Agency Costs: When Management and Owners Want Different Things

An agency cost is what it costs the owners of a company that the people running it are not the same people. Jensen and Meckling named it in 1976. The cost carries a rupee figure: Sankalp Industrial Systems Limited, an invented manufacturer, bought back 75,00,000 shares and lifted earnings per share by 3.75 per cent on the naive count and 1.76 per cent honestly. A team paid on the first number prefers it.

The relationship underneath is old and it is very simple. One person hires a second person to act for them. The second person sees more of what is going on than the first person does. And at the margin, the two of them want slightly different things. Nothing in that description requires anybody to behave badly. The cost is hard to remove for exactly that reason. It survives honesty. It survives competence. The measure the second person is judged on is never quite the same thing as the outcome the first person cares about, so the cost survives everyone in the room genuinely trying to do a good job.

What is an agency relationship, and why is a company full of them?

Consider a householder hiring a contractor to add a room to a house. The householder is paying, and will live in the room for twenty years. The contractor will be gone in eleven weeks and is paid on the day the work is signed off. The householder wants the wiring done properly behind the plaster. The contractor wants the plaster on. Neither of them is a villain. The two simply have different exposures to the same wall.

The gap between those two exposures has a name in each direction. The person on whose behalf the decision is taken is the principal. The person taking it is the agent. The arrangement between them is an agency relationship, and it appears wherever three conditions hold at once: somebody acts for somebody else, the person acting knows more about what is happening, and what the two of them want is not identical all the way down.

A company is not one agency relationship but a stack of them. Shareholders hire a board. The board hires a chief executive. The chief executive hires the head of a division. The head of a division decides which machine gets replaced. At each step somebody is spending money that belongs to somebody further up, and at each step the person spending it knows more about the decision than the person whose money it is. Lenders sit in the same picture from a different angle: they have handed over cash and cannot see day to day what is being done with it.

A street vendor who runs the cart alone has none of this. Every rupee that goes in or out passes through one pair of hands, and the person deciding is the person who eats the consequence. Leaving a cousin at the cart for an afternoon brings the whole apparatus into existence at once: who counts the takings, how the vendor knows what was sold, whether the cousin is as careful about a dropped tray as the vendor would have been. Scaled up to twenty thousand people and Rs 12,00,00,00,000 of annual revenue, that afternoon is a listed company.

Who named it, and what did they actually claim?

Michael Jensen and William Meckling set the idea out in a paper called Theory of the Firm, published in the Journal of Financial Economics in 1976. The frame is theirs.

Their claim was sharper than the loose version that circulates. Jensen and Meckling did not say that managers cheat. The claim was that once ownership and control sit in different pairs of hands, a cost arises as a matter of structure, and that this cost is the sum of three separable things: what the principal spends on monitoring, what the agent spends on bonding, and the residual loss that is left after both. The insight that made the idea useful was not that conflict exists but that the conflict has a size, and that the size can in principle be written down in money.

The second half of their claim is the half people forget. If the cost has a size, then reducing it also has a cost, and there is a point past which more watching costs more than the watching saves. A company that audits everything twice has not eliminated its agency cost; it has converted a large residual loss into a large monitoring bill and possibly made itself worse off. The answer to an agency problem is almost never simply more oversight.

What are the three parts an agency cost is made of?

Take the three in the order a company meets them. Monitoring is what the owners spend to watch, bonding is what the agent spends to be trusted, and the residual loss is everything still given up once both bills have been paid.

Monitoring is the easiest to see because it arrives as an invoice. An external audit is monitoring. So is a monthly reporting pack that nobody would prepare if the person deciding and the person paying were the same. So is a second valuation commissioned on a purchase, and so is the hour a board spends on a paper it could otherwise have taken on trust. In household terms it is the day taken off work to stand in the half-built room and look behind the plaster before it is closed up.

Bonding runs the other way and is paid by the agent. A management team that accepts a five year vesting periodThe stretch of time that has to pass before shares or units granted to an employee actually become theirs to keep or sell. on its share awards is spending something real: freedom, and the option to leave. A chief executive who agrees to a covenantA promise written into a loan agreement that limits what the borrower may do, with a consequence attached if the limit is crossed. that caps borrowing has given up room to manoeuvre. The contractor who offers a two year guarantee on the wiring is bonding. In every case the agent narrows their own choices in order to be believed, and narrowing those choices is never free.

The residual loss is what is left. Both bills have been paid, the audit is clean, the vesting is long, the covenant holds, and the owners are still worse off than they would have been if they had somehow been able to take every decision themselves with full information. The gap is real, it is denominated in money, and nothing in the accounting system will ever show it.

