Systemic Risk: When One Failure Becomes Many Others

Systemic risk is the risk that a loss arriving at one institution is carried onward to others instead of stopping where it landed. Systemic risk is a property of the connections between balance sheets rather than of the size of any single one of them. A loss confined to one balance sheet is idiosyncratic, and the same loss turns systemic the moment a connection carries it further.

Start with the loss that has to be absorbed. Everything else follows from it. A loss lands somewhere. Whoever it lands on meets it out of whatever they have available to meet it, and the meeting either finishes or it does not. If what is available is larger than what arrived, the loss stops right there and no third party ever learns it happened. If what is available is smaller, the remainder does not quietly cease to exist. The remainder is a promise that could not be kept, and a promise that could not be kept is held by somebody who was counting on it.

Transmission is not a separate event that happens after absorption. Transmission is what absorption looks like when it runs out. That is worth reading twice, because a reader arriving here usually pictures two distinct things: first a failure, then a spreading. There is only one thing. A loss meets an absorbing surface, and whatever the surface could not take is passed to whoever was holding the unmet claimA promise to pay, sitting as an asset on one balance sheet and as a liability on another at exactly the same moment. The treatment on each of the two sides is settled separately.. Nothing more mysterious than that is going on, and once it is seen as one process, the rest of this guide is arithmetic.

What makes a risk systemic rather than a large risk at one institution?

The test is transmission, and the test is not size. A very large loss that stops at the institution where it arose is a very large idiosyncratic risk and nothing else. A modest loss that reaches four more balance sheets is systemic even though the arithmetic at the place it started is unremarkable. Two things that feel like they should sit on the same scale are being measured on completely different axes, and mixing them up is the single most common way this word gets used wrongly.

Here is the same thing in a form that can be felt. A shopkeeper who loses a whole month of takings has taken a very heavy loss, and if it is met out of savings, absolutely nobody else on that street is affected. Now take a much smaller shopkeeper who cannot pay this month's bill to the wholesaler, who then cannot pay the transporter, who then cannot pay the driver. The rupees involved are a fraction of the first case. The number of parties whose position changed is four. The word systemic describes the second situation and not the first, and the amounts are the wrong place to look for the difference.

The mechanism is built from what a claim actually is, rather than from a list of institutions going over one after another with dates against them and a name for the week it happened. A claim, on its own, is enough to build the whole mechanism from first principles. The definition of a claim is the sturdier foundation of the two. A claim has not changed in centuries, and any episode is a story about conditions that have.

Idiosyncratic vs Systemic Risk: which of the two travels, and how?

A reader who has heard only the first term usually assumes the second is simply a bigger version of it. Systemic risk is not a bigger version of idiosyncratic risk. The second term names a different property of a different thing.

Idiosyncratic risk is risk whose realisation stays on one balance sheet. A borrower of one lender stops repaying. One position turns out to have been valued wrongly. One institution's systems go down for a morning. Each of those is a real loss and each of them belongs to whoever was holding it, and crucially each of them arrives on its own timetable, independent of the others. Independence of timing is the whole reason a holder of many exposures is better placed than a holder of one: the bad ones do not turn up together, so the good ones are paying while the bad one is not.

Systemic risk is not a bigger member of that class. Systemic risk is the property that a realisation at one balance sheet arrives at other balance sheets. Nothing about it is measured on the institution where the trouble began. The measurement sits on the connections leading away from that institution. An institution can therefore look completely sound on every figure it publishes and still be carrying a great deal of this risk. None of those published figures is about the connections. Idiosyncratic risk is a fact about a balance sheet, and systemic risk is a fact about the wiring between balance sheets, so no amount of staring at one balance sheet will ever reveal it.

Now the trap sitting inside that comparison, and it is the reason the two are worth teaching together rather than separately. The very thing that defeats the first risk is itself a connection of the second kind. Spreading holdings across many different things is the correct answer to idiosyncratic risk, and every institution knows it. So every institution does it. But if all of them spread across everything available, then all of their holdings start to look like one another's, and a movement in any part of that shared mixture reaches every one of them on the same morning. Each institution reduced its own exposure to any single thing. The set of them, acting entirely sensibly and entirely separately, manufactured a shared exposure to the whole mixture that none of them chose and none of them can see on its own books. How a holder of many investments spreads risk across them, and precisely what that does and does not remove, is covered separately.

