« Monetary stimulus vs financial stability is a false trade-off | Main | Macroeconomics when all goods are non-rival »


Feed You can follow this conversation by subscribing to the comment feed for this post.

Amish quilts?

How about car making in the UK. Output per hour worked collapsed from 111.1 in 2008:2 to 93.1 in 2009:1. Output per hour has since increased to 141.4 but this is entirely due to foreign demand (British exports of cars have been booming).

Surely this is evidence of labour hoarding during a demand shock rather than a productivity shock? Posen made this point at the time.


Interestingly, the largely non-tradeable wholesale and retail services sector's output per hour fell from 103.3 in 2008:2 to 98.1 in 2009:1. However, retail productivity has not recovered and was 99.7 in 2012:4.


Could low productivity be caused by low demand?

James: neat! Maybe the whole Amish sector? But I wonder if the data is there.

a: Yep. It just seems so implausible (to me, and others of the AD approach to business cycles) that they suddenly forgot how to make cars as easily as they had the year before. (Though technology can go backwards, like the Romans could make cement, but it took centuries for it to be rediscovered.) And it looks much more plausible (to me) that a fall in demand, plus labour hoarding, is what caused measured productivity to fall.

Nick, the industry that comes to my mind is something like child care, where the basic technology is pretty easy to describe (adult looks after child). That wouldn't work for any number of reasons - regulation of production techniques (max # of children per caregiver), heterogeneity in quality of care etc. But some kind of care or service job is probably the best place to find unchanged production techniques.

I suspect that this is a very good idea in theory, but impossible to implement in practice, and this is why: Any industry that has the characteristics you describe - unchanging productivity - is almost certainly going to be a constant or diminishing returns to scale industry (e.g. child care). It's also going to be one with a fairly straightforward production technology, because the technology doesn't change, and because otherwise individual workers' productivity would be hard to measure. I suspect the labour hoarding doesn't tend to occur much in those types of industries. It would be much more likely to occur in large firms where it's hard to see what each individual person contributes, or how productive any one person is - just think about trying to work out where to make a 5% cut in a university budget (or don't, because it's a nice day, and why ruin it?).

Frances: yes, hmmm. To make it work, it would need to be an industry that is just like other industries, *except* for technology never changing. Because otherwise we could say "OK, there is/isn't labour hoarding in that industry, but that industry is special, and doesn't tell us much about whether there is/isn't labour hoarding elsewhere."

I wonder....what if we took whole countries? Do we get the same sorts of measured productivity fluctuations in business cycles in LDC countries where technology shocks are implausible (but which still have a market/monetary economy)?

I was wondering about hairdressers. I don't think technology has changed there much. (Which is why the relative price has been increasing over time, and is higher in more advanced countries). In recessions, when the demand for haircuts falls, do a lot of hairdressers spend more time still "working" in the shop, but actually just sitting there readings mags and sweeping the floor for the umpteenth time while waiting for a customer?

Nick: "In recessions, when the demand for haircuts falls, do a lot of hairdressers spend more time still "working" in the shop, but actually just sitting there readings mags and sweeping the floor for the umpteenth time while waiting for a customer?"

I would suspect so, yes. But aren't hairdressers paid on a per customer basis? So if output falls, earnings fall - doesn't this change the labour hoarding story? It's not really a firm hoarding labour, it's more like a bunch of independent contractors hanging out together waiting for some customers to arrive.

Frances: I don't know. I thought they got an hourly wage, plus tips. Except the self-employed ones. But I could easily be wrong.

What matters, though, is not how they are paid, but how StatsCan measures their employment (hours worked). Because if StatsCan includes the hours they spend hanging around in the shop waiting for a customer as "hours worked", then measured productivity will fall, even though what's really happening is that they have very short bursts of unemployment.

Nick, WCI does seem to specialize in obscure tangent discussions, doesn't it? Yesterday it was the appropriate per-unit pricing for pizza (diameter, area, volume), today it's hair stylist salaries.