Why does the residual loss appear in no expense line?

Because nothing happened. The absence is the whole of the answer, and it repays a moment of attention, because it explains why this component is the one that gets managed last and costs the most.

An expense line records a transaction. Money moved, a supplier was paid, a service was received, an entry was made. The residual loss is not a transaction. The residual loss is the difference between the decision that was taken and a decision that was never taken by anybody, in a company that never existed. There is no counterparty, no invoice number, no date and no signature. Nothing was stolen, no rule was broken, no statement was misstated, and the owners are still poorer than they would otherwise have been.

Consider what a forensic accountant would be able to find. Every rupee that left Sankalp Industrial Systems Limited in the buyback below can be traced: the bank instruction, the price, the share count, the cancellation. None of the alternatives happened, so none of them can be traced: not the interest the money would have earned somewhere else, not the machine it would have bought. An absence cannot be audited.

Where do the interests of managers and owners actually part company?

Vague suspicion is useless. Saying that management might not be aligned supplies nothing that can be checked. Each axis produces a different kind of decision and each one can be interrogated separately, so the idea only becomes a working tool when the divergence is broken into named axes.

There are three places the two sides part: how long they are each exposed, how much of their own future rides on one outcome, and whether they are rewarded for a bigger company or a better one. Take them one at a time.

Divergence one, horizon: this year against the years after it

A person running a division has a tenure. The tenure might be four years, it might be nine, but it has an end, and the record they will be judged on is made of the years inside it. An owner holding the shares has no such boundary. The value of the shares today already contains every year the company will have, including the ones after the current team has gone.

So put a real decision in front of both. Sankalp Industrial Systems Limited, invented, plans to commit Rs 1,00,00,00,000 of net new capital in each of the next five years, and the return on that money is expected at 18.00 per cent against capital costing 12.00 per cent. Neither of those two percentages is a fact about anything. Both were chosen so the example has something concrete to work on. Now suppose a particular tranche of that spending only starts producing in year eight. To an owner the money is worth what it is worth whenever it arrives. To a person who will have handed over the division by year six, the same tranche is a cost with somebody else's benefit attached.

Nothing about this requires bad faith. A perfectly honest person facing that shape will weigh it differently from an owner who holds throughout, and paying that person on the years inside the tenure does not merely allow the difference, it pays for it.

Divergence two, risk: a career against many holdings at once

An owner who holds shares in forty companies experiences one disappointing year at one of them as one disappointing year out of forty. The person running that one company experiences it as their year. Their salary, their standing, their next job and quite possibly their sense of themselves all ride on the same single outcome.

The consequence runs in a direction people often get backwards. The problem is not that managers gamble; it is that managers are usually more cautious than the owners would want them to be. A decision with a good expected outcome and a visible chance of looking foolish is a bad decision for one career and a good decision for a spread of holdings. The household running on one salary refuses a promotion that involves moving cities. A cousin with four sources of income takes the same chance without blinking. Neither is wrong. The two are facing different exposures to the same event.

Divergence three, scale: a larger company against a better one

Pay, standing and the interest of the next employer all tend to track how much a person is running, not how well the money spent on it has done. Nobody designs that on purpose; it is simply what happens when size is easy to observe and returns are not.

So a purchase that makes the company larger and slightly worse can look like a triumph from inside and a loss from outside. Sankalp Industrial Systems Limited holds Rs 12,00,00,00,000 of revenue and a business in industrial valves, precision castings and the aftermarketThe parts, servicing and replacement work sold to a customer after the original equipment has been delivered, usually over many years. parts and service that go with them. A purchase that added Rs 3,00,00,00,000 of revenue at a return below what the capital costs would make every size measure move up and the owners poorer. The number scale improves is the number everybody quotes, so scale is the divergence that hides best.

How Agency Costs Can Affect Capital Allocation

Everything above is a description of a tension. Capital allocation is where the tension turns into money, and it is the reason the subject is worth studying at all. The decision about where the next rupee goes is large, discretionary, hard to check afterwards, and scored by a measure that responds differently to each of the available answers. Agency costs bite hardest at exactly that point.

Start with the concrete case. Sankalp Industrial Systems Limited holds Rs 1,20,00,00,000 of cash. Of that, Rs 40,00,00,000 is what keeps the doors open: the float that absorbs a bad collection month, and the payroll that has to clear whatever else is going on. The other Rs 80,00,00,000 is excess. The excess is not doing anything the business requires, and it necessarily has somewhere else to be.

