Why Are Some Companies Efficient While Others Are Not?
Articles,  Blog

Why Are Some Companies Efficient While Others Are Not?


[MUSIC PLAYING] CHAD SYVERSON: Thank you
all for coming today. I appreciate the invitation to
visit with you about the work that’s summarized
in the title, trying to understand where differences
in efficiency across companies come from. And you’ll see in a second,
they’re big differences and they add up to
important things. So in the end, the speed
limit on economic growth is productivity growth and
where does productivity growth come from? Well, it comes from producer
by producer making decisions about how they’re
going to operate. And so if we want to
understand economic growth, we, at the core of it, need to
understand some of the things we’ll be discussing today. I’m going to be going over a
whole bunch of different work, some done by myself, but much
of it also done by other folks looking at similar issues. We could talk for the rest of
today about this literature, but I’m not going
to make you do that. I’ll just give you a sort of
overview over the next hour or so of the things
that I and others that interested in efficiency
differences are our study. OK, so to get started,
a little background– so, over the past
25 years or so, really for the first
time ever, economists have gained access to systematic
data on company’s operations. This data is usually
being collected by national statistical
agencies and they’ve been doing this for decades, but
usually what would happen is, they’d go take a survey
or a census of companies– tell us about your outputs,
tell us about your inputs, so on and so forth . They’d add up all the numbers
and say, this is total output, this is total input and
then throw the data away, or put it in a file somewhere
and never let it be seen again. About three decades
ago, people realize there’s a lot to learn
in this micro data, forget just adding
it up, let’s look at the individual
companies’ numbers and moreover, let’s
link them over time, so we can follow companies over
time and see how they evolve. Much of the work
today is going to be facilitated by the recent
availability of that data. So in the US, most of this data
comes from the Census Bureau. We all know about
the population side– the demographic side that
happens every 10 years, but they also have
an economic census that’s taken every five
years with some surveys done in between. But the US data is not
the only data that’s driving this body of work. Many, many companies have now
made their data available, usually with some sort of
confidentiality restriction. So, you’ll have to use the
data in a secured facility and you can’t report any
individual company’s responses in any work you do. Everything sort of has to
be statistically hidden or aggregated over. But here’s just an
example of some countries, it’s certainly not
exhaustive, whose data has been used to
look at the things we’re going to talk about today. And you can see, it’s
from all over the world, every continent, it’s from
countries with over a range of development profiles. So you have some wealthy
economies, some middle income economies and some still
developing economies– all of which people
have been using to learn about where efficiency
differences come from. Also, economists have
gotten a little bit better at talking companies
out of their own data, so we can study it. It’s still not the
easiest thing to do, but every once in a
while, you get lucky and you can get production
data from an individual company at a super detailed
level and I’ll show you some work done
with that today as well. So, people use this
company level data to look at all sorts of things,
but a big portion of that work is the work we’re going
to talk about today and that is that
focusing on performance. The most commonly used
performance measure in economics is productivity. OK, what’s productivity? Well, it’s how much
output a company gets out of a set of inputs. OK, so you can
think of it as it’s a ratio of output to inputs. How much stuff you get out per
unit of factors you put in. OK, you can read that as– and then as I’ve already been
using the language efficiency, it’s efficiency in production. When I talk about
productivity, you also think of it as profitability. If you think about profits– profits as revenues
minus costs– well, take output multiplied
by the price of output. Take inputs and multiply
by the price of inputs. This is sort of like the
ratio version of profits as opposed to the difference
version of profits, which we were used to using. So, there is a
strong correlation in the data between
productivity and profits. They’re not perfectly
interchangeable and we don’t need to go
into that now why exactly, but they’re highly correlated. OK, so a few facts– one, no matter where
anyone has looked– and this is now we’re talking
about dozens of countries, scores of industries and
many, many different time periods, so thousands of
studies now in total– people find really large
productivity differences across producers, even within
what you might think are quite narrowly defined industries. OK, and again, this
is a ubiquitous fact. So, what do I mean when I say
a narrowly defined market? Well for example, I’m talking
about a market like saw blade manufacturing. Not manufacturing, not even
tool manufacturing, but making saw blades. That’s pretty specific. You might think how
different can companies be at their proficiency
in making saw blades. You stamp some steel. You make some cuts and there
you go you have a saw blade. White pan bread bakeries,
is another example. Not any old bakery– white pan bread. Not wheat bread, that’s
a different industry. Not roles, that’s a
different industry. This is like Wonder
Bread coming out loaf by a loaf by the thousands
out of a factory. That’s what you should
have in mind when I talk about white pan bread bakeries. Lars talked about my history
with ready-mixed concrete. It’s been an industry
that I’ve been looking at for a long time. Again, you might
think, oh that’s about as pedestrian a production
process that you could imagine. I like to say it’s like making
mud pies at a really big scale. You mix up some stuff,
put it in a truck and deliver it and pour it out. OK, those are
manufacturing industries but you can look at these
things outside manufacturing. So say, retail– an example
of a narrowly defined industry in retail would be bookstores
or gas stations or whatever. That’s the sort of scope
we’re talking about. If you want to get out of
retail and go into services– something like business
accounting services, not personal accounting
or tax accounting, but accounting for businesses. That’s what I mean
by narrowly define. Now, what do I mean
when I say people always find large productivity
differences within these industries. Well, a typical 90-10
percentile productivity ratio, within one of these industries– 4 digit means define
like I describe– is about 2:1 in
US manufacturing. So what does that mean? That says, let’s go find every
ready-mixed concrete producer, or go look at every saw blade
manufacturer, or whatever, and you line them up
within their industry in terms of their productivity. So, we’ve counted their outputs. We’ve counted their inputs. We’ve taken the
ratio and we’re just going to line them
up by that ratio, from least efficient
the most efficient. Let’s throw out the tails,
the top and bottom 10%, just because maybe
they’re special or weird or whatever– there’s
