotted in Figure 2.4 are the daily hashrate (in black) and a rolling 7-
day estimate of the volatility of the hashrate changes above and
below its rolling 7-day average (in red). The proportional increase in
volatility suggests if we were to model hashrate using a random
stochastic process, then we could model the volatility of its daily
differences using an appropriately scaled Brownian motion with
positive correlation with the hashrate process, indicating a positive
skew.
Distribution of Hashrate
We can evaluate the evolution of hashrate by removing the rolling 7-
daily average. This leaves us with a daily series of values above and
below this rolling average. Then, scaling by a 2-week rolling estimate
of its standard deviation gives us a mean 0, variance 1 series of daily
price changes. Comparing this distribution to a standard normal
allows us to confirm the assumption of normally distributed
increments. We find strong evidence for normality in the daily
changes in hashrate.
MODELLING HASHRATE
We can evaluate the evolution of hashrate by removing the rolling 7-
daily average. This leaves us with a daily series of values above and
below this rolling average. Then, scaling by a 2-week rolling estimate
of its standard deviation gives us a mean 0, variance 1 series of daily
price changes. Comparing this distribution to a standard normal
allows us to confirm the assumption of normally distributed
increments. We find strong evidence for normality in the daily
changes in hashrate.
Figure 2.2 (a) Histogram showing distribution of detrended hashrate daily changes. (b)
QQ-Plot of quantiles of detrended hashrate daily changes against the quantiles of a
Normal(0,1) distribution.
Autocorrelation Behaviour
Somewhat unexpectedly, we also find significant evidence for
autocorrelation of -0.4 at a lag of 1 day in the hashrate series,
meaning that increases in hashrate today are linked with decreases
in hashrate tomorrow.
Figure 2.3 (a) Autocorrelation of hashrate returns up to 20-day lag, showing significant
correlation at 1-day lag (99% CI in light red). (b) Scatterplot of daily hashrate changes
against the previous day's changes.
MODELLING HASHRATE
It is difficult to explain this behaviour in terms of miner activity. It
appears to suggest that, outside of the longer-term trend, hashrate
follows a daily cycle of growing and shrinking. As noted earlier,
expected profits are not affected by changes in hashrate until the
difficulty is changed at the end of a 2 week period, so it is
unexpected to see this behaviour on the daily scale.
This behaviour persists when we instead consider the daily log
multiplicative returns (log of hashrate today as a percentage of
hashrate yesterday). If correct, this could contradict an important
assumption about the independence of increments that impacts the
pricing of any derivatives with hashrate as the underlying asset. Such
behaviour is worth further analysis, which we wish to explore further
in a later report.
HISTORICAL DRIVERS OF HASHRATE
HISTORICAL DRIVERS
OF HASHRATE
Weather Seasonality
We find that most of the gains in hashrate occur between mid-May
and mid-October. The chart below shows the daily hashrate above
the hashrate recorded on Jan 1st for the past 6 years.
Figure 3.1 Hashrate above/below the hashrate recorded on the 1st Jan for the past 6
years.
The most striking aspect of this graph is the near-constant growth of
the network, which we noted previously is near exponential. From
2016 through 2020, the largest gains in hashrate coincide with the
monsoon season in China, which saw a yearly migration to the
Sichuan region of Mainland China as miners chased the cheap
energy provided by an excess of hydroelectricity. Additionally,
renewable energy sources are often most abundant during these
months, driving down the price of electricity and making mining
more profitable for more miners.
The large drawdown in June 2021 (yellow curve) is the result of a
large-scale crackdown on mining operations in China. Despite the
large-scale exit of miners from the country, we again observe
sustained growth throughout the summer of 2021. We believe that,
rather than a continuation of the previous trend, this is due in part to
the drop in difficulty supporting a recovery in hashrate and the rising
price of Bitcoin - both factors that attract miners to the network.
HISTORICAL DRIVERS OF HASHRATE
Chinese dominance
Prior to 2021, mining activities in China were responsible for up to
75% of newly mined blocks - a real threat to the decentralised ethos
of the currency. During the summer months, the Sichuan region
alone controlled up to 37.5% of the global hashrate. Why did so many
miners choose to set up their operation here? As noted earlier, the
nature of bitcoin mining means that operational costs are largely just
electricity and hardware, and China has both in abundance. The
country has strong production lines for every part needed in a mining
rig and a surplus of hydroelectricity during the summer monsoon
season. In the winter, relocation to Northern regions could reduce
cooling costs, boosting efficiency of their mining equipment.
