Too Big to Fail: The Problem and the Attempted Fixes

Too big to fail names a position rather than a size. An institution reaches it when the people funding it stop believing it would ever be allowed to stop, so they stop charging for the possibility that it might not repay them. The advantage that follows makes the belief more reasonable than it was. Every fix tried against it works on the belief and not on the balance sheet.

Much of that phrase is misleading. The word big points towards a total. The word fail points towards an event. Neither of those is where the subject lives. The phrase is actually about something held in the heads of the people who hand an institution their money: what they think would happen on a day that has not arrived.

Nothing about the mechanism requires a rescue, a refusal to rescue, or an institution anywhere in the world that ran into trouble. The argument runs entirely from what a funder believes and from what that belief does to the price of a claim, and that price is where the problem lives anyway. An account of some past decision would describe conditions that have since changed; the mechanism itself has not changed and will not.

What does the phrase actually name, if it is not a size?

Here is the correction, stated once and flatly. Too big to fail is a belief, held by the people who fund an institution, about what would happen if that institution stopped. The belief is not a measurement taken inside the institution, and it is not a category anybody applies to a balance sheet. Size is one of the things that can produce the belief, and it is neither necessary for it nor sufficient on its own. An institution can be modest in every figure it reports and still carry a service that would simply not be there tomorrow morning if it went, and the belief attaches itself to that just as readily.

There is an everyday version of this. Consider the one workshop in a small town that can repair a particular kind of pump. The workshop employs four people and turns over very little. If it shuts, every borewell in the district that uses that pump waits, and the people who supply it on credit know that perfectly well, so they extend it terms they would not extend to a larger workshop that competes with six others. Nobody has promised the pump workshop anything. The terms simply reflect what everyone can see about what its absence would cost. SubstitutabilityWhether anybody else could provide tomorrow what this institution provides today. What makes one institution central enough for its trouble to matter more than another's is worked out in full separately. is doing the work, and the size of the workshop is doing none.

The entire mechanism of too big to fail sits in a price and a belief, and both of those can be reasoned about from first principles with nothing having happened to anybody.

Where does the problem live, if it is not inside the institution?

The problem lives in a price, and specifically in one part of one price. The price a funder charges an institution has a part that pays for the use of money over time and a part that stands against the chance of not getting the money back. A funder who is convinced that a particular institution will never be allowed to stop has nothing left for that second part to do. The second part does not shrink because the institution has become sounder. Instead it shrinks because, in that funder's head, somebody else has taken care of what it was there for.

The effect of protection on behaviour is covered separately, and it settles this much: once a price no longer tells two institutions apart, the price has also stopped putting any pressure on either of them. The argument here narrows that from a class of institutions to a single one. The belief here is not about deposits in general or about protected claims in general; it is aimed at one named institution, and the advantage it confers is measured against every other institution that does not attract it. The relative quality of the advantage matters more than it looks. An advantage everybody has is a level of interest rates. An advantage one institution has is a reason it can do things at a price its competitors cannot match.

The price is not the end of the story, so now watch what it does over time. The price is the first step of a loop that turns on its own. The belief lowers what the institution pays for its funding; a lower funding cost lets it do things an equivalent institution cannot do at the same price; and doing more of what others cannot makes the belief more reasonable than it was when it started. Three steps, and the third one hands the first one a stronger case than it had. Nothing has to happen for the loop to turn. Nobody has to make a mistake, take a risk they should not have taken, or run into trouble. The loop is powered by ordinary competent behaviour under a funding advantage. A structural problem, then, rather than an episode somebody could be blamed for.

Why does the belief exist when nobody has promised anything?

This is the part readers find hardest, so it is worth stating as bluntly as it can be stated. An authority that has never said it would step in, that has said clearly and repeatedly that it would not, and that means every word of it, still faces the belief. The belief is not assembled out of anything anybody said, so saying something else does not take it apart. The belief is assembled out of what any reasonably attentive person can see for themselves about the consequences of the alternative.

Work through the funder's reasoning. Funders are not reading a statement and deciding whether the speaker is honest. Instead they are looking at what would be interrupted if the institution stopped, asking who else could pick those functions up and how quickly, and forming a view about what whoever is deciding would actually want to do when the choice is in front of them rather than in a speech. If the answer to that last question is obvious, the statement is beside the point. The gap is not a failure of communication, and no amount of clearer drafting fixes it. The listeners are reasoning about the speaker's position rather than about the speaker's sincerity, and they are entitled to.

Why is the threat to let an institution stop not believed?

Here is the test that decides everything, and it is one line long. A threat is believed only where carrying it out at the moment of decision is better for the person making it than not carrying it out. Not better in advance, when it is being announced and costs nothing. Better on the day, with the actual choice in front of them and the consequences of each branch visible.

