friday speedrun

Backtest of Qualifying Barts

Yields up, stocks shrug, Bart’s a continuation pattern.

September 11, 2026

Hello. It’s Friday. Thanks for signing up. I’m Brent Donnelly.

The About Page for Friday Speedrun is here.

Here’s what you need to know about markets and macro this week

Global Macro

Iran looks to be attempting to establish escalation dominance into midterms as the votes are starting to be cast by mail and the TV ads are getting locked in for the October sprint to the November 3 finish. This week was a bit of a step change as Iran attacks U.S. military assets directly and aggressively while the Bab el-Mandeb story gets worse by the day. The nothing ever happens view is still doing okay, surprisingly, even as yields in various countries make new highs not seen since various years pre-dating the Y2K scare.

Y2K was twenty-seven years ago!

Global bond yields don’t care much about Scott Bessent’s $4B or $6B of long-end buying as economic normalization, high inflation, huge global deficits, and mega corporate issuance to fund AI capex all conspire to push yields higher again this week.

This week was a continuation of the structural theme that has dominated macro for the past three years. Deficits are huge, inflation is above target, there is no credible way out, and stock markets don’t really care. I mean… I guess they care about this much:

CPI and PPI came out pretty much on the screws, with an imputed PCE number of roughly 0.25% and the September meeting has become the exact sort of nightmare Warsh warned us about in 2025 (from his Commanding Heights Lecture):

I do not find the current Fed policy of ‘data dependence’ of much real value. We should care little about two numbers to the right of the decimal point in the latest government release. Breathlessly awaiting trailing data from stale national accounts– subject to significant, subsequent revision– is evidence of false precision and analytic complacency.

And yet! An FOMC decision that will impact the economy with a six-to-eighteen-month lag has now come down to extrapolating the third decimal of a future PCE reading from two releases that come out one week before the Fed decision. File this under: What are we doing here?

A glimpse into the absurdity:

“To give you a sense of how close call the September rate hike decision is, my team is operating on 1000th decimal points for our [PCE] inflation forecasts,” wrote Bloomberg chief U.S. economist Anna Wong on X last week.

The decision whether or not to hike comes down to one question.

Does the Fed target the rate of change of CPI and/or other personally preferred inflation metrics, or do they target a particular level of PCE? They target the level of PCE inflation. Not CPI or PCE change. You will see many people on Twitter saying something like: “CPI is the lowest it’s been since 2021! Fed should not hike!” This mistakes a rate of change for a level, and confuses CPI with PCE. That is, the Fed target is a level of ~2% or so for PCE. Not “CPI is falling to its lowest level since the biggest inflation spike of our generation.”

They have been missing the target for 65 months and when they cut six times for no reason in 2024 and 2025, they lit a fire back under inflation. Inflation at a 5-year low is not the target. 2% is the target. PCE is currently 3.7%. It will go moderately lower due to some rejigging of inputs and such, but with oil above $105, copper all-time highs, and ags ripping higher, there is no reason to think 3.7% is going to be 2% anytime soon.

The rate cuts in 2024 and 2025 are like the guy that stops taking the penicillin because he feels better. When you look at inflation, see it coming down, and say Mission Accomplished, this is what happens. It simply speaks to the asymmetry at the Fed: They will always make a dovish mistake before they make a hawkish mistake.

At this point, after a hawkish June FOMC, dovish July FOMC, hawkish August Jackson Hole and on-the-screws inflation data, the Fed could legitimately do whatever it feels like doing. But the Warsh Fed is not looking any different than the Yellen or Powell Fed to me. They will now hike in September, purely because the market has priced in the hike.

Does a single rate hike, followed by maybe one or two more hikes matter for the real economy or for financial markets? Unless you’re a STIRT trader, I would hazard that the answer is no.

Stocks

Stocks grudgingly go down when oil rips higher and yields make new multi-decade highs, but they don’t look very happy about it. Shorts press and get squeezed and the market has a lot of trouble finding the conviction to stay bearish when all the scary things going on are six months old. For a real bear move in stocks, you need scarier news flow.

Bond market vol, an indicator I consider most important as we try to gauge the fear around rising yields, is pretty subdued still. That said, it is approaching the top of a recent range and could possibly break out.

Until then, bears need something more to chew on. Perhaps a shot of bearish seasonality might help?