Figure 3.2 (a) Mining activity in mainland China during the winter. (b) Mining activity in
mainland China during the summer. Legend shows hashrate as a percentage of national
hashrate. (both https://ccaf.io/cbeci/mining_map)
Exiled Chinese miners have settled in countries with the same
properties they found in China: reliable, cheap electricity and access
to affordable mining equipment. This explains the growth of mining
activities in Kazakhstan and Russia, which have seen a surge in
hashrate above the growth seen before the Chinese crackdown, and
to some extent the flourishing mining community in the US and
Canada - both energy hubs with great hardware manufacturing
capabilities.
With the settlement of miners in these countries, we expect to see a
different seasonal trend, dependent on the seasonal availability of
cheap electricity in these new mining hubs. If miners in the US
continue their interest in renewable energy, for example, then they
may continue the trend of strong growth in the summer months.
Long hours of direct sunlight, coupled with lower demand for
heating could drastically reduce running costs.
HISTORICAL DRIVERS OF HASHRATE
Bitcoin Halving
Mining Economics
The reward for mining a block, currently 6.25 BTC, also operates as
the sole method by which new coins are minted. This is the only way
new coins can enter the market. The network is designed to cap this
supply at a finite value, at 21 million BTC, by halving the number of
BTC awarded to successful miners every 210,000 blocks. Again, the
target block-time of 10 minutes means that this is on average every 4
years.
If the main incentive for mining is halved, what happens to the
hashrate? Is it still profitable to mine? How many miners remain on
the network willing to take that chance? Let’s answer this by looking
at the (scaled) behaviour of the network’s hashrate a month either
side of each halving date.
Figure 3.3 Hashrate activity above/below the hashrate on the day of a halving event.
These values are scaled by their standard deviation.
After each halving the hashrate drops significantly: many miners
choose to stop mining as the lower rewards mean it is no longer
profitable for them to continue. Despite having a larger share of the
global hashrate, miners who remain on the network do not see an
increase in the number of blocks that they expect to mine. This is
because whilst their probability of winning each block lottery is now
higher, there are proportionally fewer block lotteries held in a given
time frame, maintaining the number of blocks that they expect to
mine per hour.
HISTORICAL DRIVERS OF HASHRATE
Around two weeks after the halving event, the difficulty is lowered to
match the lower computing power on the network, increasing the
number of block lotteries held per hour and reducing costs for
miners who kept mining and attracting more miners back to the
network.
Not all halvings are created equal
Following the second halving in July 2016, we saw an 11% drawdown
in the global hashrate over the next month, whilst the latest halving
in May 2020 saw hashrate drop by more than 25% in the same
timeframe. Can we explain this?
The answer lies in an omitted variable: the BTC price. The rewards are
quoted in BTC, while the cost of electricity is charged in local fiat
currency. What this visualisation does not take into account is that
during the first and second halvings, the reduction in reward was
from $625 to $312.50 and from $16,185 to $8,092.50 per block
respectively. In May 2020, the reward per block mined dropped from
$109,500 to $54,750.00.
While this does influence the effect of a halving event on mining
activity, it is important to note that mining costs have also increased
near-exponentially during this period (proportional to hashrate on
the network), and so the breakeven point for miners will depend on
the delicate balance between both sides of this coin.
Mining Technology
Global arms race
The global arms race for hashrate dominance is not without its arms
dealers. Early miners adapted commercially available CPUs and
GPUs, which were able to produce a competitive hashrate in such a
small network. However, the rapid growth of the network quickly
made this strategy untenable. The release of the first dedicated
miner (ASIC - application-specific integrated circuit) in 2012 marked
a decisive switch to specialised equipment, which now dominates
the global hashrate.
HISTORICAL DRIVERS OF HASHRATE
Release Dates
There is no doubt that the near-exponential growth in hashrate on
the network is driven primarily by the advent of increasingly powerful
specialised equipment. The current figure of around 150 million TH/s
(150 quintillion guesses per second) would require 1.25 trillion units of
the most powerful GPU on the market.
To quantify this effect, let’s look at the increase in global hashrate 2
months after the release of a selection of ASIC miners. We limit this
analysis to announcements of one brand (Bitmain Antminer) of
equipment and evaluate only those announcements that increased
profitability by some rough estimate. Note that the hashrate of each
model is quoted in Terahashes (Th/s).
