February and March construction spending show two leading sectors in decline; only AI spending holding up the economy

 

 - by New Deal democrat

It has become increasingly likely that the Boom (or maybe Bubble) in spending on the construction and operation of AI data centers may be the only thing that has kept the economy out of a recession. This morning’s release of both February and March monthly numbers for construction spending (thus bringing this metric almost completely up to date six months after the end of last autumn’s government shutdown) is more evidence for this proposition. That’s because, in the past I have used construction to help track the long leading sector of housing; and in the wake of the Inflation Reduction Act, plus “Liberation Day,” it has also been useful to track manufacturing. But now, via tracking construction of water supply and power, it is also a useful proxy for construction of AI data centers.

On a monthly basis, in March all but manufacturing construction were positive, but since the last report was for January, comparing the two month change is also significant. Here’s how total construction, and each of the above sectors, broke down in March (first column) as well as the combined February and March period (second column):

Total: +0.6%     +0.1%

Residential: +1.6%     +1.5%

Manufacturing: -1.1%     -2.7%

Power: +0.2%     +0.5%

Water supply: -3.4%     -0.4%

In the below graph, I break out each of the above for the first three months of this year. But the above numbers are nominal. Measuring vs. the cost of construction materials is also important, and those are also supplied in the final line:

Deflated by the cost of materials, which rose 0.8% in March, only water supply and residential construction were positive. For the two month period including February, the costs of construction materials rose 2.1%, meaning in real terms *all* of the sectors were negative.

The importance of AI data center related spending also is apparent in the YoY% comparisons: nominally total construction was only up 1.6%, vs. a 6.0% increase in construction materials. Manufacturing was *down* -17.0% (thank you, tariffs!). Residential construction was a relative bright spot, up 3.5% (but still below the cost of materials). Only power, up 5.3%, and water supply, up 4.8%, fared better, but even those were negative in comparison with construction costs:

Notably, as shown in the graph below, in nominal terms even spending on water supply construction peaked last October, and total and residential construction spending peaked in December. Only power construction spending has continued to increase (but even that not in real terms, as discussed above):

And in real terms, with the exception of a brief rebound this past autumn, residential construction spending has been declining since May of 2024, although like other housing metrics, there are signs it may be bottoming out:

In other words, spending in the two leading sectors - housing and manufacturing - have been in decline for well over a year. If AI related spending rolls over as well, it is difficult to see how the US economy remains out of recession. 

Jobless claims, the most positive data of all, continues to augur for lower unemployment

 

 - by New Deal democrat

The most positive metric in all of economic metric-dom continues to be very positive. Initial jobless claims rose 10,000 to 200,000 last week, still among the lowest readings over the entire past 50+ years. The four week moving average declined -4,250 to 203,250, also among the lowest in the past 50+ years. And continuing claims, with the typical one week delay, declined -10,000 to 1.766 million, the lowest since January 2024:

For all intents and purposes, nobody is getting laid off.

On the YoY% basis more important for forecasting, initial claims were down -12.3%, the four week average down -10.5%, and continuing claims down -5.9%:

This is very positive. In fact, YoY values this much lower have in the past been associated with economic expansions when they were very strong, as shown in the below graphs which norms the current YoY% changes to 0:

Interestingly, the only exception was 1973, when the Arab Oil Embargo caused the economy to reverse quickly.

Finally, the big decline in jobless claims over the last few months *very* strongly suggests not only that the unemployment rate will not increase from its 4.3% level in March, but potentially could decline all the way to 4.0% in the next several employment reports:

We’ll see how that plays out for April in tomorrow’s jobs report.

The positive, noisy monthly March new home sales report masks underlying trend weakness

 

 - by New Deal democrat

New home sales, which were finally updated yesterday for March (and so still are about three weeks behind their regular schedule now 6 months after the end of the government shutdown!) are perhaps the most leading of all indicators for housing, itself a long leading indicator. In fact, they are so leading that they are more of a “mid-cycle” indicator, as their monthly peak usually happens closer to the mid-point of a cycle than they end. But as I almost always point out, they have two major problems: (1) they are very noisy, and (2) they are heavily revised. Which means you must always take the latest month’s number with several grains of salt under the best of circumstances.

