‘Level two’ chaotic systems, polling, and election results

 

 - by New Deal democrat

There’s no important economic data today, so let me post something that I meant to write about over the weekend before I was distracted by real life.

In his book “Sapiens,” Yuval Noah Harari writes that 

[M]any people [ ] prefer history to be deterministic. Determinism is appealing because it implies that our world and our beliefs are a natural and inevitable product of history….

[But h]istory cannot be explained deterministically and it cannot be predicted because it is chaotic…. [H]istory is what is called a ‘level two’ chaotic system. Chaotic systems come in two shapes. Level one chaos is chaos that does not react to predictions about it. The weather, for example, is a level one chaotic system….

Level two chaos is chaos that reacts to predictions about it, and therefore can never be predicted accurately. Markets, for example, are a level two chaotic system…. [because the better the history of a forecasting system, the more it will become a ‘nowcast,’ as people front-run the prediction].

Politics, too, is a second-order chaotic system…. Imagine … some political scientist in association with a computer wizard have developed an infallible algorithm [that predicts the political future. A politician on the losing end of that prediction, for example, would take pre-emptive measures to ensure that the prediction does not come to pass….]

(Sapiens, pp. 239-241)

This is exactly what happens with political polling, and the fate of poll aggregators is a case in point. Nate Silver was initially very successful. But soon, players with other interests reacted to his system. First, and with increasing frequency, partisan pollsters ‘flooded the zone’ with junk polls, forcing him to weigh pollsters by accuracy, and even to develop a slew of alternate and inconsistent forecasting systems. But those junk polls also influenced voters to decide that either a race was competitive, or that one candidate was sure to win or lose. And now we have seen the rise of outright fraudulent polling, to the same end.

The net result has been that these poll aggregations have been increasingly off-target. For example, Sam Wang famously said in 2016 that “It is totally over. If Trump wins more than 240 electoral votes, I will eat a bug.” (and to his credit, he ate a honey-dipped cricket on CNN four days after the election. 

What happened in 2016, essentially, was that “everyone knew” Hillary Clinton was going to win. And because “everyone knew” that, some voters felt free to stay home. Others felt free to cast a protest vote for Trump. Clinton herself felt free not to campaign in the “safe” upper Midwestern States. And most notoriously, FBI Director James Comey felt free to hold a press conference announcing a renewal of a criminal investigation into Clinton while early voting was already underway!

Now in 2026 we are being told that Democrats are heavily favored to take the House, and more likely than not to take the Senate. We are being told that early turnout is heavily in favor of Democrats, and that especially Hispanic voters are turning out in those areas.

But we are not just the passive recipients of these messages. We should fully expect that both potential voters and likely non-voters are going to *react* to that message as well. For example, might some Democratic voters in swing States relax? Will GOP operatives flood likely GOP voters’ mailboxes and social media feeds with horror stories about the big early Democratic turnout, thus driving more tenuous GOP voters to the polls to counter that early trend? Or conversely might a significant chunk of the public decide that they want to show up to be on the winning side, or turn a wave into a tsunami? 

We don’t know. But what we *should* know is that the very act of publicizing all these polls and early turnout models is going to significantly affect the ultimate result. Ask Brazil, where reporting and polling after the first round of balloting that Bolsonaro was almost sure to be elected drove a massive outpouring of votes for Lula.

For my part, I am hopeful. But I am thoroughly discounting all of these polls and dissections of early voting.

A century ago, early industrial psychologists ran a study of factory workers. They found that, e.g., increasing the light on the factory floor made the workers more productive. But it turned out that, more than anything else, it was the act of *observing* the workers that was the chief reason that the altered their behavior. Because, as a fan of Star Trek, to put Harari’s observation in Ferengi terms, “Hoo-mans are a clever species. When you observe the hoo-mans, they always observe back.”

Weekly Indicators for October 19 - 23 at Seeking Alpha

 

 - by New Deal democrat

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

While the usual recent troika of stock market valuations, low initial jobless claims, and strong consumer spending are holding up the short leading and coincident forecasting time frames, enough deterioration has occurred in the long leading time frame to turn that neutral.  This doesn’t mean that the economy ‘will’ stall in the next 12 months or so, but it does mean that the conditions are in place for a stall to be a significant possibility.

As usual, clicking over and reading will bring you up to the virtual moment as to the state of the economy, and reward me with a penny or two in my pocket for gathering and organizing the information for you.

No, we are not DOOOMED: real usual weekly earnings edition

 

 - by New Deal democrat

As widely reported in the past week, real weekly earnings declined a sharp -2.1% in Q3. For example, here is Menzie Chinn at Econbrowser:

And here, more histrionically, is Mike Koncxal on X:

Is this an important danger sign?

No. Why? To cut to the chase, because even after seasonal adjustment it is the noisiest wage earnings metric there is (light blue in the graph below), and specifically in comparison with average weekly earnings as measured in the monthly jobs report (dark blue), and even more importantly in comparison with the quarterly employment cost index (red). Here’s what they look like since before the onset of the pandemic:

You can easily see that the blue bars not infrequently have both upside and downside quarterly swings well in excess of average weekly earnings, and positively wild swings in comparison with the employment cost index. Note, for example, that real weekly earnings had an outsized increase in Q2 of this year that almost completely cancels out their Q3 decrease. Perhaps more importantly, while both usual weekly earnings and the employment cost index are median numbers rather than an average like the monthly weekly earnings figure, the employment cost index norms by type of employment. Thus, in the immediate wake of the pandemic, many more low wage earners were laid off compared with higher wage earners, causing the other indexes to increase sharply even as unemployment soared as well. Then, as the pandemic gradually subsided and the low wage earners were hired back, those series decelerated back to a more “normal” change. But the employment cost index, which accounted for the change in the mix of employment, remained fairly steady.

