More on why jobless claims aren’t yet signalling double dip
Jobless claims have been quite elevated of late, averaging around 450-480,000 for the past few months. While these are worrying figures, they do not in and of themselves point to recession. In my last post I used a chart of the year-on-year change in initial jobless claims to point out that every recession in the last 40 years has been preceded by a sharp rise in jobless claims. And a sharp rise (+50,000) in initial claims has always meant recession.
Right now, 4-week average initial jobless claims are still almost –90,000 lower than they were at this time last year. This doesn’t look like the kind of employment data that has presaged recession in the past 40 years.
A reader asked:
If the economy continues to shed 420K to 490K jobs a month, won’t this steady "drip-drip-drip" loss of jobs accomplish the same thing – aka, a recession [or at least something that feels like one if not "technically" a recession] – as a big spike down of 600K or more lost jobs? The "drip-drip-drip" loss of jobs scenario will take longer, but the effect will be the same. Won’t it?
My short answer is no. Here is a longer answer.
First, we have to keep in mind that we are working from a lower base now. Last year, when talking about recovery I said:
Before the Great Depression in 1929, the U.S. had nominal GDP of $103.6 billion. By 1933, this had dropped to $56.4 billion due to deflation and a decrease in production. Over the next four year, GDP growth soared. It was 17.0% in 1934,11.1% in 1935, 14.3% in 1936 and 9.7% in 1937. That’s some serious growth, right? Well, GDP was only $91.9 billion in 1937, a full 11.3% lower than it had been 8 years earlier. In fact, it wasn’t until 1941 that we attained the nominal production output of 1929.
What this should illustrate is that working from a lower base makes an increase in GDP comparatively easier than when working from a higher base. Yes, it was a powerful recovery, but it did not get the United States back to the same productive level for many years. In that sense, recovery does not mean recovery immediately.
Recoveries by definition start from a point of diminished output because recessions are periods of diminishing output. So output in the initial period of any recovery is always lower. That’s how the math works – and also why I continue to stress that this is a technical recovery.
But, the important part to remember is how the business cycle works and how the recency effect creates self-reinforcing declines or recoveries in output.
Increases in income lead to increases in retail sales which lead to increases in output and inventories which lead to more jobs and thus a further increase in income. This is a virtuous circle that defines the upward path of a business cycle. The same is true in reverse as the business cycle tops out. Decreases in income lead to decreases in retail sales which lead to decreases in output and inventories which lead to fewer jobs and thus a further decrease in income. The self-reinforcing nature of the swings in the cycle owes much to the calculation of businessregarding anticipated future demand in creating goods and services and staffing up or down.
In terms of the demand for goods and services, personal consumption expenditures make up some 70% of GDP (see chart below). When more people earn more money, collectively we have more disposable income. We can therefore spend more on goods and services. (Many may even leverage up or eat into savings and buy more than they earn, causing retail sales to increase by more than disposable income.) This causes retail sales, and thus GDP, to increase. So, by-and-large, GDP is dependent on jobs. So the labour market is crucial.
In terms of the labour market, think of demand for labour. Let’s say you are a business person who owns 15 print shops. Now, if business is so good that you can’t keep up with demand, you hire new people, probably temps at first and then full-time. If this continues for a while, you continue to add staff in anticipation of future business.
Here’s the thing though: eventually businesses miscalculate and extrapolate growth that doesn’t come to pass. Maybe the government raises taxes, maybe income growth slows. In any event, mid-cycle there is always a slump when industry recalculates. Initially, this recalculation doesn’t mean huge layoffs for your print shops because laying off workers needlessly when demand is still increasing is a killer for future business. However, the larger the miscalculation, the larger the recalculation.
We saw these kinds of larger recalculations mid-cycle in 1976-1977 in 1985 and again in 1995-1996.In each case jobless claims rose precipitously and many worried the economy would fall back into recession. However, businesses were able to recalibrate enough that the damage was limited and the cycles continued their upward path.
