4.1Bagehot's rule
Walter Bagehot's Lombard Street: A Description of the Money Market (1873) describes the money market as it worked in his day, and its Chapter II and Chapter VII carry the rule a panic calls for. The reading note that accompanies the work states the rule plainly: in a panic, the holders of the final reserve should lend freely, at very high rates. The note does not define the final reserve, and this study keeps the phrase as the note gives it rather than setting a definition of its own in its place.
Bagehot's own Chapter II sentences give the remedy and then the choice. Of the remedy he writes, "Very large loans at very high rates are the best remedy for the worst malady of the money market when a foreign drain is added to a domestic drain." Of the choice he writes, "Either shut the Bank at once, and say it will not lend more than it commonly lends, or lend freely, boldly, and so that the public may feel you mean to go on lending."
The rule therefore contains several things at once: an institution named as the actor, the holders of the final reserve and the Bank; the circumstances of the remedy sentence, a foreign drain added to a domestic drain; a description of the moment, the worst malady of the money market; and two acts between which it decides. What it does not contain is a number. Neither "very large" nor "very high" is given a figure, and nothing in the note tells the reader when a drain has become bad enough to be called a worst malady. This study reads the rule as a decision written in advance that settles the question and leaves the magnitudes to whoever is acting in the hours it governs.
Two features of that form matter for the records later in this chapter. The rule fixes the act and leaves the size of it to the hour, which means that the person in the middle of a panic is not choosing between policies but applying one under conditions the rule did not spell out. The rule is also addressed to an institution rather than to any named individual, so the question of who signs the decision falls to arrangements the note does not describe. This study reads both features as the reason the rule can be stated before the event and still be applied inside it, and marks that reading as its own. Both features reappear in the four records in a smaller form. Each of those records states a question and the evidence an answer needs, and leaves the decider, the signature and the hour unstated, which is why the ledger's cells for those matters come back unsettled or implied. A rule meant for the worst hours of a money market and a record meant for one trading account meet in the same gap: what is fixed in advance and what is left to the moment of acting.
One further caution about the rule as it is usually met. The compact form that most readers carry away, of lending freely at a high rate against good collateral, is a later summary of Bagehot's rule rather than his own sentence. It gathers the rule into a single formula, and the wording is a later condensation rather than Bagehot's own sentence. A reader who quotes the formula as Bagehot's has taken a later summary for his own sentence.
The two acts Bagehot sets side by side carry different costs, and those costs arrive at different times. This study reads the first act as stopping the lending at once; Bagehot's own sentence describes the other act in terms of what the public may feel, that it mean to go on lending. Lending at very high rates imposes a cost of rate on whoever borrows at it, and the note does not say when that cost falls. The waiting between question and answer in a panic is the panic itself: the hours in which reserves are being drawn and no one knows whether the drain will stop. This study reads the rule on the axes it uses for decision records: it gives a ground, an observable condition of drainage; it names a decider only as an institution, never as a person; it writes no deadline into the sentences quoted here; and its two acts are not easily undone. Shutting the Bank is hard to recall once the public has felt it, and credit extended at very high rates is not taken back in the same hours, which is this study's reading of the two acts rather than a statement the note makes about them.
Who bears the waiting is not written into the rule. The note names the holders of the final reserve as the actors and does not name those who wait or pay while a panic runs. This study reads the rule as aimed at limiting the losses a panic causes, without making that its only object, and the question of who bears a wait is one the rule leaves open. That is the shape this chapter will look for in the records that follow, where the same question appears at the level of a single trading account: a decision taken on a calm day, whose cost arrives later, on a bearer the records have not named and at a time they have not written down.
4.2Crises in sequence
Two accounts of how instability gets into a money market belong to this chapter, and they differ in what they put first. Charles P. Kindleberger's Manias, Panics, and Crashes (1978) gives a recurring sequence: displacement, boom, euphoria, distress and panic. For a reader new to the sequence, this study reads displacement as the change or event that starts the episode; the rest is its unfolding order. Kindleberger's book also carries the lender of last resort into the account, as the institution whose lending in a panic is part of the story, and that is where Bagehot's rule of the previous section belongs. Hyman P. Minsky's Stabilizing an Unstable Economy (1986) puts the cause inside the good years. Its financial instability hypothesis holds that in good times finance moves from hedge to speculative to Ponzi positions, and that stability breeds instability. The reading note gives the sequence and the movement that good times produce, and not definitions of the three names, so this chapter keeps to the sequence and does not set definitions of its own beside it.
