Wednesday, June 12, 2019

Is bitcoin getting less volatile?

I'm going to make the following claim. The price of bitcoin is inherently volatile. Even if bitcoin gets bigger, its core level of volatility is never going to fall.

Bitcoin's hyperactive price movements prevent it from becoming a popular medium of exchange. Merchants are too afraid to accept bitcoins. If they do, they could experience large losses. Consumers who hold bitcoins are loath to spend them. Many of these hodlers are trying to change their financial lives by getting exposure to the very same roller-coaster ride that merchants are trying to avoid. If they use their bitcoin to buy stuff, they risk losing out on the opportunity for life-changing returns.

Why is bitcoin's high volatility intrinsic to its nature? Bitcoin is a rare example of a pure Keynesian beauty contest. Players in a beauty contest gamble on what John Maynard Keynes described as what "average opinion expects the average opinion to be." No matter how big the game gets, the best collective guess—bitcoin's current market price—will always by hyper-volatile.

By contrast, other assets like stocks, gold, commodities, and banknotes have a fundamental value that helps to anchor price. This ensures that their prices can't travel very far as time passes.

But the standard deviation is falling!

In response to the claim I've just made, people have given me a version of the following: as bitcoin gets bigger and more popular, its volatility will inevitably fall. This eventual stabilization is one of the assumptions at the core of Vijay Boyapati's bubble theory of bitcoin. Bitcoin guru Andreas Antonopolous has also adopted this viewpoint, noting that "volatility really is an expression of size."

Manuel Polavieja provides evidence for this view by tweeting a chart of the 365-day standard deviation of bitcoin daily price changes.

The general slope of the curve in the chart seems to be declining, the inevitable conclusion being that bitcoin's price isn't intrinsically frenetic. As bitcoin has become more popular, its volatility has been retreating.

Sure, but bitcoin's median absolute deviation isn't falling

Manuel has chosen to illustrate bitcoin's price dispersion with its standard deviation. But the standard deviation of an asset's daily price change isn't the only way to get a feel for its volatility. There are other measures of dispersion  that can flesh out the picture, particularly for distributions that are characterized by extremely large outliers.

One problem with standard deviation is that it amplifies the influence of extreme price changes. The calculation for standard deviation squares each day's difference from the mean day's return. By their nature, outliers will boast the largest differences. Squaring them has the effect of causing the extremes to have a disproportionate influence on the final score. The calculation further promotes outliers by taking the average of the squared deviations from the mean. But in distributions such as bitcoin daily returns, the average return will always be skewed by a few crazy daily fluctuations.

Median absolute deviation is one way to reduce the influence of outliers. It calculates the differences from the median daily return, not the mean. And rather than squaring the differences, and thus amplifying them, the calculation simply takes their absolute value (i.e. it gets rid of all negative amounts). It then locates the median of these absolute differences. The advantage of using the median difference is that—unlike standard deviation, which locates the average difference—the median can't be influenced by insane values.

Below I've recreated Manuel's chart of bitcoin's 365-day standard deviation of daily returns and overlaid it with bitcoin's 365-day median absolute deviation of daily returns. The contrast is quite striking.

Standard deviation of bitcoin returns, the blue line, has been falling since 2011. But median absolute deviation of bitcoin returns, the green line, has stayed constant. What I believe is happening here is that the craziness of bitcoin's outlier days have been steadily falling over time, and thus the standard deviation has been declining. But a typical day in the life of bitcoin—i.e. the usual price volatility experienced by bitcoin holders, its non-outliers—hasn't changed since bitcoin's inception. A regular day, as captured by the median absolute deviation, is about as frenetic today as it was back in when bitcoin was a fraction the size.

What is happening at the ends of the distribution?

We can get an even better feel for the dispersion of bitcoin's returns by splitting them into quartiles and percentiles.

Let's look at the blue line first, the 25th percentile (or first quartile). This measure gives us a feel for what a lethargic day is like in bitcoin-land. Out of a sample of 365 days of bitcoin returns, 25% of them will fall below the blue line. If bitcoin is indeed getting more stable, we'd expect the 25% most lethargic bitcoin days to be getting even more lethargic. But this isn't the case. Rather than falling, the blue trend line is flat (and even slopes up ever so slightly). It seems that lethargic days are getting a bit less lethargic as time passes.

The median (already discussed above) shows a similar pattern. The middle-most day's return shows no sign of slackening, despite bitcoin's incredible growth over the last decade.  

Let's look at the top two lines. 25% of all bitcoin daily price changes are in excess of the red line, the 75th-percentile. Unlike the median, this line has been steadily falling. This means that the 25% most frenetic bitcoin days have been getting a little less frenetic. The purple line, the 90th percentile, shows an even steeper decline. The 10% craziest bitcoin days are quickly becoming less crazy.  

The interpretation of this chart seem pretty clear. The typical bitcoin trading day is not getting more subdued. It's the outliers, those outside of the 90th percentile, that have mellowed. The softening of bitcoin's extreme price fluctuations, the purple line, explains why bitcoin's standard deviation has been trending downwards. But if we only focus on standard deviation, we'll fail to see that the typical day—i.e. the median day—is just as hyperactive as before.

What about Netflix?

It's always nice to get some context by looking at how a similar data series behaves. I've chosen Netflix. Like bitcoin, Netflix has gone from nothing to billions of dollars in market capitalization and millions of users in the space of a few short years.

