Forward Guidance: Managing Expectations as a Policy Tool

Forward guidance is an authority saying what it expects to do later. People act on the words now, before anything has been done. Words reach rates because almost nothing is priced off tonight's rate: a loan or a bond depends on a rate over years, and that is an average of expected future rates. Guidance changes the average without touching the rate that is set.

Two things from earlier in these notes sit underneath that. The first is that there is a policy rate, a single number an authority sets and can change at a meeting. The second is that the authority also carries a policy stanceThe direction an authority says it is leaning in, tightening or loosening, as distinct from the level of the rate it has actually set., a direction rather than a level. Guidance is what happens when the direction is said out loud with enough detail that other people can act on it before anything is done. Everything worked here runs on the Republic of Sankhya, an invented country used across these notes, whose policy rate is 6.00 per cent and whose every figure is illustrative.

Why can words move a rate that nobody has touched?

The objection comes first, and it is the right one to have. An authority sets one number. The authority did not change that number. All it did was say something about what it might do later. How can that possibly move anything? The usual answer is about confidence and persuasion, and that answer is wrong, or at least it is a description of a mood rather than a mechanism. The real reason is arithmetic. Every later claim about guidance rests on it, so it is worth doing slowly.

The number the authority sets is a very short rate. In most arrangements it governs borrowing for a single night, and a rate on one night of money is called the overnight rateThe rate on money lent for one night, usually between banks. Overnight borrowing is the shortest there is, and it is agreed again the next day.. Now look at the decisions that actually matter to people, and ask how long a period each of them depends on. A household in Sankhya borrowing Rs 30,00,000/- to buy a flat is committing for twenty years. A workshop borrowing Rs 2,00,00,000/- to put in new machinery is committing for five. A retired teacher placing Rs 5,00,000/- in a deposit is committing for three. Not one of those decisions is priced off a rate that expires tomorrow morning. Almost nothing anybody cares about is priced off tonight's rate, so the rate the authority sets is not, by itself, the rate that decides anything.

So where does a twenty year rate or a two year rate come from, if not from the one number that gets set? The long rate comes from the same number, expected forward. Think about what a lender is choosing between when it lends for two years. The lender can lend for two years at one agreed rate. Or it can lend for a night, take the money back, lend it again the next night, and keep doing that for two years. If those two courses left the lender in wildly different places, everybody would pile into the better one until they stopped being different. Otherwise the two ways of lending the same money for the same period would not be worth the same, so a two year rate has to sit near the average of the overnight rates people expect across those two years.

Read that sentence again with the emphasis on the word expected. The word expected is the whole of the door. Nobody knows what the overnight rate will be eighteen months from now. Everybody has to work with what they expect. The long rate is built out of expectations, and expectations are made of information, and words are information. An authority that changes what people expect the overnight rate to be in eighteen months has changed one of the inputs to the two year rate. The two year rate then moves. Tonight's rate has not been touched at any point in that chain.

What does that look like on the Sankhya numbers?

A mechanism that has not been counted is a mechanism that has only been agreed with, so figures come next. The Sankhya policy rate is 6.00 per cent, invented for teaching and stated to two decimals throughout. A two year period splits into four half years, and each half year carries a market expectation of what the overnight rate will be.

In the first case, the authority has said nothing beyond setting the rate. The market looks at conditions, decides that the rate will hold for a year and then come down, and carries the expected path 6.00, 6.00, 5.50, 5.50. The four expected rates add to 23.00, and dividing by four gives 5.75 per cent. The two year rate the market is working with is therefore 5.75 per cent, and note that the figure sits below the policy rate even though the policy rate has not moved and nobody has cut anything.