It’s amazing how the bearish equity seasonal has worked so well out of sample for so many years. I was curious whether it matters if you short stocks in the second half of September or wait until after Triple Witching. Answer is: It doesn’t seem to matter.

This chart shows the cumulative P&L of going short S&P 500 on September 15 or going short at the close on Triple Witching (for this year, that would be next Friday). If you have ever done any backtesting of S&P 500 strategies, you can appreciate how extraordinary this is. It’s almost impossible to find any persistent bearish event test for stocks.

Shorting the S&P over any random two-week window loses 0.4% to 0.5% on average and wins only about 40% of the time. Shorting the second half of September has a 58%-64%% win rate and an average return of +1.2%. Pretty mind blowing, especially as this thing was unearthed in the late 1960s. Back then, Walter Merrill identified September as the worst month of the year in his 1966 book: The Behavior of Prices on Wall Street.

Shorting stocks in the back half of September does not work every time, obviously. And it always seems way too obvious, like, “It can’t possibly work again this time!” With VIX at 17, you are not paying much to take your shot. Here are the stats for the two strategies.

I am away for the next two weeks on a family trip, so I bought 25SEP VXX calls yesterday as a simple way to get some leverage for a possible stock swoon. I don’t need to tell you why stocks could dump. There are plenty of reasons right now. Then again, there are always reasons.

Here is this week’s 14-word stock market summary:

I’m bearish ‘til October 1. Just because it’s obvious doesn’t mean it won’t work.

One side note for stock pickers and basket weavers:

Could skyrocketing state-sponsored stocks spark skepticism as November nears? Many single stocks have shot skyward due to explicit or second-derivative state-sponsored support from the Trump admin. Pending a potential flash of Blue at midterms, the market may start to price the reduction of support for those lucky winners picked by the U.S. government in 2025/2026.

Below is a basket of booming beneficiaries that might become busted bets as they are bound for a bout of underperformance if Big Blue breaks through in the midterms.

I am not an equity analyst, and the twelve stocks I chose are fairly obvious but debatable choices. This is more of a behavioral call than a policy or election view. The market is soon going to sniff out Red Wave winners and lighten up, or even position for them to become Blue Wave losers.

Something to think about.

Interest Rates

The US 10-year yield continues to shoot higher, though you could plot almost any developed market country yield of most tenors and get a chart that looks something like this:

Scott Bessent’s attempts to get yields lower in late August pushed 10s to 4.65% and we are now trading near 5.0% despite a bigger buyback. The Treasury Secretary calling himself the house is meant to be clever but in the context of the performance of his boss’s casinos, the flex doesn’t hit quite as hard as he might hope. Most casinos are winners, but some go bankrupt.

The thing about government intervention, though, is that it’s very hard to bet against because they will keep doubling down and if you call their pair of eights with AA, they will sit there frozen for a while and then they’ll make a phone call or two and the pit boss will come over and tell you:

“Yes, sorry sir. But eights beat aces. Aces are low because they are only worth one, can’t you see. Right there on the card. Just one.”

And you will be skeptical, but he will hand you an information card.

So while you can play the rise in yields, and short bonds has been an excellent trade for a long time, once the government starts to push back on a trade, it becomes very difficult to monetize because you are always one headline away from a negative P&L shock. There are infinite tweaks and a long list of major actions the U.S. Treasury can take in an attempt to cap yields. What they really need is an economic slowdown, and with a bit of fiscal drag coming, AI Capex growth slowing, and supply shocks stacked upon supply shocks, they might get a soft patch by Q1 2027.

For now: Inflation is high, nominal growth is high, consumers are rich, and yields are normalizing back to 1990s levels as the great stagnation is officially over and we remain in search of bond market equilibrium.

Over in Japan, you have some concerted action to stabilize yields, too. They are working on various ways to get GPIF, the GPIF sister funds, other Japan pension funds, and individual Japanese savers to start investing more in Japanese bonds. This is a long process, but I think it will likely lead to bond market and yen stabilization and this is why I have been saying for a few months now that unhedged long JGBs will end up as a super high Sharpe trade over the next year or two.

Fiat Currencies

This section is excerpted from today’s am/FX.

I have written extensively about how FX correlations to commodities broke down post COVID, but the reality is that the correlation story is constantly in flux. Here’s a chart of the 26-week rolling correlation of changes in AUD and CAD (combined) vs. copper and oil. Because this is a 26-week metric, it masks the extremely high correlation in the past few months.