Kindleberger's sequence is a story about events, and Minsky's hypothesis is a story about positions. They meet in the same place: the years before the panic are where the instability is made. This study reads both as placing the question that a decision record must face while the calm lasts. On a day when nothing is wrong, the relevant question is not whether a panic will come, which no one can settle, but what the position currently held is doing while nothing happens. A leverage setting, a pool of collateral, a transfer between accounts: each is a fact about the present, and each determines what the account will be able to do when the sequence reaches the stage at which margin becomes short.
The two accounts differ in what they let a decision-maker hold in advance. Kindleberger's sequence gives an order of stages, and the note gives it as a recurring sequence in the history of crises rather than as a timetable: nothing in the note says how long a stage lasts, or how anyone standing inside one would know which stage is running. Minsky's hypothesis gives a direction of movement inside the good years, and the note gives it as a tendency of finance rather than as a test that can be applied to one account on one day. This study reads the two together as placing the making of an exposure in the quiet period and leaving its timing unsettled, and marks that reading as its own. The consequence for the four records is narrow and worth stating: each of them can say what an account depends on, and none of them can say when that dependence will be tested.
Charles Mackay's Extraordinary Popular Delusions and the Madness of Crowds (1841) stands at a different angle again. It is a popular history of episodes such as the Mississippi scheme, the South Sea Bubble and the Dutch tulip mania, and its Preface offers the sentence: "Money, again, has often been a cause of the delusion of multitudes." Later historians have questioned the scale and the stories of the tulip mania in particular, and the study's own list of works warns against using Mackay as a source of numbers. His value here is as a record of how such episodes were told. This study reads Mackay's telling as one made after the episode, with the outcome already known to the teller, and marks that as its own reading rather than as a claim the note makes about how his histories are built.
The distinction matters for this chapter because of the difference between a story told from its ending and a question left open. Mackay's episodes are read by people who already know how they ended; the four records of the next section are read at a point where nothing has ended yet. This study therefore takes Mackay as a caution about the direction of telling, and not as a source for the four settings or for anything that happened in them. What such a telling cannot give a reader is the position of someone deciding before the ending is known, and that limitation is what the next paragraph puts in the terms of this study's method.
That limitation is worth stating in the terms of this study's method. A decision record keeps the question open, holds an example reading, and ends with a condition for revisiting; it is written at the moment when the answer is not in. Mackay's stories, on the reading given above, do the opposite: the teller stands after the outcome rather than at the moment when the answer is not in. The problem of judging a decision after its outcome is treated properly in Chapter 7, which is where hindsight belongs in this study. The problem this chapter takes up is the one that sits on the other side of the outcome: what a rule, a position or a record has to say while the sequence is still running and nothing has yet happened.
None of the four records in the next section is set inside a crisis. All four are synthetic settings in the present, published by WSTiger for study, and each is written about a position that exists on an ordinary day: a contract quantity and a leverage setting, several positions on common collateral, collateral pledged in a reference currency, funds sitting in another account. Each asks what a loss, if one came, would do to the rest of the account. Read against Kindleberger and Minsky, the records sit at the point where the sequence would arrive rather than where it begins.
4.3Four records of margin and collateral
Synthetic scene: the four settings below are synthetic records published by WSTiger for study, set in the present, with no real firm, trade or market event behind them; the questions and example judgments quoted are the records' own words, and each judgment is WSTiger's reading of its record, dated and revisable.
Leverage Multiple and Actual Risk opens on a proprietary trading account in the present. The contract quantity, the entry price and the stop-loss price are already fixed; the only thing anyone changes is the leverage setting on the account screen. The record's own assumption is that only the leverage setting was changed, while the contract quantity and the entry and stop-loss prices were kept the same. The evidence it requires is the contract quantity, the notional exposure, the margin mode and the remaining capacity, and losses under conditions up to stop-loss or liquidation. The record's question is whether the setting is a fact about the risk:
Does lowering the displayed leverage automatically reduce the price-movement risk of the position?
WSTiger's example judgment reads the record as follows:
Displayed leverage, notional exposure, remaining margin capacity, and expected loss at the stop-loss are examined separately.
If the contract quantity is the same, the sensitivity of profit and loss to price changes may be unchanged, so a reduction in risk is not reported based on the setting value alone.
