The Half-Time Show: The State of Bitcoin Network Security After the Halving
Block rewards are currently the primary source of revenue for miners. A reduction in this reward due to the halving causes some miners to exit the network. In the short term the sudden drop-off of miners can potentially leave the network more exposed to security threats like 51% attacks.
The distribution of hashpower among different types of mining hardware also has an impact on the network’s security properties. The availability of old hardware on secondary markets poses a potential threat to the network.
Novel techniques involving nonce distributions allow us to numerically estimate the amount of hashpower provided by certain types of hardware, including older hardware like S7s and S9s.
It now appears that a significant number of formerly-offline S9s have been turned back on. Currently, the Antminer S9 family of miners is responsible for about 32% of Bitcoin’s hashpower.
While the amount of hashpower that could be created by offline S9s is nowhere near enough to 51% attack the network, the change to the network’s security dynamics caused by their presence is significant.
Halving a Good Time
On May 11th, for the third time in Bitcoin’s history, the amount of new coins issued per block was cut in half. This event, known as the halving, occurs every 210,000 blocks, or approximately four years, until issuance is eventually rounded down to zero.
In the most recent halving, the block reward was reduced from 12.5 to 6.25 BTC. The period leading up to the halving was marked by pronounced market volatility, which has somewhat subsided since the reduction occurred. The impacts of the event on the network’s security are nuanced.
Block rewards are currently the primary source of revenue for miners, so a reduction in this reward causes some miners to exit the network. With less revenue to go around, margins are tightened and less efficient miners may suddenly find themselves operating at a loss. In the long run, these miners are typically replaced by more efficient operations as the market rebalances. In the short term, however, the sudden drop-off of miners can leave the network more exposed to security threats like 51% attacks.
State of the Network Issue 44 reasoned about the impacts of the halving on miner economics from first principles. This piece will address a similar topic, focusing on the implications of the halving on security and the economics of running old mining hardware. In the process, we’ll use nonce data to estimate the prevalence of certain types of hardware on the network today, and discuss how the presence of old hardware impacts Bitcoin’s security model.
To Halve and to Secure
Bitcoin miners are compensated through both block rewards, which are directly affected by the halving, and transaction fees, which are not. Transaction fees are generally a function of demand for block space, and therefore tend to spike during periods of congestion and high traffic.
Currently, fees make up a small percentage of total miner revenue. Over the past five years, only about 4.4% of miner revenue has been generated from fees.
As Bitcoin’s block reward continues to halve approximately every four years, transaction fees will need to increase in order to sufficiently incentivize miners to secure the chain. Since slightly before the halving, fees have surged to make up about 17% of total miner revenue. While this effect has been intensified by the reduction in the block reward, transaction fees themselves have also increased to levels not seen in almost a year.
This increase in fees may have been amplified by the reduction in hash rate that has taken place since the halving. This reduction, in turn, is caused by less efficient miners leaving the network. The drop in hashpower has increased the time between blocks, therefore reducing the amount of available block space.
To reduce the variance of their payouts, miners often aggregate into mining pools, which are loose coalitions of miners that are organized by an operator and share revenue, typically according to hashpower contribution. Individual miners often switch between pools depending on several factors, most notably fees charged by operators.
As long as no single dishonest entity controls more than half of Bitcoin’s hashpower, the network is secure. In a process known as a 51% attack, an adversary who controls more than half the network’s hashpower can censor transactions and perform double-spends.
The minimum number of pools who would need to collude in order to 51% attack the network is known as the Nakamoto coefficient. At this time, the top 4 pools would need to collude in order to 51% attack the network. This number has generally gone up over the course of Bitcoin’s history, indicating a steady increase in decentralization.
The Nakamoto coefficient is not a perfect metric, and makes Bitcoin seem significantly more centralized than it is. Individual miners, who face large up-front expenses on capital like hardware, are disincentivized from attacking the network. These miners would likely defect from a maliciously-operated pool.
