Covenants: The Constraints and What a Breach Triggers

A covenant is a test written into a borrowing contract. The test measures something about the borrower, on a stated definition, at stated times, against a stated level, and failing it gives the lender rights it did not have before. A covenant prevents nothing. A covenant turns a slow deterioration into a dated event, and a dated event is the only thing a lender can act on.

Here is the problem a lender is actually trying to solve, and it is not the one people assume. Borrowers weaken. A lender takes that much for granted and prices for it. The awkward part is that weakening arrives without a date on it. There is no morning on which a business becomes noticeably worse than it was the previous evening, and a right that has no trigger is a right nobody can pick up and use. So a lender does the only thing available: it picks one measurable quantity, writes down how far is too far, and attaches a consequence to the crossing. A gradual change then acquires a moment at which somebody has to do something.

The idea is that simple. The difficulty is how much can hide inside the words chosen for that one measurable quantity: a single test, with a single number written into it, can produce two answers that are both correct and a long way apart, and the gap between them is worth measuring in the only currency that matters to a borrower under pressure.

What does a covenant actually do?

Less than most readers expect, and the gap between what it does and what it is imagined to do is where careless lending starts. A covenant does not stop a borrower weakening. A covenant does not put cash into a business that has run short of it. No wording causes a payment to arrive on a day when the money was not there. Not one of those things is inside a lender's reach through a document, and no wording has ever changed that.

A covenant arranges a schedule of looking instead. On a stated date, a stated measurement is taken. If that measurement has passed a stated level, the lender acquires, on that date, rights it did not hold the day before. A covenant is not a protection at all; it is an early date, and the date is the thing being bought.

A familiar arrangement runs the same way. A landlord who inspects a rented flat every six months has not made the tenant tidier. The tenant's habits are whatever they are, and an inspection clause has never scrubbed a kitchen. The landlord has arranged finding out sooner: an inspection in month six rather than a surprise at the end of a three year lease. The damage, if there is any, is identical in both worlds. The difference is how much of the lease is left to do something about it.

The alternative reading sounds strange once it is stated plainly. If a covenant genuinely made borrowers safer, then writing a tighter one would make a borrower safer still, and the strongest lending in the world would be the lending with the most tests attached to it. In practice, tests cluster where lenders were least comfortable to begin with. The clustering runs the opposite way to that reading. The presence of a schedule of measurements says something about how the lender felt, and nothing at all about how the borrower will trade.

The place where the line meets the level repays attention. The line crosses between the fourth date and the fifth, and nobody learns about the crossing until the fifth. Somewhere in that gap the borrowing was, on the test's own terms, no longer within its terms, and every party to it carried on as though nothing had happened. The gap is not a flaw in the drawing. A dated measurement works exactly that way, and what a covenant cannot see between its dates is set out below.

Which kind of test is it, and when does the lender find out?

Covenants come in two shapes and the difference between them is not strength. The difference is the trigger for looking. A maintenance covenant is tested at fixed dates, whether anything has happened or not. The quarter ends, somebody prepares the reading, and the reading is compared with the level. Nobody has to have done anything to bring that about. A borrower drifting quietly, taking no decisions and announcing nothing, still meets a test on schedule.

An incurrence covenant is tested only when the borrower wants to do a specific thing. Borrow more. Pay money out to its shareholders. Sell a substantial asset. The test sits in the document like a gate on a road that is only ever closed when somebody tries to drive through it, and a borrower who never tries to drive through it is never tested at all. On that borrowing, in the two years during which nothing was attempted, no measurement was ever taken and no level was ever compared with anything.

The two are not strong and weak versions of one another: one of them buys dates and the other buys a veto over actions, and a lender who needed dates and holds vetoes finds out at the end rather than along the way. That is a design question settled when the document is drafted, and it is settled once. Nothing about the level chosen, or how demanding it looks on paper, changes which of the two things the lender has actually bought.

The everyday version is a car. One arrangement says the car goes in for a service every six months regardless of how it has been driven. The other says nothing about servicing at all but requires an inspection before the car may be lent to anybody else. The second is not a weaker version of the first. The second answers a different question, and a car driven hard and never lent out is a car nobody looks at, however carefully that arrangement was drafted.

Which of a covenant's five parts carries the work?

A covenant is usually quoted as a single number, and a lender who can quote only that number has read about one fifth of it. Pull one apart and there are five separable parts, each of which can be argued over independently and each of which changes the answer in a different way.

The five parts are: what is measured, on what definition, how often, at what level, and what a failure entitles whom to do. Their weights are very unequal, and the one everybody quotes is not the heavy one. Take them one at a time.