Every rupee of that Rs 80,00,00,000 has exactly three destinations. The money can go back into the business, as more machines, more capacity or more working capital. It can go to the lenders, reducing the Rs 6,00,00,00,000 of gross debt the company carries. Or it can go to the owners, as a dividendA payment a company makes out to its shareholders, usually in cash and usually declared as an amount for each share held. or a buybackA purchase by a company of its own shares from the people holding them, after which those shares stop existing and the count outstanding falls.. There is no fourth door. Nothing else can be done with a rupee that a business does not need.

Now notice what the three doors do to the person standing in front of them. Putting the money into the business is the door whose benefit arrives slowly and outside a short tenure. Paying down borrowing helps quietly and shows up in nobody's headline. Returning it to owners moves a per share measure immediately and fits on one slide. The three doors are not equally visible, and the one that is most visible is not the one that is most often right.

The same logic scales up from the Rs 80,00,00,000 sitting there to the Rs 1,00,00,00,000 of fresh capital the company plans to put in each year. Over five years that is Rs 5,00,00,00,000 of discretionary money, all of it allocated by people who are measured on something. Whether that money should go where the plan says is a separate question with its own arithmetic and it is covered separately. One narrow claim is enough on its own: the chooser has an interest in the answer, and that interest is not identical to the interest of the person whose money it is.

Why does the measure a person is paid on become the measure the company manages?

The link between a measure and a decision is the hinge of the whole subject, and it is more mechanical than cynical. People do not respond to what is wanted of them. People respond to what is counted.

A delivery rider paid for each parcel produces speed. The same rider paid for each hour produces care and fewer parcels. Nobody in either arrangement has decided to be a worse or better person; a number was chosen, and behaviour arranged itself around the number. Every measure a reward is attached to becomes a target, and every target quietly reshapes the decisions taken underneath it.

Inside a company the measures that get attached to reward are usually the ones that are easy to compute and hard to argue about: revenue, an operating margin, a profit figure, or earnings per shareThe profit belonging to shareholders divided by the number of shares outstanding, reported as an amount for each share.. Each of those has the property that a comparison against an alternative is not part of the calculation. Any of them can be raised by taking an action whose alternative was better, and the measure will not say so.

Earnings per share is the measure the worked example below runs on. The example is not a story about somebody misleading a board. A figure can be entirely correct, prepared by a competent person in good faith, and still be the wrong figure to have looked at.

Can an agency cost be measured in rupees?

An agency cost can be measured in rupees, and here it is. Everything from here on runs on one transaction at Sankalp Industrial Systems Limited, taken as given.

At the end of the year two years before the base year the company repurchased 75,00,000 of its shares, paying Rs 80.00 for each of them. The bill came to Rs 60,00,00,000. Afterwards the count stood at 20,00,00,000 where it would otherwise have stood at 20,75,00,000. How a buyback is actually executed, and what it does to the balance sheet, is covered separately. Only the two numbers the transaction produced are needed below.

The first number is the one a board paper prints, and it is 3.75 per cent. Year 0 profit attributable to owners is Rs 1,38,00,00,000. Spread over the 20,00,00,000 shares that now exist, that is Rs 6.90 a share. Spread over the 20,75,00,000 shares that would otherwise still have been outstanding, it would have been Rs 6.6506. The rise is exactly 3.75 per cent, and the reason it is exact is arithmetic rather than luck: 20,75,00,000 divided by 20,00,00,000 is precisely 1.0375.

The second number is the one that is honest, and it is 1.76 per cent. The Rs 60,00,00,000 did not vanish. The money was spent, and spending it stopped it doing whatever else it was going to do. One alternative is a deposit earning 6.00 per cent before tax. The 6.00 per cent is an assumed rate rather than an observed one, and a real deposit rate would sit above or below it. The honest figure moves with whichever rate is used. The cash would then have added Rs 3,60,00,000 before tax. Less the 25.0 per cent effective tax rateThe proportion of pre-tax profit a company actually ends up paying in tax once every allowance and adjustment has run through, rather than any headline rate. this company assumes for itself, a figure invented for this example, and Rs 2,70,00,000 is left. Profit would have been Rs 1,40,70,00,000, and spread across the 20,75,00,000 shares that gives Rs 6.7807. Against that comparison, the buyback's true effect on earnings per share is 1.76 per cent.

Neither figure is a lie. Both are arithmetically correct. The two readings differ on one thing only, and it is not a number: it is the choice of what to compare against. An improvement figure is always a comparison, and whoever chooses the comparison has already chosen most of the answer. The rupees behind the 1.99 percentage point gap are the Rs 2,70,00,000 a year of after tax income that no longer arrives, and that is what an agency cost looks like when somebody finally writes it down.