Still, pools select the blocks that their constituents mine, and barring defection exercise a certain degree of control over them. It also may be possible for an attacker to censor transactions with less than half of the network’s hashpower through techniques like feather-forking. It’s therefore useful to have a pessimistic metric like the Nakamoto coefficient to quantify the degree of centralization among miners.
Stratum V2, an implementation of Betterhash with modifications to the original protocol, suggests letting individual miners select the blocks that they will mine, rather than the pool operators doing so. This potential improvement to the way pools are operated would put more power in the hands of individual miners, further decentralizing the network.
In addition to the distribution of hashpower among different entities, the distribution of hashpower among different types of mining hardware has a significant impact on the network’s security properties.
To add a block to the blockchain, Bitcoin miners attempt to find a nonce, or arbitrary value, that causes the block header to hash to below a certain target. The rate at which these hashes are computed and verified is known as hashpower, and a nonce satisfying this condition is called a golden nonce. Golden nonces are theoretically uniformly distributed throughout the space of potential nonces and valid blocks. The threshold that the hash of the block header must satisfy is set by the network’s difficulty parameter, which is periodically adjusted according to the rate at which blocks have been accepted to the chain.
While mining was initially performed with CPUs, the process was parallelized and made more efficient early-on through the adoption of GPUs. Today, almost all mining is performed using mining rigs that contain specialized chips known as ASICS. These devices are significantly faster, better at parallelization, and more energy-efficient than other hardware.
Purchasing these devices requires a large up-front capital expenditure. This benefits the security of the network by requiring miners to lock up capital in an illiquid asset and therefore disincentivizing them from acting maliciously.
The presence of old mining hardware changes this security model, since it tends to require smaller up-front investment at the expense of higher operating costs. While there are practical and logistical barriers to starting a mining farm aside from the cost of hardware, the presence of old hardware allows entry into the market with significantly reduced capital expenditure.
Due to secrecy in the mining industry, it’s generally difficult to discern which types of mining hardware are being used to secure the network. However, novel techniques allow us to numerically estimate the amount of hashpower provided by certain types of hardware.
Bitcoin’s nonce distribution offers hints at the types of hardware being used to mine on the network. By combining this data with information on the prices of hardware on secondary markets, we can quantify the degree of risk posed by the existence of inexpensive, slightly dated hardware. For a detailed explanation of our analysis of nonce distributions, see our series “The Signal and the Nonce” Part 1 and Part 2.
Since golden nonces are uniformly distributed throughout the nonce spaces of all potential blocks, we’d expect a plot of the winning nonces over time to look like random static. Bitcoin’s nonce distribution doesn’t.
Near the left-hand side of the plot, nonces are concentrated in the lower ranges of the distribution. This is a result of a sampling technique used by miners in the CPU-mining era, which involved iteratively testing values starting from zero and incrementing upward.
Bitcoin’s nonce distribution also has a characteristic streaked pattern that first appeared in late 2015, and has recently begun to fade. The striations in question start out broad, and then suddenly narrow out, before gradually fading away. There are four distinct streaks, each of which can be specified in terms of its narrow and wide bands.
These streaks were noted in State of the Network Issue 23, and their source was identified in State of the Network Issue 45. The striations were found to come from the way in which nonces are sampled by the Bitmain Antminer S7 and S9 mining rig lines. Each of these rigs was at one point the dominant miner on the network, with the S9 having recently been supplanted in this role by the Antminer S17.
The wide and narrow bands are attributable to the sampling techniques used by the S7 and S9 families, respectively. We can use this knowledge to numerically estimate the proportion of the network’s hashpower provided by S7s and S9s.
According to these numerical estimates, the proportion of hash rate provided by S7s and related hardware peaked in May of 2016 at about 61%. Today, S7s are not responsible for a significant portion of hashpower. The proportion of blocks mined by S9s and related hardware peaked in May of 2018 at about 78%. Today, about 32% of blocks are produced by S9s.