But, business cycles end when the miscalculations are largest. In the post-war era, usually it was the Fed raising interest rates which has set off the slump that leads to recession. The anticipated future business, inventory and staffing overhang are so large that they result in a huge and sustained upswing in jobless claims. Households can only maintain their desired spending level for so long in such an environment. Eventually, the collective cut back in retail sales and deleveraging is so large that it induces a further retrenchment in staffing and output and the cycle goes into reverse.
In sum, there is a tipping point in every business cycle that sends the economy into recession. In the past, it has been preceded by higher Fed Funds rates and higher jobless claims. The jobs picture is key in the chain of events which leads to recession as it takes a large increase in job losses to trigger the change in spending patterns that induces the leg down.
So, I don’t see a 450,000 jobless claims number as recessionary in isolation. I think we need to see an uptick in claims for the jobs picture to be the critical piece in inducing recession – something that I could see happening in 2011 (50-60% odds I say).
This cycle will probably be different though. Right now consumers are continuing to reduce their debt loads even though nominal GDP has turned up. This is true deleveraging as Household debt to GDP is declining. See Comstock Partners’ The Significance of Consumer Deleveraging. Could this induce a double dip without our seeing jobless claims tick up too? I doubt it because I see jobs as the critical factor. But I will be watching the numbers.
Uh no. A few things.
First, looking at yoy change at this point is silly. That graph is useless. Look at the level, or the change in level of late. A year ago is hardly a reasonable comparison.
Second, if we are still shedding jobs at this high level, near the level which was the peak for the last recession, and the employment base is now (sadly) about two million jobs LOWER than it was heading into the last recession… that’s also relevant.
Third, we came down to about 450k earlier this year, then started rising, not surprisingly when stimulus started fading, along with the economy in general. That we’ve bounced down a bit of late is good news, and confirms your point that these number dont presage recession. However, they remain far to high for recovery. If they start going up again, as they were doing a little while back, then, Houston, we have a problem.
Ultimately, you have to ask yourself WHY these numbers are either going to go down, or up. Why would people stop laying off workers when many many businesses have not yet adjusted to the new low level of demand? And why would anyone start hiring when demand is faltering?
Geoff, you sound like you have experience with the data given how sure you are in your statements. But my analysis is based on how the data has performed. I have written about this extensively in the past. It is the change in jobless claims which is relevant. Here, Robert Gordon, who sits on the NBER and helps make a determination of recession dating, says the same thing:
The reason your logic doesn’t work here is because GDP growth is a first derivative statistic i.e. it measures the change from one period to the next. You can’t just look at the level of claims, you have to look at the change in level. For instance, in the 1982 recession claims peaked above 600,000 while they were only above 400,000 in the 1990-91 recession.
By the way, if you look at job losses from the non-farm payrolls data set, for example, you would also see that the economy lost jobs well into 2003 without triggering a double dip recession (fiscal and monetary stimulus helped).
The point it is the change in jobless claims actually are the best indicator, having been correct for every recession and recovery since 1967 without false positives.
sorry, but you just arent going to get me to agree with your YOY argument. The change IS relevant. You said that, I said that, and RG said that. But the YOY isnt the right change to look at. You have to look at the trend change. Im probably nitpicking here, but i see the Year ago change stuff throw out all the time, and most of the time, its misleading. Its ok if the data isnt seasonally adjusted, but I believe this data is. Again, if you start seeing jobless claims go up from here, on an adjusted basis, that is, for the size of the now diminished emplyoment base, then yes, we’ve got a problem. The year ago change depends entirely on the trend a year ago. If for some reason last year at this time, there was a decline, then even if claims are flat now, you’ll get a rising yoy change. Is this starting to make sense? :)
Geoff, one more point: you said,
“First, looking at yoy change at this point is silly. That graph is useless. Look at the level, or the change in level of late. A year ago is hardly a reasonable comparison.”