measurement problems. So we’re just going to
focus on the middle 80%. So, we have the 10th percentile
to the 90th percentile. Well on the average, within
one of these narrowly defined industries, that 98th percentile
company in the industry is going to be getting
twice as much output from the same inputs as the
10th percentile in this firm. OK, same inputs,
in other words, I’m going to give both companies
the same number of workers, the same amount of capital
and the same total value of intermediate materials,
yet the standard thing is one company will be able
to get twice as much output from those inputs as the other. OK, that is the typical case. That’s not the crazy tail case. That is the average ratio
in US manufacturing. OK, now nothing is special
about this large dispersion in US manufacturing. If you look at
Chinese manufacturing, the average 90-10 ratio is 3:1. If you look at India, it’s 5:1. If you look outside
manufacturing, it’s typically a little
bit bigger than 2:1. If anything, US manufacturing
is on the small side in terms of productivity dispersion. So this is what
we’re talking about, these sort of really big
differences in performance. Two companies operating
in this same narrow market at the same time, yet they
have very big differences in their efficiency
of operations. Another fact that people find is
that these productivity levels are persistent. If you’re an efficient
company this year, you’re considerably more likely
to be efficient next year than inefficient. If you’re inefficient
this year, you’re likely to be inefficient next
year if you stay in business. I’ll get back there in a second. So for example after
five years, a third of businesses in the
top 20% are still there. If productivity is just a
random thing, of course, we wouldn’t expect any
persistence at all. Those in the second 20%
are more likely to be there than anywhere else and
so on and so forth. So, you get this
sort of stickiness. Now, it’s not
perfectly persistent– companies can affect
their productivity levels and we’ll talk about
how that happens, but it’s not an easy thing. And it’s certainly not
the case that if you’re having a bad year,
you can just wait and things will probably
get better next year. Things will probably not
get better next year. Moreover, companies
that are inefficient, like I said, not only
do they stay efficient, they typically quite
often go out of business. So, productivity is
literally a matter of survival for businesses, OK? And again, this is
something people have found no matter what the
industry time period or country you look. The higher productivity
companies in an industry are more likely to survive. So just to give you an example,
again terms and numbers in US manufacturing, the bottom
fifth are 2 and 1/2 times as likely to go out of
business over the next year, as a firm in the top fifth
within their narrowly defined industry. High productivity
companies are also more likely to grow
faster in the future. It’s good to work for a
high productivity company because their workers
are paid more. Now of course there,
you might think there’s an issue of causality
is it that productivity leads to higher pay
or paying workers more or leads to higher productivity? Well, that’s a little bit
of both as it turns out. We’re not going to have time
to go into it too deeply, but it actually does look
like causality goes both ways. And by the way, consumers
are often better off when productivity goes up
too because if companies can produce things
more efficiently, that means they have
lower costs and typically lower costs, or at
least partially, passed on to consumers in
terms of lower prices. And I hope motivates the
importance of productivity and understanding why or
understanding the fact that there are big
productivity differences, so let’s get to talking
about why those productivity differences exist. What things determine
whether some companies are efficient or less efficient. So, I wrote a paper that tried
to summarize what everyone was thinking about along these
lines a few years ago and I divided explanations
into two broad sets of factors. So, the first set are
things that companies, at least in theory, have
some ability to control. These I call the
levers, all right. Then, the second set of things
are aspects of the market environment that the
company operates within that can also affect productivity. Now, companies don’t have direct
control over these things, but these might
affect their incentive to pull the levers
in category one. We’ll talk about some
of the aspects of both of these sets of factors, but
let’s go in order and start talking about the levers. Now, I came up with six
categories of levers. The first is managerial
practices or managerial talent. OK, you think of the manager
or the management team in a company is like a
conductor of an orchestra, even though as a good
conductor makes an orchestra sound better, a
bad conductor can make an orchestra sound
worse, even if the instruments don’t change. That’s one way to think
about what management does. You can have the same inputs,
yet really poor performance because the managers don’t
know how to coordinate the actions of those inputs. A second lever are differences
in the quality of labor capital inputs. So, that gets a little
bit at this wage issue. If you can hire
workers who are somehow better than other
workers or better matched to the type of
operations you’re doing, you can get more
output per worker than if you hire poorly
matched workers– same thing with capital inputs. A third category is
information technology and R&D, or intellectual property. If you’d like, you could think
of this as 2a rather than three because in some sense, these
are specific kinds of capital. The fourth is
learning by doing– learning by doing
is as it sounds. It’s becoming more proficient
through the very act of operating. The fifth is product
innovation– so, a lot of times when we think
about productivity in terms of process. It’s about, OK, I
got these inputs and I got to
combine them to make this thing I’m going to
sell, but you can also innovate rather than
on the process side, on the product side. What if you change the product
you’re making in a way that raises its value to consumers. Now, you are making per unit
import a more valuable thing. We would probably
want to think of that as being more productive,
even though it’s all the same physical unit
that it always was. It’s a more valuable version
of that physical unit and so in some sense you’ve
become more productive. And then, the final thing are
firm structure decisions– so this is about
vertical and horizontal. So vertical is within
a production chain. These are decisions
about where you’re going to get your inputs. Are you going to buy
them on the market or are you going to vertically
integrated and make them yourself? How are you going to distribute
your outputs as a company? Are you going to use other
companies to do that or are you going to distribute
your own products? That’s the vertical
aspect and then the horizontal aspect is how
many different kinds of product markets do I want to operate
in and how closely are they related to each other. Are they very different
kinds of products or are they more similar kinds
of products– and choices about the horizontal
and vertical scope of companies interact with
their productivity levels. All right, now I’m
not going to have time to talk in detail
