Firms
Why Companies Exist At All
Introduction
There is a contradiction sitting at the center of every market economy. The case for markets is that decentralized prices coordinate strangers better than any planner could. Nobody decides how many shoes get made; the price system works it out. Yet most people in a market economy spend their working lives inside organizations where prices do essentially nothing and someone tells you what to do. Your manager does not bid for your afternoon. Nobody quotes an internal price for the accounting department. Inside the firm, the market has been switched off and replaced with a small command economy.
In 1937 a young economist named Ronald Coase asked the obvious question that the field had somehow skipped: if markets allocate resources so efficiently, why do these islands of central planning exist at all, and why are they the size they are? Why does a car company employ its designers rather than contracting each design separately? And if planning inside an organization works so well, why does one firm not simply absorb the entire economy?
The answer he gave has held up for close to ninety years, and it explains far more than corporate structure. It tells you why outsourcing waves arrive when they do. It tells you why some industries consolidate while others fragment, and why your employer keeps some functions in-house and buys others. Most usefully, it tells you what would have to change for any of that to change again.
Using the Market Is Not Free
Coase's insight was that economics had been treating market transactions as costless, and they are not. Every time you buy something rather than doing it yourself, you have to find out who supplies it, discover the price, compare offers, negotiate terms, write something down that covers what happens if things go wrong, and then monitor whether you got what you paid for. Those are transaction costs, and for anything complicated they are substantial.
Once you see them, the firm stops being mysterious. An organization exists wherever it is cheaper to give an instruction than to negotiate a contract. Hiring an employee is a way of buying a bundle of unspecified future work at a standing price, precisely so you do not have to price each task. When your manager asks you to look at something on Thursday, no contract is drafted, no price is quoted, no comparison shopping happens. That saving is the entire reason the organization exists.
This gives a clean, testable rule for where the boundary of any firm sits: activities move inside when the cost of transacting exceeds the cost of managing, and move outside when the reverse is true. It is genuinely predictive. Every large shift in what companies do in-house over the last fifty years - the outsourcing wave, the rise of contract manufacturing, the gig platform - happened after something made market transactions dramatically cheaper.
The Boundary Moves When Coordination Gets Cheaper
The mid-century corporation was enormous and did almost everything itself. General Motors made its own steel, built its own components, and ran its own finance arm. Standard Oil owned wells, pipelines, refineries, and filling stations. This was not corporate vanity. In an era when a long-distance call was expensive, contracts took weeks to negotiate, and monitoring a distant supplier meant sending someone on a train, the transaction costs of buying anything complicated were genuinely enormous. Ownership was the cheaper option.
Then a series of unrelated technologies collapsed those costs simultaneously. Containerization made shipping a component from another continent trivially cheap and reliable. Cheap telecommunications made supervising a distant supplier possible. Standardized interfaces meant a component could be specified precisely enough to buy rather than build. Enterprise software made coordinating across organizational boundaries almost as easy as within them. The predictable consequence, arriving right on schedule, was the great unbundling: firms shed everything that could be specified in a contract and kept only what could not.
The clearest illustration is a company that owns almost none of its production. Apple designs devices and controls the interface with the customer; the physical manufacturing is contracted out. That structure would have been unthinkable in 1960 and is unremarkable now, and nothing about human nature changed in between. The cost of specifying and monitoring a manufacturing relationship fell far enough that ownership stopped paying for itself.
The Hold-Up Problem, or Why Some Things Are Never Outsourced
Cheap coordination did not push everything outside the firm, and the reason is a specific vulnerability that Oliver Williamson built much of his career on. Some investments are useful only within one particular relationship. Build a component plant next to a single customer's assembly line, tooled specifically for their product, and you have created an asset with essentially no value to anyone else. Williamson called this asset specificity, and it changes the balance of power the instant the money is spent.
Before you build, you have alternatives and the negotiation is even. Afterward, your customer knows your plant is worthless without them and can renegotiate the price downward, confident you will accept anything above your operating cost. This is the hold-up problem. Everyone can see it coming, which is why the plant often never gets built - the investment that would have made both parties better off does not happen, because the party making it cannot be protected by any contract they can realistically write.
Common ownership is the standard solution: if the plant and the assembly line belong to the same firm, there is nobody to hold anyone up. This explains a pattern that transaction costs alone do not. Firms outsource generic inputs freely and integrate relationship-specific ones almost regardless of cost, which is why a car manufacturer will buy commodity fasteners from anyone but tends to own or tightly bind the supplier of a bespoke component designed around its own platform.
Ownership Is Authority Over What the Contract Did Not Say
Sanford Grossman, Oliver Hart, and John Moore sharpened this into a definition of ownership that is worth carrying around, because it applies far beyond business. Every contract is incomplete: the world contains more contingencies than anyone can enumerate, so no agreement can specify what happens in every future state. Ownership, on their account, is the right to decide in exactly those unspecified cases. It is residual control - authority over everything the contract forgot.
This is why the question "who owns the asset" has real consequences rather than being an accounting formality. Whoever holds residual control captures the value of unanticipated opportunities and bears the cost of unanticipated problems. Both parties know this in advance, so it shapes how much each is willing to invest before anything is signed. The recommendation that falls out is precise. Ownership should sit with the party whose non-contractible effort matters most, because that is the party whose incentive to invest you most need to protect.
The same logic explains employment itself. An employment contract is remarkably vague about what you will actually do, which looks like sloppy drafting and is in fact the point. It is a purchase of residual control over your working time within broad limits, and it exists because specifying the work in advance is exactly what nobody can do.