There are basically two models out there.

One is that hair stylists are paid on commission, e.g. this job ad. In Ontario, the employer has to bring that up to minimum wage (http://www.labour.gov.on.ca/english/es/pubs/guide/minwage.php) but I don't know if that's true in other parts of the world. From what I see on-line, it doesn't seem to be.

The other model is straight self-employment. E.g., according to an Ontario government web site "An increasing number of Ontario Hairstylists rent chairs in salons." (http://www.ontarioimmigration.ca/en/working/OI_HOW_WORK_STYLIST_CM.html)

It may *look* like a fall in measured productivity, but I don't really see it as a straightforward test of the labour hoarding model.


Isn't it possible to investigate all this by exploiting the fact industry-specific demands respond differentially to the business cycle? If the AD-plus-labour-hoarding hypothesis is true you'd expect to see labour productivity decline far more in industries where demand is known to be highly responsive to short-term fluctuations in aggregate income. If RBC theory is true...not so much. Or at least RBCers would then have to argue that the productivity shocks that cause recessions are structured so that the biggest shocks always occur where demand is most income-elastic. (Let me guess...that's what they do argue!)

Frances: OK. It's not like labour hoarding in the traditional sense, if you are self-employed or work on commission. But what matters more is whether there would appear to be a measured productivity shock even when we know there isn't. That the decline in measured productivity is a consequence of the fall in demand, rather than an exogenous shock that causes output to fall.

Giovanni: Hmmm. Dunno. I would need to think about that.

In other words, a movement along (and maybe off) the production function, rather than a shift in the production function.

Nick "But what matters more is whether there would appear to be a measured productivity shock even when we know there isn't."

Well, since productivity shock theory was what made me switch from macro to micro ("this is your theory of the business cycle? really????? this is so obviously implausible") I have no problems with that! We can agree it's not conventional labour hoarding, and find a different demand-type story to explain Okun's law - search or matching or rent-seeking or something.


Productivity is typically defined as goods produced per unit of time, which is fine except it ignores the money influence. Employees, suppliers, etc. don't get paid in what they produce, they get paid in money. Since money (in credit based economies) begins as a debt, productivity on a macro scale should be described as output per unit of debt or Real GDP / Total Debt Outstanding. Expressed that way you can see that:

< img src="http://research.stlouisfed.org/fred2/graph/fredgraph.pdf?&chart_type=line&graph_id=&category_id=&recession_bars=On&width=630&height=378&bgcolor=%23b3cde7&graph_bgcolor=%23ffffff&txtcolor=%23000000&ts=8&preserve_ratio=true&fo=ve&id=GDPC1_TCMDO&transformation=lin_lin&scale=Left&range=Custom&cosd=1950-01-01&coed=2013-01-01&line_color=%230000ff&link_values=&mark_type=NONE&mw=4&line_style=Solid&lw=1&vintage_date=2013-05-04_2013-05-04&revision_date=2013-05-04_2013-05-04&mma=0&nd=_&ost=&oet=&fml=a%2Fb&fq=Quarterly&fam=avg&fgst=lin">

U. S. productivity has been in free fall since 1950. That is what perennial budget and trade deficits do to a country.


Along the same lines as above. Suppose it could be shown that industry-specific labour productivity declines more in some industries than others in a typical recession. Then if RBC theory is true - that is, if the different declines in productivity reflect different shifts in production functions - wouldn't we expect relative prices to move inversely with respect to observed relative productivity declines. Whereas an AD shock/sticky prices/labour-hoarding interpretation would suggest either no short-run correlation between relative prices and relative productivity declines, or possibly a positive correlation (since industries where demand is most profoundly depressed by the onset of recession would tend to have both the largest labour hoards and the greatest incentive to cut their prices as a means to maintain their sales).

I seem to remember a paper a while back that argued productivity figures were't very good. IIR one of the major reasons was the valuation of imports and exports - questions like:

How much of the value of an Iphone is added in the US?
How much of the value of a copy of MSWord sold in France was created in the US?

and of course

How do you measure productivity in the financial sector?