What waiting costs, on the ledger's axis A3, is set by the fact that the setting changes instantly and the exposure does not. The exposure is there from the moment the position exists, and the record's own next-verification line asks for profit and loss under each price change together with the margin status; that calculation can be made for hypothetical price changes as well as observed ones, so the record writes no deadline into it. On A4, what can be undone: the setting is the easiest thing in the record to change and the least able to reduce anything, and the ledger codes reversing it as costly rather than free, so this chapter does not treat a change of setting as an act without consequence. This study reads reducing quantity as the option that acts on the exposure itself: once a position exists, cutting quantity means closing part of it, and closing realises whatever that part has done. The three options sit at three different distances from the thing that is hard to undo. Adjusting the setting alone leaves the exposure where it is and asks only that the risk be seen differently; recalculating on exposure gives the account a better picture of itself and moves nothing; reducing quantity changes what the account stands to lose and costs the part of the position that is closed. Who waits under each of the three, and who pays if the answer does not come, is left open by the record, and the ledger's bearer cell for it is unsettled. The ledger's coders left the decider and horizon cells unsettled as well, which is what this chapter keeps meeting: the record writes the mechanics clearly and writes none of the people or hours into it.
Risk Propagation of Cross-Margin opens in the same kind of account, with several positions on different instruments sharing one pool of collateral. Losses on one instrument have already reduced the account's margin headroom. The record names the evidence an answer needs: the margin sharing scope, the position-level profit and loss and maintenance margin, and the account-level stress results. It does not say who produces those numbers or when they arrive. What is waited for is the next mark on the account, and the answer to how much of the pool the individual stops have already claimed:
Does a loss on one position affect the remaining survival margin of other positions?
WSTiger's example judgment reads this record as follows:
Manage the stop-loss risk of individual positions together with the account risk of common collateral.
Do not consider the impact of collateral deficiency blocked merely because other trades are profitable.
The second sentence carries the timing. Its point is that the profitability of other trades does not by itself block the impact of a collateral deficiency; it is a caution about what a profitable trade establishes, not a statement that nothing the account does can affect the deficiency. On A3, the record's own next-verification line names what must be measured while the waiting goes on: the path by which account collateral is depleted before individual stop-loss measures take effect, which is the question the record leaves open and which a reader would have to answer from the account's own history. The record's condition for change names what would end the question: if account risk comes within the limit through actual exposure adjustments or confirmed collateral changes, the operating range is re-evaluated. On A4, once the pool has been drained, none of the three options creates collateral. Adding account-level stress testing changes what the desk can see; continuously monitoring collateral headroom changes what the desk can see faster; managing individual stops changes the order in which losses are cut. All three are acts on information and sequence, and the record writes no option that restores headroom after it has gone. Managing individual stops alone keeps each loss inside its own line and leaves the shared pool unmeasured; adding account-level stress testing gives the desk a number that reflects the shared pool and stops nothing by itself; continuous monitoring brings the warning earlier and stops nothing by itself either. What each gives up is time or sight: the first gives up sight of the pool, the second gives up the speed of a live limit, the third gives up the comfort of a number that was adequate yesterday. Who waits between the loss and the measure is not named in the record, and the ledger's bearer cell for it is unsettled. The ledger's coders left the horizon and reversal cells for this record unsettled as well, and coded its decider only as an implied team or organisation.
Co-Deterioration of Collateral Assets is set around the collateral itself rather than the position. The collateral sits in a reference currency, with a recognised ratio per collateral type, and the scenario assumes that the value of the collateral declines together with losses on the held positions. The record names the evidence an answer needs: the reference currency and the recognised ratio by collateral type, the concurrent movement with position losses, and the actual availability of alternative collateral. It does not say who gathers those numbers or when. The price being watched is not only the price of the position. The record's question is what that joint movement does:
When the collateral assets pledged alongside a position also decline as the position incurs a loss, how does the risk change?
WSTiger's example judgment reads this record as follows:
Reflect losses on positions and declines in collateral value at the same time.
The fact that the collateral quantity is unchanged must not be interpreted as meaning that available margin is maintained.
