I just got around to posting this paper on SSRN, although it was written a couple of years ago. I need to cite it for other work I’m currently doing, so it has to be out there, somewhere. It is a more concise version of the theory than previous renditions and stays closer to the main point.
What it shows:
There is a simple explanation for why firms exist, why they have the boundaries they have, and why they are organized as they are, which is superior to the alternatives—and it has nothing to do with transaction costs or anyone whose name begins with the letter C.
This theory is implicit in much of the management literature, especially strategic management.
It’s based on the same math as fitness landscapes, but it doesn’t draw on evolutionary theory.
It exemplifies a more general methodological approach that de-emphasizes hill-climbing (optimization theory derived from concave programming) and emphasizes instead hill-finding. There are many potential applications in economic theory, but the theory of the firm stands out.
For the life of me, I don’t understand why this approach to the economics of the firm isn’t universally accepted. Hardly anyone even knows it exists. It strikes me as too obvious to take credit for or be proud of.
Here's the abstract:
This paper proceeds from the assumption that economies are characterized by a high degree of interactive nonconvexity in most activities and at most scales. The consequence is nonconvex production and preference sets and the corresponding inefficiency of myopic algorithms. One application of this perspective is the theory of the firm. Conventional theories explain the existence, boundaries and internal organization of firms on the basis of contracting costs that impede the otherwise optimizing properties of market decentralization. I propose instead an approach in which the motive for organizing production within rather than between institutions is to internalize nonconvexities, thereby obtaining the benefit of explicitly coordinated plans. A useful device for representing this problem is the profit landscape, understood to be nonconvex in the sense that fitness landscapes are in evolutionary theory. Firms face three types of challenges, optimizing with respect to a particular profit hill (the problem analyzed in standard microeconomics), selecting a desirable hill, and achieving flexibility to transition between hills in the face of environmental change. These entail tradeoffs, which are reflected in the diversity of personnel, organizational, and innovation strategies observed in actual enterprises. While the use of the landscape metaphor in coordinated activity theory resembles a similar deployment in evolutionary economics, the two approaches differ in the questions they ask and the units of observation and analysis they employ. The applicability of the coordinated activity model is underscored by its congruence with the bulk of management literature, which can be understood more readily in terms of hill-selection than, or in addition to, the hill-climbing paradigm of conventional economics. In this sense, the existing management literature already provides a body of empirical and applied support for coordinated activity theory, although not generally for the socially-founded objectives of economics.
"For the life of me, I don’t understand why this approach to the economics of the firm isn’t universally accepted. Hardly anyone even knows it exists...."
ReplyDeletePossibly the problem is that the writing is terribly obscure. I would like to understand what this paper is about but the writing is too discouraging to try beyond writing this complaint. I will surely read a clear version, should there be one.
Anon: This is an abstract of a paper written for an audience of other economists, which is who I would imagine care about a topic like "the theory of the firm". It's true there's a pop econ/airport book to be written which applies this approach to day-to-day business and political problems, and I'd love to write it if the opportunity presents.
ReplyDeleteI supposed that would be the response, however I am a professor at a top school and find the writing obscure and beyond. Again, I would thoroughly read the work if only the abstract were accessible as a guide to the paper. I am a professional, and this writing is needlessly inaccessible.
ReplyDeleteSo be it...
"It's true there's a pop econ/airport book to be written...."
ReplyDeleteHow droll.
Sorry if my pop/airport idea came across as flippant. Actually, I think this is a legitimate channel, and many authors I admire have written books like this, but I can see how you might interpret it differently.
ReplyDeleteAs for your main point, well, it's an abstract argument, so abstracting it gets meta fast. If there are particular passages that are needlessly opaque, I'd really be grateful for having them pointed out. The text of the article is as clear as I knew how to write it. Like you, I have no use for people who try to show how smart they are by being difficult to understand.
Out of curiosity, are you an econ prof?
I will read with great interest.
ReplyDeleteIn the last few decades work in high-dimensional gradient descent (learning systems and evolutionary systems) has tended to show that even rough fitness landscapes have "escape routes" from local plateaus, though they may be hard to find. Sometimes these have favorable (but shallow) gradients, sometimes they are flat (these are called neutral networks).
