Click to enlarge.
I know that just a few billion dollars might not seem like much on a short-term chart, but look what happens when we zoom out a bit.
Click to enlarge.
$1 trillion per month in total US inflation adjusted trade in 2025! Or bust!
Isn't my optimistic way of looking at things so much better than just looking a dreary chart without any analysis? I'd like to think so.
Click to enlarge.
There's certainly no happy in that chart.
Green is a healing color! And red parabolas make mine look so Christmassy by comparison!
Lip service, baby. That's what I'm talking about.
Source Data:
St. Louis Fed: Custom Chart
How do you find these parabolas (serious question, not a lead in to a sarcastic remark).
ReplyDeleteDo you "eye-ball" the data first or do you have a parabola-finder program you run against the data?
That's a good question.
ReplyDeleteI do not have any sort of parabola-finder program. It's just me looking at the data and deciding if it is worthy to drag into Excel.
The data in the FRED database is easiest to work with, since I can use the natural log function to confirm an exponential trend easily (straight lines on log charts show exponential growth). If the log of the data is curving, then it may be a parabola instead.
Since I mostly care about the long-term trends, if I see interesting curves in a data set then I usually start by smoothing the data to remove the noise (using moving averages). That requires taking it to Excel.
I then take that data and experiment on it. I setup starting and stopping points and try fitting an exponential curve. If it doesn't fit well because the trend can't curve enough, then I try a parabola. If the parabola doesn't fit well, I often give up and move on to the next data set. I am most interested in exponential and parabolic trend failures out of curiosity.
Once I've determined which curve fits the data the best, I then fine tune the starting and stopping points to maximize the r-squared value. This fine tuning determines the actual point of failure (if there is a failure), which is clearly very useful.
As a side note, I was all over the place when I started this blog in 2007. I didn't really have a system. I mostly just looked for exponential trends because I knew they were out there.
I had no idea there were so many parabolic trends though, so I never really looked for them much. Many were hidden in the noise. Smoothing the data was sort of a breakthrough moment for me. It can make all the difference. There's something almost magical about finding a parabola with an r-squared of 0.999 or more. There's much doubt that's a parabola. Can't get that with noisy data.
Parabolas get a lot easier to find with practice. Just knowing they exist makes me want to find more. These days, I can often spot a parabola just by looking at the raw data, even if it's fairly noisy.
There's much doubt that's a parabola.
ReplyDeleteThere's not much doubt that's a parabola. Oops.