Bitcoin Hash Rate Flash Crash Addressed By Researchers Questions were raised over what was labelled a “flash crash” in Bitcoin hash rate last week. Data from Blockchain.com shows the hash rate dropping from 98 EH/s to 67 EH/s. Other data sources such as Coin.Dance report an even more pronounced drop with the figure moving from 98 EH/s to 58 EH/s over the course of a day. With hash rate used to secure proof-of-work networks, the drop has sparked concern among some analysts. Larry Cermak, Director of Research for The Block, posted a tweet looking for potential explanations for the drop. Does anyone know why did Bitcoin's hash rate drop all the way to 57.7 PH/s yesterday? That's a 52% drop in just the last week. It has not kind of recovered to about 80 PH/s but still significantly down. Genuinely curious in what could have caused such a dramatic decline. pic.twitter.com/ATAxky7qlk— Larry Cermak (@lawmaster) September 24, 2019 Emin Gün Sirer called for businesses to increase the number of confirmations they require for payment in lights of the drops. Simple exercise: suppose you see 51% of the hashrate disappear. What should you do? Stop accepting payments What about 49%? Number of confirmations should be INFINITY. 48%? Thousands of blocks. 40-something? Hundreds.You get the idea. 6 isn't a magic number.— Emin Gün Sirer (@el33th4xor) September 24, 2019 Hash Rate Is Unknown Several researchers have stepped in to point out that the drop may not be as concerning as it first appears. Firstly, hash rate is an unknown metric. The metric displayed on websites such as BTC.com, Blockchain.com, and bitinfocharts.com is inferred from both the interval between blocks and the difficulty level. If block intervals increase, it is inferred that the hash rate has dropped and vice-versa. Block times are Poisson distributed which means that we know the average block time is ten minutes but the interval between one block and the next is random. It could be one minute but it also it could be over one hour. On average, it will be ten minutes but a series of blocks found in quick succession will infer a high hash rate whereas longer-than usual intervals between blocks will infer a lower hash rate. This is why when we observe raw values for inferred hash rate figures, the graph consists of large spikes up and down to reflect the variance in block time intervals. The seven-day moving average is commonly used instead as an estimate of how much hash rate is deployed. “herein lies the essence of the problem. If the measurement interval is too short, the estimate becomes vulnerable to the inherent variance in block times.”Chris Bendiksen, Head of Research at CoinShares Just Variance? Chris Bendiksen has noted the drop to be likely just variance. A number of analysts did a deeper dive on the numbers. Nic Carter, co-founder of CoinMetrics, graphed the time interval of the 144-block moving average with block height. He found that there has only been a slight increase in the 144-moving average of block time. This is the last week's worth of blocks ordered by interblock time. The orange line is the 144-block moving average. That little bump is what people are panicking about pic.twitter.com/XYohIOWs85— nic carter (@nic__carter) September 25, 2019 James Prestwich noted the outsized impact of long-block intervals compared to short block intervals when estimating the hash rate. Hash rate has since recovered indicating that the most likely explanation for the observed drop in hash rate was a number of statistical outlier blocks which had highly unlikely long interval times. A number of these outliers can be observed on the graph which Nic Carter shared with one block interval being over 70 minutes. The "hashrate crash" isn't real. Long block intervals are vastly overrepresented in fixed-period sampling This is because the interval is long. A 18-minute block has approximately the same likelihood as a 2-minute block, but affects time-based sampling 9x more (!!!)— James Prestwich (@_prestwich) September 24, 2019 It is also worth noting that rainy season in Sichuan province is drawing towards an end. Sichuan’s rainy season takes place from April to October each year resulting in abundant energy consumption. As rainy season draws to an end, mining farms operating in the region have informed MinerUpdate that energy production drops by roughly 66% and energy prices rise as a result of this. With Sichuan estimated to account for roughly 50% of Bitcoin hash rate, it is feasible that miners coming offline in the region can contribute to a reduction in the hash rate. However, when it comes to hash rate flash crashes, variance in block intervals is the highly likely answer.