Chart doctor: the mysterious music of the yield curve

Chart doctor: the mysterious music of the yield curve


These days, you can make
music from anything. I’m going to show you how
to take 100,000 numbers and turn them into
something musical. This is the mysterious
music of the yield curve. It’s a chart that
analysts use frequently to predict what might
happen to the economy. And it’s actually
quite a simple chart. The y axis just shows
the interest rate. And the x axis shows us for each
bond what its maturity date is. So on the left-hand side, it
can either be very short – one month, three
months, six months. On the right-hand side,
it can be very long-dated, up to 30 years. And this yield
curve here from 1992 is fairly typical of
what you would expect from a regular yield
curve in that it’s the longer-dated bonds that
generate higher returns. The thing is with
the yield curve is that these rates
change on a daily basis. And so the shape of the
yield curve can change. So one of the things that
can happen is it can flatten. And so this is the
yield curve from 1989. And you can see that
actually the yields are very, very uniform. It’s a very, very
flat yield curve. And this is a sign
that people are starting to express
uncertainty about what might happen to the economy. And in extreme cases, the yield
curve can invert like this. So on the left-hand
side now, you’ll see it’s the shorter bonds
that are generating the higher yields. And these inversions,
many analysts consider them very,
very important. Every recession in the
US since world war two has been preceded by
one of these inversions. But because the yield curve
changes on a daily basis, tracking it over time
can be difficult. And one of the
things that we can do is we can use animation
to show us what’s happening on a daily basis. So here what I’m
going to do is I’m going to set the yield
curve to animate from 1979. And you can see that five days
animating per second here. So we get a sense of how the
yield curve changes over time. If I speed that right up, you
can see that actually the yield curve is very, very
active, particularly at the shorter
end of the yields. There’s a lot of movement. With the yield curve
animating so quickly, it can be difficult to remember
how it’s travelled over time. So one of the other
things that we can do is introduce some
ghosting of outlines from these peak inversions
so we can remind ourselves of where the yield curve has
been at different points. So you can see
here, October 1979 was an inversion three
months before a recession, and similarly in 1980. This generates an animation of
about three and a half minutes. And it’s eerily silent. So could we actually
use the data to produce sound as
well as a visual? And that’s where this
process of data sonification comes into play. And what we’re going to do with
sonification is we’re going to take the y axis – so that’s
from 0 per cent to 15 per cent on the yields of our bonds –
and we’re going to map that into a four-octave scale
of musical pitches. And if I take the yield curve
back to its regular shape, and we introduce the
sonification element, you get to hear what it
might actually sound like. You can see that that’s
the rising pitch. As the yields rise,
so does the pitch. And that’s an upward
ascending arpeggio that represents a fairly
regular yield curve. And going back to the different
shapes of the yield curve that we looked at
earlier, we can start to hear what they sound like. So the flatter yield
curve sounds pretty much as you would expect. It’s the same notes repeated
again and again because they’re more or less identical yields. And then for our inversions –
so those are the points where the short-term bonds
are the highest – then that ripples downwards. And in the case of this
inversion from 1980, you can notice that the overall
pitch of the yield curve is higher because the
yields are very high. They’re up at 15 per cent
at this point in time. And the only problem now
with our sonification is it’s reflecting what we’re
seeing on screen very nicely. But at two seconds
per day, this would generate an incredibly long
animation and piece of music. And if we speed it up, then we
get a little bit of a problem. Is that now animating
five days per second, the music of the yield
curve is quite difficult to pick up on the arpeggio. It’s too fast for us to discern. So instead, what we can do is
actually speed the animation right up, but only play
every 30 days the music. And that generates something
a lot more melodious. And you can hear those
changes in pitch. The level of the pitch and the
direction of the pitch really following what we’re
seeing on the screen now in terms of the animation. And in many ways
now the yield curve feels alive with
both sound and audio being generated from the data. One issue now is
that in audio space, we don’t have the equivalent
of our year markers or our date markers. We don’t know what the passage
of time is in the audio space. So the first thing that we can
do is introduce a bass drum. And all the bass
drum is going to do is it’s synchronised
to the month. So when you hear a bass drum,
the gap between the bass drum is one calendar month. And you can see that’s perfectly
synchronised to the date on the screen. What we can also, therefore,
do is introduced… so we’re tracking the month. Let’s try and track the years. This time, what we’ll
do is we’re going to add on a year in vocal form. ’82. ’82. ’82. ’82. ’83. ’83. ’83. You can see that I’ve
placed a slight delay, a repeating echo on the
voice, so that it fades out through the year. So you get a real sense
of progression over time. And then to provide a final
join between what we’re seeing on screen and
what we’re hearing, we can also play a sound effect
whenever those peak inversions appear on screen. So I’m going to take it
right back now to 1979 with four layers of sound on. 1980. 1980. 1980. 1980. ’81. You can see now that there’s
a perfect join between what we’re seeing on screen and what
we’re hearing in the audio. So we really have
sonified this yield curve. Now, why might that be
a useful thing to do? Well, there is an
argument that actually by hearing what
you’re seeing, it helps you to reinforce
that experience generally. So it could make it much
more memorable for you if you can hear the
data as well as see it. But also for those
people who can’t see, this is a tremendously
useful technique. Because actually if
you close your eyes, you’re actually
able to parse this yield curve data
using audio alone. The data that I
used on this project is freely available on
the US Treasury website. And I’ve posted
details of the tools that I used in the
online article, along with links
to a new tool that allows you to
create sonifications without any coding. Feel free to post links in the
comments to any visualisations that you’ve made yourself.

9 Replies to “Chart doctor: the mysterious music of the yield curve”

  1. Interesting applying music to time series. By mapping music to market activity, data sonification can also reveal rhythmic patterns in stock price activity.

  2. I love sound, but even so I'm not convinced of how significant this is, and I'm extremely doubtful sonification "might just be the next big thing in data presentation." My views on data sonification changed after I watched this video from Tantacrul: https://www.youtube.com/watch?v=Ocq3NeudsVk who makes many good points.

  3. If that's what my FT subscription money is contributing to on the margins, I don't mind the mild hikes! Brilliant!

  4. thought this audience might find this interesting too? https://www.oreilly.com/ideas/cameron-turner-on-the-sound-of-data

Add a Comment

Your email address will not be published. Required fields are marked *