The Day Oil Prices Went Negative — Experiments in Sonification
This is a story about the day oil prices went negative, on April 20, 2020. To be more specific, West Texas Intermediate crude oil futures contracts for May 2020 plunged to -$40 a barrel (just typing that felt awkward). I first began paying attention when this thread began unfolding in a Slack channel with former energy reporter colleagues:
Then I hopped on Twitter, where people were wondering…could they, would they, go negative ? A question that was answered soon: Yes, yes they absolutely can.
Oil prices eventually stabilized back in positive values the next day. And as I write this, in mid-May, WTI crude futures contracts are around $26/barrel. (Shifting baselines are a curious thing…when I was an energy journalist in 2014, it was major news when oil went from over $100/barrel to around $50/barrel over the course of 5 months.)
But on the day oil prices went negative, people were reacting with shock and confusion. With oil prices dancing below zero, we collectively wondered: What do negative oil prices even mean? As economists and journalists explained this phenomenon, my data-nerd/audiophile mind turned to a different question: How do you sonify negative values?
What do negative values sound like?
Sonificiation is a technique for representing data with sounds — graphs for your ears. (For more on sonification generally, check out The Sonification Handbook, ICAD, or my last blog post.) Sonifying a negative value presents a mental pickle: How can you have a negative sound? But when I started to think about the equivalent problem in visualization, I realized that visualizing negative values is similarly bizarre. Heck, take a step back and the whole concept of negative numbers is counterintuitive. But because I’ve been thinking in negative numbers and reading them off charts for more than 30 years, I don’t remember what it was like to use negative numbers for the first time. It must have been confusing, even disorienting. (If you recently had to teach a child this concept, I would love to hear your take.) We’ve developed standard conventions for representing negative values visually — in spreadsheets: negative signs, red text, or parentheses; in charts, lines or bars dipping below the x-axis and color cues.
But we don’t have standard conventions for representing negative numbers through sound. Or, if they do exist, most of us haven’t learned how to listen for them. So here are some experiments exploring how we might sonify negative values, using crude oil prices as a test case.
But first, what does the data look like?
** Data munging diversion ** Turns out, it’s hard to get intraday prices (minute by minute change) of a commodity, especially after the fact. Some financial intelligence services have this data, but it’s often hidden behind a paywall. However, I was able to read the data for 30-minute intervals off of an interactive graph from Trading Economics. Some values were missing, which I either imputed them (using the mean of the values on either side) or estimated them based on a graph from the Energy Information Administration. I decided to leave the gaps where trading was closed. Below are graphs of the data from the EIA, and the data I was able to reconstruct:
The graph I created on the left uses 30-minute intervals, whereas the EIA one has even finer time-series data, thus there are some subtle differences . However, for the purposes of this sonification exercise, where my goal is to convey the strangeness of negative values, the 30-minute intervals will do the trick.
Finally…the sonification part!
First, to establish a baseline for our listening, let’s try a straight-up vanilla sonification: a simple audio graph of crude oil prices, with price mapped onto pitch. I created this using TwoTone.io, a free web-based sonification tool. The 48-hour period of April 20 and 21 is mapped onto 24 seconds of sonification time. Remember, we have trading data from every 30-minutes, so that means you’ll hear roughly 2 hours (4 tones) compressed into each second of the sonification. I’ve chosen a marimba and a 3-octave range to capture the $52-spread from the maximum ($16.72) to the minimum (-$39.55).
You can hear how the price plunges and then recovers, but there’s no indication of when the price crosses over into negative values, or when it nears zero. So how can we make the negative values — which are the critical element of the dataset we want to emphasize — more noticeable?
First, let’s try changing the timber — a musical term for the quality or texture of the sound — of the notes by changing the instrument. When the prices go negative, the notes will switch from a marimba to a trumpet. Why those two instruments? For the very practical reason that they have overlapping octave ranges, which means the frequency/pitch of the notes can change seamlessly.
Ok, this gets a bit closer to capturing the unsettling nature of the data. But we can do better: Let’s try changing the key from major to minor when the price goes negative. I got this idea from Agoston Nagy, who used major and minor keys to sonify positive and negative emotions of Tweets. Now, I’m not a musician (in fact, I’m embarrassed by how little music theory I retained from my years of violin lessons), thus I wasn’t sure what keys to use. So I turned to this delightful translation of Schubart’s musical key characteristics from 1806. The description of F# Minor stood out to me:
“A gloomy key: it tugs at passion as a dog biting a dress. Resentment and discontent are its language.”