With those caveats out of the way, let’s take a look at yesterday’s numbers. The headlines were that sales increased sharply, from 635,000 annualized to 682,000. The median price on a non-seasonally adjusted basis declined to 387,400, a -6.2% YoY decline. And inventory declined ever so slightly, by -2,000 annualized to 481,000 units. 

Here’s a more in-depth look.

First of all, home sales react to mortgage rates, as shown in the first graph below of the last three years:

As mortgage rates (red, right scale) generally declined, especially later in 2025, new home sales increased (blue, left scale). With the increase in mortgage rates over the past two months, I expect the three month average of new home sales to start to decline again.

The trend is somewhat clearer when we average new home sales quarterly (light blue line below). Although they comparatively lag a little, the signal is also distilled quite clearly from the noise by single family permits (red, right scale):

Permits had just started to pick up in the last few months, a few months after new home sales, before the increase in mortgage rates hit. But the Q1 average of new home sales did pick up a decline. We can expect this (noisily!) to continue in Q2.

The next metric to follow sales are prices, shown below historically both monthly (light blue) and quarterly (dark blue) to help smooth out some of the noise:

Prices do go down, or at least advance at a decelerating rate, late in expansions as typically higher interest rates kick in. Now here is the post-pandemic view:

We can see that prices peaked in Q3 2022 and have generally been declining on a consistent basis each quarter thereafter. Faced with a downturn in sales (due to high prices after the pandemic and then a big increase in mortgage interest rates), builders adapted to the market. As of Q1 of this year, the median price of a new home was -8.9% lower than its Q3 2022 peak. While the trend in mortgage rates before the Iran war had been downward, and might yet be reflected in a temporary firming in prices, it seems very likely that the downward trend in prices will resume with the higher interest rates, and in some cases costs of material, due to the Iran war disruption.

The final domino to fall is new houses for sale, i.e., inventory (orange in the graph below). The historical look shows the inventory ultimately always follow sales (blue) with a significant delay, and almost always have started to decline significantly before the onset of recessions:

The post-pandemic view shows that inventory did indeed peak in March and May of last year and declined since, down -4.6% as of March:

But this last graph also shows something interesting — namely, inventory stabilizing. This, by the way, is something that has also been true in the last several months for single family houses under construction (blue):

The historical look suggests that units under construction turns before inventory:

In the present case, it appears that inventory may have made a bottom a month or two before units under construction. If we go back to the long term historical graph above, inventory has typically bottomed after the end of recessions. In other words, as recessionary as housing has been for the last year, this suggested that the danger may have been passing, i.e., the economy may have dodged a bullet.

But again, if the Iran war continues to cause interest rates to be elevated, then sales will turn back down, price reductions will continue, and inventory will continue to decline. 

March JOLTS report: reverting to 2025 averages

 

 - by New Deal democrat

The JOLTS report is low on my list of useful tools, but it does break down the labor market further than the jobs report, and it does have several slightly leading components, so let’s at least take a brief look at this series, which *finally* was updated to its normal last report (for March) on what would be on schedule prior to the government shutdown  last autumn.

Below are job openings (blue), hires (red), and quits (gold) through March, all normed to 100 as of the onset of the pandemic:

Job openings seem to get the lion’s share of attention from most commentators, but I treat them as somewhat fictional. In any event, they declined -56,000 to 6.866 million, the lowest since the pandemic, except for last December. On the other hand, both actual hires and quits rebounded in March, by 655,000 and 125,000. In the case of hires, it was the highest reading in two years after its post-pandemic low last month; in the case of quits it was about average for the past year. 

Layoffs and discharges, on the other hand, rebounded sharply from their lowest numbers of the past 12 months at the end of last year, up 153,000 to 1.867 million:

This is in stark contrast too the extremely low level of new jobless claims we have seen since November, including a new 50+ year low last week. In this case I put more stock in jobless claims, which are both more timely and much less noisy.