That’s actually not true. There have been numerous periods in which claims increased modestly mid-cycle. I discussed a few (1977, 19885, 1996). We saw a similar rise in 2003. But in none of these cases was the rise significant enough to trigger recession. What you need is a rise that is large over a significant enough time frame to induce a retrenchment in consumption that feeds into business expectations.
Again, your choosing mid cycle rises does not refute my point. What you say is accurate, yet, you need to fous on the level, not the year ago comp. Yes, we’ve seen this before. Yes, we saw it a few months ago, yes, we could see it again. The mid cycle rises are not necessarily recession inducers, true. But that is only because they arent sustained, or dont continue to rise.
I agree with your overall point that “jobless claims aren’t yet signalling double dip”, and I liked your detailed explanation of mid cycle slowdowns that may not lead to a reversal in the cycle.
However, while the “sharp rise (+50,000) in initial claims has always meant recession” is a valuable observation, I’m not sure why the year-on-year chart has any value in this context. Given that we measure GDP quarter-to-quarter, the year-on-year change is a lagging indicator of recession (the chart even confirms this visually), and doesn’t say much about whether aggregate income has begun dropping in the last couple quarters, or will begin dropping shortly. Now I see your discussion with geoff, and would just add… yes, not every mid-cycle increase in claims leads to recession, but the YoY measure still lags the start of the recession in the cases where a recession DOES ultimately unfold, so it doesn’t help much in discussing whether recession is imminent or has recently started.
Also I’m not certain I would have used this terminology: “I think we need to see an uptick in claims for the jobs picture to be the critical piece in INDUCING recession” since there are many possible recession catalysts right now, but we’re probably on the same page anyway as clearly job losses have to come into play as a result of these first-stage catalysts (even different ones than over-optimism by business), and then the losses become a new catalyst adding to the effect in the subsequent time periods. I would argue there are situations where the first-stage catalysts are powerful enough to start a recession on their own, especially if growth momentum is slow to begin with, but maybe that’s just semantics or simply irrelevant, as we can’t really measure the time periods with that much granularity.
Incidentally (since you showed a table of GDP components) I recently graphed contribution to change in real GDP of household consumption, government spending, net exports, investment, and inventories for the US and post-1989 Japan, and you might find the graphs interesting (post-1989 Japan doesn’t look as bad as conventional wisdom suggests, hence double-dip is far from inevitable for the US today, though there are still some unique circumstances today that could change our outcome):
hbl, the year-on-year change is a coincident to leading indicator actually. See here:
The reason this is a coincident to leading indicator is that it’s the change in the number we care about. The level of the change determines whether it is leading, coincident or lagging For example, if you told me the year-on-year change had risen +20,000 I would say that a further rise could spell recession. However, if you told me the year-on-year change had risen +75,000 I would say that recession had likely already begun. So again, the y-o-y measure does not lag. It depends on the level of change you measure.
When I started looking at this metric in 2000 I used the 6-month figure but I have started to use the year-on-year because I prefer the unadjusted numbers.
Either way, the effect comes about because of the change in spending habits and the change in business decisions they induce. When these changes reach a certain “breaking point” for more than a few weeks which I am pegging at +50,000 a recession has been a lock every time.
I should also add that measuring quarter-on-quarter jobless claims numbers is actually going to induce a lot of false positives. I’ve run the numbers. The time period simply isn’t long enough to take the lagging effect of job losses to business decisions, inventories and output into account.
Thanks for the linked post with numeric data points. It does support your description of jobless claims as a “coincident to leading” indicator in the majority of cases. Theoretically I still think it would be possible for a recession to start for which YoY jobless claims were a lagging indicator (to a greater extent than your 1970 example of 3 month lag), but it does appear to be unlikely in practice based on your examples.
Also, remember, I could just move the bogey down a notch to get date correct. When I came up with the 50,000 number I did so because it was a good round number that people could remember and to be able to say with certainty a recession was coming/already here (I was doing the economics for a jobs company).
The reality is that if you said the number was 35,000 or 40,000, that would be more accurate. And then you would get coincident to leading data every time except 1981 which lagged because of the uniqueness of the double dip.