Applied to this subject, it comes apart immediately. If letting an institution stop would interrupt functions that other parties depend on, then on the day, faced with the real choice, whoever is deciding would rather not, however firmly and however honestly they said otherwise beforehand. And the uncomfortable part deserves to sit for a moment rather than be hurried past: the promise is disbelieved before anything happens, and it is disbelieved for a good reason rather than out of cynicism. The funders are not accusing anybody of dishonesty. The funders are simply working out, correctly, what the ranking of preferences will look like when the moment arrives, and pricing accordingly today.

The everyday version arrives fully formed. A parent two hundred kilometres from home announces that the car turns round at the next argument. Nobody in that car believes it, including the parent, and not because anybody thinks the parent is lying. Turning round would cost the parent four hours and the entire purpose of the journey, and every passenger can work that out as well as the parent can. The threat was sincere when it was made and unattractive the moment it would have to be carried out, and that gap is visible from both seats. The general shape of this problem, a promise that the promisor will not want to keep once the moment arrives, belongs to Kydland and Prescott, Rules Rather than Discretion, 1977.

What are the three kinds of fix, and which one touches the belief?

Everything tried against this problem falls into one of three shapes, and it is worth being able to name all three because they are constantly confused with each other in ordinary conversation.

The first is to make stopping survivable. Arrange matters in advance so the functions can continue without the institution, and that is what resolution planningWriting down beforehand what would be done, and in what order, so that doing it on the day interrupts as little as possible. What those decisions are was worked through earlier on this subject. means. Have claims already standing that can absorb a loss where they sit rather than having to be repaid. The first fix behaves differently from the other two, and here is why. Once carrying out the threat no longer interrupts anything, carrying it out becomes the better choice on the day, the preference ranking stops reversing, and the threat becomes worth something. Making stopping survivable attacks the credibility directly rather than working around it. The decisions themselves, and the order they are taken in, are covered earlier in this subject.

The second is to make the institution less central. Build capacity elsewhere so that fewer things stop if this one stops. Whether that is possible at all depends on the function; some are easy to duplicate and some are not.

The third is to make the position costly. Place extra requirements on an institution that is in it, so that whatever advantage the belief hands it is at least partly taken back. The requirements themselves, what has to be prepared in advance and filed with an authority, and the conditions attaching to a loss-absorbing instrumentA claim issued on terms that let it take a loss where it stands, instead of having to be repaid in full before anything else can happen. The terms it must carry are set by an authority. are all set by the Reserve Bank of India at rbi.org.in.

Why is every one of the three fixes incomplete?

None of the three finishes the job, and presenting this as a solved problem would misrepresent the most interesting part of it. The three are worth taking in order.

The first one replaces one belief with another belief. Arranging in advance that the functions can carry on without the institution only works if the arrangement is believed, and if it actually works on the day. Neither of those can be verified before the day arrives. The only test that would settle it is the event itself, and the event is precisely what everybody is trying to avoid. So the funders are now asked to believe something new: not that the institution would be let go, but that letting it go would be manageable. A manageable failure is a better belief to be asking for, and it is still a belief.

The second one runs into functions that genuinely have very few providers. Building a second provider of something is slow, expensive, and in some cases not possible at all inside any period that helps. A second settlement route for a payment people make forty times a week cannot be conjured into existence by deciding that there ought to be one. Where a second provider can be built, this fix works quietly and well; where it cannot, saying it should be built is not a fix but a wish.

The third one is the uncomfortable one, and it deserves to be said without softening. Identifying an institution as a systemically important institutionOne that an authority has formally identified as an institution whose failure would matter to the system as a whole. The tests used and everything that follows from being identified are set by an authority and are not stated here. is a public statement that its failure would matter. The same act that imposes the extra cost also publishes the reason for the belief, and publishes it with the signature of the body best placed to know. The funders did not have to work it out for themselves any more; it has now been confirmed for them. The tension is genuine rather than an argument against identifying anybody, and what any authority should do about it is a judgement rather than a fact about the mechanism. A reader who has not seen the tension will think the third fix is simpler than it is.

How would anybody know whether a fix was working?

The measurement is easy to describe and that is what makes it worth describing carefully. Compare what claims on the institution attracting the belief cost against what claims on an otherwise similar institution cost, and watch the gap over time. If the fixes are doing anything at all, that gap narrows. The gap is exactly the right measurement, it is the one an analyst would reach for, and there are two reasons it is much harder than it sounds.