For going from positive to negative values, we’ll use a type of key modulation called relative modulation, which takes us from A Major (“This key includes declarations of innocent love, satisfaction with one’s state of affairs; hope of seeing one’s beloved again when parting; youthful cheerfulness and trust in God.”) to F# Minor. And for going from negative back to positive, into a world that is not quite the same as before, we’ll use parallel modulation from F# Minor to F# Major (“Triumph over difficulty, free sigh of relief uttered when hurdles are surmounted; echo of a soul which has fiercely struggled and finally conquered lies in all uses of this key.”). Fun fact: This modulation that is used in the song “Happy Together.” Here’s what it sounds like:
I don’t notice these changes as much; that could simply be my untrained musical ear, or the fact that I am not modulating on the right chords. Perhaps this could be more clear by including the essential chord of each key in the background behind the marimba and trumpet melodies. Musicians out there, please chime in if you have ideas on how to improve this!
Let’s add a couple of final elements to really make the switch from positive to negative values noticeable: a hissing sound in the negative range (inspired by Stephen Hetzler’s sonifications of mathematical functions) and a record scratch when values cross from positive to negative or vice versa. Here’s the result:
Which of these techniques, if any, did you find effective? How might you improve on any of these concepts? For me, this is a good start, but it still doesn’t fully capture the shock and confusion that the data evoked.
Let’s get a little weird with it.
This time, we’re going to throw mapping prices onto pitch out the window. Instead, we’ll use a familiar song clip and modulate the tempo as the price gets higher or lower. This idea is inspired by Kuang Keng Kuek Ser’s project for PRI using “Ice Ice Baby” to convey projected sea level rise. No shade to Vanilla Ice, but we’re going to use much better artist: The Talking Heads. Specifically, the song “Nothing But Flowers,” because it’s about ambivalence toward a post-apocalyptic world. Also, it has a line that goes:
From the age of the dinosaurs / Cars have run on gasoline.
The tempo of the song will correspond to the magnitude of the price, scaled up or down from $16/barrel. That is, as values get close to $0, the song is going to get very, very slow. If the price were to get up to $32/barrel, it would be double-speed. Instead of just speeding up and slowing down a song, we’re also going to reverse it. So all of the negative values will be played backwards. Negative values close to $0, like -$5/barrel, will be slow and backwards; negative values far away from $0, like -$39/barrel, will be fast and backwards.
We’re going to focus on a 16-second clip, looped over and over and distorted according to the data, so that you can hone in on the data-driven changes without being distracted by the many tempo and key changes throughout the rest of the song.
Ready? Here’s a version with me narrating what’s happening (I recommend listening to this first…it’s also my favorite sonification experiment):
And for the extra brave souls among you, here’s a non-narrated version:
Weird, huh? I find it both discombobulating and mesmerizing — which sums up that mood of that day pretty well.
One final experiment.
Let’s take a moment and focus on what negative values truly mean. In the context of crude oil prices, it meant that there was not enough room to store the oil being produced, because the demand for oil had dropped. Thus, people with oil were willing to pay others to take it off their hands. It was a switch in polarity, a change in direction — things flowed one way, and then they flowed the other way.
What might that mean in the context of sound? Perhaps when prices are positive, sounds are played for you. But then, when they become negative, you, the listener, now have to create the sounds. It’s a switch in the direction of flow of sound production. This funky concept is not really meant to be a representation of the data that you might actually use to read off a number or make a calculation, but rather an experience to make you stop and think about what the data actually mean.
Because this type of sonification requires listener participation, I’ve written it in the form of a score:
Play the video clip below. When values are positive, listen to the notes. When values are negative, perform the notes. Lower negative values correspond to lower tones.
Here’s my attempt at performing the score:
As I was doing this, I realized that I was relying on my visual senses to see the data and then translate it, in as near to real time as I could, into sound — something that was extremely difficult, and that I’m sure I butchered. Intuitively, I wanted to make the sounds louder, instead of lower, as the values got more negative. But the quality of the sounds I made wasn’t really the point — the point was the experience of having to produce them.
Where else can these experiments take us?
How else might we convey negative values with sound? Perhaps by using sounds that have some conceptual metaphorical connection with polarity, like the sounds of breathing in and breathing out, stacking up Jenga blocks and then removing them, or writing on a white board and then erasing it. My favorite part about doing these sonifications is that I get to hear your reactions and ideas for taking these experiments further. So what do you think negative values should sound like?