The commodity vs. currencies correlation cracked when the War in Iran started because risk aversion combined with higher commodity prices is not clearly beneficial to cyclical currencies. But now, as oil rallies are not consistently accompanied by risk aversion, CAD and AUD can benefit from higher commodity prices.

You can see it in my next chart, if you put on your glasses. AUD, CAD, copper, and crude are all moving as a giant blob again. I think the main explanation for this is that equities are not particularly hard hit by Iran headlines these days, even when oil spikes, and therefore the risky / cyclical currencies are able to benefit from the export boost potential of rising commodity prices. Furthermore, the moves in commodities have been broad-based as ags and precious have done pretty well in recent months, too, and broad-based commodity rallies are more supportive for Australia and Canada than idiosyncratic, geopolitics-driven rippers in oil.

While my job is to have a strong view on the USD at least most of the time, I really do not have a dollar view right now. The ECB is hiking into a potential winter energy crisis, but nobody cares. AUDUSD is crowded just as dividend season ends and copper turns, but who’s going to short AUD? NZD looks like a bad currency to own as rate hikes there are overpriced. EM carry is okay but do you want to own it into the worst two-week period of the year for risky assets? Hard to find clear bullish non-USD stories other than the JPY.

On the other side of the ledger, the Big Random U.S. Policy Wheel spins every morning and sometimes lands on tariffs, or war, or trillion-dollar economic rebates, or government calls for Fed rate cuts into full employment, or whatever. And Scott Bessent is increasingly eager to participate. As Jared Dillian pointed out today in the Dirtnap: The U.S. government is probably not far from enacting some new measure to cap bond yields as prior measures fail. It’s very hard to get excited about long dollars, especially as midterms will bring more policy confusion and fiscal drag looms.

USDJPY shorts remain in control as the busted 155.00 level has become monster resistance. Anyone that missed the selloff in USDJPY so far is camped up there somewhere hoping to sell 154.30/70 ahead of the massive support that broke and is now resistance. My 08SEP view that USDJPY is set for a 152/155.50 range remains and yen vol still looks overpriced to me. Next week’s BOJ is going to be a humdinger as there is plenty of room for disappointment if the market overestimates how much the Japanese actually care about what Scott Bessent thinks.

Here’s the calendar:

My new book is out:

TRADE OUTSIDE THE BOX: Advanced Thinking for Professional Traders. You can buy it on Amazon right here. Here’s the one-pager.

The book is written for experienced traders. If you are new to the game, I would suggest you start with Alpha Trader. If you have been trading or investing for more than five years, this book is for you (and you don’t need to have read Alpha Trader to enjoy it).

Crypto

The debasement trade is taking a breather as global yields shoot higher and the idea that Bessent can cure the bond market fever was short-lived. Still, when Bessent put out the bond buyback release in late August, bitcoin was $64k and now it’s still holding above $78k. The giveback has been minimal in BTC, ETH, and the DATs.

Last Friday Speedrun (August 28), I wrote about the extremely elevated RSIs in BTC and ETH and how I expected a consolidation or perhaps a correction as low as $72k in bitcoin. The consolidation view has been fine, but the pullback has been milder than I expected as we have thrice bottomed on a high seventy-six handle. I’m not a big fan of the Bart Simpson formation but it’s a perfect one, so here you go.

The weird thing about my attempt to put Bart into that bitcoin chart is that the image looks nothing like Bart Simpson, really. But it does?

But it doesn’t?

Anyway: I am going to put this chart and a bunch of daily BTC data into Claude and ChatGPT and ask them to backtest the Bart Simpson formation. Whatever the results, I am going to report them here. I am going to do this now; I’ll be back in a bit less than twenty minutes. But due to the way writing and reading work; you will not have to wait any time at all! You will fly forward through time like a time machine flying forward person.

Okay. I am leaving this document to go do the backtest. See you in 20 minutes or so.

TIMESTAMP 1:39 P.M. EASTERN TIME

TIMESTAMP 1:55 P.M. EASTERN TIME

I’m back. It didn’t take as long as I expected. And I went to get pretzels, too.

I think the results here are quite interesting, so I am going to take the time to reproduce a significant chunk. Claude’s response was similar to ChatGPT numerically, but the explanation and such was not as good so I will use a combination of Claude and ChatGPT outputs.