The first sentence is a statement about when the numbers must be read together; the second is a warning about what a steady quantity can hide. On A3, waiting costs the account twice, because the value of the collateral moves with the value of the position, and the required evidence the record names includes the concurrent movement with position losses and the actual availability of alternative collateral. The scenario assumes that the two decline together, and it gives no speed or price path for either movement, so the joint timing is assumed rather than measured. On A4, the asymmetry among the three options is sharp. Stressing the account for co-deterioration is a calculation, and it can be redone at any hour. Preparing alternative collateral is the option that could act on the shortage, and the record's required evidence asks for the actual availability of alternative collateral without saying at what point in the decline it can still be arranged. The record's own verification line asks for the case in which collateral prices and position prices move adversely together, which is a calculation again rather than an action. Maintaining the assessment on collateral quantity is the cheapest of the three to hold and the one the example judgment warns against, because a quantity that is unchanged says nothing about the value behind it; stressing for co-deterioration changes what the account sees and can be run again at any hour; preparing alternative collateral is the only one of the three that could change what stands behind the position. Who waits while the value falls is left unstated by the record. The ledger's coders left the horizon cell unsettled for this record, coded the reversal cell as unknown, and the human place as unknown as well.
Screen Balance and Funds Available for Immediate Use is the most concrete of the four. The account's total assets are sufficient; part of them are in another account, where withdrawal, transfer and deposit delays apply. The record names the evidence an answer needs: the available balance in each account in real time, the actual transfer and deposit status, and the required time and alternative routes. It does not say who gathers them or when. The required time is the record's own unit of waiting: the time by which funds must be in place. Its question is short:
If funds are in another account, can the margin required now also be treated as covered?
WSTiger's example judgment reads this record as follows:
Funds owned are distinguished from funds usable at the right time.
The margin headroom of the current account is not calculated using scheduled transfer amounts whose arrival has not been confirmed.
On A3 this record is the one of the four whose horizon the ledger's coders read as days to months, implied rather than stated; the record itself supplies its own clock in the required time and the transfer and deposit status. The cost of waiting is measured against the deadline the margin requirement imposes, and the ledger's bearer cell for the record is unsettled: it does not say who pays if the transfer arrives late. On A4 the three options differ in what they count and when. The total-asset basis counts funds that the record's own assumption says cannot reach the account that needs them by the required time, so it reports headroom that cannot be used at a call. The basis of funds available for immediate use counts only what the account can use at the required time and leaves out funds that are owned but not yet available. Reflecting the margin after transfer confirmation recognises a balance only after its arrival is confirmed, which shows less headroom in the meantime and recognises the account's whole position later. The record does not say who bears a call that arrives while the funds are still in transit. The record's condition for change makes the point precisely: funds count only once they have been deposited into that account and recognised as collateral. The ledger's coders left the decider, reversal and human-place cells for this record unsettled.
Read together, the four records have one shape that this chapter reads as shared. Each begins from a fact of the present that is easy to state and expensive to discover late: a multiplier on a screen, a pool shared by several positions, a collateral value that moves with the position, a balance that belongs to the account but sits somewhere else. Each then asks what a loss would do beyond the thing being watched. Each names the evidence an answer needs and leaves the answer itself to whoever applies the record later. On this reading they are not four kinds of risk but four versions of one question: what the account can still do once the number it relied on has stopped being the right number. The records themselves state their questions and their evidence and stop there, which is why the reading belongs to this chapter rather than to them.
Across the four records the options divide by what they act on. Some act only on what the desk can see: a recalculation, a stress test, a monitor, a basis of valuation. Some act on the position or on what is pledged: reducing quantity, managing individual stops, arranging alternative collateral. What a reversal of either kind would mean is left open in these four records: the ledger's reversal coding for them is unsettled, costly or unknown, and this chapter draws no general rule about undoing from them. None of the four records names who decides, by when, or who bears the waiting if the answer does not come; the ledger's coders read those cells as unsettled, implied or unknown rather than as stated in the records' own words. The ledger's agreement figures give the context: the horizon cell was settled by both coders in 30 of 53 records, the reversal-constraint cell in 11, and the reversal-action cell in 13. The records write their mechanics and leave their timing to be inferred.
4.4The debate: is instability made inside markets or a sign that they adapt?
The plan for this study names a debate on its own pages: is instability made inside markets, or is it a sign that they adapt? The first position belongs in this study to Minsky and Kindleberger, whose work has been set out in section 4.2. The second is named in the plan beside Andrew W. Lo's Adaptive Markets (2017).
On that side of the debate this study must say what it can and cannot do. The reading note supplied for Lo's work, in the list this study works from, describes the history of Long-Term Capital Management, from its founding to its losses in 1998 and the recapitalisation organised by the Federal Reserve Bank of New York; that is the same description the list gives for Roger Lowenstein's When Genius Failed. The note therefore does not set out an argument about markets adapting. This study cannot write Lo's position into this chapter on its own authority, and where a work's note is silent the study stays silent. What it can record is that the plan names the question and that no position on the adapting side of it is available from the notes of the works this study may use. That side of the debate therefore stays open in this chapter, and nothing is written into it on this study's own authority.