These estimates are based on the assumption that S7s and S9s do not sample within their respective excluded bands and that all miners sample uniformly outside of any excluded regions. These conditions are violated in the CPU-mining era, but appear to hold from the GPU-mining period onward. The excluded regions are determined manually, and estimates are corrected for any extrapolated values outside of the 0-100% range and normalized.
The estimates are subject to a certain amount of noise, which is visible toward the left of the graph. The small bump in the estimated proportion of S9 hashpower in 2015 could be due to noise, or may be a sign of something else such as the testing of experimental hardware.
These figures are consistent with other estimates of the hashpower output of these types of hardware. A CoinShares report on Bitcoin mining released in December 2019 estimated that S9s made up about two thirds of the hardware in their equivalence class. In March, the founder of Beijing-based Spark Capital estimated that S9s provided 20 to 25 percent of Bitcoin’s hashpower.
The proportion of hashpower provided by each type of mining rig provides further perspective on the threat posed by old hardware.
The most salient insight from this plot is the exponential growth in the hashpower securing the network.
The estimated amount of hashpower provided by S9s reached its peak in August of 2019, when they generated about 52 exahashes per second. In February of 2020, the estimated hashpower generated by S9s reached the bottom of a valley at about 21 exahashes per second.
It now appears that a significant number of formerly-offline S9s have been turned back on, likely as a result of a recent appreciation in the price of Bitcoin. This hardware now computes about 37 exahashes per second.
Due to rapidly changing market conditions, this effect may not be sustained. However, it illustrates the degree to which mining with old hardware may be viable given favorable conditions, and the ease with which this less-expensive hardware can be deployed.
S9s are being sold on secondary markets for a fraction of their retail price. The miners can be purchased for between $20 and $80, compared to an original price of about $3000. Given today’s economic climate and the inexpensive electricity brought on by China’s rainy season, miners have found it possible to operate these devices profitably.
While the amount of hashpower that could be created by offline S9s is nowhere near enough to 51% attack Bitcoin, the change to the network’s security dynamics caused by their presence is significant. This effect may be felt more acutely by other platforms that use Bitcoin’s proof of work algorithm, including Bitcoin Cash and Bitcoin SV, which are currently secured by about 2.5 and 1.8 exahashes per second of computational power, respectively.
The Epitaph of the Third Epoch
In anticipation of the halving and on optimism related to increased institutional interest, the price of Bitcoin increased dramatically before giving up some of its gains. Since the halving, volatility has subsided somewhat, but price has continued to trend upward. The halving has also accelerated an increase in transaction fees and precipitated a slight drop in hashpower.
Halving-related sentiment will continue to impact the market, and the halving itself will continue the test of whether Bitcoin can successfully transition to a model where miners’ revenue is predominantly based on fees. The long-term effects of this event remain to be seen, but its impact on the economics of mining and the market as a whole are already pronounced.
Network Data Insights
Bitcoin (BTC) and Ethereum (ETH) transaction fees continue to climb, despite a relatively flat week for most other metrics.
Bitcoin’s median transaction fee reached $2.88 on May 14th, its highest level since June 2019. Similarly, Ethereum’s median transaction fee reached $0.25 on May 14th, its highest level since August 2018. Median transaction fees tend to surge when blocks are relatively full. The causes for this surge are explored in today’s Network Highlights section.
Bitcoin’s hash rate has dropped to 81.66 TH/s following the halving, about a 40% drop from pre-halving highs. As noted in this week’s Weekly Feature, this hash rate drop-off is to be expected as less efficient miners exit the network. It will likely recover after a period of churn where efficient miners replace less efficient operations. However, it is unclear exactly how long this turnover period will last.
As a result of the hash rate drop, the average interval between Bitcoin blocks has risen to its highest levels since late 2018 (excluding the period around March 12th 2020, where block interval shot up due to the sudden drop in Bitcoin price and subsequent hash rate drop-off).
Since there are less blocks being produced there is more competition for block space, which has led to the increase in transaction fees. Paying a higher transaction fee leads to a higher chance that miners will include the transaction in a block. So median fees tend to surge during periods where block space is at a premium.