Why is the second trouble not a repeat of the first?

The cause of the second trouble is the step most often got wrong. The first institution's trouble has whatever cause it has: a borrower who stopped repaying, a position valued on an assumption that turned out to be wrong, a mistake somewhere in its own arithmetic. The second institution's trouble has a completely different cause, and the cause is the first institution.

A defence built against the first cause does nothing whatsoever about the second one. Do not soften that. An institution that has never lent a rupee to the kind of borrower behind the first trouble is not protected from the second by that fact. The second trouble does not arrive dressed as a bad borrower. The trouble arrives as an unpaid claim on a counterpartyThe party on the other side of an agreement, who has to perform before the other side gets what it was promised. Who stands on the other side of which arrangements is worked through separately. it did deal with, and there was nothing in its own lending record that would ever have flagged it.

The street version is exact and worth carrying. A shop that keeps its books immaculately, extends credit to nobody it does not trust, and has never had a bad month is not protected from the wholesaler two doors down closing. Its own discipline was aimed at its own risks, and it worked. The trouble reached it along a completely different route, from a party it dealt with rather than from a decision it made, and no amount of care inside its own accounts would ever have shown the exposure. No failure of discipline is involved. The exposure sits in a category the discipline was never pointed at.

The consequence is uncomfortable. Every institution can be individually well run, individually careful, individually defended against everything it can see, and the second trouble still arrives. Nothing in that says anybody was careless. Carefulness is aimed at causes, and the second cause is not a cause anybody at the second institution had any way of aiming at.

Why does a defence that works for one institution fail when every institution uses it?

Holding things that can be sold quickly is a sensible preparation. If money is needed at short notice, some of them are sold, what has to be met is met, and the business carries on. Any careful institution does this, and any careful householdUsed here in the ordinary sense of the people who share a home and a budget. The household stands in for any single decision maker weighing a choice on its own terms. does the equivalent by keeping some savings where they can be reached in a day.

Now have all of them do it on the same morning. The preparation assumed there would be a buyer at roughly the price the holding was valued at. When every holder becomes a seller at once, the buyer they were all counting on is the person standing next to them, who is also selling. The price they are all selling into is a price they are all making. The preparation is individually sound and collectively self-defeating, and neither half of that sentence is a criticism of anybody.

The everyday version is exact rather than approximate, and that is why it is worth using. One person standing up at the back of a crowd sees the stage better. Everybody standing up leaves everybody seeing exactly what they saw before, and now standing. Nobody made an error. The first person's reasoning was correct and stayed correct. The reasoning simply described a situation in which one person stands, and that is not the situation that came about.

The phrase fallacy of compositionThe error of assuming that whatever is true of one participant acting alone stays true when every participant does the same thing at the same time. The fallacy is a general point about reasoning rather than an idea belonging to any one author. names this shape of thinking, and the words are worth having. The shape then becomes visible everywhere in this subject. A rule that is correct for one participant is not automatically correct for all of them, and the step from one to all is where the reasoning has to be redone rather than assumed.

What decides how much of a loss reaches the next balance sheet?

Two quantities and nothing else: what arrives at a link, and what is there to absorb it. Everything this guide does with chains, controls and stopping conditions is built from that pair and nothing is added to it.

If what is there is larger than what arrives, the loss is met in full and nothing continues past that point. If what is there is smaller, the difference is not destroyed. The difference is a claim that could not be met, and a claim that could not be met is an asset on somebody else's books that has just changed value. The net worthWhat is left over on a balance sheet once everything owed has been counted against everything held. How it is read off a set of statements is settled below this subject. of an institution is the most familiar of these absorbing layers, though it is not the only one, and its size relative to what is heading towards it is the whole of what matters at that link.