By the way, there was some sort of auto strike in 1977 that saw the number rise precipitously (but only for a few weeks). That reinforces the notion that the feed through happens with a lag.
I have run the same test on the change in unemployment and it too could be a coincident indicator depending on the change threshold you use.
hbl, just to clarify, I am saying the following:
1. Sudden large rises in jobless claims for an extended period (two months) induce a pullback in consumption
2. If the pullback is long enough, this induces a pullback in output as revenue stagnates and inventories rise.
3. If these effects are great enough all around, another round of layoffs results and recession takes hold through the down cycle of reduced income and inventories.
4. In looking at the data over the past 10 years, I have found that 3-month change data are too noisy and lead to false positives while 6-month and 12-month data do not.
5. Originally I used the 6-month but I prefer the 12-month now because it allows me to use the unadjusted series and strip out seasonality (admittedly this skews the data to being more coincident than leading)
6. Every time there is a recession, we see the data for 12-mo and 6-mo rises confirming this. On only one occasion was the 12-mo data lagging. There were no false positives.
I also did this exercise with continuing claims with the threshold at 200,000 eliciting the same results, which you can see here:
hbl, one more thought:
if we do double dip, the 1980-82 experience says the jobless claims numbers may lag. Everyone knows unemployment numbers are lagging. What isn’t usually lagging is the first and second derivatives. And that’s what I was trying to get at here. It’s tricky because I’m really trying to catch the first uptick in job losses that causes the consumer retrenchment before the dip.
The thing is that a double dip is so unique. And that’s what we saw in 1981 when the claims numbers were still relatively low when — bang — the second dip happened and they shot up overnight. In 1981, businesses were not done purging and fixed investment fell dramatically in the 2nd quarter, even though it rose in Q3. The damage was done and things fell apart in a hurry.
So, all of this is to say, the change in jobless could well lag because the contours of this cycle just don’t fit well. I am realistic about that. Moreover, this quarter already is at about zero growth. What happens in Q4 then? I am still pretty worried.
Thanks for the follow up thoughts! It’s useful to see historical specifics in particular.
It does seem like the slower the current growth rate of GDP, the smaller the increase in jobless claims that would be needed to indicate a contraction (hope I have that straight…). And I agree the way things look now there’s still a lot to worry about in Q4 and beyond…
By the way, good for you taking well deserved time off last month!
I am little confused on the double-dip points you mentioned. Can you pls make it clear? Would appreciate.
Are we heading for double-dip or not? If not, when we can see the growth?
If heading for double-dip by when we can see? What implications it would have on economy and on people? Would you see this make US an undeveloped country as famous trend researcher, Gerald Celente says? Is this double-dip different from past incidents? If so in what way..
The last sentences outline my view best:
“So, I don’t see a 450,000 jobless claims number as recessionary in isolation. I think we need to see an uptick in claims for the jobs picture to be the critical piece in inducing recession – something that I could see happening in 2011 (50-60% odds I say).
This cycle will probably be different though. Right now consumers are continuing to reduce their debt loads even though nominal GDP has turned up.”
The U.S. economy is already operating at stall speed with consumers still in deleveraging mode. States and municipalities are shedding workers. Stimulus has been the only thing holding things up.
Normally jobless claims start to rise before a serious decline in output comes to pass. Unlike the unemployment rate, the change in claims are coincident or lead rather than lag the onset of recession. Businesses at cycle ends are conditioned by events and reduce output and staff only with a considerable lag.
However, I anticipate jobless claims will be lagging like they were in 1981.Now they have not been conditioned and I believe they will act rather quickly to any dip in output. What could cause that dip: housing decline, tax increases, protectionism, etc. And if that dip occurs, businesses would react and the resultant consumer retrenchment would induce a severe decline in output and rise in unemployment.
So I will be watching the numbers but this cycle they may not tell me anything until it’s too late. For all of us, that means we have to be cautious and look elsewhere for signs of recovery.
Comments are closed.