The decrease in the overall number of blocks has also led to an increase in the size of each individual block. Bitcoin mean block size reached a new all-time high of 1.32 MB on May 17th.
Ethereum median transaction fees have also shown signs of growth since the Bitcoin halving. However, Bitcoin Cash (BCH), Bitcoin SV (BSV), Ripple (XRP), and Litecoin (LTC) median fees have not shown any significant increases.
Market Data Insights
This marks the third consecutive week that Bitcoin has outperformed other cryptoassets and forms a trend that cannot be ignored. Although it cannot be ruled out that this trend is simply a byproduct of the random walk of prices, one plausible explanation is that Bitcoin’s store of value properties are increasingly needed in today’s market environment. And as we observe the emergence of adoption by institutional investors, Bitcoin is the logical first choice as the gateway asset that may lead to the eventual adoption of cryptoassets as a distinct asset class.
A Call Option on Inflation
There is an emerging narrative that Bitcoin is needed in a market environment of unparalleled monetary and fiscal policy by global central banks and governments. To examine this phenomenon, we show the year-over-year change in the Fed’s balance sheet highlighting the speed and magnitude of the Fed’s reaction to COVID-19. The Fed’s policy response has already exceeded the balance sheet expansions seen in the three previous quantitative easing programs following the financial crisis.
Bitcoin’s strong returns lately and the renewed interest in Bitcoin as a store of value in a rising inflation environment is remarkable because all indicators are still showing that inflation is not a problem, despite the strong growth in money and credit.
The shutdown of large swathes of the economy represents a demand shock which is deflationary by nature -- energy prices reaching unprecedented lows are a prime example. Although monetary and fiscal policies are effective at getting money into the hands of U.S. businesses and households, the velocity of money has simultaneously declined. The most recent print for the U.S’s core inflation, which excludes food and energy items, fell 0.4% over the previous month, the largest monthly decline in the history of the series, according to the Bureau of Labor Statistics.
Similarly, inflation expectations either from survey-based indicators, such as the University of Michigan’s Survey of Consumers, or market-based indicators, such as 5-year, 5-year forward inflation expectations derived from TIPS, are well-anchored.
How can we reconcile the fear that Bitcoin will be needed in a rising inflation environment with the data that shows that realized inflation in the short-term is non-existent and inflation expectations over the medium to long-term are low? According to the Fed’s Survey of Consumer Expectations, median expectations for inflation have not meaningfully changed in response to the pandemic, but the level of uncertainty and disagreement across respondents have seen unprecedented increases.
One framework to bring clarity to this question is to view Bitcoin as a call option on inflation and to examine its greeks: the sensitivities of the price of the option based on the parameters of the underlying. What we are seeing now is an increase in the implied volatility of future inflation even if median expectations for the future level of inflation remain unchanged.
Standard option price theory indicates that increases in implied volatility of the underlying should lead to an increase in the price of a call option. Therefore, we can potentially attribute the recent increase in price of Bitcoin to the increase in implied volatility of inflation rather than the increase in the expected level of inflation.
CM Bletchley Indexes (CMBI) Insights
All CMBI and Bletchley Indexes had very good weeks, ending between 5% and 15% higher than the previous week’s close. Following the biggest news of the week, the Bitcoin Halving, the CMBI Bitcoin Index was the strongest performer, returning 14.4%. The CMBI Ethereum Index also had a strong week, closing 11.9% higher. Despite these two strong performances though, it was the Bletchley 40, small-caps, that was the best of the market cap weighted indexes, closing the week 12.1% up.
Coin Metrics Updates
This week’s updates from the Coin Metrics team:
Coin Metrics is hiring! Please check out our Careers page to view the openings.
As always, if you have any feedback or requests, don’t hesitate to reach out at firstname.lastname@example.org.
Subscribe and Past Issues
Coin Metrics’ State of the Network, is an unbiased, weekly view of the crypto market informed by our own network (on-chain) and market data.
Check out the Coin Metrics Blog for more in depth research and analysis.