Published market calls
published forecast week · Sep 1 – 8, 2026
Original probabilities registered before their outcomes, with every result posted. A 30% call still lands 3 in 10. See the record →
Our track record0.146 Brier · lower better77% called right · 22 resolvedReality Check →
This week’s regimeThe week the hike came back. Chair Warsh used his first Jackson Hole keynote to say the quiet part — core PCE at 3.7% and running 4.1% annualized over six months, labor at full employment, policy "not restrictive" — and September hike odds roughly doubled on it: futures now price about two-in-three for a hike, prediction markets near 70% for one this year, and the two-year jumped nine basis points in a day. Everything on this board trades off that repricing: gold has given back almost 5% from its record, Nvidia surrendered its entire earnings pop, and the long end is backing up into Friday's payrolls — consensus +60K after July printed negative. The FOMC itself meets Sep 15–16, just outside this window; this is the positioning week, shortened by Labor Day. Last week's board went five for seven, and the two misses are the story: our 87% lira call lost by a tenth of a percent, and energy gave back its blockade premium. Every number below is the model's, not an analyst's — and the honest read is the same as last week: two of the six sit on their own twenty-year base rate, one sits below it. The model's edge this week is a momentum tilt and one structural crawl. We would rather you read that than not.
as of 2026-09-01
MacroGuru near-term modelResolves at the Tue Sep 8 close; Turkish August CPI lands Thu Sep 3.
USD/TRY finishes the week higher than it started (48.27 at issuance). Still the only number above 70% on the board — one week after the same call missed.
Why we think soOver 1,008 weekly windows USD/TRY closed higher 87.0% of the time, and the model prints 86.8% — the crawl, not our judgement. Last week is why you should read this one carefully: the same 87% call missed, and it missed by a tenth of a percent — from Wednesday's final close the lira ended the week marginally stronger, though from the price on the screen at issuance it ended weaker. That is what an 87% losing week looks like: not a crisis averted, a rounding error against us. The mechanism is unchanged — deeply negative real rates and a managed crawl — and the model's own graded record on this market is 22 of 22 (mean Brier 0.011), because it grades from issuance prices. Our published record now carries the miss, and we would rather show you both numbers than pick the flattering one. What breaks it: an intervention big enough to reverse a whole week, or another photo finish.
Moves↑ TRY
Research lens A structural bleed, not a trade on the week's Fed story — the crawl doesn't care about Warsh.
How it resolves · our edge · sources
Resolves 2026-09-08 — TRUE if USD/TRY's daily close on 2026-09-08 is higher than its 2026-09-01 daily close (the lira weakens over the week); FALSE if it closes flat or lower. Settled on MacroGuru's daily price cache — the same series both dates.
⊕ Our edge — Repeating a high-conviction call the week after it misses is the whole point of publishing a model instead of a mood.
MacroGuru near-term modelPayrolls Fri Sep 4; resolves at the Tue Sep 8 close.
The 10-year Treasury yield closes the week higher (4.78% at issuance), extending the Warsh repricing — hike odds near two-thirds and payrolls landing Friday.
Why we think soThe 10-year closed higher in 55.1% of 1,008 weekly windows; the model prints 56.2%, a one-point momentum tilt — the yield is up 3% over seven days and the expected move (+0.3%) points the same way. The macro story writes itself: Warsh told Jackson Hole policy isn't restrictive and the short end repriced overnight. But here is the uncomfortable number: at this exact seven-day horizon the model's rates record is 4 of 10 — worse than the coin — and 16 of 25 across all horizons (Brier 0.239). Rates are its weakest market and we publish the call anyway, labelled, because the mechanism is the week's central story. What breaks it: a soft payrolls print Friday that takes the hike off the table — consensus is +60K after July's negative print, so the bar for a downside surprise is low.
Moves↑ 10Y Treasury yield
Research lens The week's master variable. Watch Friday 8:30am ET — payrolls either arms the September hike or disarms it.
How it resolves · our edge · sources
Resolves 2026-09-08 — TRUE if the US 10-year Treasury yield's daily close on 2026-09-08 is higher than its 2026-09-01 daily close; FALSE if flat or lower. Settled on MacroGuru's daily price cache — the same series both dates.
⊕ Our edge — We show you the model's worst market with the label on, rather than only its best.
MacroGuru near-term modelResolves at the Tue Sep 8 close.
Gold closes the week higher ($4,387 at issuance) after giving back nearly 5% from its record — a drift-up base rate against a freshly hawkish Fed.
Why we think soGold closed higher in 58.5% of 1,008 weekly windows and the model prints exactly 58.5% — pure climatology, to the decimal. That is worth being honest about: after a −5.4% week the momentum signal argues down, the drift base rate argues up, and they cancel. So this is not a bottom call on the Warsh flush; it is "gold usually drifts higher" with nothing added. The record matches the humility: 14 of 22 graded gold forecasts, mean Brier 0.224, barely better than a coin. What breaks it: the same thing that caused the flush — another leg of hike repricing pushing real yields up. A strong payrolls print Friday is the direct threat.
