译者按：2024年3月6日，IMF官网发布了SIGNE KROGSTRUP、THOMAS SANGILL和METTE VON SICARD合写的文章，发表在IMF《金融与发展》3月刊。本文简要地分析了人工智能驱动的银行挤兑的风险及其机理，分析了以科技方法应对这种新型风险的利弊得失，展望了CBDC应用情形下对挤兑速度的影响。三位作者中，KROGSTRUP是丹麦央行理事会成员，SANGILL是丹麦央行国际经济与关系部门负责人，SICARD是EU和BIS有关监督事务的高级顾问，三人的观点具有启示意义。现将这篇文章全文翻译如下，供参考。
CONTAINING TECHNOLOGY-DRIVEN
BANK RUNS
SIGNE KROGSTRUP, THOMAS SANGILL, METTE VON SICARD
MARCH 2024
AI, social media, and mobile banking may bring more bank runs; safeguards of yesterday may not be sufficient tomorrow
The sudden withdrawal of bank deposits—accelerated by digital technology—contributed to the failures of Silicon Valley Bank, Signature Bank, and First Republic Bank in the United States and Credit Suisse in Switzerland in the spring of 2023. While a complex set of factors led customers to lose confidence in these banks’ financial health, the spread of rumors on social media and access to deposit withdrawals with the click of a button in mobile apps contributed to the speed with which customers moved their money out of the banks. The speed was unprecedented: in earlier episodes of bank runs, such as during the global financial crisis, social media and mobile banking apps were unheard of or barely existed.
Banks differ, and the reasons customers may suddenly question a bank’s viability vary. As events in 2023 illustrate, however, the risk of sudden bank runs may generally be affected by advancing digital frontiers in banking.
A bank run occurs when many customers simultaneously withdraw their deposits because they are worried about the bank's financial health. Although some deposits are often insured through national deposit guarantee programs, uninsured deposits may be withdrawn when there are concerns about a bank’s health. Even if the bank is fundamentally healthy, suspicion of problems can potentially be self-fulfilling if the bank does not have enough liquid funds to meet customer withdrawals.
In the worst-case scenario, a bank that would otherwise survive may collapse if concerns trigger a bank run. The effects of bank runs can go beyond the cost to banks’ owners and remaining creditors and can become a financial stability concern. Bank runs can be contagious, and adversely affect real economic growth. For this reason, financial authorities and regulators have set up a governance framework for containing this risk.
Speedy bank withdrawals
A typical bank funds itself mainly with deposits from its business and household customers. The bank holds on to a fraction of these deposits to meet potential withdrawals. The rest is used to generate income for the bank. For instance, the bank will offer loans to individuals or businesses in need of funding.
This business model relies on depositors not withdrawing their money all at once. In this case, the bank cannot pay it all back because the deposits are now tied up in longer-term lending to other bank customers.
In normal times, deposit funding of bank lending activities is rather stable. Depositors usually keep a certain balance in their accounts to pay for such expenses as housing and groceries. Individual depositor account fluctuations generally cancel out over time across the many account holders of a bank.
But if rumors emerge that a bank may be at risk of collapse, widespread deposit withdrawals may occur. If withdrawals happen slowly, the bank has time to find funding elsewhere or sell assets to raise funds. Fast withdrawals, on the contrary, can bring down a bank before it can secure funding alternatives.
The potential speed of withdrawals is hence crucial. As the digital frontiers of banking further advance, the speed with which bank customers can withdraw deposits may further increase. Without appropriate adjustments to banks’ management of such funding risks, this could pose a potential threat to financial stability.
While the future impact of AI and tech innovations in banking is uncertain, it is possible to imagine that the frequency of bank runs may increase significantly.
Quick transfers between banks
One way that bank withdrawals can become faster is through easier and faster transfers to other banks. Historically, transfers of deposits between banks have been somewhat limited. Among the reasons is that many bank customers typically have accounts with only one bank, notably because it is time-consuming to collect information about terms and conditions and open an account with a new bank. It may also be expensive to change banks.
New technology may eliminate some hurdles. For instance, online and mobile banking services have made it easier for customers to transfer money between banks 24/7. Increasing access to cheap instant payment systems is reducing the time it takes for a customer to transfer money from one bank to another. It is also likely that personal banking relationships, and the associated loyalty to a bank, play less of a role when banking relationships are increasingly digital.
Artificial intelligence may also accelerate bank withdrawals and transfers. Today, AI-powered tools can analyze an almost unlimited amount of data at high speed, including banks' terms and conditions and news flows from both social media and more traditional media, such as newspapers. Based on such analysis, future AI-powered tools may help bank customers automatically and instantly reallocate deposits across different banks, based on criteria set by the customer. Such criteria could include the interest paid on deposits, perceived bank safety, or the customer’s wish for diversification across banks.
Regulatory requirements, such as the requirement to verify the identity of their customers (know-your-customer requirements), may not prevent AI-powered tools from opening accounts on behalf of a customer. Once a customer uploads the necessary documents and mandates to an AI-powered tool, it may be able to engage in dialogue with several banks and confirm the customer’s identity.