The first is that the comparison needs two institutions alike in everything except the belief, and no two institutions are alike in everything. One has a different mix of borrowers, a different mix of funders, a different cost base and a different history of how it has been run. Any measured gap between the two contains all of those differences plus the one being isolated, and nothing in the number says how much of it is which. None of that is a reason to stop measuring; it is a reason to hold the answer loosely.

The second reason is an absence rather than a difficulty: nothing in this material records what a claim on either institution here costs. Not a coupon, not a spread, not a cost of funding. The measurement can therefore be named to the last detail and not performed. When somebody offers a number for the value of the belief, the first question is which two institutions were compared and what else was different about them.

What do the four questions yield at two invented institutions?

So put the whole thing to work. Here are the four questions worth asking about any institution, in the order they are worth asking: what stops if this stops; who else could do it by tomorrow morning; who is funding it, and would those funders be able to price the risk if the belief were absent; and what has been prepared in advance for the day it is needed. Now run them over the two invented institutions this material carries, and watch carefully where the answers stop coming.

Suvarna Commercial Bank Limited carries Rs 1,92,000 crore of deposits on a balance sheet of Rs 2,40,000 crore, so deposits run to 80.0 per cent OF ASSETS. Out of them, current and savings balancesMoney held in accounts that can be drawn on at any time, without notice and without a fixed term. What these products are and how they differ was settled earlier on this subject. come to Rs 80,640 crore, or 42.0 per cent OF DEPOSITS, and that is money people use to pay each other with on ordinary weekday mornings. Rukmini Finance Limited takes no deposits at all. Rukmini Finance funds itself in the market instead: Rs 14,400 crore of borrowings on Rs 18,000 crore of assets, so borrowings too run to 80.0 per cent OF ASSETS, and it holds nobody's transaction balances. The first question can be answered from those figures for both institutions, and the other three cannot be answered here at all.

A blank with a reason attached teaches more than a filled cell without one, so each blank is worth a line. Answering the second question needs to know who else provides the same thing, and this material contains no other institution in any sector at all. Answering the third needs to know what a claim on each one costs, and no such figure exists in this material. Answering the fourth needs what has been filed with an authority, and that is set by the Reserve Bank of India at rbi.org.in. One column filled and three drawn empty is the honest result rather than an unfinished one, and neither institution is being called central or not central here. A verdict like that needs a whole system, and this material holds one bank, one finance company and nothing running between them.

Does the same share of funding mean the same thing stops?

Look again at those two figures side by side, because there is a trap sitting in them. At Suvarna Commercial Bank deposits run to 80.0 per cent OF ASSETS, and at Rukmini Finance borrowings run to exactly the same 80.0 per cent OF ASSETS. The same figure, down to the one decimal place either of them is printed to. A reader who stopped at that line would conclude the two balance sheets are the same shape and would be badly wrong, because the shares are the same and what sits inside them is not. One of those funding sides is other people's transaction balances and the other is nobody's, and that single difference changes the whole answer to what stops if it stops.

The two are different in another respect the base rule makes visible. Net worthWhat is left over after everything owed has been subtracted from everything held. On a lender's balance sheet it is the layer that meets a loss first, and it was settled below this subject. stands at 10.0 per cent OF ASSETS in the first case and at 20.0 per cent OF ASSETS in the second, so leverageHow many times an institution's assets stand against its net worth. Settled earlier as a multiple of equity, it gets used here and not built again. is 10.0 times equity at one and 5.0 times at the other, from funding shares that read identically. There is a further layer of liabilities at the bank, standing between the deposits and the net worth, which this material says exists and never puts a figure on, and that layer is part of why one identical share produces two different multiples.

Why is no probability that anything fails stated here?

A loop, a credibility problem and three fixes now stand, and the question a reader naturally asks next is how likely any of this is. There is no answer to that here, and the reason is worth stating. A probability that a particular institution fails is a claim about the world, this material contains no event in the world, and a figure produced anyway would be an invention wearing the clothes of a measurement. Such a figure would carry a decimal point, it would sit in a box, it would look exactly like something that had been computed, and it would be a guess.

A set of questions stands in its place, and the set is more useful than a number would have been. The questions can be carried to any institution at all. Four questions, all answerable from what an institution publishes about itself: what stops if this stops, who else could do it by tomorrow morning, who is funding it and on what terms, and what has been prepared in advance for the day it is needed. The whole apparatus above turns on those four questions, and the answers belong to whoever does the reading.

Who sets what, and why is every value left blank?

Four things named here carry no value anywhere. Every one of them is settled by the body printed alongside it, every one gets changed from time to time, and a printed value would be inaccurate rather than simply old on the morning it changed. The emptiness does double duty here: it is the standing discipline, and it is also the argument itself. The level of the requirement was never the interesting part; the credibility was.