about each of these. I’m just going to
go through a few, with some interesting examples
of things that I and others are looking at along these
lines about the relationship between these levers
and productivity levels. So, let me first talk about
that first thing I mentioned, which was managerial practices
or managerial talent. Until about 10
years ago, I think it’s fair to say management
had the highest ratio of people talking about it as a source
of productivity differences relative to actual evidence that
it was a source of productivity difference. OK, now you could always have
gone to some airport bookstore, and there’s a whole
row of books about how to be a better manager,
but it’s usually just someone’s bad
intuition or storytelling about their own
experience of business. There’s really very
little systematic evidence on what particular
management practices are done and then which ones are
related to better productivity, but that started to change. Now, there are more
systematic efforts at trying to collect data
on management practices. The best known and the first
is the world management survey. This is run by an economist
at Stanford, Nick Bloom, and at MIT, John
Van Reenen, and they have, over the
past decade, hired folks to survey companies around
the world in phone interviews. They’ll last 45
minutes to an hour where they talk about
operations with these companies. Now, the companies know they’re
talking to someone about how their company operates and
it’s a research project, but they don’t know that while
they’re discussing what they’re doing, the interviewer is
scoring them and systematically walking him through a number
of different practices and then on each practice,
based on their responses, they’re getting scored from
low to middle to high quality management practice
and then they put all that data together. This effort has expanded. It’s not just the world
management survey anymore. For example, now the US
Census Bureau actually sends out a survey on
management practices to about 50,000 companies
every few years. And so, we’re getting more
and more data on companies management practices. So what do you find? Well, here’s some data. I’ll just show you some
basic numbers from the world management survey data set. As you can see
now, the survey is extended into dozens of
countries and hundreds of companies within
each of these countries, lined up from highest scoring
i.e. best management practices to least best management
practices at the bottom. It sort of lines up with GDP per
capita or economic development. So, that’s sort of
interesting, maybe it’s informative– it’s
hardly causal. It’s probably more
than just a matter of getting Mozambican companies
to manage themselves better to make Mozambique
a wealthy country, but we know those things,
as we can see in the data, are highly correlated. So, the US, Japan, Germany,
Sweden, Canada are the top five and then you can see the
mix within other countries in the data. Now, these are averages. If you go within
any given country, you’ll see a big spread
around that average. So just like there are big
differences in productivity within industries, there
are big differences in management practices
within industries. Here, this shows
the distribution for, for example, countries
US, Brazil, China, and India. So what it’s done is the
raw data is the histogram, the sort of yellow chartreuse
color histogram there. And then, the red curve
is a kernel density fit to that
distribution for the US. And then, the other countries
overlays the kernel density of the US management
score distribution upon the raw data of
Brazil, China, and India. OK, so a couple of
points to take away here is one, like I said,
there’s a big spread, even within countries in
terms of management practices. Two, you can see
the mean differences in terms of shifts
of the distribution across the country. But three, you can see that
there’s actually differences, not just in the mean
across countries, there’s differences in sort of
the weights of the distribution across different quantiles. For example, if you compare,
say China and India, China actually has a
fairly relatively speaking small number of really
poorly managed companies but a big bulk of sort of
mediocre management companies. On the other hand,
India has got a larger set of really poorly
managed companies and hardly any very well
managed companies– same thing with China. The US, of course, has
very few poorly managed and a fairly large chunk
of really well-managed. So, this just goes to show
that there’s a big spread and perform in
management practices within and across
countries and these come, not just for mean
shifts, but from other parts of the distribution as well. Now, I showed you how big those
average management practice differences are
across countries, but one interesting
thing shows up when you break out
companies within countries into whether they’re part
of a multinational or just a domestic company. And you can do this for a
bunch of different countries and the data– and you can
see that multinationals more or less are
basically well-managed regardless of where they are. OK and this is about
management I’m saying, well, people think
management is related to productivity differences
and indeed you see that. So, this just fits a curve,
a non-linear relationship between total
factor productivity. Total factor
productivity, by the way, is just output over
a bunch of inputs combined in a particular
way as implied by the production function. So, it’s labor, capital,
and materials inputs smushed together into a
single amalgamation input. So, there is a strong positive
and monotonic relationship between productivity and
management practice scores. It’s also highly
correlated with good stuff, just like productivity is. So, companies with higher
management practice scores grow faster and they’re
more likely to survive. They’re more profitable,
their workers get paid more and
so on and so forth. Everything so far is
just about correlation. We’ve seen productivity
related positively to management practice. We haven’t really shown
that it’s causally related. In other words, if
I could actually go in and tweak management
practices in a given company, should I expect then that
productivity will go up. Now to learn that, really with
like gold standard evidence, you’d have to do some
sort of crazy experiment where you take a random
set of companies, you go in and change their
management practices. You have another
random set that’s held out as a control
group, and then you see how productivity changes
in the treatment versus control groups. OK, but that’s kind of crazy
to imagine– going and doing experiments with the
actual companies in terms of management practices. Well it turns out,
economists are crazy. So, we have gone and
run these experiments. I’m going to talk
about one of them and it’s not a big
experiment, partly because it’s really expensive
to do these experiments. This is one done on 20
cotton fabric plants, all in Mumbai, India, and these
are not tiny little mom and pop operations. The average employment
at one of these factories is 300 employees and they’re
selling $7 million of goods. The treatment
group of plants got five months of management
consulting advice, where their management can
sign– some of you are going to go work
for this company. I can’t say who it is,
but I bet some of you are going to go work
for this company. And every week, the
