Bitcoin’s seasonal affective disorder

In the past couple of years Bitcoin has been settling into a regular seasonal growth pattern, in which growth stops during the summer and we add users during the autumn/winter months (in the northern hemisphere).

It’s a little hard to see on this 2-year graph because Bitcoin traffic is easily affected by press cycles and the accompanying bubbles. When the price is skyrocketing a lot of coins move in/out of exchanges and this results in a volume spike.

However the pattern has been getting clearer with time. After the late 2013 bubble traffic grew steadily from January 2014 to around March, then fell a bit and then didn’t do very much over the summer. Starting around August 2014 traffic started to grow steadily again throughout the autumn and winter months with a sharp and obvious drop for Christmas, before steadying out again around March at around 100,000 transactions per day (1.15 txns/sec average)

Assuming a quiet summer I suspect we’ll see a slight fall as we head into April/May and organic growth will resume around August.

This pattern should not surprise or worry us. The reason I am predicting this so confidently is that the “summer slump” is, in fact, entirely normal for large internet services. I’ve seen it before and so seeing it crop up once again in Bitcoin traffic graphs isn’t any great shock.

Interestingly,’s estimate of volume in dollars hasn’t changed much. It’s an open question what this means: is traffic growing because of a static userbase changing how it uses transactions, or does it represent a real increase in number of users?

Given the same seasonal pattern is seen on other web services, and given that the unique addresses used graph shows a similar pattern, I suspect this is real usage growth. The static dollar-denominated volume might reflect either an issue in how that graph is calculated, or a shift in the demographics of bitcoin users away from the “nouveau riche” who can move millions of dollars in minutes and towards smaller, newer users who have less individual impact. It’s also worth remembering that these graphs are looking at money movements, which is not the same thing as actual trade.

It is sometimes said that the Bitcoin network tops out at 7 transactions per second. That’s not correct anymore, and this is my fault. The 7 tx/sec calculation originally comes from an article I wrote on the topic of scalability in 2011. When I wrote that paragraph in March the “sendmany” command was still a month away, so transactions were almost all small. Since then the size and complexity of our transactions has gone up, so the true limit is probably more like 3 transactions per second. That would be about 260,000 transactions per day sustained, except that in practice it’s very difficult to redline a distributed system without it encountering serious problems before long. So the true limit is likely lower.

(Note: that’s just talking about block chain size. There may be other scaling limits we aren’t quite aware of yet)

In the last growth season we slightly-less-than doubled average block size. It seems likely that after the next winter growth season, we will be straining at the seams. As it will take time for people to upgrade once the required hard fork is scheduled, it seems prudent to aim for having patches tested, merged and ideally about to ship by the end of the summer. That then gives plenty of months in which miners and merchants can upgrade before the upgrade locks in, with some headroom for unexpected traffic spikes as well.

Show your support

Clapping shows how much you appreciated Mike Hearn’s story.