Why is the first number the one that gets printed?

Because it is the one that requires no extra work, no extra assumption and no extra argument. The 3.75 per cent falls straight out of two share counts and one profit figure, all three of which are already in the accounts. Nobody has to decide anything.

The 1.76 per cent requires somebody to state, out loud and in writing, what the Rs 60,00,00,000 would otherwise have done. A statement of that kind is an assumption, and assumptions can be challenged. A person who puts one in a paper has handed a committee something to argue with, and has done so in order to make their own transaction look worse. Meanwhile a paper carrying only the 3.75 per cent is complete, correct, and impossible to attack on its arithmetic.

So the naive figure wins not because anybody prefers it dishonestly, but because the honest figure costs the person preparing it something and pays them nothing. That is the residual loss in operation: no rule broken, no rupee misplaced, nobody to blame, and a real gap between the decision taken and the decision that was available.

Two things are worth taking from the control. The first is that the overstatement is not a fixed quantity. The overstatement shrinks steadily as the return the cash would have earned rises. The size of this particular agency cost therefore depends entirely on an assumption that the board paper never made. The other thing to take away is that the crossing is nothing like arbitrary. The improvement vanishes at exactly 11.50 per cent before tax, because 11.50 per cent after tax is 8.625 per cent, which is precisely the Rs 6.90 of earnings sitting on the Rs 80.00 that was paid. The line moves with the price paid for the shares, and with nothing at all about the business.

The curve draws an identity in the arithmetic and nothing more. The identity does not establish when repurchasing shares is worth doing. A per share earnings figure is not a measure of what a business is worth, and the two can move in opposite directions on one transaction.

How does anyone actually use this?

Three readers use the idea in three quite different ways, and none of them needs a model to do it.

A lender uses it to decide what to write into the agreement. A lender has handed over money and cannot see the decisions being taken with it. The arrangement is the agency relationship in its purest form. The response is bonding purchased in advance: a covenant capping borrowing, a restriction on how much may be returned to shareholders while the loan is outstanding, a requirement to be told before a large purchase. Every covenant in a loan agreement is an agency cost that somebody decided to pay in order to avoid a larger one. Sankalp Industrial Systems Limited carries Rs 6,00,00,00,000 of borrowing, spread over three tranches, and the lenders behind each of them sit on exactly this kind of paperwork.

An analyst uses it as a reading habit rather than a calculation. When a company reports an improvement, the first question is against what. When a company holds cash it does not need for several years running, the question is not what the treasuryThe part of a company that manages its cash, its borrowing and its banking arrangements day to day. function is doing with it but why the choice keeps being deferred. When incentive pay is tied to a measure, the analyst reads that measure as a prediction of which decisions the company will find easy.

An investor, or anybody deciding whether to hand money to somebody else to manage, uses it as a checklist of three: how long is the person deciding exposed to the outcome, how much of their own future rides on this one thing, and are they rewarded for size or for return. The three questions come straight off the three axes, and they work on a company, on a small business run by a relative, and on a household deciding whether to let one member handle the joint account.

Notice what none of these three does. None of them tries to remove the agency cost. The cost cannot be removed, and every attempt to remove it converts residual loss into monitoring cost at some exchange rate nobody has computed. The three readers size it instead, name where it sits, and price it into what they are willing to pay or lend.

What is not an agency cost?

The term is useful in proportion to how narrowly it is used, and it is routinely stretched until it means nothing more than a disappointment.

A project that fails because demand turned out lower than forecast is not an agency cost. The failure is a bad outcome. Everybody wanted the same thing, everybody worked from the same forecast, and the world did something else. The fix for it is a better forecast or a smaller commitment, not a different incentive.

A decision that turns out badly is not an agency cost either, and neither is a decision that was simply wrong. Incompetence costs money and it is a real problem, but the person deciding was trying to achieve exactly what the owners wanted and failed. There is no divergence of interest to point at, and treating it as one leads to the wrong repair.

The cost of running a company is not an agency cost. Salaries, offices, an audit fee that would exist in any structure: these are the price of operating, not the price of separation. And the plain cost of information is not one either. Even a person deciding entirely for themselves has to spend something to find out what is going on.

An agency cost requires a divergence of interest between the person deciding and the person whose money it is, and where that divergence cannot be named on one of the three axes, what is in front of the analyst is something else. Calling every disappointment an agency cost drains the term of exactly the meaning that makes it worth having.