management consultants would come in and talk with
the management of the plant and say, OK, let’s work on
this, let’s work on that. We’re going to implement
this thing, that thing. We’re going to do this, this
way now, instead of that way, and then start
tracking performance. Now, what you really like
to do with the control group is nothing, right,
just nothing happens. Now, it’s not that easy. They actually had to measure
how well the control group of factories was performing. To do that, they had to go
install computers and record keeping things and
stuff like that. So, that process
took a month and so for a month, the
consultants were actually in the control group factories. So, it’s not like
a perfect control. So, we’re really looking
at the difference between five months of
intensive management consulting and one month of sort of
light management consulting. So, they offered advice on 38
specific management practices tied to operations, quality
and inventory control, and then they collected
data on the operation of these factories
over two years. And so, if you look
at what happened to productivity in these
plants, it looks like this. So, this is over the first
year or so of the experiment. So, week zero is the week that
the management consultants showed up for the first time. You can see the red line
here is the productivity level of control plants. The black line is the
productivity level of treatment plants. And you can see, over the
course of the experiment, and it didn’t happen
immediately– it was sort of a gradual increase. The treatment plants got
more and more efficient, and let’s say, nine months
after the intervention started– that’s a better way to say– the treatment plants
productivity levels, even though they’re basically
the same before as the control plants, were 20% higher
than the productivity level in the control plant. And because, of course, the only
difference between treatment control is one just got randomly
assigned to get the management consultant for five
months rather than short visits for a
month, we can interpret this differences being causal. That actually implementing
some of these 38 practices, by the way, not every
company actually, and even in the treatment
group, implemented every one of the 38. On average, they implemented
about 2/3 to 3/4 of them, but actually implementing these
management practices causally raised productivity. All total, the productivity
gains that we saw should save a plant about
2 to $3,000 per year. Now, that’s off of $7
million of revenue, and probably a profit level of
something in that neighborhood. So, it’s almost a
doubling of profit. OK, and that’s net, by
the way, of the cost of putting in the computers
and keeping track of the things that the management consulting
say to keep track of. OK, so it seems from
this experiment, other people have done
other experiments like it, sometimes with mixed results. It seems that certain kinds
of management interventions are more effective than
others, but on average, they seem to be effective at
raising productivity. That means, if we
can get companies that are poorly
managed to figure out how to make themselves
operate better, engage in better management
practices, they could raise their
productivity level and average productivity levels
in the economy would grow up and we would get economic
growth, that’s important. The problem with this
is, the first step in getting poorly managed
companies to recognize they need to become
more efficient, is to make sure they recognize
they’re poorly managed. It turns out companies have no
idea how well or poorly managed they are. So, what the survey did is
they asked this company, score yourself in terms of
the quality of your management practices relative to everyone
else in your industry. On a scale of 1 to
10, where are you at. Well it turns out, if
you plot the responses on the horizontal axis versus
their actual productivity on the vertical axis, there’s
basically no relationship. Really efficient companies
aren’t any more likely to say they’re well managed
than poorly managed, and inefficient companies don’t
realize they’re poorly managed. So, that’s a big
issue and we really don’t have a good
idea of how to just break it to companies
that they’re not as good as they think they are. I mean I’m willing to go in
and tell anybody anything, but they’re not going
to listen to me. We’ve got to figure out
sort of how to demonstrate this in a way that hits home. OK, so that’s
management practices, let me move on to another
lever, learning by doing. This is something I did
with Steve Levitt and John List, who are also in
the econ department. We studied learning by
doing in an auto assembly plant over the course
of a production year. So, what we had in
our data is, the step by step assembly
of every car made in that plant over
the course of years. This is about 200,000
cars and there’s hundreds of steps that go
together to assemble a car. And we got to see for
every one of them, whether that step went
right or whether it resulted in a defect. OK, so we were able to
count up every defect in every single car and look
at how that changes over time. This is actual data that we use. This is an example, so what we
see here, each line is a step. OK, this is all one car. So, here’s a VIN– VIN is a vehicle
identification number. Every car made
gets a unique VIN. So, this is one
car and, in fact, this is just a
dozen observations of out of hundreds for
this particular car. What do we see? Well, we see the
process that’s happening and a description related to
the outcome, the department and the zone, and the
team of the factory that’s doing that operation. So, where and whom did
the operation and then we see a time stamp. So, this operation happened
on the 18th of April at 10:09 in the morning,
and there was no defect when this process happened. However, when someone tried
to attach the air conditioning line to the shock tower,
it didn’t go on tightly. And then it was short on the
other side– shy means short– and then the other one it
wasn’t ceded to the block. OK, so whoever was doing
the air conditioning line was having a bad car. And so, we have
three defects there. Here is another one where
the left a post plug is shy and so on and so forth. So, we see this for
every single car. If you look at the
number of defects per car over the course of the
production year– and this shows the entire year. This is weekly data,
average number of defects per car over the week– this is what it looks like. So this, turns out,
was a classic learning by doing pattern. You have really
high defect rates or really low
productivity, saying you think of quality as sort
of the inverse of productivity. At the beginning, you
get some quick gains and then they start to slow
down but you get gradual gains after that, and
that’s what we found. So, they got defects
per car down, from 80 per car in this week,
to by seven weeks later, it was down below 30. So, they had about a 70%
drop in defects per car over the first eight
weeks of operations. By the way, we threw
out the first couple of weeks of production
where they’re only making a few dozen
cars, but those had several hundred
defects per car. And those, by the way,
they get they get sold. So, if you’re
going to buy a car, and there’s a stamp
that will tell you– there’s a sticker when
that car was made. So, most cars have
a major redesign every five years or so. If it’s the first
year after redesign and you see it’s made in July or