Practically, for evolutionary / economic systems this implies that there are likely to be viable (cash flow positive) paths to higher plateaus but that they are hard to find.
I'm not sure if this makes any difference having only scanned your paper but at least it is maybe an interesting sidebar.
There are also other potentially more consequential issues that make these ideas even more important, but I'll wait to bring those up until I read the paper.
Thank you for the patience.
ReplyDeleteI have read completely through the paper, and you explained the concepts carefully though technically so that I understand what was written. Now I will set aside the paper and return with the questions "what have I learned and what are the applications?"
Yes, though technical you were clear but I do not yet understand what purpose of the paper is. I will return.
jed,
ReplyDeleteNeural networks are not defined as having flat gradients. Some do have those, or very shallow ones, which can be a problem in deep forward machine learning algorithms. But they are not necessarily linked. Really neural networks are complicated nonlinear regression systems.
@rosserjb: I don't think we disagree, and I wasn't talking specifically about neural networks. I'm just saying that empirically nearly all (apparent) local maxima in real high-dimensional landscapes turn out to have "escape routes" that are positive or neutral gradients leading to even higher apparent maxima. This shows up in biochemical / genetic networks, neural networks (as you mention), other hill climbing optimization techniques like kernel machines, and so forth. It seems to be a general property of "real world" high dimensional fitness landscapes.
ReplyDeleteSewall Wright.
ReplyDeleteI actually knew him. I quoted his last paper in my 1991 book. Oh yes, he invented the whole fitness landscape paradigm. I happen to own certain crucial manuscripts of his from the 1920s. Yeah, boring, so what.
ReplyDeleteI read through the paper again, and understand the paper and appreciate the internal clarity of the language, but why the paper is of importance is just not clear to me. I need a simple explanation of the importance of the paper, which I was frustrated about from the beginning. Writing on the firm should be interesting to me, so I spent the time but...
ReplyDeleteNo matter, I appreciate your patience.
If you feel neoclassical economics is ignoring something obvious, that's probably because it is. What it is ignoring in the case of the neatly drawn set of production possibilities, all convex and such, is uncertainty.
ReplyDeleteTo get the firm on that smooth frontier of production possibilities, it is simply assumed that all the firm's technical and managerial problems have been solved within the limits of available knowledge, knowledge that is in some sense complete, perfect and symmetric. That is, knowledge is such that these problems can be solved to some limit known with certainty: the profit-maximizing firm has applied the universe of available technical knowledge to its maximal effect and knows that it has done so.
By sweeping away all the problems requiring engineering and technical ingenuity, the economist has focused her attention exclusively on the problem of allocative efficiency.
And, then, with typical arrogance, declared allocative efficiency to be all there is of any interest or importance.
In a world of uncertainty, where the firm does not know where the bounds of technical knowledge are exactly, and where it is possible to learn (!!!) by doing, allocative efficiency is neither the only thing, nor plausibly, the most important thing.
I submit to you that business firms are typically pre-occupied with the control of production processes, and hedging about being wrong. Technical and managerial efficiency in the control of processes in a cybernetic sense, with feedback from errors, completely dominates allocative efficiency.
Taking this view, as you say, the production function is not even wrong: unless the firm is on the frontier of production possibilities, output is NOT a function of inputs. So, unless the firm has completely solved the technical problems of production, the production function is useless as a representation of the economics of production. In the real world, it would certainly seem odd to discover a firm confident that it did not need to strive and learn and take risks (!) to better control its production processes.
The economic intuition, it seems to me, is not that the set of production possibilities might possibly be non-convex as a given, but, rather, that the firm, in pursuit of managerial/technical efficiency and learning, has strongly circumscribed incentives to locally discover and expand the set of production possibilities in ways that tend to make that set non-convex.
