By Jon Danielsson, Director, Systemic Risk Centre at London School Of Economics And Political Science. Originally published at VoxEU.
Financial crises usually inflict the most damage when banks suddenly shift from pursuing profits to survival. This column argues that such drastic behavioural changes render statistical analyses based on normal times ineffective. That is why we cannot predict the likelihood of crises, or what banks will do during those crises. Since this behaviour arises from a natural desire for self-preservation, it cannot be regulated away.
In times of extreme stress, banks instinctively prioritise self-preservation to weather the storm. Whereas this is understandable from their perspective, it leads to perhaps the most significant harm caused by financial crises.
Milton Friedman’s controversial criterion states that a business’s objective is to make money for its owners (see Kotz 2022). When applied by a bank CEO, this principle manifests in two distinct behavioural regimes.
Most of the time – perhaps 999 days in a thousand – banks focus on maximising profit through regular borrowing and lending activities.
However, on that rare one day in a thousand, when a major upheaval strikes and a crisis unfolds, short-term profit takes a backseat to survival. Banks halt the provision of liquidity and start hoarding it, triggering runs, fire sales, and a denial of credit to the real economy. This is usually the main economic damage of crises. It is difficult to predict or prevent – and impossible to regulate – because it arises from self-preservation.
These two vastly different behavioural regimes frustrate investors and regulators, not least because statistical models based on normal times fail to capture them.
The One-in-a-Thousand-Day Problem
The buildup to a crisis and the recovery afterwards are prolonged processes that can span years or even decades. But the actual crisis erupts suddenly, catching almost everyone off guard. It is as if we go to bed one night and wake up the next morning to find ourselves in a crisis.
Fortunately, crises are rare. According to Laeven and Valencia’s (2018) financial crises database, the typical OECD country experiences a systemic crisis once every 43 years. Given that the high-intensity phase of a crisis is relatively short, it is reasonable to say that a country is not in an acute crisis 999 out of a thousand days, but in crisis on that one remaining day.
The intense phase of a crisis is driven by banks striving to survive. Profit becomes irrelevant because they are willing to incur significant losses if it means securing their future. Critical decisions are made for entirely different reasons than usual – and often not by the usual people.
Survival hinges on having as much liquidity as possible. Banks minimise liquidity outflows and convert their liquidity into the safest assets available – historically gold; today, central bank reserves. When investors ‘went on strike’ in August 2007, they were motivated by survival.
This drive for self-preservation leads to fire sales and runs. Entities dependent on ample liquidity face hardship or even collapse, while the real economy suffers as credit lines are cancelled and banks refuse to lend. These outcomes constitute the main damage from crises and explain why central banks inject liquidity during such times.
Collectively, this indicates two distinct states: the usual 999 days when banks maximise profit, and that critical last day when they focus on survival. Roy’s (1952) criterion aptly describes this behaviour – maximising profit while ensuring they do not go bankrupt. Thus, these two behavioural regimes are a direct consequence of aiming to maximise shareholder value.
Speed Is Essential
The shift from pursuing short-term profits to survival happens almost instantaneously. Once a bank decides it needs to weather a storm, acting quickly is crucial. The first bank to withdraw liquidity from the system stands the best chance of survival. Those who hesitate will suffer, and even fail.
This was evident when the Hong Kong family office Archegos Capital Management could not meet margin calls. Two of its prime brokers – Morgan Stanley and Goldman Sachs – acted almost immediately and mostly avoided losses. The other two – Nomura (which lost about $2 billion) and Credit Suisse (which lost about $5.5 billion) – hesitated, held lengthy meetings, and hoped for the best.
Implications for Risk Measurement
The one-in-a-thousand-day problem signifies a complete structural break in the financial system’s stochastic processes because the 999-day regime differs fundamentally from the crisis regime.
Each 999-day regime also differs from others. Crises occur when risks are ignored and accumulate to a critical point. Once a crisis happens, that particular risk will not be overlooked again, and new hedging constraints will alter how prices evolve. This means we have a limited ability to predict price movements after a crisis.