August or September, pass, OK? One interesting thing
that this factory did is it ran two shifts, but it
didn’t start the second shift until after seven weeks of
operating on just one shift. So, we broke out defects
per car depending on what shift was operating when
the car was being put together. And very interestingly, I told
you about these huge drops in defect rates that occurred
during the first seven weeks. That was all on the first shift. Now, the second shift starts. The second shift workers had
not made a car to that point. OK, they’re only training was
to watch the first shift for one week before they started. But somehow, as soon as
they start operating, they don’t have a
bunch of defects. Everything that was
learned by the first shift, somehow, got
immediately transferred to the second shift. And that tells us something
about the nature of learning and where this knowledge
stock resides in a company. At least in this
company, it doesn’t seem to be with an
individual worker because these workers are
different than these workers. If it’s about the worker, these
guys should have to relearn everything, but they didn’t. Somehow, whatever was learned by
the first shift workers got put into the production
process in a way that second shift
workers could just show up and operate at the
new lower defect levels that the first
shift had achieved. However, if you look
at model by model, and this factory
made three models over the course of the
year, but also staggered the start of production
of each model, there was re-learning when
new models were started. So when the second
model starts, there is a little bit of a
spike in diffract rates. They don’t start where the
first model was at, at the time. They start and then when the
third model version of the car starts again, they have
another spike in defect rates. And as you can
see interestingly, if you squint a little bit, you
can see defect rates actually rose in model one a little bit
when they started model three and there’s a reason why. Model three is sort of a
special version of model one. I can’t talk about
what model it is, but it’s a special kind of car
that’s related to this one. And they had so many
problems with this, they were basically
taking resources away from quality control
and model one and applying it to
trying to fix what was going wrong with model three. Absenteeism, also
you might think is related to learning by doing
and we found that, but turns out it’s really small. So, absenteeism in this
factory is no small thing. Average absenteeism was 14%,
that means on any given day, one out of seven
workers was not at work. However, even if you could
cut absenteeism down to zero, turns out the relationship
between the speed of learning and absenteeism is slow. You’d only cut
defect rates by 5%. OK, so by the end
of the year, you’re talking about defects
per car of 10. You know even if you got rid
of absenteeism completely, that would just drop
that to 9 1/2 per car. It’s not that important. So again that
suggests, it’s not so much about the workers keeping
the learning in their head, the factory has
processes that get what the workers learn out of
their heads into the production process itself. So, any given worker can
show up, start operating on the factory floor and
gain all the knowledge that other workers who have done
the same things before them, enjoy the gains of that learning OK, so now let me move
on to external factors. This is just a
couple of the levers. Let’s move on to the sort of
market environment things. I talked about four
different kinds of environmental factors– external factors in that
paper I was talking about. One is productivity
spillovers, you can think of that as
learning across companies. OK, I see what another
company is doing and I figure out how
to do it, now I do it. Or I hire one of
their former employees and they bring their
processes into my company. That’s what we’re talking
about, spillovers. The second is competition. Competition, both within
markets domestically, as well as perhaps through trade. Third is regulations of
the regulatory environment and fourth is input
market flexibility. How easily labor and capital
can move around the economy. And again, I’m not
going to be able to talk about each of these in
detail, but I’ll just go over some examples. So, let’s focus first on
competition and productivity. This is a big thing for
economists, of course. We think there are
a lot of good things about competitive markets. One of which is, it encourages
efficient operations. Now, there are two mechanisms
through which this can happen. One is that, when you’re in
a really competitive market, you better figure out
how to be competitive because if you’re not,
your competitors are going to take business away from you. So, your incentive to become
more efficient, we think is typically higher in
more competitive markets. The second mechanism is a sort
of Darwinian process, where even if a company can’t
change its own productivity level, if competition grows
the more efficient companies and forces the less
efficient out of business, average productivity
in the market will go up, even if no
individual company is getting more efficient. OK, that’s sort of
a selection process. Or if you’d like to think about
it in terms of statistics, the first one is
the within process and the second one
is a between process. All right, now both
mechanisms matter. How much each one matters
depends on the sector. So for example, typically,
if you look in manufacturing, it’s about 50-50. Of total average manufacturing
productivity growth, about half comes from individual
manufacturers becoming more efficient. The other half comes from the
market re-allocating activity away from less
efficient manufacturers and towards more
efficient manufacturers. If you look at retail
on the other hand, it’s almost all
through selection. So, it’s rare to see a
retail store actually become more efficient over time. They’re as efficient
as they’re ever going to be when they
open up for business for the first time. A lot of retail
productivity growth comes instead from the efficient
retailers growing and the less efficient ones shrinking. All right, so let me
give you an example from my favorite industry,
ready-mixed concrete. So again, concrete is concrete–
it’s a big clump of mud, in some sense. It’s just made with
special ingredients. So, what limits competition? There’s no brands of concrete. No one cares about– there’s no
really big quality differences. What matters, in terms
of competition, is space. This stuff is really heavy,
relative to its value and it’s perishable. Once you add the water,
you’ve got about 90 minutes to get out of the truck
or the truck is ruined. OK, so this stuff
isn’t going very far. So, competition is
limited by density. When you have a lot of producers
of concrete in a given area, then buyers of concrete
have many choices. When you have less density,
they have fewer choices. So, density ought to be
related to competition. OK, so let’s look at an
example here of two markets. The blue dots are
concrete companies. Market A is a dense
market, Market B is a less dense
market, let’s suppose you’re a construction company. It turns out construction
companies are the biggest buyers of concrete,
not surprisingly, and you’re located
at that square in each of those markets. OK, but because of the
big transport costs, you can’t buy concrete
from anywhere. They have to be close