As a result, when we measure the changes YoY rather than quarter over quarter, the huge swings tend to even out:

The monthly metric of average weekly earnings for all workers, averaged quarterly, is higher YoY on a nominal basis almost by the same percentage as median usual weekly earnings, but with much less drama.

And the same holds true on a historical basis. Note that the above graph, as well as all those below, show *nominal* changes, to account for the fact that different deflators may be used in the normal reporting of the series.

Here is 1979-99:

And here is 2000-2019:

Again, the “usual median weekly earnings” series is much more volatile, as you can see from the wider swings in the blue bars compared with the others.

This is not to say that Q3 real wages for ordinary workers have been good. As I’ve reported for a number of months now, on a YoY basis, real earnings have been flat or slightly negative. But take the noisy Q3 report with multiple grains of salt. It is likely to revert to the mean compared with the other reports in Q4.

Jobless claims continue near historic lows, forecast sub-4.0% unemployment rate

 

 - by New Deal democrat

Let’s take our regular weekly look at jobless claims. As a brief reminder, I always look at this because, especially in conjunction with stock market performance, it is an excellent short leading indicator, and also leads the monthly unemployment rate. 

And the news continues to be very good especially on a historical basis. Initial claims declined -6,000 to 194,000, one of the lowest numbers in 50+ years. The four week moving average declined -1,000 to 197,750, also one of the lowest numbers in 50+ years. With the typical one week delay, continuing claims rose 25,000 to 1.728 million, which remains also well behaved:

On the YoY% basis more important for forecasting purposes, initial claims were down -16.0%, the four week average down -13.2%, and continuing claims down -11.2%:

Note that these comparisons have been getting *even more* positive in the past several months, indicating a strengthening jobs market.

Finally, we’re far enough along in the months to update what this likely portends for the unemployment rate (orange, thin line) for the next few months. Since the unemployment rate is a rounding of the actual ratio of unemployed to the labor force, I’ve also included that (dark red):

We’ve had a slight contratrend move in the last several months in the unemployment rate, likely reflecting the upward tick in new jobless claims earlier in summer. But it is almost certain that the downward trend in the unemployment rate will continue, likely under 4.0% in the next few months.

In short, despite the continued chaotic economic lunacy coming out of Washington, it is extremely hard to derail the US economy. The bigger danger over the next months is that AI will prove to be a bubble after all, the popping of which would drag both consumer spending and also job growth down with it.

The inflationary expansion is still expanding, and is likely even more inflationary

 

 - by New Deal democrat

There were a couple of reports last week about the manufacturing sector that I didn’t cover then. Let’s take a look at them now. What I am interested in is whether the rebound in manufacturing is continuing, is it more or less inflationary, and is it supported by sales?

First up, both the NY and Philadelphia Fed manufacturing indexes were reported. The below several graphs average the two to minimize noise. Ideally I would like to show the three month average, but unfortunately it is one of the few things that FRED does not indulge.

The headline index average (blue in the graph below) shows that the rebound is very much in place, and if anything has accelerated in the past few months. The more forward looking new orders (red) suggests that the rebound, if not ebbing, might be a little past its peak:

In other words, still strongly positive, and set to continue for a few more months, all else being equal.

The prices paid (blue) and prices received (red) index averages tell a less positive story:

Not only are inflationary pressures still very widespread (in the prices paid subindex), but the indication is that they are increasingly being passed on to customers (in the prices received subindex).

So if there is an industrial expansion, it very much remains an inflationary one - and the inflationary aspect may be accelerating even more.

Now let’s look at the second report from last week: business sales and inventories.

The first thing for me to repeat is that sales lead inventories by several months at least, as shown in the below graph of the absolute levels of sales and inventories (/1.5 for scale) in log scale (note these are both nominal, but since we are interested in the trend, we don’t need to worry about the proper deflator):

Sales always turn up and turn down first. Once business see that their sales have turned, they either increase or decrease inventories in reaction.

But an even better way to show this same data is YoY, as below:

This makes it easy to see that inventories change, up or down, after sales do.

Now here is a close-up of the last few years in absolute terms:

Note that sales have stalled since May, declining slightly and then, as of August, the last month reported, only 0.5% higher. While this isn’t a downturn, and there have been a number of similar sideways moves before, if this were to continue several more months I would expect inventories to react as well - which would mean a cutback in production and, if pronounced and long enough, in employment.

Which brings me to my final graph, which is the average of the employment subindex in the two regional Fed reports:

Manufacturing employment has continued to grow at the best pace since the immediate post-pandemic Boom, if at a slower pace than at its June peak. The three month average is slightly down from there, but still quite positive.

In summary, the business sales and inventory report, and the NY and Philadelphia Fed manufacturing reports, show that the inflationary expansion is still both an expansion and inflationary, with employment growth as well. But new orders suggest that in a few months it might begin to decelerate somewhat, exacerbated by customers reacting to increased costs; and if that shows up further in sales, then production and employment might then begin to flag as well.