The warning light went red for one day. A rule change, not the market, turned it back.

Where we are

The market-risk light reads YELLOW today. It turns RED only when two things are true together: households are gloomy (the University of Michigan consumer survey below 55) and the money the world's four biggest central banks hold, counted in dollars, is shrinking. The first is true. The second, as the light now reads it, is not.

It was RED on one day this week. On September 24 the weekly money reading came in as shrinking, and under the rule at the time one reading was enough, so the light went red. On September 25 we changed the rule: the money reading now changes only when two weekly readings in a row agree. The light went back to YELLOW that day and has stayed there. The history keeps September 24 as RED, the colour we published that day. This was our change, not a move in the market, and it is why this edition says so first.

Why the change: the weekly money reading keeps flipping. Over the last five Mondays it read shrinking, flat, growing, flat, growing. A light that follows each of those turns red and back on noise. The two-read rule holds the last thing two readings agreed on, which is flat, and this week's 'growing' is one of the two readings it would need to move. The margin is thin: over four weeks the money total is up 0.02%, about as close to the line as it gets.

Households got gloomier

The University of Michigan consumer survey fell from 51.7 to 48.1 on September 25, the lowest reading in the twelve weeks on file here (it was 49.5 in July and 55.2 in August). That is the first condition of the RED light, and it is now further from the line, not closer.

It also moved the growth score. Jawz scores growth from three measures: factory output, new claims for unemployment benefits, and consumer sentiment. Last week two were improving and one was flat. This week claims are still improving (the four-week average of new claims is 202,000), factory output is still flat, and sentiment went from improving to worsening. The growth score dropped from the best of its three grades to the middle one, and the tool's confidence in its read went from high to moderate.

The economy label did not change. It has read 'growth stalling, prices rising' (label FALL) since September 21, ten days confirmed. Growth's direction is still exactly flat, which is why the tool keeps marking it as sitting on the line with 'economy running hot' (label SUMMER). It has crossed that line three times in twelve weeks: July 20, August 24, September 21. Prices are still rising: the Fed's preferred inflation measure, excluding food and energy, is 3.3% over a year (July's figure, the latest on file), and the consumer-price version is rising too. Inflation remains the dominant risk.

The money: the dollar is doing the work

Over twelve weeks the four-bank total rose $196 billion, to $25.06 trillion. Split it and the two parts point opposite ways: the central banks' own balance sheets shrank by $71 billion, and a weaker dollar added $267 billion by making the European, Japanese and Chinese balance sheets worth more when counted in dollars. The dollar's part is the bigger one. The rise is the ruler, not the money.

The US-only measure — the Fed's balance sheet minus the Treasury's own account at the Fed minus cash parked overnight at the Fed — is $5.77 trillion, down $72 billion over twelve weeks, and reads as shrinking. The Treasury's account stands at $977 billion. This number explains the backdrop; it is not part of the light.

The trend this brief trusts most — the Fed, the European Central Bank and the Bank of Japan, measured the same way at both ends — went from $17.66 trillion on July 6 to $17.62 trillion now, down 0.23%. It fell $0.18 trillion this week alone: the ECB's part from $6.86 to $6.76 trillion and the Bank of Japan's from $4.19 to $4.11 trillion. The Bank of Japan figure is monthly and still August's, so its drop is the yen weakening (156.87 per dollar, the latest rate on file, from September 18), not new policy. China's statement on file is for August, 58 days old; September's is due between October 16 and November 5.

Borrowing got more expensive again, this time from the rate itself

The real cost of borrowing for ten years — the ten-year US government bond yield minus the inflation the market expects — rose from 2.61% to 2.85%, the highest of the thirteen weekly readings on file. It was 2.25% on July 6. This is the seventh edition in a row naming it as the one measure that keeps moving the same way.

Last week it rose because the market's inflation forecast fell. This week the forecast did not move (2.33%) and the borrowing cost itself jumped, from 4.94% to 5.18%. Same direction, the other cause.

Credit noticed a little. Riskier companies now pay 2.80% extra to borrow, up from 2.70% last week and 0.17 points over four weeks; still calm by the tool's measure. Safer companies pay 0.79%. Expected stock-market volatility fell from 15.4 to 14.2, calm. The dollar rose from 118.2 to 119.5 on the broad index in the latest week, and is still down 1.1% over twelve weeks.

What to watch

• The next two Monday money readings. The light turns RED again only if two in a row come in as shrinking, with the four-week change sitting at +0.02% today. It moves to 'growing' if next Monday agrees with this one.
• The ten-year borrowing cost at 5.18%, and whether riskier companies' borrowing costs keep following it up. For most of the summer they did not.
• China's September statement, due October 16 to November 5, and the consumer survey's next final reading on October 30.
• Data quality is clean: every source current, nothing stale or unavailable, the collector healthy, all thirteen weekly rows in each history recorded live. The dollar and exchange-rate figures are as of September 18, the latest the source has published. Two instrument notes. The weekly history samples September 21 and 27, so it does not show the one RED day; the daily record does. And the direction of financial conditions is on a knife-edge: three of five measures say tightening, two say easing, and the rule needs a margin of two, so the live read says 'stable' while this week's history row, taken earlier, says 'tightening'. Neither is an error; both are the same rule reading nearly identical numbers.

The frame, not the trade

The way to be wrong this week is to read the light going back to yellow as the market improving. The market did not improve on these numbers: households got gloomier and borrowing got dearer. What changed is how many readings the light waits for before it believes the money has turned.

The general lesson: when a number you follow changes, check first whether the way it is measured changed. We changed ours on a Friday and said so. Most rulers are not so polite.

So the question for your own book: do you have a rule that acts on a single reading — a stop, a rebalance trigger, a 'sell if it closes below' line? Look at the last twelve weeks and count how many times it would have fired, and how many of those it would have reversed a week later. If the count is high, the rule is reacting to noise, and you might write down how many readings in a row you would need before you act.

None of this is advice and none of it is a forecast. It is a week's data, the reasoning applied to it, and a change to our own instrument printed where you can see it.