Consequently, models based solely on the 999 normal days – an almost unavoidable practice – cannot forecast the likelihood of a crisis or its developments. Attempting to do so leads to what I have termed ‘model hallucination’ (Danielsson 2024).
This also explains why market risk techniques such as value-at-risk (VaR) and expected shortfall (ES), which focus on relatively frequent events (for VaR, one in a hundred days; for ES, one in forty days), are inherently uninformative about crises.
After the 2008 crisis, I organised an event with senior decision makers from that period. Tellingly, one of them remarked: “We used the models until we didn’t”.
Policy Consequences
The one-in-a-thousand-day problem leads to significant misunderstandings about crises.
Excessive leverage and reliance on ample liquidity are the underlying causes of crises. But the immediate crisis trigger and the ensuing damage result from financial institutions simply trying to survive.
Therefore, when analysing crises, we must consider both factors: leverage and liquidity as the fundamental causes, and self-preservation as the immediate cause, which influences the likelihood and severity of a crisis.
We can regulate leverage and liquidity through macroprudential measures. However, we cannot regulate self-preservation. Banks’ behaviour during a crisis is not misconduct or excessive risk-taking – it is the instinct to survive.
In fact, financial regulations can inadvertently exacerbate the one-in-a-thousand-day problem.
Imagine all financial institutions prudently adhere to regulatory demands. Regulators increasingly instruct them on how to measure and respond to risk. When an external shock occurs – such as a virus outbreak or war – all these prudent institutions perceive and react to the risk similarly because they are following the same instructions from the authorities. The result is collective selling in a declining market and uncontrollable fire sales. These prudent banks are not permitted to put a floor under the market and halt the fire sales. Only central bank liquidity injections do so.
This is the fallacy of composition in financial regulations: making all institutions prudent can actually increase the likelihood and severity of crises.
The Impact of Artificial Intelligence
The growing use of artificial intelligence (AI) exacerbates the one-in-a-thousand-day problem (Danielsson and Uthemann 2024).
In banks, one of the primary users of AI and advanced computing is the treasury function – the division that manages liquidity. When the treasury AI detects rising uncertainties, it swiftly decides whether to profit by supplying liquidity and stabilising the market, or to withdraw liquidity, which might trigger systemic stress.
Here, AI’s strengths – speed and decisiveness – can be detrimental.
In a crisis, the treasury AI acts swiftly. Stress that might have unfolded over days or weeks now escalates in minutes or hours. AI’s ability to handle complexity and respond rapidly means future crises are likely to be much more sudden and vicious than those we have experienced so far.
Conclusion
A common belief holds that one stochastic process governs how banks and other financial institutions behave, regardless of the underlying conditions – maximising short-term profits within set constraints. If this were true, we could use data from normal times to model not only bank behaviour during stress but also the likelihood of crises.
However, this view is incorrect.
There are two states: routine profit maximisation for about 999 days out of a thousand, and self-preservation on that one critical day.
In crises, banks disregard short-term profits to focus on survival. This means that normal-time behaviour cannot predict actions during a crisis or the likelihood of one occurring. It also implies that post-crisis behaviour and market dynamics will differ from previous patterns.
The survival instinct explains why crises can be so suddenly triggered and become so severe.
As we increasingly adopt AI for liquidity management, future crises may become particularly swift and intense, unfolding in minutes or hours rather than days or weeks.
Recognising the one-in-a-thousand-day problem allows authorities to mitigate the damage caused by crises and enables investors to hedge risks or even profit. Otherwise, they risk being blindsided, exacerbating the resulting harm.
References available at the original.
Do we have any independent data on the extent to which banks use something accurately describable as “AI” to manage liquidity?
They should have put in the effort to describe the AI. It’s such a general term, just signalling not describing.