But I think you may be able to answer your question without using productivity numbers.

Have you considered looking at investment shocks?

You could look for the signature of known supply shocks. The Arab oil embargo comes to mind.

You might also be able to find clean examples in places like 17th century Venice which had good records, simpler economies and lots of shocks.


I agree with Frances that the specific evidence you're after will be tough to find. Looking at hairdressers was my first instinct too, but demand for haircuts is quite inelastic, and I doubt that there will be enough variation to detect productivity changes from labor hoarding, especially once you consider the massive errors in high frequency estimates for data at this level of disaggregation. (In the US, the source data used to construct quarterly hairdresser output is the Quarterly Services Survey, which recently stated that the standard deviation for the 2011Q4-2012Q4 percent change in "personal care services" is 21.2%. And that is still a much bigger industry than haircuts alone.)

And in general, there seems to be a strong correlation between "straightforward industry with little technological change" and "industry with mostly acyclical demand", which makes this exercise tough.

We can, however, scrutinize the distribution of measured productivity shocks across industries, to see whether patterns arise there that look suspiciously like labor hoarding rather than true technological change. For instance, once you throw away the computer industry, productivity in durable goods manufacturing does not grow any faster than productivity in nondurable goods manufacturing, suggesting that the incidence of technological shocks should not be much greater in the former than the latter. If we see much more volatile (and cyclical) productivity in the former, it's a strong hint of cyclical utilization.

One (inexplicably obscure) paper that put this into practice was Bils and Klenow's 1998 JPE, Using Consumer Theory to Test Competing Business Cycle Models. They find that durables and luxuries have more procyclical productivity, consistent with a demand-side view and suggestive that these productivity changes are not truly exogenous. At this point, however, there are still two widely discussed explanations: either (A) cyclical utilization or (B) increasing returns to scale. Bils and Klenow's strategy for ruling out (B) is to look at relative prices, since (A) implies procyclical prices for durables and luxuries while (B) implies countercyclical ones. It turns out that durables and luxuries do have relatively procyclical prices, leaving the utilization story triumphant.

One early proposal for using price data to purge cyclical utilization from productivity data is Hulten's 1986 Journal of Econometrics paper, Productivity Change, Capacity Utilization, and the Sources of Efficiency Growth. The idea is that with standard assumptions on production functions, you can measure productivity in two ways: either the standard "primal" method using quantities, or a "dual" method using prices. Under additional assumptions, the difference between these two measures will reflect capacity utilization. Informally, if a productivity shock measured using quantities is legitimate, we should also see it in lower prices for output relative to input.

Unfortunately, industry productivity data is such a mess that it's hard to put this into practice. (Misspecification and data issues abound. The meta-point here is that when our data sucks, we shouldn't be so cavalier about interpreting residuals as structural shocks!) But in individual instances, I think this can be quite useful. For instance, when measured auto industry productivity plummets during a recession, if the shock is real we should see a dramatic increase in prices, which as far as I know has never happened. (Technically you'd need to look an input prices too, though, which could be a big deal if factors are immobile and very inelastically supplied.)

Frank: "Productivity is typically defined as goods produced per unit of time..."

No it isn't. Productivity is goods produced per unit of input.

Giovanni: I think you might be right.

Peter N. Scott Sumner did a post once, on the effect of the tsunami on Japan. The tsunami was obviously exogenous, which makes it a very good natural experiment. And it was big, and did a lot of damage. But you could barely see it at all in Japan's GDP figures.

Yes, productivity is hard to measure. And I expect the question I'm trying to address here is whether the apparent shocks to productivity are really just measurement error.

As businesses operate with excess capacity, productivity itself will be driven by aggregate demand. As an extreme example, take a software company that can, in principle, produce an infinite number of units with the same fixed labor input (writing the software), but only produces as many units as it can sell. There are aspects of this with many other business as well.