ReplyDeleteThe firm isn't hill-climbing so much as it is poking at the frontier, pushing it outward locally. The landscape may be better described by the concept of economic rent than profit. Think about simple examples of a firm controlling production processes: arbitrarily, a baker. If the baker owns ovens, the baker may manage the process to make more intensive use of the resource he owns. Highly specialized equipment and organization, dedicated to a particular scheme, if that scheme reduces the waste of time and intermediate goods (consumables, like ingredients), may earn an economic rent.
There are a number of implications of accepting that the firm is engaged in imperfect control of production processes, in a technical and managerial sense, where the economies achieved are not strictly allocative in nature, which I will not try to squeeze into a blog comment. Increasing returns are certainly symptomatic of these circumstances, but the increasing returns formulation is a projection back onto the reduced dimensions of allocative efficiency: instead of a 3-D object moving thru time, the economist is taking a 2-D snapshot, wrongly calling it stasis and inferring things from a single perspective.
In terms of your "interaction effects" and myopic hill-climbing, I think you might profit by thinking in terms of control by feedback. The conceit of the Hayekian system of markets coordinated by price is that there is local knowledge, but lots of knowledge isn't local at all. It is in control of the process (say the reduction in error and waste) that economies from managing interaction are realized.
Also, I think you should recast your profit landscape as a landscape describing economic rent for another reason. The pursuit of economic rents, rather than simple efficiencies, is a better description, imnsho, of strategic management. Firms, in pursuit of technical efficiency in the control of processes, are led to make sunk-cost investments and they need political power to earn a return -- a quasi-rent -- on a sunk-cost investment. This is generally what is meant by trying to find a viable "business model" that will reward sunk-cost investment. In this sense, all business is "rent-seeking" rather than profit-maximizing. If you are building a rail line, you may need to go into property development to capture the returns or get the government to give you some access to the stream of revenue from property taxes -- just to give one non-myopic example.
Gross substitution, which is implied I think by drawing convex production possibilities, also cannot survive seeing production processes as essentially under control. There's only one seat and one steering wheel on a farm tractor. Not really a practical option to have three.
Hope this helps.
ReplyDeleteSince I am so far removed from or out of this kind of topic I tried to understand the abstract. I sort of immediately associated 'nonconvexity' with nonlinearity so I googled it----in fact i think i first saw the term of use in Arthur's classic paper on 'increasing returns'. Wikipedia has an article on nonconvexity in econ , and a google search on that turns up some interesting ones (one by H Stahl of UCB which appears to have been written in late 70's, more about firm location).
So as the article says this idea has been around (I guess Arthur attributed it to Marshall from early 1900's )---to me all it means is there are multiple steady states, equilbria, or a rugged fitness landscape.
Maybe explictly applying it to supply side rather than demand or preferences is new ---ie predators use the same strategies as the preys---they select among the available hills and look for best one to hide or seek (or market their qwertys or search engines and apples, or choose which product is best).
Pattern selection is a well known issue in biology---and sort of still a mystery ---but more detail is available now due to cognitive sciences---there are 'mechanisms' for choice beyond flipping a coin--all kinds of biases and algorithms. In biological view my impression is organisms themselves are sort of like firms--not all organisms will exist and we dont know why some do and some don't just like firms. But now one may be able to make better guesses about which will occur.
My impression is most neural nets of any complexity are non-convex---have multiple optima. (Hopfield long ago noted the relationship between neural n ets and spin glasses---perhaps the canonical example ffrom physics of a fitness landcape (eg Kauffman's NK model). The first associative nets (maybe Anderson or previous people) were close to linear, or were because they linearized them to make them solvable, but that's like saying population genetics was linear (sure, it was, since they just expanded it via Taylor's series and only took the first 1 or 2 terms---could prove 'survival of the fittest' held (Fisher's fundamental law of Natural Selection) .
Basically the same thing done in economics. The minute they started going past the simplest systems (eg 1 consumer, 1 product economy, or 2 specie ecologies, or diploid genetics, or added endogenous or exogenous interactions ) you got the same things (in econ first were seen i think in 60's, in genetics it may have been 1970's or even later).
Your nonconvexity point is good and "obvious", but applies to the existence of production organizations as economic entities.