What Goes Wrong Inside
Switching off the price system solves the transaction-cost problem and immediately creates a different one. In a market, a supplier who performs badly loses the business, and the signal is unambiguous and automatic. Inside a firm there is no such signal. Someone has to notice, judge, and act - and that someone has their own interests, limited attention, and imperfect information about what everyone below them is actually doing.
This is the principal-agent problem operating at every level simultaneously. Shareholders cannot fully observe executives; executives cannot fully observe divisions; managers cannot fully observe the people they manage. Each layer responds by installing measures, and each measure gets optimized against, which is Goodhart's law arriving on schedule. Most of what people find absurd about large organizations - reporting that nobody reads, targets that reward the wrong thing, processes that exist to demonstrate that a process exists - is not incompetence. It is the visible cost of running an economy without prices.
There is a second cost, less discussed and often larger. Because internal resources are allocated by decision rather than by price, it becomes rational for people inside to spend effort influencing those decisions instead of doing the work. Economists call these influence costs, and they are the formal name for office politics. They are not a cultural failing that better hiring would fix; they are the predictable consequence of any system where the way to get resources is to persuade someone rather than to outbid someone.
Why One Company Does Not Swallow Everything
Those internal costs answer Coase's second question. If firms exist because internal coordination is cheap, what stops one firm from absorbing the entire economy? The answer is that management does not scale linearly. Each additional activity brought inside adds monitoring load, lengthens the chain between the decision and the information the decision needs, and increases the number of people whose effort is easier to spend on internal persuasion than on output.
The equilibrium is where the marginal transaction cost saved by bringing one more activity inside exactly equals the marginal management cost of running it. That is the size of the firm. It is why conglomerates built on the theory that good management is a transferable general skill have repeatedly underperformed and been broken up, and why the twentieth-century assumption that bigger is inherently more efficient turned out to be true only up to a limit that varies enormously by industry.
This is also the strongest available critique of central planning, and it is a more interesting one than the usual argument. The problem with planning an entire economy was never that planners were stupid or malicious. It is that the informational and influence costs that make a firm stop growing at ten thousand people do not disappear at a hundred million - they compound. A national economy is a firm that grew past the point where anyone can see what is happening inside it.
Who Is the Firm For
Everything above describes what firms are. What they are for is a separate question and a genuinely contested one, and the two positions are usually caricatured by their opponents, so it is worth stating each at its strongest.
The shareholder-primacy case, associated with Milton Friedman, is not that shareholders matter more as people. It is that shareholders are the residual claimants - everyone else is paid a contracted amount and shareholders receive whatever is left - so they are the group whose return is tied to whether the enterprise actually creates value rather than merely transfers it. Adding other objectives, on this view, gives managers a defense for any decision whatsoever, because some stakeholder always benefits. A single objective is what makes managers accountable.
The stakeholder case is not that profit is unimportant. It is that shareholders are not in fact the only party bearing uncompensated risk. A worker with firm-specific skills, a town built around one plant, a supplier who made a relationship-specific investment - each of these holds a claim that no contract fully protects, which is exactly the incomplete-contracts logic above applied to people other than owners. If residual risk is the argument for control rights, the argument does not stop at shareholders.
The honest reading is that the theoretical dispute is less decisive than it appears. Both sides converge in practice on the same hard problem: whatever the objective, someone has to be able to tell whether management is pursuing it. Single-objective governance is more measurable and easier to game in narrow ways. Multi-objective governance is harder to game and much harder to check. The evidence does not settle which failure mode is worse, and claims that it does, in either direction, run ahead of what has been shown.
The Firms That Broke the Pattern
Coase's framework assumes management costs eventually bite, which is why firms stop growing. A category of modern company appears to have escaped that constraint, and understanding why is the most economically consequential open question about firms today.
When the product is software, the cost of serving one more customer is close to zero, so the usual brake on scale never engages. Network effects push in the same direction: the value of a marketplace, a social platform, or an operating system rises with the number of people already on it, so the leading position becomes self-reinforcing rather than self-limiting. And the internal coordination costs that historically capped firm size are precisely what information technology has spent forty years reducing. Every force that used to stop a firm growing got weaker at once.
The visible result is a rise in industry concentration across most of the developed world. With it came what David Autor and co-authors call superstar firms: highly productive, extremely profitable, and employing far fewer people per dollar of revenue than the firms they displaced. That last property connects directly to the falling share of national income going to labor. An economy whose output shifts toward firms with structurally low labor shares will show a falling labor share even if no individual firm changes anything. Whether this represents earned efficiency or entrenched market power is the central live dispute in competition policy, and the answer plausibly differs by industry.
Where the Boundary Goes Next
The framework makes a conditional prediction rather than a forecast, which is the most useful thing it can do. If a technology substantially lowers the cost of specifying, monitoring, and coordinating work, the boundary of the firm should move outward again, and more activity should be bought rather than employed. If instead it lowers the cost of managing large internal operations, firms should grow.
Current machine-learning systems arguably do both, which is why confident predictions in either direction should be discounted. Work that could not previously be specified precisely enough to contract out may become specifiable, which pushes activity outward. At the same time, the ability to monitor and coordinate enormous internal operations at low cost pushes the other way, and the firms best positioned to deploy these systems are the ones already largest. The one thing the framework says with confidence is that the boundary is not a fixed feature of capitalism. It is a running calculation, and it has been recomputed several times within living memory.
A company is not a natural object. It is the answer to a cost comparison that a market economy performs continuously: is it cheaper, right now, to negotiate this or to order it? Every reorganization, outsourcing wave, merger, and spin-off is that calculation being run again with new numbers. Once you can see the calculation, corporate life stops looking arbitrary - and the same lens turns out to work on any organization that coordinates people without prices, which includes governments, armies, universities, and most of the institutional world.
