5 minutes before US markets open, I give you a plot showing the trailing 15 days percent return of the S&P500 since 1928 versus the one-day return from the 15th to 16th day.

Notable because yesterday our 15-day return was -19%, so this graph is centered around that (-17% to -23%), and overlaid on it is the linear fit. Which, for the vision-impaired, is flat: even odds that the market will pop up or continue to crater down.

I don't have time to redo analysis considering opening futures…

FYI!, if I plotted today on this chart, it’d be more negative than every other point shown on this graph!

That is, of all fifteen-session windows where the SNP500 dropped between -13% to -28%, the one that ended yesterday was followed by the worst single-day drop ever, in the data going back to 1928.

What a glorious amazing beautiful invigorating day!

Show thread

Here's a redo of the plot from earlier, with a bugfix and a bright red dot for today.

Early today, the 15-session % change for SNP500 was almost -19%. That's the x-value of the red dot.

Today we lost almost 10%. The red dot gets a y-value of -10%.

Which is the lowest of all points in this graph of %-changes between -25% and -15%.

The second chart shows the full zoomed-out plot: all 15-day returns from -37% to +55%, and all one-day returns of -20% to 16% are here! Which is your fave???

Show thread

But why have the fifteen-session trailing returns predict just the one-day future return?

Why not see all one-YEAR future returns?

That's what's show in this plot. Each dot in the previous plot is still here with the same position along the x axis (trailing 15-session returns) but the y axis shows the returns over the subsequent year.

We're at -25% on the x axis, and the data supports anything from another -35% decline in spring 2021 or a healthy +45% recovery.

Show thread

The eye-popping 150% annual returns at the top-left of the plot all come from the later stages of the 1929-1933 brutal bear market and depression.

I really need to show a second predictor here to indicate that we're just coming off the top of a bull market, and are not in the late stages of a collapse.

Poking around with the resultant linear regression models with various other predictors like "six months trailing return" also… lots of uncertainty.

Lots of opportunity.

Show thread

From limited experience, Fridays are complicated. It reminded me that the # of editors who edit Wikipedia has a strong day-of-week pattern too (github.com/fasiha/wikiatrisk#f).

So I made this. Same data as above (updated to include Friday) but plots five different linear regressions for each day of the week.

It's not looking good for Monday! Jokingly from the full linear regression's slope.

On Friday we were at -16% on the x axis. And out-of-the-money puts on S&P500 were exxxpensive!

Show thread

(15 was interesting to me when this thread started because the peak was Wednesday, Feb 19, fifteen days before.)

Hussman Funds had this plot overlaying several bull market tops over each other and showing the subsequent declines—how deep and how long it took. It was interesting and I can't find it again. I… hell, I can just make it myself.

Show thread

It's an ugly chart, far more busy than my memory of Hussman's and needing deep introspection, but here's the seventeen bull market tops that were followed by drawdowns of 20% or more to their bear market bottoms.

Split into two sections just to have a hope of seeing each of the lines.

Each is quite a story.

(SVG at github.com/fasiha/historical-o for nicer zoom)

Show thread

After staring at this for a few hours—

1) My hypothesis that most bear markets started with a big crash is wrong. Most (2007, 2000, '80, '73…) just wander down to -50% or whatever over months and months. This month's trapdoor only matches 1987 and… 1929.

2) I'm not counting the '32 and '33 trapdoors because the runup to their tops was so steep, and they were '29's aftershocks. February was at an eleven year raging bull market.

3) 2007 saw a trapdoor, ~11 months after the top. '38 had one too.

Show thread

These are nominal prices. I've been staring at this chart, of nominal vs real (inflation-adjusted) prices of the S&P also. Log-plot.

Staring at the nominal curve gives a wrong feeling about the post-WW2 boom. And underemphasizes how shitty the 1965–1980 era of stagflation and oil shocks were for equities.

The inflation-adjusted curve shows the post-WW2 era as a more natural extension of the pre-WW1 era. Interpolating straight lines (→exponential growth in log plot) to today gives ~1500~2000.

Show thread

Might as well post the drawdowns chart that includes this pre-1928 monthly data from Robert Shiller.

I'm thinking a lot about 's short-term vs long-term debt cycles from "How the economic machine works in 30 minutes" economicprinciples.org because just eyeballing this chart, it seems that outside the world wars and great depressions we have >20% drops every ~10 years, i.e., the short-term debt cycle.

Wars and depressions overlap with the long-term debt cycles (~75 years)—coincidence?

Today we hit 20-session trailing returns of -30% 😳!

But enough about the past! What about the next three days? What does history say it might go up or down to till Monday close?

As usual, these charts show the trailing returns, so find -30% on them. Then one shows the WORST drop over next 3 days, the other the BEST spike over 3 days. Color is year.

Depending on whether you've bought put or call options expiring Monday 😊.

What an incredible ride, in a ghastly, Army of Darkness sort of way.

Show thread

I feel this Matplotlib dark style with Viridis plot colors is very Manhattan in the rain, very bright lights big city. Makes me feel like 8-bit pixelart Neuromancer.

Show thread
Sign in to participate in the conversation

The social network of the future: No ads, no corporate surveillance, ethical design, and decentralization! Own your data with Mastodon!