Here is my prompt (same prompt in both models):

I ran this on the 4,304 daily Bitcoin closes from 1-Jan-2014 through 11-Sep-2026. The main result is fairly clear: the classic Bart Simpson implication — “vertical up, chop sideways, then dump” — does not work in this sample. If anything, the ordinary Bart was a continuation pattern.

How a Bart is defined (the strict version). I checked each rule against the example in your screenshot (the Aug–Sep 2026 setup), and the detector picks that one up.

Face: a jump of at least 10% within 7 days that is also at least 2.5× normal volatility for that length of move

Neck: the 20 days before the jump must be flat, with a range no more than 60% of the face’s height. This rule removes V-bounces out of crashes and slow staircase climbs.

Hair: the next 10 days chop sideways near the top. The pullback can’t exceed 35% of the face, the push above the top can’t exceed 25%, and the whole band has to fit within 40% of the face.

Signal and costs: the signal fires at the close on day 10 of the hair, costs are 10 bp per side, and only one trade is open at a time.

The inverse Bart is the mirror image: a crash, then a flat bottom, then go long. The detections chart shows every hit, so you can check for yourself that they all look the part.

I deliberately made the definition restrictive and did not tune the pattern parameters on subsequent P&L. A qualifying Bart needed a 2–6 day vertical move of at least 8%, at least 2.5 standard deviations versus prior 30-day volatility, and a directional efficiency ratio ≥0.68. It then needed 7–21 days of genuinely sideways “hair”: narrow range relative to the vertical leg, very little net drift or regression slope, repeated direction changes, and price staying near the new level. I also required a reasonably quiet pre-jump baseline and imposed a 30-day cooldown so the same formation could not generate multiple signals. I visually checked the resulting windows against the Bart silhouette.

That very strict screen produced only 6 Barts and 5 inverse Barts, which is probably about right if the objective is “only things that actually look like Bart.”

The six strict normal Barts occurred on 11-Apr-2019, 4-Nov-2019, 24-Dec-2020, 24-Mar-2023, 28-Jun-2023, and 3-Nov-2023. The inverse signals were 14-Aug-2016, 4-Oct-2019, 4-Mar-2020, 25-Aug-2023, and 11-Jun-2026.

The striking result is the normal Bart. Shorting it lost money at every holding period I tested. At 10 days the average short lost 5.2%; at 20 days it lost 11.9%. The December 2020 Bart was particularly destructive because Bitcoin simply resumed its rally. In other words, the “right side of Bart’s face” frequently never arrived.

Inverse Bart is more interesting immediately after the pattern. The next-day long was 5-for-5, averaging +1.54%, and the three-day long averaged +1.83%. But with only five observations, I would not call that a statistically reliable anomaly. By days 5–15 it disappears completely. The March 2020 inverse Bart is also a useful warning: the supposed rebound setup instead ran straight into the COVID crash.

I also relaxed the morphology filters enough to increase the sample to 15 Barts and 8 inverse Barts. That did not rescue the folklore. Normal Bart shorts averaged -1.2%, -0.1%, -1.5%, -3.0%, -4.3%, -7.4%, -9.4%, and -10.2% over 1/3/5/7/10/15/20/30 days respectively. So the failure of the short-Bart thesis is not just an artifact of the tiny strict sample.

Bottom line: I would not use “Bart Simpson formation” as a bearish Bitcoin signal. The data actually suggest that a clean upside Bart is more consistent with post-breakout consolidation/continuation than impending collapse. The one piece worth investigating further is the very short-term 1–3 day rebound after inverse Barts, but five strict examples is nowhere near enough to establish an edge

A FEW OF CLAUDE’S BART DETECTIONS – they are different from ChatGPT but both LLMs come to the same conclusion:

Bart’s a continuation pattern.

CLAUDE’S ROBUSTNESS CHECK

So: Qualifying Barts are bullish! Qualifying inverse Barts are bearish. It’s a continuation pattern with a small sample. Good to know.

I love how the AI doesn’t skip a beat and starts making up terms like: Qualifying Barts. Sure, AI might kill us, but this is the kind of pleasure we can experience from AI before it determines we are simply obstacles.

Humanity: It’s not just our maker; it’s an obstacle.

Oh, and Zcash keeps going.

Commodities

Big narrowing wedge pincer thing in copper. If you have some other reasons to be bearish copper, it’s a nice setup from a risk/reward point of view because you can simply stop out if it makes a new high (above $7.00 on HGZ6) and if you’re right, you catch three or four ATRs.