I don't think this should be classified as labor hoarding. Labor hoarding also makes sense when finding skilled labor is costly or there are substantial training costs, but there is a productivity effect due to changes in aggregate demand as well.

Imagine a donut shop where the owner makes donuts in a small oven. With a larger market, they could have many stores and a centralized bakery that mass produces the donuts and trucks deliver them to the stores with a smaller per unit cost and smaller labor input per donut produced. Make the market even bigger, and the business may have a Donut-o-polis with robots making hundreds of thousands of donuts per hour. Labor is now devoted to programming and maintaining the donut making machines rather than directly making the donuts -- in other words, you have substantial productivity gains contingent on very large quantities being sold.

But now shrink the market for donuts and the firm will shut down the donut-o-polis and revert to the less productive technique.

This actually happened with Webvan, which was founded in the era of easy money in the dot com boom. They spent billions on building robot operated factories for home grocery deliveries, but the business failed because they couldn't find enough customers to cover their fixed costs. With enough customers, Webvan would have succeeded because their per unit costs were low, the convenience of not needing to grocery shopping is valuable, and the total turnover in the market is high. In a recession, a lot of these ventures fail and a lot of existing firms delay plans to increase capacity by adopting more efficient techniques.

I think strong aggregate demand -- having a broad market of buyers with substantial disposable income to sell to -- is a key driver for productivity gains in both the short and long term.

Did my comment get swallowed?

rsj - it's in spam. Probably Nick will get to it tomorrow morning.

Even hairdressing has problems, at least in the UK

So you’re being badgered by the job centre to go on an A4e compulsory course on how to wipe your own arse or they’ll cut your bennies. Now, if you’re self-employed, you’re eligible for the various Gordon Brown negative-income-tax schemes if you’re earning less than the thresholds. So here’s the trick: you declare that you’re self-employed and you’re picking up a few shifts at your mate’s newly launched hair salon. And you sign off, and sign up for Working Tax Credit.

From Alex Harrowell

Not proof, but surely evidence for 2008. First consumer sentiment falls, then the rate of consumer borrowing falls, slowly, at first. Then the business and financial sectors' borrowing falls, with the financial sector falling the most. It's still falling, though it shows signs of going positive.

Consumer demand seems to lead the way into recession in this case, and non-financial business leads the way out.

Interesting to read Matt Ronglie's description of the Bils and Klenow paper (of which I did not know). Encouraging to see my thinking is only 15 years behind the curve!


Frank: "Productivity is typically defined as goods produced per unit of time..."
No it isn't. Productivity is goods produced per unit of input.

Okay, what I was referring to is total factor productivity (have to remember to be specific).


On the basis of dimensional analysis, TFP is criticized as not having meaningful units of measurement. The units of the quantities in the Cobb–Douglas equation are:
Y: widgets/year (wid/yr)
L: man-hours/year (manhr/yr)
K: capital-hours/year (caphr/yr; this raises issues of heterogeneous capital)
α, β: pure numbers (non-dimensional), due to being exponents
A: (widgets * yearα + β – 1)/(caphrα * manhrβ), a balancing quantity, which is TFP

Total factor productivity is expressed in units of widgets per year ( aka quantity of goods per unit of time ). There is an argument to be made that this method of measuring productivity suffers from the Cambridge Critique. See:


Instead from a macro point of view, why not use a unitless measure of productivity like Real GDP (measured in dollars) / Total Debt Outstanding (also measured in dollars).

Bottom line - even if you choose the general definition of productivity as output per unit of input, what units of measurement for output and input are you using?

Found both rsj and Matt R in spam. And one comment not in spam that endlessly repeated a word with 4 letters beginning with P!