What is measured. A leverage reading, a coverage reading, a cash balance, a total of capital spending. The choice of instrument decides what kind of trouble is even visible. A test built on a leverage reading looks at the size of the obligation against a year of earnings; a test built on interest coverageA reading that sets a year of earnings against a year of interest, so it asks how many times over the interest bill is covered. How it is constructed from a set of statements is settled elsewhere on this platform. looks at whether a year of earnings clears a year of interest. Leverage and coverage can move in opposite directions at once, and a document that carries only one of them is blind to whatever the other would have shown.

On what definition. The exact meaning of every term inside the measurement. The definition carries more of the work than the other four parts and is skipped more often than any of them. A definition takes many lines to write down; a level takes four characters.

How often. The test dates, and whether the measurement covers one period or several added together. A reading taken on a single quarter and a reading taken on four quarters added together are different instruments with the same name, and one of them is far harder to have a bad month inside.

At what level. The number the reading is compared with, and whether that number stays put over the life of the borrowing or steps down as the years pass. The level is the part everybody quotes.

What a failure entitles whom to do. The consequence. Note the shape of that sentence carefully: it names a right and a holder of the right, not an outcome. A great deal of confusion about breaches comes from reading this part as though it described an event that occurs by itself.

Why can one test have two correct answers?

The definition matters more than the other four parts put together, so it is worth working on a set of amounts declared for the arithmetic and describing no borrower. Debt of Rs 800 crore. Cash of Rs 300 crore. Interest of Rs 40 crore a year. Then earnings before interest, tax, depreciation and amortisationWhat a business earned in a year before financing costs, tax and the write-down of long-lived assets come out of it. Building it from a published set of statements is settled elsewhere here. running at Rs 200 crore a year. And one test, also declared here and taken from no agreement: a leverage reading compared with a level of 4.50 times a year's earnings.

The test looks like a complete instruction and is not one. Leverage has two ordinary meanings, both in daily use, and the level does not say which is meant. The gross reading sets the whole debt against a year of earnings. The net reading subtracts the cash first, on the reasoning that money already sitting in the account is money the borrower could throw at the obligation tomorrow, and sets the remainder against the same year of earnings.

Work both. Gross is Rs 800 crore over Rs 200 crore, or 4.00 times a year's earnings. Net debtThe debt with the cash held against it taken off first. Which cash counts, and whether all of it counts, is settled elsewhere on this platform where the two leverage readings are built. takes the cash off first, so Rs 800 crore becomes Rs 500 crore, and the net reading sets that Rs 500 crore over the same Rs 200 crore to give 2.50 times a year's earnings.

One test, one number written into it, two answers that are 1.50 turns apart, and neither of them is a mistake. The mistake, if there is one, was made long before the reading: it was signing a test whose definition had not been settled in writing, and that mistake is invisible on every single day until the day the reading gets close to the level.

Carrying that identity turns a mystery into a quantity. Somebody who cannot see why two lenders disagree about a borrower's leverage can now ask one question instead of arguing: how much cash is there, relative to a year of earnings? A borrower holding a year and a half of earnings in cash will have two leverage readings a year and a half of turns apart. No amount of care about the arithmetic closes that gap. The gap is not an arithmetic error at all, but the answer to a different question.

What is headroom, and how should it be written down?

Headroom is the distance between where the reading sits today and the level at which the test is failed. Headroom is naturally written in turns, meaning units of the same multiple the reading is stated in, and on the declared amounts it comes to 0.50 turns on the gross definition and 2.00 turns on the net definition. Both come from a subtraction already done: 4.50 less 4.00, and 4.50 less 2.50.

One more question turns that number into information. How far can the earnings actually fall before that distance is used up? Turns do not answer it. A turn is a unit of a ratio, and a borrower under pressure does not lose turns; it loses earnings. So convert.

On the gross definition the reading reaches the level when the earnings have fallen to Rs 800 crore over 4.50, or Rs 177.7778 crore a year. The fall is Rs 22.2222 crore, and set against the declared Rs 200 crore that is 11.1111 per cent. On the net definition the reading reaches the level when the earnings have fallen to Rs 500 crore over 4.50, or Rs 111.1111 crore a year: a fall of Rs 88.8889 crore, or 44.4444 per cent measured on the same Rs 200 crore. Check both back, since a reader who cannot reproduce a number will not trust it. Rs 800 crore over Rs 177.777778 crore is 4.500000, and Rs 500 crore over Rs 111.111111 crore is 4.500000.

The same test, at the same level, permits four times as much of the earnings to disappear under one definition as under the other, and that difference is what the definition was worth. A borrower on the gross definition is one bad tenth of a year from a dated event. The same borrower on the net definition can lose nearly half of a year's earnings and still hand in a passing certificate.

The shortcut, and the trap sitting next to it

There is a one-line route to those percentages. Derive it rather than memorise it: the derivation shows exactly when the route stops working. With the reading as the debt over the earnings, and the level as a fixed multiple, the earnings reach the level when they have fallen to the debt divided by the level. So the fall, as a fraction of where the earnings started, is one less the reading divided by the level, and rearranging gives the headroom divided by the level.