Say the arithmetic in one line so that it survives without any drawing at all: what arrives at the next link is what arrived here, less whatever was absorbed here. Run that along a chain and a property falls out that most readers have not expected. The amount arriving at the third link is not the amount that left the first, and whether it is smaller or exactly the same size depends entirely on what each link in between absorbed rather than on how large the start was.

Two absorbing layers show how far apart such a layer can sit between two institutions that are both standing perfectly comfortably. Suvarna Commercial Bank Limited, an invented bank, has net worth of Rs 24,000 crore against total assets of Rs 2,40,000 crore. Net worth is then 10.0 per cent of assets, the same fact as saying its assets are 10.0 times its equity. Rukmini Finance Limited, an invented finance company, has net worth of Rs 3,600 crore against assets of Rs 18,000 crore, or 20.0 per cent of assets, with assets at 5.0 times equity. Read a fall of 5.00 per cent of assets against each of those. The fall comes to Rs 12,000 crore at the bank and Rs 900 crore at the finance company, or 50.0 per cent of net worth at the first and 25.0 per cent of net worth at the second. The figures are arithmetic about leverageHow many times over what an institution holds is funded by what it owes rather than by what is left over for it. Worked in full where the economics of an institution are settled. and nothing else. Neither institution is in any trouble at either figure. The same fall simply eats half of one layer and a quarter of the other.

How can institutions with no dealings between them carry the same risk?

There is a second route, and it needs no agreement between any two parties at all. Two institutions holding the same kind of thing are both valued off the same price. Neither has ever lent to the other, neither has any arrangement with the other, and neither would appear anywhere on a list of the other's counterpartyThe party on the other side of an agreement, who has to perform before the other side gets what it was promised. exposures. When that price moves, both of their balance sheets move together anyway. The connections themselves, taken one at a time, are worked in full at the start of this subject.

A list of who owes whom will never measure this risk, so any measurement built only on claims between institutions is measuring one route and reporting it as the total. The point is short and sharp. The route being measured and the risk being described are not the same, and the gap between them is invisible in the measurement itself.

Does the same arithmetic over an afternoon and over a year describe the same risk?

No pair of situations with identical arithmetic is the same risk if one of them unfolds across a year and the other across an afternoon, and speed belongs with the mechanism rather than with the separate subject of handling a crisis. How fast something is felt decides how many links it reaches before anybody can act on it. Speed is part of the transmission rather than a topic sitting beside it.

An absorbing layer is only useful if it can be reached in time. Net worth is not a pile of cash sitting in a drawer; converting it into something that meets a claim takes decisions, sales and often days. Give a link a year and the layer at it is fully available. Give the same link an afternoon and a good deal of that layer may as well not exist. Nothing can be turned into anything before the next claim falls due. The arithmetic did not change. The time available to execute it did, and that is enough to change the answer.

The interesting part is that the speeds are written into the instruments themselves rather than into anybody's behaviour. A claim that settles on a fixed date arrives at a moment everybody knows in advance and can prepare for. A balance repayable on demand arrives whenever the person holding it decides to ask. The asking may come all at once, and it may come at four in the afternoon. Nobody chose to make the second one fast. The speed comes from the instrument itself, and what a runWhat happens when the holders of balances repayable on demand all ask for them at the same time. The mechanics of one at a single institution are settled in the banking sequence. on a single institution looks like was settled earlier and is used here rather than rebuilt.

Why can no single institution price this risk for itself?

An institution deciding how much of something to hold weighs what it stands to gain against what it stands to lose. The calculation is complete and honest about its own position, and a well run institution does it carefully.

One quantity stays outside that calculation: what the institution's own holding adds to everybody else's exposure through the price they are all valued at. No institution can see that quantity, and nothing invisible can be weighed. Every extra rupee held of the same thing makes the shared movement slightly larger for everybody holding it. The cost is real, a decision creates it, and it lands on parties who were never in the decision and were never asked. The word for a cost shaped like that is an externalityA cost created by one party's decision and carried by parties who were not in it and did not agree to it. The term belongs to economics generally rather than to this subject..