ReplyDeleteThe "transaction cost" approach applies to explaining why production organizations take the specific legal form of a set of (not fully) open-ended, in both time and scope, contracts as in a company charter and a set of employment contracts rather than a sequence of close-ended (transaction specific) contracts.
It is is not widely known that in ancient times companies did not exist, but what happened were called "ventures", specifically in the form of a "commendam" for a single project, like a shipping of specific cargo from A to B. The "venture" was a set of contracts for a specific project in practice, with passive members (the commendantes) and usually one active member (the commendatarius, often the captain of a ship). BTW "Venture capital" is so called because it is organized as a rule as a set of close-ended funds (with the investors as "commendantes" and the venture capital firm as the "commendatarius", but with a different liability structure from the ancient one), typically 5-10 years in duration, each of which is a "venture".
Anyhow, the ancient practice was that each "project" was specifically "chartered", like today in "chartering" a ship for a specific voyage.
The question Coates tried to answer, whether he realized it or not, was why (mostly) open-scope (mostly) permanent charters make sense, and in effect his explanation is that rechartering an organization for every project every year is quite dispersive and expensive in various ways.
I have actually a much better explanation than Coates, one that is fairly close by not quite the same as nonconvexity, about a specific issue that makes the rechartering option really not that good, that there is a massive qualitative leap between fixed-term accounting and open-term accounting, which is much bigger than that between single-project and multi-project accounting: depreciation.
That's the key: long lived capital can only be properly funded and managed if there is depreciation, and depreciation can only be carried on the books of an organization if it is permanently chartered, because depreciation is open-ended.
In ancient times (and sometimes even today) this was handled by closing the books and virtually the liquidating "venture" every year by assessing the capital values at the end of the "venture", and chartering a new "venture" carrying over the capital from the older one. This is of course still reflected today in the yearly balance sheet, but the company is liquidated and rechartered every time the balance sheet is issued.
Depreciation is essential in accounting of long term capital, and is only compatible with open ended charters, as close-ended ones create too many possible problems, such as conflicts of interest (depreciation is very controversial even in open ended companies), as well as costs, to be cost-effective for industrial organization revolving around long-term capital plant.
Rechartering new ventures every year as a new bundle of contracts is instead far more practical for traditional merchant ventures like the "spring resupply voyage for silk to the levant".
In those times the capital good that needed long term stewardship were for example ships, and they sometimes belonged to the state, or temples, which were permanently chartered entities, which leased them to merchants.
«I have actually a much better explanation than Coates, one that is fairly close by not quite the same as nonconvexity»
ReplyDeleteThe reason why depreciation and nonconvexity are related is that depreciation as a rule *creates* nonconvexity and indeed *as a rule* is *the* source of nonconvexity. Depreciation introduces time and time-dependent decisions and path dependence into economic/business calculations, as it is sensitive to duration and discount rates and risk and in particular uncertainty. Depreciation exists because capital is not perfectly malleable/"leets", and that creates nonconvexity. I have some sympathy for the view that the nonconvexity of "nature" is what makes physical capital not perfectly malleable, so in a sense physical capital is what links the non-convexity of "nature" to the non-convexity of economic organizations and of the political economy, and indeed neoclassical "Economics" is full of insane attempts to eliminate that link.
This also suggests why "bundle of close-ended contracts" organizations do happen particularly in financial services: it is much easier to wave hands and build a cloud of ad-hoc contracts around malleable financial assets than around physical capital, and why financial speculators in particular conglomerates builders try to reduce every organization, even those that rely on physical capital, to non-coasian bundles of transaction contracts.
Regardless of the theory of the firm as an economic (as opposed to legal) entity, depreciation is also in an important way the reason for much offshoring:
ReplyDelete* When depreciation matters it is easier for worker unions to have power over management by striking, precisely because depreciation carries with it a need for an open-ended, wider-scope set of contracts, and strikes are particularly effective. Consider car factories, steel plants, manufacturing operations in general. It is indeed part of the union-buster knowledge that industries with significant amounts of long-term capital are particularly susceptible to unionization.
* In most offshore locations worker unions or strikes are legally or effectively impossible. In particular in China the "communist" government has outlawed independent worker unions and strikes are pretty much impossible.