Moves↑ Gold
Research lens The anti-Warsh trade: it works if the hike repricing stalls, and Friday's payrolls decide that.
How it resolves · our edge · sources
Resolves 2026-09-08 — TRUE if gold's daily close on 2026-09-08 is higher than its 2026-09-01 daily close; FALSE if flat or lower. Settled on MacroGuru's daily price cache — the same series both dates.
⊕ Our edge — We print the model's 'no view' as 58% because that's what it says — not manufactured conviction either way.
MacroGuru near-term modelResolves at the Tue Sep 8 close.
NVDA closes the week higher ($216.41 at issuance). The stock beat, popped 5%, then gave the entire move back — it now sits below its pre-earnings close.
Why we think soNVDA closed higher in 59.6% of 1,008 weekly windows; the model prints 59.2% — effectively the base rate, with a small markdown for the −2.0% final session before issuance. The tape context matters more than the number: Nvidia delivered the beat, closed up 5% the next day, and has since surrendered all of it — the post-earnings round trip is complete and the stock sits fractionally below where it traded before the print. The call says drift reasserts from here; the expected move is slightly negative (−0.4%), meaning more small up-weeks than large ones. Record: 19 of 25 graded, mean Brier 0.213. What breaks it: the hike repricing — at 45% realized vol, a rates-driven risk-off week swamps any drift.
Moves↑ Nvidia
Research lens The tell for whether AI demand still buys the dip when the Fed is the headwind rather than the tailwind.
How it resolves · our edge · sources
Resolves 2026-09-08 — TRUE if NVDA's daily close on 2026-09-08 is higher than its 2026-09-01 daily close; FALSE if flat or lower. Settled on MacroGuru's daily price cache — the same series both dates.
⊕ Our edge — A beat that gets fully sold tells you positioning was the problem, not the quarter — the model leans on drift anyway.
MacroGuru near-term modelResolves at the Tue Sep 8 close.
The energy sector (XLE) closes the week higher ($64.52 at issuance) with Brent near $93 and the US naval blockade of Iran still in force. The same call missed last week; the momentum signal got stronger, not weaker.
Why we think soEnergy closed higher in 56.5% of 1,008 weekly windows; the model prints 57.8% — a 1.3-point momentum tilt, with XLE up 4.0% over seven days and the expected move (+0.5%) pointing the same way as the probability. Full disclosure: this exact call missed last week — XLE gave back 2.3% while Brent held its bid. The sector then reclaimed the entire drop in four sessions, which is why the model re-issues the lean rather than retreating from it. It remains our best-scored equity market: 21 of 25 graded, mean Brier 0.206. What breaks it: crude. The whole tilt is a blockade premium pass-through, and one de-escalation headline out of the Gulf unwinds it in a session.
Moves↑ Energy sector
Research lens Still the counter-current trade: everything else on the board is a Fed story; this one is Hormuz.
How it resolves · our edge · sources
Resolves 2026-09-08 — TRUE if XLE's daily close on 2026-09-08 is higher than its 2026-09-01 daily close; FALSE if flat or lower. Settled on MacroGuru's daily price cache — the same series both dates.
⊕ Our edge — Re-issuing a missed call with a stronger signal is what a model does and a pundit doesn't.
MacroGuru near-term modelPayrolls Fri Sep 4; resolves at the Tue Sep 8 close.
The S&P 500 closes the week higher (7,637 at issuance) — a mild long through Friday's jobs report, with September hike odds near two-thirds.
Why we think soThe S&P closed higher in 59.5% of 1,008 weekly windows and the model prints 57.7% — 1.8 points below its own climatology, the only call on the board marked down from the base rate. The markdown is momentum (−0.5% over seven days) plus a negative expected move (−0.7%): the model expects a higher close to be more likely than not, but the average path to be down — a right-skewed grind, not a rally. Payrolls is the catalyst that decides it: consensus +60K after July's −23K, and this tape has been treating good news as bad news since Warsh spoke. Record: 16 of 25 graded, mean Brier 0.232. Not independent of the Nvidia call — one AI leg decides both, and we disclose that rather than double-count it.
Moves↑ S&P 500
Research lens If payrolls lands near consensus, drift carries it. A hot print is the risk now — strong data feeds the hike.
How it resolves · our edge · sources
Resolves 2026-09-08 — TRUE if the S&P 500's daily close on 2026-09-08 is higher than its 2026-09-01 daily close; FALSE if flat or lower. Settled on MacroGuru's daily price cache — the same series both dates.
⊕ Our edge — Reads with the Nvidia call by design; a soft-landing print can decide both.
How we score this. Every call is logged, graded against what actually
happens, and published — wins and losses. Odds built from history; market odds from Polymarket & Kalshi.
Reality Check →