enough to whatever you’re building for you to
be able to buy from them. So, in Market A, you
can buy from anyone within the shipment radius. Same in Market B,
but the difference is, because of the differences
in density, in Market A, you have six choices. In Market B, you have two. So, we would think Market A is
more competitive than Market B. And if our hypotheses
about the relationship between productivity and
competition is right, we ought to see more highly
productive concrete companies in Market A than Market
B. And it turns out, that is exactly what you find. This is the distribution
of the productivity levels of every concrete producer
in the United States and there’s about
over 5,000 of them. I’ve just divided it into
the densest half of markets and the least dense
half of markets. It’s just above and
below the median. A simple cut of the data. You can guess which distribution
is the denser market even without the labels. It’s the solid curve, and you
can see that in denser markets, there are systematically fewer
less efficient companies. Zero as the average
productivity level producer in a given year– and there is a larger number
of more efficient companies in denser markets. There’s still a big spread,
even within dense markets, even within less dense markets,
but on average, productivity is higher in denser
markets because of the competitive effects
we were talking about. Here’s another example of
competition and productivity and this is a trade story. So, this is about iron ore. So, most iron ore
that comes from the US is mined in northern
Minnesota, in the Iron Range of north and west of
Lake Superior, that’s put on boats in Duluth and
sent to various buyers, usually on the Great Lakes,
including the Gary Ironworks just down the lake here. So, this is the price of iron
ore in the US between 1970 and 2004. And you can see
throughout the ’70s, iron ore prices
were rising steadily and then something
happened in 1983. They sort of turn on a dime and
then fell for several years, then kind of leveled off
Well what happened in 1983, is for the first time, it became
economical to take iron ore from Brazilian mines,
ship it up the Atlantic, down the St. Lawrence Seaway
through the Great Lakes and then supply to Gary
Ironworks and other places like it, from
Brazilian iron ore, rather than from
northern Minnesota. OK, so that limited the
ability of US producers to raise their price
any more and you can see it was basically capped
by the Brazil-based price. But it did another thing– this ability of
iron users in the US to now switch to
Brazilian iron ore, forced US mines to
become more efficient. So if you look at productivity
levels in US mines, that’s the blue line here. So what you see is
from 1970 to 1982, there was no change
in productivity. It was two tons per worker hour. And in fact, I found an old
US Geological Survey book from 1950. Do you know what iron
ore mine productivity was in the US in 1950? Two tons per worker hour. It didn’t change for 30 years. And then, the Brazilians show
up and within five years, it went up to four
tons per worker hour. 30 years of nothing, then in
five years after competition starts, productivity
has doubled and it continued to rise after
that until the six tons per worker hour by 2004. Now it turns out, this
isn’t the Darwinian effect, of the less efficient
mine shut down, the more efficient mines grew. It was all within mines. So, every individual mine
raised their productivity level once they faced competition
from the Brazilian mines. OK, let me move on to
regulation, another sort of environmental factor. Regulatory policies
can impose barriers to efficiency or
incentives to be efficient. Now, regulations have
many useful things too. We all know about externalities. A lot of regulations are
there to sort of fix problems with externalities,
but that doesn’t mean they won’t have
implications for productivity, or sometimes
they’ll, by accident, have implications
for productivity because people aren’t
thinking about what incentives they might be posing for
productivity differences by creating these regulations. Now, there is a an alternative
hypothesis out there– this is sometimes called
the Porter hypothesis. The Porter hypothesis is
having new regulations put on a company that sort of
forces a kind of reckoning. This goes back a little bit to
the management practice story, that the idea of the
Porter hypothesis is well, OK, now
we have to abide by this new regulation that’s
going to change how we operate. Since we have to
change how we operate, why don’t we just think about
our entire process from scratch and redesign it in
the best way possible. OK, and that sort of
redesign, from the ground up, could actually
raise productivity after the regulations
are put in place. That’s the Porter
high hypothesis. So in work with Michael
Greenstone and John List, also here at Chicago,
we’ve studied the effect of the US Clean Air Act
amendments on US manufacturers in terms of their
productivity level. So the Clean Air
Act amendment says, look you can’t pollute too
much ozone or sulfur dioxide or carbon monoxide or a total
set suspended particulates, and if you do,
we’re going to force you to put on
abatement equipment to reduce your emissions. We find that companies that are
subject to these regulations, do in fact, see a drop in
their productivity level. They have to take some inputs. They used to be put
into making stuff and now those inputs,
workers and capital both, are being applied to abide by
the tenets of the regulation. You add all of that up across
all the affected manufacturers and that suggests
that the Clean Air Act amendments cost $21 billion
in terms of lost manufacturing output. Now keep in mind, that’s
one side of the seesaw. The other side is
the health benefits from having less pollution. We don’t try to measure
that in our work, but other people have tried
to do that and suggest that those might be in the sort
of low triple digit billion. So, this suggests actually,
even though there is a cost here and it’s a real cost $21 billion
bucks a year, on balance, we’re actually
getting more benefit out of these regulations
than it’s costing us in terms of lost productivity. Here’s another example of
how a regulation– this is sort of like
accidental productivity effects of regulation. This is from the US Sugar Act. So the US Sugar Act was
passed during the New Deal. It’s one of the
programs of the New Deal and, of course, as
you might guess, it was about how sugar is
produced in the country and it lasted as past 1934. It lasted in 1974. The way it worked was, it was
basically a big money shuffling scheme where farmers would
be paid a subsidy based on the total amount of
sugar in their beets. So, this isn’t about sugar cane. This part is about sugar beets. By the way, sugar beets
are not like the things you get in your salad. They’re about like
that big and red. Sugar beets are about
that big and white, OK. And they’re all about 15% to
20% sugar, and a bunch of sugar comes from beets. Other sugar comes from
cane, but in the US, I think it’s about
half and half. So, they subsidized farmers
for growing sugar beets and the total subsidy was
just the gross amount of sugar and all the beets you farm,
were going to pay you a subsidy. How did they pay
for that subsidy? They, being the government,
they taxed sugar refiners. OK, so they said, all right for
every unit of sugar you make, we’re going to take