The other main reason for offshoring is that if the economics of the organization are not much affected by depreciation, then it is merely easy to offshore as the organization is easier to decomposed into a pattern of contractual transactions, e.g. by staffing with casual/temp labour.
I’m not allowed to say everything in one comment, so I’m splitting it up.
ReplyDeleteI’ve been busy with other stuff and have just now returned to read the comments that have been piling up in this thread. As I would have guessed, there are a lot of interesting perspectives behind them and much to chew on. Here are a few preliminary responses
Jed: Hey, are you *the* Jed Harris? Very long time, no see. (Still no see in fact.) We have a lot of interests in common after all these decades. It would be interesting to connect sometime.
As for your point about flat or shallow escape routes, I’d be curious in seeing the evidence and the speculations on why this might be a pattern. It might be just a random distribution of inflection signs and magnitudes, no? So if you had enough functional complexity you might discover, ex post, that there was typically some non-(very-)diminishing way of getting from A to B. I can think of practical examples of this in cases where we have pretty good knowledge of the functional interdependence and the problem is mostly one of implementation. (I believe switching energy systems in the face of climate change could be like this, since our knowledge in this sphere is far greater than what we are currently acting on.) Of course, it’s in the nature of hill-finding that you don’t really know that the hill you have a nice escape path to is the one you most want to arrive at.
All of this is highly relevant because the positive function of an economics of hill-finding is to improve hill-finding in real-life applications. My presumption is that abstract reasoning can take you only so far, and the better way to proceed is to burrow into specific cases, draw lessons for theory, re-burrow, etc. etc.)
Which brings me to anonymous: Thanks for giving this topic your time and curiosity. There are negative and positive purposes to this project. The negative is critiquing mainstream economics at what I believe is a fundamental level. I won’t go further into it here, but all the Paretian/welfare stuff, the Ramsey stuff and so on is highly vulnerable to this line of critique. I think that’s important in itself.
But it’s not enough to shoot holes in weak theories; we also need a positive agenda for figuring out how to move forward. This is what underlies my response to Jed. To analyze policies and institutions you have to have a sense of what the questions are, what you want them to do. Hill-climbing ideology supplies that and is the basis for existing policy analysis and normative comparative systems. I do policy analysis too all the time, and thinking about what socialism might be and why we might want it is also central. I want to be able to do these things using a better (more appropriate) theory, and that’s the goal of studying hill-finding.
Comment continued:
ReplyDeleteBruce: I agree with the centrality of control issues, and in a more comprehensive treatment of the theory of the firm that plays an important role. The longer paper (which was linked in an earlier econospeak post) that the latest one is a step backward from combines the nonconvexity stuff with Stafford Beer. It tries to envision a spectrum of discovery moves that range from discovering the process you already have (to control it better) to discovering a process remote from what you have. Organizational strategies differ with the importance given to different points along this spectrum
I should mention that activity interactions I’m referring to in this paper are not only technical/mechanical but also organizational, marketing, finance, the whole thing.
Also, the notion that the firm is maximizing quasi-rents on fixed investments has an honored tradition in economics. Edith Penrose is a great example! The resource-oriented theory of management is essentially all about this.
Blissex: Lots of interesting stuff here to think about. I am not sure the word “depreciation” is the right one for the point you’re getting at. In a sense this is about maximizing the value of existing fixed assets, so capital gains come into play along with both depreciation and capital losses. Your argument is not so far from Alfred Chandler’s, is it? But Chandler couldn’t quite explain where the line might be drawn between market and organizational mechanisms for accomplishing these ends. That’s what I’m trying to do, among other things.
To summarize, the theory of the firm I’m advancing is that it exists to establish the capacity to formulate and implement plans. This means we need a theory of when planning is called for and what it accomplishes, in a private context as well as a public one. Neoclassical economics is tightly organized around the explanation for why planning is not necessary (under suitable conditions) for hill-climbing purposes. I’m saying that the move to rugged landscapes and hill-finding introduces a straightforward explanation for the role of planned (coordinated) activity.