Finally, the quits rate (blue) tends to lead the YoY gain in hourly nonsupervisory wages (red). Here is the post-pandemic close-up:

With quits increasing back to their 2025 average, this counts as good news, since it suggests that YoY wage growth is likely to stabilize. At the same time, I have to caution that wage growth is a lagging indicator, remaining low after recessions until there is a sufficient decline in unemployment, and remaining at expansionary levels until employment has decidly rolled over.

Economically weighted ISM services + manufacturing indexes strongly stagflationary

 

 - by New Deal democrat

We got a bunch of data this morning. I’ll discuss the JOLTS employment data from last month, and the ISM services report today. Since new home sales are a long leading indicator, I think I can safely wait to discuss that report tomorrow.

Let me start with the ISM services report in this post. I’ll put up a separate one to discuss the JOLTS report.

To begin with, recall that services make up roughly 75% of the US economy. As a result, downturns in manufacturing aren’t nearly as important as they used to be. So for forecasting purposes I use a weighted average of the ISM manufacturing (25%) and services (75%) reports, especially the three month average. For the last few months - including this month - that hasn’t been important, because the manufacturing index has rebounded above 50 into expansion territory. But what did happen is that several aspects of services became more negative.

Let’s start with the headline number, which declined -0.4 to 53.6. The three month average declined slightly, by -0.1, to 54.6 (gray). In the below graphs I also show the equivalent number from the ISM manufacturing report in blue:

New orders declined -7.1 to a still expansionary 53.5, while the three month average actually rose 0.2:

The first piece of bad news was that the prices paid component remained at 70.7, the highest level in three years, and the three month average rose 3.4 to 68.1. This means that price increases in the services sector, just like the manufacturing sector, are widespread:

The combined average is the highest since the first half of 2022. For comparison, here is YoY CPI for the past five years:

In mid-2022, CPI was running over 1% *each month(!)* and 8% YoY; and even ex-shelter was over 9% !

The second piece of bad news was that while employment rose 2.8, it remained in contraction at 48.0, while the three month average declined -0.7 to 48.3:

Since the three month average for the manufacturing index was 49.0, this means that the economically weighted average is back in contraction for the first time since it briefly was last summer. For comparison, here is nonfarm payrolls for the past five years:

Last summer the economy started shedding jobs for several months.

In summary, while the headline and leading new orders number in the ISM services index remained positive, stagflation was very much front and center, in both stagnation in employment and higher inflation in pricing.

This was not good.

Long leading indicators in Q1 GDP point in absolutely opposite directions

 

 - by New Deal democrat

The advance report for Q1 GDP was a study in contrasts, both in terms of the leading indicators and also the sources of strength.

Let me start with the general information: on a nominal basis, GDP in Q1 increased a mediocre 1.4% at an annual rate (blue). After adjusting for inflation, in real terms it only grew 0.5% annualized (red):

This is one of the poorer performances since the pandemic. On the other hand, since Q1 of last year was even worse, on a YoY% basis, growth in the GDP improved a little both nominally and in real terms:

But GDP growth was even more mediocre than the top-line measure makes it look, because in Q4 we had the lengthy government shutdown. That spending resumed in Q1, so government spending bounced back sharply:

Only in four quarters in the past five years did government spending grow more sharply than in Q1. Had the spending growth only been at “normal” post-pandemic levels, real GDP growth in Q1 would have been basically *0%*.

But wait, there’s more!

I’ve been writing for many months that the housing sector has been outright recessionary. That was further confirmed in this report. In nominal terms, private fixed residential investment (the GDP version of housing) declined -0.9%. In real terms it declined -2.1%. The best way to look this as a long leading indicator is as a share of nominal (blue) or real (red) GDP:

Either way, this was the lowest contribution to GDP since the recession. To beat this particular dead horse one more time, it was recessionary.

So, in addition to the rebound in government spending, what saved Q1 GDP? Corporate spending and profits. 