While the future impact of AI and tech innovations in banking is uncertain, it is possible to imagine that the frequency of bank runs may increase significantly.
Fighting tech with tech
Although new technology can increase the risk of bank runs, banks may also use technology to reduce the risk. For example, AI tools may be developed to improve liquidity management and monitor withdrawal patterns, which could help lower bank-run risk.
Other tools to reduce run risk include adequate funding—such as more equity financing or more liquid assets on banks’ balance sheets—that can be used quickly to raise funds to repay depositors. Banks may themselves find it optimal to use more equity financing or hold more liquid assets—or they may be required to do so if authorities are concerned about systemic financial risk. Authorities may also extend deposit insurance and access to central bank funding facilities in case of a crisis. In addition, appropriate recovery and resolution regimes can help restore confidence in the financial system after a bank failure to avoid the spread of bank runs.
The set of tools is not a panacea, however. Each tool can have undesirable side effects that should be balanced against their benefit of reducing risks. Take a potential requirement that banks place more of their deposits in assets that can be sold immediately at no cost, such as high-quality government bonds. In case of a run, the bank can quickly sell the bonds and pay back its depositors. In the extreme, all deposit funding could be placed in highly liquid safe assets, which would effectively eliminate bank runs. However, it would also mean that bank lending to households and businesses would have to be financed by other means, notably equity or long-term borrowing on the part of the bank. There is a risk that it would reduce lending to the real economy, temporarily or permanently. It could also affect the balance sheets of central banks and governments, as well as asset prices, because of higher demand for safe liquid assets.
Expanding deposit insurance programs would also reduce risks. However, depending on how the insurance is financed, high coverage could impose unacceptably high costs on the public purse in case of bank failures: a sufficiently large prepaid insurance fund in many cases would be difficult to put in place up front. Such coverage could also interfere with banks’ incentives to behave prudently (creating moral hazard). Similar effects are present if banks’ access to central bank emergency lending is extended. This move could place the central bank at risk of a financial loss, and it may lead to risky behavior of banks or disruption of the interbank loan market.
In employing such tools, the potential adverse side effects should always be balanced against the benefit to society of risk reduction, and some risk will always remain. Our point is that with new technological advances, this balance may have to shift.
Central bank digital money
The increased use of electronic payments in place of cash has led a growing number of central banks to consider introducing a central bank digital currency (CBDC). A CBDC would allow households and firms to convert their deposits at commercial banks into deposits at a CBDC account—that is, at the central bank.
A deposit with the central bank would in most cases be considered very safe. Depending on how they are designed, CBDCs might shift the dynamics of bank runs, transforming them from runs between commercial banks to runs from commercial banks to the central bank's balance sheet. If there are no constraints on the account at the central bank, customers of a commercial bank perceived as risky might opt to shift all their money to the central bank. The concern is that this option may in itself increase the risk of, or exacerbate, bank runs. Notably for this reason, some central banks considering the introduction of CBDCs are contemplating constraints on how much money a household or firm can deposit in a CBDC account.
However, if future technology in banking substantially increases the speed of possible deposit withdrawals, the speed of bank runs may be unaffected by the presence of even unlimited CBDCs.
We do have tools available to address bank-run risks, but it is crucial to acknowledge that there is no silver bullet.
Each tool comes with its own set of advantages and drawbacks. However, given the unpredictable nature of technological breakthroughs and their uptake in financial markets, it is important to follow developments closely and consider how best to adjust the toolkit. What ensured the safety of the financial system yesterday may not prove sufficient tomorrow.