Table 3.1 The release of mining technology usually leads to a spike in global hashrate
around 2 months later. Data in this table sourced from
https://www.asicminervalue.com/.
Since January 2017, the average increase in global hashrate over any
2-month period was +15%, showing that with the exception of
November 2018 and May 2020, technology releases lead to a spike in
hashrate. The two dates that buck the trend highlight a different
result. Both release dates coincided with events that had a much
stronger influence on miners' decisions to mine: the former with a
near 50% drop in BTC' price, and the latter with a bitcoin halving.
These events have a much larger effect on hashrate than the release
of new technology, showing the hierarchical dependency of hashrate
on these driving factors.
HISTORICAL DRIVERS OF HASHRATE
Effect on miners
The announcement of new equipment is usually detrimental to
existing miners’ profit margins. New technology expedites the
redundancy of old equipment, allowing others to mine more
efficiently and increasing their chances of winning the next block’s
lottery. Running outdated equipment can even have negative
profitability if the value of the BTC rewards does not exceed the
costs. This leaves miners with the choice to update equipment at
considerable cost or accept being squeezed out of the market.
Increases in network hashrate only reduce each miner’s expected
profit two weeks later, at the time of the next difficulty update. This is
because the increase in hashrate causes a proportional decrease in
block time, meaning there are proportionally more blocks on offer in
a given time frame, and keeping the expected profit per second
constant. The effect on profitability only takes hold when the
difficulty is adjusted to bring block time back to 10 minutes, and
locking in a miner’s reduced expected profits and increased costs.
Too many miners exiting the market in anticipation of their
decreased share of hashrate could spark a drop in difficulty two
weeks later which, if large enough, could lower running costs enough
to make old machinery profitable again. Additionally, miners
expecting BTC to continue to appreciate may take the bet that coins
mined at a loss now will be profitable in the future.
China Exodus
The biggest factor influencing the mining landscape in the last
twelve months has been the swift regulatory action taken by the
Chinese government. Citing “financial and environmental concerns”,
officials strengthened previous restrictions on the industry by
banning outright all mining operations in May 2021, and outlawing
all cryptoasset activites in the following September. At the same time
as accomplishing the stated aims, the announcement allows the
government to limit extra-state financial activity and clear the
market for its own digital currency, backed by the central bank.
HISTORICAL DRIVERS OF HASHRATE
Figure 3.4 7-Day rolling average of total hashrate on the network throughout 2021.
Despite heralding a decisive end to mining operations, the
announcement was just the latest in a long line of restrictions on the
industry. By May 2021, China’s share of hashrate had dwindled from
its ATH of 75% (Sep 2019) to 45%. The resulting exodus merely
accelerated an existing migration to more friendly climates like
Russia and North America, where operations had been scaling for
around 2 years.
The drop also marked a rare occasion where professional miners with
cutting edge mining equipment were responsible for the drop in
hashrate, rather than the attrition of miners with outdated, inefficient
mining tech. In cases like these, smaller mining outfits benefit from
their sudden increase in the proportion of hashrate that they control.
This increase in income is supplemented two weeks later by the
resulting drop in difficulty, which further increases small miners’
profit margin by lowering computing costs.
The swift recovery after such a large, sudden drop can be explained
in part by the return of smaller miners, along with the continued
growth of the industry in North America. Regardless of where the
power originates, this is compelling evidence for the robustness of
the mining network, and contextualises the implications of any
future drop to 25%.
HISTORICAL DRIVERS OF HASHRATE
Kazakhstani Regulation
Kazakhstan has seen the largest increase in mining activity as a result
of the ban. Whereas in June 2021 it contributed a respectable 10.6%
of global hashrate, that number has now increased to 21.9% (as of the
end of August). Like China, its cheap energy is particularly attractive
to miners, and it is a cheaper destination to relocate to than the US
or Canada.
However, miners have not been welcomed by all in the country. Their
exorbitant drain on the national energy supply has caused the
shutdown of at least three major power plants in the north of the
country. In response, the state grid operator has begun rationing
energy to the 50 miners registered in the country, meaning that they
are the first to lose power when the grid is strained.
A tax on cryptocurrency mining profits, planned to be introduced in
January 2022, only darkens the prognosis for operations in
Kazakhstan. It is difficult to see the current growth continuing, and
we can expect further movement out of the region towards Russia
and the US.