In the second case, the authority says that it expects to hold the rate where it is until a stated condition is met, and the condition will not be met inside two years. The market now carries 6.00, 6.00, 6.00, 6.00. Now the four add to 24.00, and dividing by four gives 6.00 per cent. The two year rate has gone from 5.75 per cent to 6.00 per cent, a move of twenty five basis pointsOne hundredth of one per cent. A shift from 6.00 per cent to 6.25 per cent is twenty five basis points, and a full one per cent is a hundred., and the policy rate is 6.00 per cent in both cases because nothing was done to it.

One honesty note before any of this is carried elsewhere. A real two year rate is not a plain average of four numbers. Lenders ask for something extra to tie money up for a long stretch rather than rolling it over night after night, and that extra is called a term premiumThe extra return a lender asks for committing money over a long stretch instead of lending it again and again for short ones. The term premium sits on top of the average of expected short rates.; the periods are not equal in weight; and the whole structure is built as a curve rather than as four buckets. The simple average is a simplification, chosen to isolate the one thing at issue: a long rate is built from expected short rates and therefore moves when expectations move. Adding the extra bits would change the level and would not change that.

What kinds of guidance are there, and which one is strongest?

Guidance is not one thing. Sort it by what it hangs on. Nothing else predicts how guidance behaves under pressure. There are three kinds and the ranking between them surprises most readers.

The first kind hangs on a date. The authority says the rate will stay where it is until a stated month. A dated statement is beautifully clear, and a market can price it in an afternoon. Now ask what happens when conditions turn before that month arrives. There are exactly two exits. Keep the rate where it is, and hold a rate that conditions no longer support. Or move it, and the words were broken. There is no third door. A date cannot change and conditions can, so a dated statement has no honourable exit.

The second kind hangs on a condition. The authority says the rate will stay where it is for as long as something specified continues to hold, and names the something. Now conditions turn. The stated thing stops holding, the guidance ends because its own terms say it ends, and nobody was misled. Everything the authority said was accurate, including the ending. Conditional guidance is the strongest kind precisely because it can be abandoned honourably, and readers routinely have this backwards because they assume the tighter promise is the stronger one.

The third kind hangs on nothing. The authority says it expects rates to stay low for a while, or that it sees no case for a change at present, and attaches no condition and no date. Such a statement gives a reader nothing to test, nothing to check it against and no way of ever showing it was departed from. Guidance that hangs on nothing is the weakest kind, and it survives because it is the most comfortable thing to say.

What is guidance for, when a rate change already exists?

If an authority wants borrowing to be dearer, it can raise the rate. So what is guidance adding? The answer is not volume. Guidance is not a rate change said more loudly. Guidance reaches a different part of the structure.

Picture the whole set of borrowing costs at every length of time, from overnight out to ten years, drawn as one line; that picture has a name, the yield curveA single line showing what borrowing costs at each length of time, from overnight out to many years. The curve is the shape of rates across periods rather than one rate., and it serves here only as a picture. The policy rate pins the left hand end of that line, and it pins it hard. A move in the policy rate moves the very short end almost fully, almost at once. But a ten year rate is an average over forty half years, and a change to the first of them barely shifts that average, so the further right along the line, the less the near end has to say about it.

So the two tools do not compete: the rate change works on the near end and guidance works further out, and an authority that wants the far end to move has to say something because there is no lever there to pull. This is also why guidance gets reached for when the near end has run out of room. If the rate cannot usefully go lower, the only part of the structure still available is the part built out of expectations, and the only way in is words. The money marketThe part of the market where borrowing and lending runs for very short periods, days or weeks rather than years. The money market sits at the short end of the structure. at the near end will follow a rate change on its own; the two and five year end will not.

What does breaking guidance cost, and who actually pays it?

Suppose guidance is given, and then abandoned, and the stated condition has not changed. Ask exactly what went wrong. Nothing mechanical. The rate is at whatever level the authority has now chosen and works exactly as a rate works. The damage is somewhere else entirely, and it is in the future.