some of that money. We’re going to pay
it to the farmers for having grown that
sugar in the first place. Now, of course the companies
aren’t going to like this. So the government said, all
right we’ll make a trade. If you let us tax
your production, we’re going to
let you fix prices and we’re not going to
sue you for anti-trust. It’s OK, we can live with that. Here we go. So in essence,
this was basically a tax on consumers to pay
farmers to grow sugar beets. So, if you think about the
incentives, and some of you might be sugar beet
farmers, but I’m going to assume you’re not– Well, if you’re a
farmer, you just want to maximize how much
sugar you have in your beets. So it turns out what you want
to do, botanically speaking, is grow the biggest
darn beets you can. The problem with that is
when you have bigger beets, you have more sugar, but
the sugar per unit weight get smaller and smaller. So, it’s harder and harder
to refine actual sugar out of your big honking beets. Productivity in the sugar
refining business goes down, and again, sugar
companies aren’t going to like that,
but remember, they’re able to fix prices. So, if they’re inefficient and
have high cost, that’s fine. They just charge
consumers higher prices. OK, so what you would expect
given this regulation is you’re going to have low
refining productivity. In other words, you’re not
going to get much sugar out of each kind of beet that
goes into the factory. This is, over the
entire 20th century, the profile of refining
efficiency in US sugar refining. So at the beginning
of the 20th century, they were able to get about
220 pounds refined crystal sugar out of each ton of beets. That went up steadily,
until guess when, the Sugar Act is passed in 1934. Then it proceeded
to fall steadily, until guess when, the
Sugar Act is repealed in 1974, at which time it turns
around and goes back up again. OK, so between 1934 and 1974,
efficiency in sugar refining fell from 315 pounds per
ton down to 240 or so– so, by about a third almost. OK, we basically lost– and in fact by the end of the
century, we’re back up to 300. We weren’t even at the
level we were at when the Sugar Act was passed. We lost basically a century’s
worth of technological progress because of regulation. And the regulation,
bear in mind, had no direct impact
on sugar refining. It wasn’t about pollution
coming from sugar factories. It was just about this scheme
to take money from consumers and give it the farmers, yet,
because it wasn’t thought out, it had these
efficiency implications that really added up. OK, the last external factor
I want to talk about briefly is this input
market flexibility. OK, I talked about
this Darwinian process, that the market can
take activity away from less efficient
producers and give it to more efficient producers. Well if that’s
going to happen, you need the market to be
able to move inputs from less efficient to
more efficient producers. Labor has to be able to move. Capital has to be able
to move and it turns out there are big differences across
countries, and within countries over time to some extent, in how
efficiently input markets work. And it turns out
that this stuff seems to be related to
productivity outcomes as well– average
productivity differences across countries and industries. So just as kind of a
motivating example, if you want to say,
well, how would we measure whether this
Darwinian process is working? One thing you could
imagine is just look at the correlation
between size and productivity. If the Darwinian
process is working well, that should be kind of big, like
the most efficient producers ought to be the largest. If it’s not working, you’d
have zero correlation. And if it’s
perversely working, it would be negative– that
the most efficient producers would be the smallest. The least efficient
would be the largest. If you look at the Eastern
European countries, around the time
of the transition, out of being a planned
communist economy and into being a
market economy, you see a pretty
systematic relationship between the correlation
between productivity and size. At the beginning,
there was actually, in most of these countries,
a negative correlation. The most efficient
companies were the smallest. The least efficient
were the largest, but over the ’90s, as they
transition to a market economy, the market started
taking inputs away from less efficient
companies and moving them to more efficient companies. That correlation
within countries ran from negative to positive. The exception, turns
out, is Estonia, which actually was reasonably
positively correlated before and actually fell a
little bit over the ’90s, but most of the Eastern
European transition economies, you saw this shift from negative
to positive correlation. What are sort of
the big questions still out there that people are
working on including myself? Well one is, I talked about all
these different productivity drivers. They’re not all equally
important in every market. What is it that determines
how important each one is in each market? Where is management
the most important? Where is it the least important? Why, what is it about
markets that make management more important, for example,
in this market than that. How important are demands
factors versus supply factors? How much of this
Darwinian process is driven by consumers
being able to switch versus input markets
working better– that the labor
market has improved, so workers are more willing to
switch companies or whatever. Is it management or
managers that matter? I was talking about
management practices, but is it about the person
or is it just the practice? Can I take any old
person to just say, do these 38 things and
everything is going to be great, or is
there some interaction about the type of
person implementing those practices that matter. How badly are
resources misallocated? There are a lot of work
being done on this right now and a lot of it here
at Chicago, in particular. Me and other folks are trying to
understand why you sort of end up in these situations where,
for example, in the worst cases, you have a
negative correlation between productivity and size. How did the market get
there and why isn’t it working to sort of straighten
that correlation around and turn it positive. Moreover, can we
predict innovation? Can we tell by
looking at companies which ones are going
to have productivity growth in the future? And then, what is the
role of, or hope for, policies that encourage
productivity growth and there’s a whole set
of possible thing here and we’re just starting
to scratch the surface. OK, so that’s all I’ve
got at this point, but I’d be happy to answer
any questions you might have. That’s a great
question, finance is the worst because what
is the output of a bank? What is the output of a bank? It is not clear what the
output of a bank is, OK. This is a story I like– do
you know how they used to, the BEA, the Bureau
of Economic Analysis, puts together the GDP numbers. It is my understanding, and
I think the story is true, and even it’s not,
it’s such a good story I’m going to tell it anyway. The way they used to
measure productivity growth in the banking sector, because
it’s really hard to measure productivity in banking, is by
looking at the number of checks cleared per unit input. Because why, because
at least they can measure how many
checks were cleared, right? But to imagine that,
that is the sum total of the output of banks