HISTORICAL DRIVERS OF HASHRATE
Ethereum's switch to Proof of Stake
Ethereum, the second biggest cryptocurrency by market cap,
currently uses a very similar mechanism to create new blocks and
mint new currency units. As a result, ETH’s mining network is subject
to many of the same driving factors.
In March 2022, Ethereum is expected to complete its switch to a
Proof of Stake system, meaning that new blocks will no longer
require expensive computational work to be processed. Tickets to the
lottery will instead be distributed according to the number of coins
that miners choose to lock up (or "stake"). Without any mining, there
are no mining rewards. Without mining rewards, what will miners
choose to do?
One option for miners is to pivot to mining a different
cryptocurrency, but most profitable equipment on the Ethereum
network is specialised Etheruem’s number-searching algorithms
(called hash functions). A simple switch to mining BTC (or any other
cryptocurrency) would not prove profitable, and would at the very
least see a reduction in mining efficiency. Instead, redundant ETH
miners might choose to sell their equipment at a loss to buy bitcoin
mining equipment, or even exit the market completely.
LINKING HASHRATE TO PRICE
LINKING HASHRATE
TO PRICE
Same day changes in price and hashrate
Rolling correlation estimate
A rolling estimate of correlation between same day BTC returns and
hashrate changes shows scant evidence for a significant relationship
for most of the period considered.
Figure 4.1 Low historical correlation between daily BTC returns and daily hashrate
increases, using a lookback of 80 days. There is a period of coupling towards the end of
2021 as both increase.
However, towards the end of the period (namely from Jan 2021
onwards) we see an increasingly positive correlation between the
two, perhaps as a result of the concurrent bull run and recovery in
hashrate from the drop caused by the Chinese exodus. This could be
the result of increasing prices attracting miners to the network, eager
to take their share of the growing value of mining rewards (in
nominal terms). The historically low correlation suggests that this
coupling could be due to a common factor driving both BTC price
and hashrate upwards, but it is difficult to identify causality from this
information alone.
LINKING HASHRATE TO PRICE
Is there evidence of a causal relationship?
Hashrate leading price
it is not easy to find an answer to this question that we are confident
is significant and not merely a spurious result. When we evaluate the
correlation between all hashrate changes and BTC returns, there is
little evidence of a significant relationship between the two through
time.
However, by limiting this analysis to days where hashrate rose (and
excluding days where it dropped) we find some evidence of a
persistently negative correlation between hashrate increases and
daily BTC returns 14 days later. This means that two weeks after an
increase in hashrate, we observe a decrease in the BTC price slightly
more than we would expect if there was no relationship.
Figure 4.2 Correlation (using a lookback of 85 days) between hashrate increases and BTC
prices two weeks later. Overlaid in black is the historical difficulty series.
This may be more interpretable. Average fortnightly hashrate
increases lead to a difficulty increase 2 weeks later, and a
corresponding rise in running costs for miners. Miners may be selling
long-held Bitcoin reserves to hedge their operation and cover the
larger costs, causing selling pressure on the market and driving the
price down. The same reasoning does not apply to drops in hashrate,
as a lower difficulty would not necessarily prompt miners to create
buying pressure on the market.
LINKING HASHRATE TO PRICE
Is there evidence of a causal relationship?
Price leading hash
Conversely, one would expect hashrate to follow bitcoin price in the
long term. Both values can be used as metrics that track the
adoption of the currency for its primary purpose - transactions and a
store of value. Additionally, the rapidly increasing price has outpaced
any halving events, maintaining the profitability of mining. As long as
there are excess profits to be made, miners are sure to join the
network or expand their operations.
Figure 4.3 Rolling correlation (using a lookback of 30 weeks of weekly data) of weekly
BTC returns with hashrate changes three weeks later.
Figure 4.3 provides some evidence for this behaviour despite some
short-term decouplings. Since 2018, changes in BTC's price loosely
correspond to a change in hashrate in the same direction 3 months
later. We would also expect to see a similar relationship at larger
time scales, as miners join the network at different stages of a bull
run.
The relationship between hashrate and BTC's price appears two-fold,
then. In the short term, miners' hedging activities in reaction to
larger hashrates can drive BTC's price down. In the longer term, we
expect the increasingly lucrative mining rewards to attract new
miners (and their hashrate) to the network. Neither is a perfect
indicator of the other, a result that we would expect from the
complex nature of the market.
27 Old Gloucester Street
London WC1N 3AX
info@blockscholes.com