The break destroys the informational content of the next statement. Before the break, people took a sentence from this authority as information about what would happen, and the sentence moved the expected path. After the break, people weigh the same sentence against the memory of a sentence that did not hold, and it moves the path less. Push further and the sentence moves nothing, at which point the tool has stopped existing. Breaking guidance does not cost the decision that broke it anything at all; it costs every statement that comes afterwards some of its ability to move a rate, and that bill is paid by the institution over years.

The temptation is therefore structural rather than a matter of anyone's character. On the day, the authority faces a real choice with a real cost on one side: hold a rate conditions no longer support, or move it and take the reputational hit later. The cost of holding lands today, on this decision, on these numbers. The cost of moving lands later, spread thin, on statements nobody has written yet. A decision maker comparing a sharp cost now against a diffuse cost later will lean towards moving more often than is good for the tool. And the tool belongs to the office rather than to the person holding it this year.

Why does a tool this cheap get used more than it should?

Lined up side by side, the shape of the problem is obvious at once. Saying something costs nothing today. There is no outlay, no paperwork, no market operation, no reserve to fund and nobody to ask. A sentence is added to a statement and the expected path moves. Withdrawing it costs the credibility of everything the authority says afterwards, and there is no way to pay that bill in advance or to pay it off early.

An ordinary version of this is a shopkeeper who tells regular customers that a price will hold until the festival season. Saying it takes a second and brings people in this week. Going back on it takes a second too, and the next time the shopkeeper says anything about prices, the regulars discount it. The second cost never appears on any till roll and it is real all the same.

Any tool that is free at the moment of use and expensive only later gets used more often than it should be, so a reader should expect guidance to be given more freely than it deserves, in more situations and in vaguer forms. The expectation is not cynicism about anyone. It follows from the shape of the incentive and would hold whoever was doing the job. Vagueness carries the smallest bill of all, and the vaguest kind of guidance is for that reason the most common kind a reader will meet.

How should a guidance statement be read?

Three questions, in order, and they take about a minute. The three questions are the same whatever the statement says and whoever issued it.

First, what is it conditional on? The words to read for are until, or for as long as, or provided that. Where the thing the statement hangs on can be pointed at, the guidance is conditional and what would end it is already known. Second, is that condition observable? A condition tied to something published, on a stated schedule, by somebody other than the authority itself, is one that can be watched. A condition tied to conditions remaining supportive is not a condition, it is a mood, and it cannot be watched. Third, what would have to happen for this to be abandoned honourably? Where that can be answered in one sentence, the statement is doing real work. Where it cannot, the statement can be abandoned at any moment and nobody will be able to say it was departed from.

Guidance with no stated condition is a forecast wearing a policy label, and the way to tell the two apart is that a forecast can be wrong while guidance can only be kept, honourably ended, or broken. Both are useful, and they are useful for different things. Treating the first as the second is where readers get hurt.

Why does a lender watch the guidance more closely than the rate?

Here is the practical version, and it is where guidance stops being an idea about communication and starts being somebody's margin. Take a lender in Sankhya with a loan bookThe whole set of loans a lender currently has outstanding, taken together as one thing rather than loan by loan. running for years and funding that has to be arranged over years as well. Tonight's rate does not decide whether that book earns anything. Rates over the next few years decide it, on both sides of the book.

So when a statement lands, the lender is not reading it for the rate, a number it already knows. The expected path is what its longer term funding and its longer term lending are both priced against, and the lender reads the statement for what it does to that path. A change in expectations reaches those numbers immediately, on the day, without any decision having been taken. For anybody holding commitments that run for years, guidance arrives before a rate change does and is already reflected in prices while the rate is still exactly where it was last week.

The same logic reaches a household, in a smaller way. Somebody choosing between a fixed rate and a floating rate on a long borrowing is making a bet about the path, not about tonight. Somebody choosing between a three year deposit and rolling one year deposits three times is making the same bet in the other direction. Neither of them needs to follow any of this to be affected by it. How a lender actually turns any of that into the rate it charges a borrower, and what it does to its own funding, is a different subject, covered separately.