The first reading can be stated from the notes, and it can be set against the four records precisely because the notes say so much. Minsky's hypothesis moves from hedge to speculative to Ponzi positions, and holds that stability breeds instability. Kindleberger's sequence runs displacement, boom, euphoria, distress and panic, and the study carries the lender of last resort as part of its account of crises. Both readings are about a market over time, and this study reads both as placing the answer in the period of quiet. That is the reading this chapter has already drawn from the two works, and the four records in the previous section bear on it in a narrow way. Each of the four is written from the side of one account: a proprietary trading system, one pool of collateral, collateral in a reference currency, funds in a transfer between accounts. From that side, instability is legible as collateral and margin rather than as sequence. A leverage setting, a common collateral pool, a recognised ratio and a transfer status are not stages of a mania; they are the facts that determine what an account can do on a day when a loss arrives. This study reads the four records as written at the point where Minsky's progression of positions would arrive, with leverage already in place. They take the position as given and ask what a loss does to the rest of it.
That narrowness is also where the records stop bearing on the debate. They cannot show whether a market as a whole adapts, because a market is not in them. They cannot show whether instability is made inside markets, because they are synthetic settings with no outcome and no market history attached; each record stops at its condition for revisiting and never reaches a result. Kindleberger's lender of last resort has no counterpart in any of the four: there is no institution standing behind the account, only transfers, deposits and recognised ratios that may arrive too late. The debate in its general form concerns whether the structure of a market produces instability; the four records concern whether one account's collateral can be counted on. The two questions are not the same, and the records cannot settle the first from within the second.
What the four records can contribute to the debate is a description of the vocabulary in which instability is recognised at the level of one account. Collateral, recognised ratio, margin headroom, required time: these are the words available on a calm day, and they are words about capacity rather than about direction. A sequence such as Kindleberger's names stages a market passes through; a movement of financing positions such as Minsky's names what a position has come to depend on. The four records name neither, and this chapter reads their silence on both as a limit of the setting rather than as evidence for one side of the debate. The debate is about markets, the records are about accounts, and any bridge between them would have to be built from evidence a synthetic setting does not carry. What a setting of this kind can show is what a decision is made of at the moment it is taken. What it cannot show is what a market becomes afterwards.
There is a third place where the records touch the debate without entering it. This study reads the two accounts as making the quiet years decisive for what follows, which is this study's reading rather than a formulation either note supplies. What the four records share with that reading is a request: each record's own verification line asks for a calculation to be made, the profit and loss under price changes, the path by which collateral is depleted, the case in which collateral prices and position prices move adversely together, the funds separated into those available for immediate use, in transfer and with usage restrictions. What each record adds is the requirement that the calm-day number be replaced by a number drawn from the joint movement of things that are currently separate. Whether that requirement can be met at all in the conditions the records describe, and whether meeting it would change the decision rather than only the report, is not something a synthetic setting can answer. That is the limit this chapter records, and the reason the next chapter leaves the market for the organisation.
4.5Hands on
What carries from this chapter into the next is a shape rather than a rule. The division set out above carries forward: an option that changes only what the desk can see acts on the account's picture of itself and changes little about a loss already in progress, while an option that changes a position acts on the exposure, and the ledger's reversal coding for these records leaves open how far either can be called back. Waiting in these records is the transfer time, the joint price movement, and the interval between a loss and the moment when the account-level number is read.
An organisation has the same asymmetry in a different object. There the thing that arrives in time is a warning: a threshold crossed, a limit exceeded, a system that behaved in a way nobody designed. Chapter 5 takes up that subject, with Turner on the incubation period in which unnoticed events accumulate, Vaughan on the normalisation of deviance, and Reason on latent conditions lining up with active failures. None of the four records in this chapter has a warning in it; each has a number and a request for a calculation. That is the difference the next chapter will have to explain rather than assume.
Two smaller handover points should be marked here so that they are not read twice elsewhere. First, the reading of what can be undone in these four records is partial: the ledger's reversal coding for them leaves the matter open, and the question of whether a decision can be reversed is taken up elsewhere with the works on waiting and irreversibility. Second, all four of this chapter's records are set in the present, which is the design of this chapter rather than a finding about the records; a setting read across past, present and future is a different exercise, and these four records have no such versions.
The method that allows these four records to be read on the same two axes was set out in Chapter 2, and the counts quoted in this chapter come from the ledger there and from "The ledger's numbers".