There are two obstacles here rather than one, and either would be enough on its own: the cost falls outside the decision, and no participant holds the information needed to price it even if every single one of them wanted to. The second obstacle is the harder one and the one usually skipped. Willingness is not the problem. An institution that wanted, sincerely and at its own expense, to charge itself correctly for what its holding does to everybody else would still be unable to. Pricing it requires knowing the whole set of holdings across every participant, and no participant sees the set. Each one sees its own books and nothing beyond them.

The limits on how much one institution may hold on another, and on how much of one kind of thing it may hold, are therefore set centrally rather than negotiated between the parties. Both of those limits are set by the Reserve Bank of India at rbi.org.in, and both of them move. The rows for them are drawn further down with the authority inside them, and each value is to be taken from the source at the time of reading.

What has to be true for the transmission to stop here?

The transmission stops at the first link where what arrives is smaller than what is there to absorb it. That is the entire condition, and everything below is a consequence of it rather than an addition to it.

Two consequences follow that are not obvious at first. First, a chain can stop at the second link or run past the fourth with exactly the same starting figure, and the whole difference lies in what each link absorbed rather than in how large the start was. Second, thickening the layer at one link protects everything downstream of that link and nothing at all upstream of it. Where the layers sit matters just as much as how thick they are.

The worked illustration runs on four unnamed links and on index numberA figure set at a convenient starting value so that movements can be read off it without implying any amount of money at all. readings throughout, and every reading below is one.

The arrival at the first link is an index of 100.0. The layer available at every link is held at an index of 30.0. Holding the layer steady is the assumption doing all the work, and no real set of institutions has the same layer at every link. At a pass-through of 50.0 per cent of what arrives, meaning that half of what reaches a link is not absorbed there, the arrivals along the chain are an index of 100.0, then 50.0, then 25.0, then 12.5. The third link is the first at which the arrival is smaller than the layer of 30.0, and the chain stops there. At a pass-through of 20.0 per cent the arrivals are an index of 100.0, then 20.0, then 4.0, then 0.8, and it stops one link earlier at the second. At a pass-through of 100 per cent the arrival is an index of 100.0 at every link and never falls below the layer at any of the four drawn.

Read that last case narrowly. Over-reading it is easy. The case says that what ARRIVES has not shrunk across the four links drawn. The case says nothing about whether any institution stops paying. Whether anybody stops paying is an outcome, and the size of an arrival is not one.

Move it and three things happen that are worth naming. Below a pass-through of 30.0 per cent the chain stops at the second link. At exactly 30.0 per cent the arrival at the second link equals the layer to the decimal, and since the condition is strictly smaller than, it carries on to the third. From there the stopping link creeps rightward far more slowly than the setting rises: it is still stopping at the third link at 54.7 per cent and moves to the fourth at 54.8 per cent, and it is still stopping at the fourth at 66.9 per cent. At 67.0 per cent and above, none of the four links drawn brings the arrival below the layer at all. The statement is about an arriving amount and not about whether any institution stops paying.

Each link multiplies the one before it, so what looks like a modest change in the setting is not a modest change in the outcome, and a reader who has moved this once will never again assume that a small difference at each step is a small difference overall.

How does somebody assessing an institution actually use any of this?

Who measures this, and who sets the limits on it?

Four of the quantities circled here are set by an authority and revised from time to time, so a stated value would be wrong rather than merely old on the morning it changed. The rows below carry their labels, the authority and its site are printed inside each row, and each value is to be taken from that source.

An empty row with a label on it and an address inside it is a working sheet, and a filled row that has gone stale is a confident error. The fourth row is the one worth reading twice. That row touches a clearing arrangementThe arrangement under which a separate entity stands between the two sides of a trade and becomes the party each of them faces. How it works is settled where the plumbing of a market is worked through. rather than a lender, and the resources kept against a member failing there belong to the Securities and Exchange Board of India (SEBI) at sebi.gov.in rather than to the Reserve Bank of India. The conditions that make an institution central enough to be identified as a systemically important institutionAn institution an authority has identified as one whose trouble would matter more than another's. Which ones, and on what conditions, is worked in full at the start of this subject. in the first place are covered separately and worked in full there.