Here the honest account departs from the version most readers expect. Because both definitions on this borrowing are tested against the same declared 4.50, the conversion divides both headrooms by the same number. The conversion rescales the two headrooms; it does not reorder them. Four times as much room in turns is four times as much room in earnings, and a lender who held only the turns figure was not, in this instance, misled about the ratio. Turns and a fall in the earnings carry the same information whenever the level is shared, and they stop carrying the same information the moment the level is not.

Watch the divisor do the damage. Take a headroom of 0.50 turns against the declared level of 4.50 times, a permitted fall of 11.1111 per cent. Now take a headroom of 0.50 turns against a second level, of 2.00 times, declared for the arithmetic alone and attached to no borrowing. The permitted fall is 0.50 over 2.00, or 25.0000 per cent. Identical headroom, written identically on both files, and the second borrowing has more than twice as far to fall before anything happens. A turns figure cannot survive that comparison, and a lender makes it constantly. Borrowings are compared with each other, not with themselves.

What does a breach entitle a lender to do?

A failed test typically gives the lender the right to demand that the whole borrowing be repaid now. Read that sentence twice. Everything turns on the difference between holding a right and using one. Nothing happens by itself on the morning a certificate comes in showing a reading past the level. The change is that somebody now has options they did not have the previous week, and options have to be exercised by a person making a decision.

There are four realistic paths and they are worth naming plainly. The lender may waive the breach and carry on, sometimes in exchange for a fee. A second path is a period during which the borrower puts the reading back, after which the breach is treated as undone. A third is new terms, typically a higher rate, a tighter test, or more collateralSpecific property pledged so that a lender can look to it if the borrowing is not repaid. What collateral adds, and how a charge over it is created and registered, is covered separately. pledged against the borrowing. The fourth is calling the borrowing due.

Call a borrowing due from a borrower that would have recovered, and a difficulty has been turned into a failure. Waive the same breach over and over, and the only dated event the lender had has been given away. Both of those are real costs and they point in opposite directions, which is why no arithmetic settles the choice. The four paths below each give something up. Which of them is right depends on facts about a particular borrower that no set of accounts records.

One point that looks like drafting detail and turns out not to be. Whoever holds the right has to be identifiable before any of this is usable. On a borrowing held by one lender that is straightforward. On a bond held by many, the rights typically sit with a trusteeA party appointed to act on behalf of the holders of a bond, so that rights belonging to many holders can be exercised by one identifiable body. The duties placed on a trustee are set by an authority. acting for the holders, and what that body may or must do on a breach is fixed elsewhere.

What can a covenant not see?

Four things, and a lender who has all four in mind reads a compliance certificate very differently from one who does not.

A covenant cannot see between its own dates. A reading that is comfortable on a test date and uncomfortable a week later is a passed test. The gap is not a defect in the drafting. A dated measurement is exactly that, and the first drawing above showed a crossing happening in a gap and being discovered afterwards. Dates set closer together make the gaps smaller; they do not disappear, and a borrower who understands the dates can, without doing anything improper at all, arrange affairs so that the dates find a tidier position than the weeks around them.

A covenant cannot see anything it does not measure. A test built on a leverage reading is silent about a customer who has stopped paying, a plant that has stopped running, or a supplier who has stopped supplying, until those things have worked their way through into the measured quantity. A borrower whose difficulty appears somewhere outside the tested quantity passes every test on the way down, and the certificates it hands in are all perfectly accurate.

A covenant cannot make the measurement honest by itself. The definition is negotiated, and it is negotiated by both sides. A borrower has every ordinary commercial reason to prefer the definition that reads better, and a lender has every reason to prefer the one that reads worse, and what ends up in the document is the outcome of that rather than a neutral fact. The accounting basis underneath it all is set elsewhere and is named among the sources below.

And it cannot tell a lender what to do. That is the whole of the previous part. A covenant buys information and rights, on dates, and that is the complete list of what it buys. Everything else a reader hopes to get from one has to come from somewhere else.

What can be read when there is no agreement?

A reader reasonably expects the worked example to arrive on a named borrowing at this point. Palash Cements Limited, an invented cement maker, has issued a five year bond carrying a 9.10 per cent annual coupon, and that coupon is the entire extent of what is recorded about it.

No borrowing agreement stands behind that bond, and so there is no test level, no test date, no definition and no reading. Without a set of accounts there is no earnings figure to set against a level, and without an agreement there is no level to set it against. Every reading above therefore runs on four declared amounts, attached to nobody, and one declared test level. A certificate filled in from memory would be an invented document, and an invented document is worse than an empty one because it looks like evidence.