I’ll run this idea up a flag pole and see if anybody salutes it. Suppose, just suppose, the US government set up their own bank which would have the full backing of – you guessed it – the US government. Call it GovBank. Banks get greedy and send interest rates sky high, GovBank offers lower interest rates forcing those other banks to lower their as customers desert them. You have a one-in-a-thousand day crisis and the banks freeze up? GovBank becomes the lender of last resort until those other banks can get themselves sorted out. In fact, the mere existence of GovBank may force those commercial banks to put a lid on their worst excesses. Australia had such a bank for decades called the Commonwealth bank until the 90s when the big wigs convinced the government to sell it for a big dollop of cash to buy elections with. Since then the other banks have been able to run free.
as a lay person whose economic expertise is largely based upon having read Yves excellent book (which was stolen and I’ve been too cheap to buy a new copy) a similar thought occurred to me. brings to mind the Bank of ND, far as I know only US public bank. worth looking into.
Isn’t that a central bank?
New Zealand has KiwiBank, set up by the Labour Govt. in 2001 ostensively to rein in the vampiric practices of the ‘big four’ banks then ruling the roost here which were all Australian. Quite who ownes and runs it I don’t know – it’s described as ‘100% Kiwi-owned’ – and by offering an alternative with lower fees and more ‘customer satisfaction’ general opinion is that it did force the big four to become ‘more competitive’, although being presumably run at least at the operational level by bankers it this year pleaded guilty to systematically overcharging customers for fees and interest rates for many years, and the current Tory Government orgasms at the thought of selling it – presumably to one of the big four.
https://en.m.wikipedia.org/wiki/Credit_rationing
A lot has been written on this topic. Maybe I am missing something new in the post above.
It’s our favorite market system exciting our banker’s inherent aggressive tendencies to save themselves during market failures.
Anybody else notice that this is once every three years!
I was willing to treat that 1 in a 1000 as simply cueing ‘rare’ but since they repeated that 13 times, I would want data for that stat. The one stat they did quote was:
that’s pretty far from 1 in a thousand days.
I don’t think the article is at all clear about this, but, if we say 43 years have 15,000 days, it seems to envisage 15 days of (‘high intensity phase of’) crisis over that length of time.
Hmm if the crisis lasts 15:days, which from my recollection of 2007/8 and the SVB one isn’t unreasonable, then really we should be talking 1 fortnight in a 1000 fortnights. Or 1 bimonth in the 1000 bimonths. One can’t increase the granularity any further than that.
There does seem to be some difference between one in a thousand day (as used in the title) and one day (fortnight or what not) in a thousand (which seems to be what they really mean). Your reference to fortnights made me think of settlement dates, which I see are now much shorter than they used to be.
Strange to not mention Minsky at all in the article.
Yes, ignoring Minsky is a major omission.
“A big boy did it and ran away” is implicit in this risk free piece’s assessment of bankster’s behavioural psychology.
They’ll even use the ‘exogenous’ excuse for ‘endogenous’ gambling practices.
The responsiblity for the “excessive leverage” and such over confidence is the Minsky Moment, but seemingly we are now in a post Minsky world.
“Excessive leverage and reliance on ample liquidity are the underlying causes of crises. But the immediate crisis trigger and the ensuing damage result from financial institutions simply trying to survive.”
Ah…the “too much liquidity made me do it” defense.
Also, good PR: using the term “simply trying to survive” to describe bank officers’ behaviour during a financial crisis instead of “trying to stay out of jail”.
And notice the using of the Archegos situation as an example and making the banks look like victims. Ole Bill went to jail and couldn’t have a “too much liquidity made me do it” defense work for him.
“Since this behaviour arises from a natural desire for self-preservation, it cannot be regulated away.”
Guess that is the bottom line – (((cannot be regulated away ))) -.with the implied deregulate
Since when is an entity, a man made entity – natural? I guess some are trying to evolve AI to have human rights and emotions—-well, if not for legislation, regulation, generated electric and a host of laws, the AI as self preserving/survival mode, let alone a bank with a natural desire for self preservation is ludicrous.
The fact that man has felt the need to offload mans con’s and frauds onto something made by man (corporations and the like / AI as human effected by mimicking emotion and only trying to naturally survive).