is crazy, and of course now checks are– the productivity and banks
now would look terrible because checks
are going to zero. Banks– no one’s
figured that out. Services you could
always go with revenue. So it’s revenue per unit input,
but there’s a problem there, and this is something
I worked on a lot. Revenue is price
times quantity, right. Now if price is about quality
differences, we’re probably OK. If you think, if you’re
making a higher quality car, even if you’re making the
same number of cars per worker as before, you
might think you’re more productive because, like
in some quality adjusted units, you’re making more. But what if price is
high because competition in the market fell? Or someone imposed
a trade tariff and you were able
to raise prices. We haven’t become
more efficient, you’re just able to
charge higher prices. So that’s tough, how do you
sort of separate price variation that’s not about efficiency
versus price variation that is about efficiency. And people are working
on ways to do that, but nothing’s
really perfect yet, but that’s one of
the things you have to deal with once you get
out of the simple world of, I can count cubic
yards of concrete. I can count loaves of
bread, but you can’t do that for a lot of producers. That’s a great question. There are definitely
good case studies of companies that have
implemented sort of, not necessarily six sigma per
se or any particular kind of [INAUDIBLE] but just sort
of systematic routinized ways of thinking about
production processes and gotten productivity
gains from that. We don’t have a lot of sort
of experimental evidence of the Indian textile
manufacturing kind, but it would point to
there being a possibility that this stuff really causally
matters, I agree with that. And I think now what’s
going on in this area is, a lot of this stuff
was sort of invented within manufacturing because
manufacturing in some senses is a very inherently
process driven thing, but now people are
taking these techniques and moving them
into other sectors. So what I know about from
my own work is health care. So, there are
actually hospitals now that run on the Toyota
production system. So, they think
about every process in a very systematic way. Where does all the
components of time come in? When we have to wait for
an hour for a lab result, why does that happen? Does it have to be an hour,
that sort of thing and sort of routinizing the process
of health care delivery. Now there’s resistance to
that among some health care providers, but
there is some, again more anecdotal at this
point than anything else, the evidence that this stuff
actually has real benefits. So you’re seeing it,
but I don’t think we’re quite at the point where
we just systematically know it works with a high
degree of certainty. Yes. AUDIENCE: Is there any
significant correlation between business that
were founded by a family business or business
with management that are trained and
have strong technical? CHAD SYVERSON: Yes, so one
of the interesting results of the world
management survey is that, on average,
family-owned companies don’t seem to be more
or less poorly managed than non-family owned company. However, if you focus in on
family owned companies whose CEO is the eldest
son of the founder, they are systematically
more poorly managed. OK, and you can
imagine why, what are the odds that the
eldest son of the founder just happens to be the
best possible manager available for this company,
probably not very high. And so, when you see companies
that actually are run that way, they don’t seem to have
this high quality management practices. If you divide it up by all
sorts of different ownership structures, the best
managed companies are the sort of publicly
held companies, where you diffuse shareholders with a
board of directors– on average have the best
management practices. Yep. AUDIENCE: Some cases you saw
that [INAUDIBLE] company, they really [INAUDIBLE] external
competition, even with the IRS. So that means they
could’ve reinforced before, and they probably [INAUDIBLE]. CHAD SYVERSON: That is
a great question too. Why are you willing
to be inefficient? So Sir John Hicks,
the economist said, the best of all monopoly
profits is a quiet life. And what I think
that means is, you don’t have to try very hard
when you don’t face competition. You might not be the most
efficient you could be, but to some extent, who cares. And if it takes effort,
even if it’s a small effort, and even if it’s a small effort
compared to the productivity gains that you could get, if
you implemented these changes, you might not do it. Moreover, you know
it’s easy for us to think about in a
sort of static world, where you know you’ve got
your average cost curve. You kind of figure out
where the bottom is. That’s where you should operate. That’s where you minimize cost,
but in reality, best practice is a moving target. It’s moving all the
time, demand side stuff supply side
stuff is always shifting around, the best
way to operate your company. And it seems to be
that just companies who don’t face a lot of competition
just don’t put a lot of effort on trying to make those
constant changes to follow best practice. Where if you do have
a lot of competition, you have no choice. And so, I think it’s
that, but you’re right. In some sense, these mines were
leaving money on the table. Well, it’s a little tricky. They were leaving money on the
table, but what was happening is the workers were basically
taking a lot of leisure on the job. That’s another way
to think about it. They were getting the
rents of that inefficiency by not having to work very hard. So that study that
those numbers are from goes into hide detail
about how the minds change their operations
and basically change rules about who can do
what when at these mines to become more efficient. And before, it was basically,
I was a repair person and I would only have to
repair one kind of truck, and if nothing broke, I
didn’t have to do anything. But then after the
Brazilian showed up, now I got to repair
anything that goes wrong and I’m working a
greater portion that day. So in some sense, the rents
got moved from workers to put into the efficiency
gains of production operation. Yeah. AUDIENCE: My question is
related to the previous one. So, why the low productivity
isn’t [INAUDIBLE]? CHAD SYVERSON: Yeah so
that’s a great question. It is in the sense that,
almost always when you look, low productivity predicts exit. So, markets are always kind of
chopping off the less efficient and you typically
see reallocation from less efficient to
more efficient. So that process is
working, but it turns out, we don’t need to go into detail. There are a lot of reasons why
that process doesn’t end up with one really big producer
who is as efficient as you could ever be and they serve
the entire market. You can still have
dispersion and productivity in equilibrium, even
in a world where the market is sort of working. If there are frictions
in either the output markets or the product markets. And some of those frictions
are policy-driven. You might change
policies, but other things are just like,
look I can’t ship. I can be the most efficient
concrete producer in the world, but if I make it
here in Chicago, I can’t send it to
someone in Miami no matter how badly they
want that to happen. OK, all right well,
thank you very much. Enjoy the rest of your day.

One Comment

Leave a Reply

Your email address will not be published. Required fields are marked *