As per usual, corporate profits (red) aren’t reported in the first advance GDP report, but a reasonable proxy, proprietors’ income (blue) is, and that income rose 1.8%:

Only three times in the past five years did it increase more on a quarterly basis:

This strongly suggests that corporate profits in Q1 rose sharply as well, something that has been apparent in the quarterly corporate profit reports to Wall Street:

And the GDP report showed that the type of spending associated with the construction and operation of AI data centers, information processing equipment and intellectual property products, both increased sharply, with the former at a near record increase:

Update: The surge in corporate spending was also apparent in the final March report for manufacturers’ spending on durable goods orders (blue), capital goods (red), and consumer goods orders (gold), which increased 0.8%, 3.4%, and 2.4% respectively, the core capital goods segment to an all-time record high:

To sum up, Q1 GDP likely would have been all but flat were it not for the rebound in government spending. Further, the two long leading indicators in the report, residential investment and corporate profits, pointed in starkly opposite directions. The former was recessionary and the latter showed a Boom. 

The rebound in government spending will not be repeated in Q2. The direction the economy takes from here will depend on how long the corporate-led AI data center Boom can continue, vs. how battered consumers are by higher inflation.

Weekly Indicators for April 27 - May 1 at Seeking Alpha

 

 - by New Deal democrat

My “Weekly Indicators” post is up at Seeking Alpha.

It is surprising how positive most of the high frequency data is, despite the continued shutdown of the Strait of Hormuz and all of the impending shortages in energy and other important commodities have failed to make a dent in most of them. Even mortgage applications have rebounded somewhat.

As usual, clicking over and reading will bring you up to the virtual moment as to the state of the economy, and bring me a penny or two for my efforts in putting it all together in a coherent format.

April ISM manufacturing: how long can the AI manufacturing Boom keep exploding prices from creating a consumer implosion?

 

 - by New Deal democrat

April data started out as usual with the ISM manufacturing index. There was some good news, but mainly bad news.

The good news was that the headline ISM number (blue in the graph below) remained positive, and was unchanged at 52.7 (recall that any number above 50.0 indicates expansion). The more leading new orders subindex (gray) did rose from 53.5 to 54.1, suggesting the modest Boom - probably 90% of which is related to AI data center construction - will continue. The three month averages, which smooth out a little volatility, were steady at 52.6 and slightly lower at 54.5, respectively:

I am convinced that all of the activity surrounding AI data center construction and operation, and the affluent consumer spending secondary to the (narrow) stock market Boom associated with it are the only things keeping the US economy from being in recession at present. In any event, this number suggests continuing expansion in the goods producing sector for the next few months.

That was the good news. Now here’s the bad news.

First, the “less bad” trend in goods producing employment over the past few months reversed in April, with the employment subindex declining -2.3 to 46.4, the lowest number in four months:

Further, as the above graph shows, the longer term downtrend in manufacturing employment may have reasserted itself.

But that wasn’t even the worst news, which was in the prices paid subindex, which rose sharply, by 6.3, to 84.6, the highest number since  May of 2022, and one of the 6 highest numbers in the entire past decade:

This is a very sharp inflationary pulse, which is going to pass right through into consumer prices for goods.

With this kind of inflationary pressure, I am skeptical of how long the AI Boom can continue to resist a possible consumer implosion.

Blockbuster initial jobless claims report suggests unemployment could decline all the way to 4.0%

 

 - by New Deal democrat

I wasn’t able to get to either the GDP report or the jobless claims report yesterday. I’m going to hold off on the GDP report until next week, because there was a lot going on, but this morning (before the ISM manufacturing Index comes out) let’s take a look at jobless claims.

Yesterday’s report of only 189,000 people making new jobless claims last week really deserves its own special mention. Why? 

Well, not only is it the lowest - by 1,000 - of the entire post-pandemic period (note below graphs subtract 189,000 so that the current number shows at 0):

It is also the lowest since September 1969, almost 60 years ago!:

That’s pretty wild.

