The print that never happened
For three weeks our American recession case rested on one number: July payrolls at −23,000, the first negative print of the cycle. This morning the Bureau of Labor Statistics revised it to +21,000. It was never negative. Then August came in at +162,000 against a forecast of 55,000. The evidence we built on has been deleted, and we are cutting US recession odds accordingly. The consolation: our Bank of Canada predictions, published three days early, came in at six from six — with one factual slip we are docking ourselves for.
We were wrong, and here is exactly how
On August 30 we raised US recession odds to 40–45% and said, in print, that America had become the riskier of the two economies. The case had two legs. One of them just broke.
The broken leg. We wrote that "payrolls went negative" and that a second consecutive negative print on September 4 would take the odds above 50%. There was no second negative print because there was never a first one. July has been revised from −23,000 to +21,000, a 44,000 upward correction, and June was lifted 11,000 to +31,000. Together, 55,000 jobs that we thought had not existed, did.
This is the second time in three weeks that a revision has embarrassed a call built on a single monthly print — the first being Canada's Q1 GDP, revised from flat to positive. The lesson is not subtle and we are going to state it plainly: one month of payroll data is not evidence of a turning cycle. It is a noisy estimate that gets rewritten twice. We knew that and wrote the sentence anyway because it fit the story we were telling.
The leg that held. The other half of the case was the Federal Reserve, and that argument is intact and arguably stronger. More on that below.
Bank of Canada: the scorecard
On August 30 we published six predictions for the September 2 decision, three days before it happened, with confidence levels attached. Here is how they scored.
| # | What we predicted | What happened | Grade |
|---|---|---|---|
| 1 | Hold at 2.25% (88%) | Held at 2.25%. We were more confident than the market's 77%, and its 22.8% hike probability did not materialise. Our error: we called it the "ninth consecutive" hold. It was the seventh. The call was right; the count was not, and we are noting it rather than quietly fixing it. | Hit |
| 2 | Statement frames tariff inflation as a two-sided risk (70%) | Macklem: "the new US tariffs and the Canadian counter-tariffs could add costs for businesses and feed into consumer prices over time" — alongside the growth warning that uncertainty "may lead businesses more broadly to delay investment and hiring decisions." Both sides, explicitly. | Hit |
| 3 | No forward guidance toward easing (75%) | "Governing Council will assess the sustainability of the economic rebound and the outlook for inflation, and is prepared to adjust monetary policy as needed." Conditional and directionless — no easing signal. | Hit |
| 4 | Loonie moves less than 0.5% on the decision (75%) | The loonie moved 0.06% on the decision — about an eighth of our threshold. The CAD story that week was the Fed differential, exactly as we argued. | Hit |
| 5 | No published tariff scenario analysis (65%) | The Bank stayed qualitative. No numbered scenarios in the statement or opening remarks. | Hit |
| 6 | Next move is a hike rather than a cut (60%) | After the decision, pricing for the next meeting still showed a hike as the live alternative to a hold, not a cut. Tracking as called; it fully resolves whenever the Bank actually moves. | Hit |
Why +162,000 is weaker than it sounds
Having just been burned by over-reading one month, we are not going to over-read this one in the opposite direction. The print is genuinely strong. The composition is genuinely not.
Two sectors are the entire beat. Food services and drinking places added 59,000; local government education added 42,000. That is 101,000 of the 162,000 — 62% — from restaurants and public school payrolls. Manufacturing added 16,000 and health care 13,000. Information lost 23,000, so tech is still shedding.
The trend rate is stall speed. The BLS notes the prior 12-month average is +31,000 a month. August is more than five times the trend. One outlier does not redefine a trend line any more than one negative print did.
The denominator is still shrinking. Labour force participation is 61.6% — up a tick on the month but down half a point since January. The unemployment rate is holding at 4.1% partly because people are leaving the labour force, which is precisely the effect we flagged in Canada and it deserves the same scepticism here.
Real wages are falling. Average hourly earnings rose 3.1% year-over-year. Core PCE is running at 3.3%. Workers are losing ground in real terms even as the headline job count impresses.
So: a labour market that is not rolling over, but is not booming either — adding roughly 31,000 jobs a month on trend, concentrated in low-wage and public-sector work, while purchasing power slips.
The Fed argument survived intact
Here is the uncomfortable part for anyone hoping strong jobs meant good news. Markets sold off because the report was good.
Rate-hike odds for September jumped from about 50% to 60% on the print before easing back toward 50% after Governor Christopher Waller struck a more dovish note. The 2-year yield hit a fresh 52-week high at 4.39%, the 10-year moved to 4.78%, the 30-year sits at 5.24%, and the Dow fell around 300 points. Chair Warsh has, in the reporting of this week, taken "a more aggressive tone, suggesting additional action is warranted to combat inflation."
And oil has gone back up. WTI is at $90.55 — it was near $82 a week ago — with the Iran conflict back in the headlines. That is an inflation impulse arriving exactly when the Fed is already leaning hawkish.
The recession mechanism we described on August 30 has not gone away; it has changed shape. It is no longer "the labour market is cracking and the Fed can't help." It is now: the labour market is just strong enough to let the Fed tighten, and inflation is just high enough to make it want to. A hike into 3.1% wage growth and $90 oil is how a decent economy gets talked into a bad one.
Odds, moved
United States — recession within 12 months
Cut, and we are executing our own rule rather than defending a position. The published correction rule says event-driven changes are capped at half a move until hard data confirms — a payroll print is hard data, so it earns a full move, and here it moved against us. Held at the upper end of the band because the Fed risk, a 52-week-high 2-year yield and $90 oil are all real.
Canada — recession within 12 months
Unchanged, exactly as the pre-committed rule specified: a hold with balanced-risk language leaves the number alone. Macklem noted the newly tariffed products are only about 5% of exports to the US — smaller than the headlines implied — but warned the uncertainty may make firms "delay investment and hiring decisions." Retaliation took effect September 8.
What we are watching
The rule we are adding
On August 30 we published a correction rule after getting a market call wrong. It worked — it is why Canada's number did not move on Tuesday. Today's miss was a different failure, so it gets a different rule:
No thesis rests on a single monthly print. Payrolls, GDP and jobs numbers are revised, often by more than the figure that made them interesting. From now on, a labour-market or GDP argument on this site must cite the trend — a three-month average or the twelve-month rate — alongside the headline, and any odds change driven by one month is capped at half a move until the following month's revision confirms it.
Had that rule been in force on August 30, we would have cited the twelve-month average of +31,000 rather than one negative month, and we would not have needed to write this section.