Corporations can have their charter removed, can go bankrupt, be dissolved (something like the human equivalent of the death penalty or decapitation) unless given extra protection like, corporations and AI proponents seem want to do.
Another question is what exactly is normal activity, given that normal activity is a legislated activity – the man-made laws and regulations are the eco-system in which these creatures of man exist –
“This means that normal-time behaviour cannot predict actions during a crisis or the likelihood of one occurring.” thus, in my small mellon, the man-made legislation and regulation (L&R) landscape define normal-time behavior and, it is the man-made L&R that is deficient in crisis which, are man-made.
I assume that if deregulating a Zoo could mean that giving a more immersive and educational wild animal experience to our children, by removing the bars that separate the audience from the stage, would be a wonderful idea, and if a crises occurs – well that would be a positive learning experience.
I gather killing off banks that exposed themselves too much would lead to a “natural selection” of banks taking less risks and actually thinking about long term survival after a few crisis cycles.
Killing off banks likely does nothing to change behavior. If the bankers make money even if their banks fail, and walk off into the sunset with their bonuses (and I assume a job at another bank).
I’m not saying kill the bankers, but until the people hiding behind the corporate veil suffer the consequences of actions nothing will change. At the least don’t profit off failure.
No, don’t kill the bankers, just put them away and take their ill gotten assets,
and place them in general pop not a country club. If they’re in California since prop 6 failed they could be forced to perform honest work, likely a new experience for them.
They wanted to get rid of Lehman Brothers….just like in high school there is the in crowd and those who are out. Lehman was out. Goldman Sachs was in.
All very interesting – but why is this a problem? It seems to me that these financial ‘crises’ have been very profitable for the right people.
It may well be that the exact timing of any given financial ‘crisis’ is not predictable by anyone – but surely that the current system will generate such crises, and that they will be very profitable for some, is eminently predictable. I mean, when we had Glass-Steagall and it was enforced we had I think zero bank failures. And banking was boring and had modest profits. This is known.
If I put a piece of material in a testing machine and double the load every 10 seconds, I can’t reliably predict to the second when the material will break, but I can with 100% certainty predict that it will break. Because that’s how I set the system up.
Perhaps the real problem is not our statistical models. but a lack of morality and concern for the society as a whole amongst our elites.
“Perhaps the real problem is not our statistical models. but a lack of morality and concern for the society as a whole amongst our elites.”
If it’s a concept that doesn’t fit neatly into statistical modelling, there appears to be an inability to comprehend the concept or they deny it’s real.
But walk into their offices dressed a certain way and one can be berated about social codes.
In China, almost all of the big banks are owned by the central government and many of the smaller ones. Also, provinces or even cities can own their own banks which can be quite big. In case of a financal crisis, the gov’t can panic but banks can’t themselves and I think the gov’t here, especially the central gov’t is not prone to panic, Mike Liston
There is above a 3.3 trillion dollar foreign currency reserve, above a 1.5 trillion dollar sovereign wealth fund, an important gold stock and a long term balance of payments surplus. Also, foreign investment in assets that can be highly leveraged is not permitted:
https://fred.stlouisfed.org/graph/?g=F7qB
January 30, 2018
Total Reserves excluding Gold for China, 2017-2024
AI, which is intended to take away entry level white collar jobs seems like it will simply crumble systems from the outside inwards, not cause system level problems from the off. It is a real time experiment in the tune of move fast and break things.
Fast payments are the new risk vector for banking and the reason the Fed can’t get reserves down. Silcon Valley Bank showed them that they need to be able to bail out a bank in real time, which means an asset swap in real time, which means reserves in place all the time. Lending policy will hence remain loose until something goes down.
There was an argument (don’t remember the source) that the short stock market slump in the early fall of 2024 didn’t evolve into a full crash because the trading bots were so quick to buy the dip. Maybe the trading bots have a stabilising effect. Predicting and modelling future market volatility could also become much harder.