Matt: thanks for that. I didn't know of those papers. Yep, they do sound very much along the same sorts of lines that Giovanni was talking about, in earlier comments.

rsj: I really like your 'software' example. It's an extreme case of increasing returns to scale. It's both similar but different to the labour hoarding case. It's similar in that there's no shift in the production function. But different in that it's a movement along, rather than a movement off. And your software guy is working the same number of hours throughout, he's just selling fewer copies when AD falls. It's like a Samuelsonian pure public good.

Frank: "Total factor productivity is expressed in units of widgets per year ( aka quantity of goods per unit of time )."

No it isn't. That is what we call "output".

What we call "productivity" is a measure of output/input.

And yes, there are different measures of productivity, and some of them are problematic, because both output and input can be vectors, and need to be converted to scalars.

Let's stop this discussion now.

See my mistake. Discussion stopped.

Generalizing the example of the programmer, is there a notion of "fixed labor" similar to that of fixed capital? I.e. it doesn't require less work for an accountant to write a smaller number in the "units sold" column. Every firm will have a marketing, legal, IT department, etc.

rsj: yes. The "fixed input" (or "overhead") could in principle be labour, land, or capital. Ot a mix of all three.

Regarding my bogus hairdresser post, it's certainly an example of demand-driven productivity, although it's influenced by arbitrage among different bits of the tax and social security regime.

Ahh, so maybe we can identify the share of variable costs as gross margin = (revenue - cost of goods sold)/revenue ?

Not sure if this is the correct metric. But if it is, and if labor is allocated uniformly to both fixed and variable costs, then since the SP 500 has a gross margin of about 40%, we would expect a 1% drop in output to correspond to a .6% drop in labor? That's assuming all companies reduce variable costs and keep fixed costs the same. But if a 1% drop in output results in 1% of the companies going out of business entirely, then it would result in a 1% drop in labor. This also assumes all sectors are hit evenly. Gross margin for retail is 25%, but for transportation it's 53% and for utilities it is about 55%.

I was trying different things in FRED, and I came up with this. I think it's definitely relevant, but I'm not sure what to make of it. It's just the ratio of Net Worth to GDP. You can make a good argument that the ratio should be roughly constant, and it was for a while, and then things changed during the great moderation.

If GDP is a proxy for the creation of new assets, then is the difference a change in valuation of existing assets or is it the result of increased financial system intermediation? Is it a problem? Is it a cause or an effect?

Peter N.,

You should notice the inflection point beginning in 1983 where net worth / GDP started rising and kept rising. That inflection point happens to coincide with the year the U. S. Bureau of Labor Statistics replaced housing with owner's equivalent rent in the CPI computation.

What you are seeing is an administrative change in the way inflation is calculated resulting in repeated credit expansions that fuel assets prices.

To answer your question, yes this is a change in the valuation of existing assets. It is a problem, for two reasons:

1. Because it places undue influence onto a particular portion of the economy (housing)
2. Because an economy becomes more recession prone when monetary policy is unequally weighted toward not reigning in the excess while mitigating the damage from the fallout.

In the above, variable costs should be 1 - (revenue - cost of goods sold)/revenue, as the subsequent text suggests.

Just a shot in the dark, but I am trying to estimate what proportion of labor is allocated towards fixed costs (e.g. writing software, administration, operating the power plant, creating media, accounting, etc.) versus the proportion of labor allocated towards creating an additional unit of output.

Perhaps 40% is too high, but I wonder what would happen if we were to take a straw poll of readers here.

Just for kicks, I was looking at the front office positions of the Greatest Baseball Team in the World (SF Giants), which coincidentally has a great selection of fixed cost labor categories: http://sanfrancisco.giants.mlb.com/team/front_office.jsp?c_id=sf


I don't see why you can't get what you want in many cases by categorizing the items of a detailed balance sheet. You'll have to make some decisions of what goes where, and these decisions won't be universally applicable.

Still, there's no percentage in reinventing GAAP.

Hi Nick, (this is not a comment to your post,) may I kindly ask whether you have an answer to the question I post here:

It would be great! Thanks in advance.

The comments to this entry are closed.

Search this site

  • Google

Blog powered by Typepad