As Netflix has grown, its median absolute deviation and its standard deviation have softened. So both Netflix's outliers and its regular days have been tempered over time. Compare this to bitcoin, where the typical day continues to be just as frenetic as before.

I believe that the contrast between the two assets can be explained by the fact that at its core, bitcoin is a Keynesian beauty contest. Netflix isn't. As Netflix has grown and its earnings have become more certain, Netflix's typical day-to-day price fluctuations (as captured by its median absolute deviation) have softened. But the failure of a prototypical bitcoin day to stabilize, even as the asset grows, can be explained by bitcoin's basic lack of fundamentals. Its price is permanently anchorless.  

Intrinsic vs extrinsic price fluctuations

So why has bitcoin's typical volatility stayed constant while its extremes have become more tame? If bitcoin is a Keynesian beauty contest, shouldn't both its typical volatility and extreme volatility have stayed high and constant?

Let's assume that there are two types of bitcoin price fluctuations. Intrinsic price changes are due to the nature of bitcoin itself. Extrinsic changes occur because of malfunctions in the unregulated third-parties (wallets, exchanges, investment products) that have been built around bitcoin. Mature assets like stocks and bonds that trade on well-developed and regulated market infrastructure tend not to suffer from extrinsic volatility.

Over the years, third-party catastrophes have accounted for some of the largest shocks to the bitcoin price. When Mt. Gox failed in 2014 it caused massive fluctuations in the price of bitcoin. But this was extrinsic to bitcoin, not intrinsic. It had nothing to do with bitcoin itself, but a security breach at Mt. Gox.

If you've been around as long as I have, you'll remember Pirate's Bitcoin Savings & Trust—a ponzi scheme that caught up many in the bitcoin community. When BST collapsed in 2012, it dragged the price of bitcoin down with it. Again, this was an extrinsic price fluctuation, not an intrinsic one.

The infrastructure surrounding bitcoin has grown up since those early days. Mt. Gox blow-ups and BST scams just aren't as prevalent as they used to be. There are enough robust exchanges now that the collapse of any single one won't do significant damage to bitcoin's price. And so bitcoin's price outliers have gotten less extreme. The declining influence of third-party infrastructure on bitcoin's price is reflected in bitcoin's falling standard deviation. As the infrastructure surrounding bitcoin reaches the same calibre as the infrastructure that serves more traditional assets, bitcoin's extrinsic price fluctuations will cease to occur. At that point the steady decline in bitcoin's standard deviation will have petered out.

Median absolute volatility screens out the effects of the Mt. Goxes and BSTs. And so it is the best measure for capturing bitcoin's intrinsic volatility. Think of this as the base level of volatility that emerges as people try to guess what average opinion expects average opinion to be. And as I pointed out earlier, this sort of volatility has stayed constant over many years. A Keynesian beauty contest is manic by nature, it isn't going to mellow out with time.


In sum, on a typical day bitcoin is about as volatile in 2019 (at a market cap of +$100 billion) as it was in 2013 (when its market cap was at $1 billion back in 2013). Which would seem to indicate that if and when it becomes "huge" (i.e. $10 trillion), it will continue to be just as volatile as it is now.

New recruits are being introduced to bitcoin on the premise that they are buying into tomorrow's global money at a bargain price. But shouldn't they be warned that they are playing a new sort of financial contest? Sure, bitcoin can be used for payments. But the underlying beauty contest nature of bitcoin will always interfere with its payments functionality. Which means that usage of bitcoin for paying is likely to be confined to a small niche of enthusiasts who are willing  to put up with these nuisances, and the de-banked, who have no choice. Bitcoin is risky, play responsible.


  1. Hi JP, thank you for your time to write a response.

    I see we agree that Bitcoin has come less volatile at its extremes. Correct?

    Regarding median daily return, I fail to see how that measure is "frenetic". 2% is defenitely not low, but I don´t think is crazy high either. If I am not interpreting it wrongly, your Netflix example is showing about the same measure than Bitcoin. I am curious about that same measure for a non top forex pair such as CAD/BRL

    1. Yep, less volatile at the extremes.

      Since 2003, Netflix's median absolute deviation over a 365-day rolling window has averaged 0.0164. Bitcoin's has averaged 0.0187 since 2011. So Netflix's 'core' volatility is somewhat lower, although not as low as gold's at 0.0056.

      If I get a chance I'll do CAD/BRL. But if I had to guess I'd say somewhere between gold and Netflix.

  2. Interesting analysis. Would be great to see the MAD only for downside moves. Nobody minds the upside risk. It's the downside vola that drives people mad.

  3. well explained. Thanks for sharing your insights

  4. U have been sold the libertarian view of Bitcoin and so with ur proper economic reasoning such an axiom leads u to believe Bitcoin has little value as a medium of exchange.

    Consider as bitcoins market cap rises to the level of gold's and the highest value transactors push the lower level transactors off of on chain transactions.

    The cost to settle between central banks and large institutions will be very attractive for them.

    They won't be speculating. And that doesn't represent a Keynesian beauty contest scenario.

    The game theory for this is all fleshed out by the master of it and he has a far broader knowledge of economics than u could ever hope to.

    U should expect the volitility of Bitcoin to asymptotically wane over time (not to perfect stability) and that central banker money will eventually be pegged to Bitcoin (and using moving averages to increase the observed stability).

    Once u consider this u will not believe the genious that Satoshi was. And that's the difficulty believing someone could have that amount of foresight.