The four week moving average declined to 207,500, which while very low, is still above several weeks earlier this year and also in January 2024. Continuing claims, with the typical one week delay, declined to 1.785 million:

Here is the YoY% view most important for forecasting purposes:

Initial claims are down a whopping -20.9% YoY, while the four week average is down -8.1%, and continuing claims are down -6.3%.

This is very positive - indeed the most positive datapoint in the entire economy.

As per usual, let’s update what this suggests about what will happen with the unemployment rate in the next several months, which will be reported next Friday for April:

One year ago the unemployment rate was 4.2%. The big downdraft in initial and continuing claims suggests that the unemployment rate, which was 4.3% last month, will decline, possibly all the way to 4.0% in the next few months.

Stagflationary real personal income, real spending on durable goods were recessionary in March, but consumers and manufacturers haven’t retrenched

 

 - by New Deal democrat

Personal income and spending are among the most important monthly indicators of all, because they give us a detailed look at consumption by the broad range of American households. And since consumption leads employment, they also give us an idea of what is likely to happen with regard to jobs in the near future. This morning’s data for March was the first that represented activity including at least part of the Iran war oil shock. 

In March, nominally personal spending rose 0.9%, while personal income rose 0.6%. Since the PCE deflator increased 0.7%, however, spending was only higher by 0.2%, while real income declined by -0.1%, the second decline in a row, taking the absolute number down to its lowest since last July:

Further, on a YoY% basis, real income was only up 0.2%. Aside from months where it was affected by tax law or stimulus programs, such a low increase has only happened during recessions. Real spending was up 2.1%, the lowest in over two years except for December:

Once we exclude government transfers — one of the important coincident metrics used by the NBER to date recessions, real personal income also declined less than -0.1% to the lowest level since last July:

On a YoY basis, this was now down -0.1%. Again, aside from comparisons distorted by tax law changes or stimulus programs, this has only happened during recessions:

Another important component of the data is spending on goods, and in particular durable goods, which is a leading indicator. Historically, the pattern has been that real spending on goods (red in the graph below) turns down in advance of recessions, and in particular spending on durable goods (orange), which tends to turn down first. Real spending on services (blue) has tended to rise even during all but the most prolonged or deep recessions. These have been flashing red warning signals. 

In March, real spending on services rose only 0.1%, but real spending on goods rose 0.6%, and on durable goods rose 0.9%. The below graph shows the post-pandemic record, normed to 100 as of March of last year:

Keep in mind that gasoline is a good, explaining the substantial increase in goods spending in March. Nevertheless, on a YoY basis real spending on goods was only up 0.7% and real spending on durable goods *declined* -0.6%:

Once again, this last number was recessionary.

The Iran war came a-callin’ for PCE inflation, as the deflator increased 0.7%, causing YoY PCE inflation to rise to 3.5%:

Meanwhile, the personal saving rate - i.e., the portion of income left over after spending, declined -0.3% further in March, to 3.6%, the lowest such rate in over three years, and a low rate which has only been exceeded during the 2005-07 period and during 2022:

Basically, in order to deal with the spike in gas prices, consumes got further out over their skis, leaving them vulnerable to any further shock, but explaining why the economy did not contract last month, as typically sharp retrenching by consumers is also something that happens just before the start of a recession.

Finally, the updating of the PCE deflator also allows for an update to another important coincident indicator used by the NBER to consider whether the economy is in recession or not; namely, real manufacturing and trade sales, which is delayed by one additional month. These increased 0.6% in February to another new all-time high:

This is of a piece with the recent rebound in manufacturing data we have seen in things like durable and capital goods orders as well as manufacturing production. 

There are two very big trends in this data. First, the consumer is in real trouble, as real income is faltering, as is spending on goods (ex-energy), and in particular on durable goods. As I noted above, this is recessionary, although consumers have not (yet) started to retrench. On the other hand, manufacturing — likely aided by AI data center construction, and as of March likely also aided by oil company profits — continues to Boom. It is likely only this that has kept the economy technically out of recession.