译稿
防止技术驱动型银行挤兑的蔓延
人工智能、社交媒体与手机银行可能会带来更多银行挤兑，以前的保障措施可能不足以应对未来的银行挤兑
突然提取银行存款——由数字技术加速的——促成了2023年春季美国硅谷银行、签名银行、第一共和银行与瑞士信贷集团的失败。尽管一系列复杂的因素导致客户对这几家银行的财务稳健失去信心，但是社交媒体上谣言四起，以及手机应用程序上点击一个按钮便可提取存款，使得客户们迅速地将他们的资金转出银行。这种速度前所未有：在较早的银行挤兑事件，比如在全球金融危机期间，社交媒体与手机银行应用程序闻所未闻，或者几乎不存在。
各家银行不同，而且客户会突然质疑银行的生存能力的原因各异。然而，正如2023年的事件表明的，突发银行挤兑的风险一般会受到银行业推进数字前沿的影响。
当许多客户同时提取他们的存款时，就会发生银行挤兑，因为客户们担心银行的财务稳健。尽管一些存款通常通过国家的存款担保项目被投保，当人们担忧银行的稳健时，未被投保的存款会被提取出来。即便银行本质上是稳健的，要是这家银行没有足够的流动性资金以满足客户的提取，质疑银行的问题也可能会自我应验。
在最糟糕的情形下，如果担忧触发一次银行挤兑，那么一家本来可以生存的银行就会崩溃。银行挤兑的影响可能超越了银行所有者与剩余债权人的成本，并成为一次金融稳定关切之事。银行挤兑可能具有传染性，并对实体经济增长产生不利影响。因此，金融管理当局与监管者已建立了防止这种风险蔓延的治理框架。
快速提取银行存款
一家典型的银行主要用来自企业客户和家庭客户的存款提供资金。这家银行持有一部分存款满足潜在的提取。其余资金被用于为银行产生收入。例如，这家银行将对需要资助的个人或企业提供贷款。
这种商业模式依赖于存款人不会一次性地提取他们的资金。在此情形下，由于存款目前与对其他银行客户提供的长期贷款相捆绑，这家银行无法全部偿还这些资金。
在正常时期，银行贷款活动的存款融资是相当稳定的。存款人通常在其账户中保持一定的平衡，支付住房和生活用品等开销。久而久之，单个的存款人账户变动会在一家银行许多账户持有人之间相互抵消。
但若有谣传说一家银行可能处于倒闭的风险中，可能就发生普遍的提取存款现象。如果提取存款缓慢，这家银行有时间在其他地方寻求融资或以出售资产来募集资金。相反地，快速提取存款会让一家银行在获得其他融资渠道之前倒闭。
因而提取存款的潜在速度很关键。随着银行业进一步推进数字前沿，可能会进一步加快银行客户会提取存款的速度。
在银行之间快速转账
提取银行存款变得更快的一个方式是通过更简便更快地转账到其他银行。在历史上，银行之间转移存款在某种程度上是有限的。其中一个原因是许多银行客户一般只有一家银行的账户，尤其是因为收集有关条款和条件的信息并在一家新银行开立账户非常耗时。变化银行也可能是很昂贵的。
新技术可能会消除一些障碍。例如，网上银行和手机银行服务使客户更容易在银行之间24×7全天候转账。越来越多的人使用廉价的即时支付系统，减少了客户从一家银行向另一家银行转账所需的时间。当银行关系日益数字化时，个人银行关系及与之相关的对银行的忠诚度也可能不再那么重要。
人工智能还可能加速银行取款和转账。如今，人工智能驱动的工具可以高速地分析几乎无限量的数据，包括银行的条款和条件，以及来自社交媒体和更传统媒体(如报纸)的新闻流。基于这种分析，未来的人工智能驱动工具可能会根据客户设定的标准，帮助银行客户自动、即时地在不同的银行之间重新分配存款。这些标准可能包括存款支付的利息、对银行安全性的感知，或者客户希望在各银行之间实现多元化。
监管要求，例如验证客户身份的要求(“了解你的客户”要求)，可能不会阻止人工智能驱动工具代表客户开户。一旦客户将必要的文件和授权上传到人工智能驱动工具上，它就可以与几家银行进行对话，并确认客户的身份。
虽然人工智能和技术创新对银行业的未来影响尚不确定，但可以想象，银行挤兑的频率可能会大幅增加。
用科技对抗科技
虽然新技术会增加银行挤兑的风险，但银行也可以利用技术来降低风险。例如，可以开发人工智能工具改善流动性管理和监控提取存款模式，这可能有助于降低银行挤兑风险。
其他降低挤兑风险的工具包括充足的资金，如更多股权融资或银行资产负债表上更多流动资产，这些资金可以迅速用于筹集资金以偿还存款人。银行自身可能会发现，使用更多股权融资或持有更多流动性资产是最佳选择——或者，如果当局担心系统性金融风险，它们可能会被要求这样做。当局还可能扩大存款保险，并在发生危机时获得中央银行的融资安排。此外，适当的恢复和处置机制有助于在银行倒闭后恢复对金融体系的信心，避免银行挤兑的蔓延。
然而，这套工具并非万灵药。每个工具都可能有不希望看到的副作用，这些副作用应该与它们减少风险的好处相平衡。比如一项潜在的要求，要求银行将更多的存款投资于可以立即免费出售的资产，比如优质的政府债券。如果发生挤兑，银行可以迅速出售债券并偿还给存款人。在极端情况下，所有存款资金都可以被置于流动性较高的安全资产中，这将有效地消除银行挤兑。然而，这也意味着银行对家庭和企业的贷款将不得不通过其他方式融资，特别是银行的股权或长期借款。它可能会暂时或永久地减少对实体经济的贷款。它还可能影响到中央银行和政府的资产负债表，以及资产价格，因为对安全流动资产的需求增加。
扩大存款保险计划也会降低风险。然而，取决于保险的融资方式，如果银行倒闭，较高覆盖率可能会给公共财政带来难以接受的高成本：在许多情况下，规模足够大的预付保险基金将难以预先到位。这样的覆盖范围也会干扰银行审慎行事的动机(造成道德风险)。如果银行获得中央银行扩大紧急贷款渠道，也会出现类似的效果。此举可能使中央银行面临财政损失的风险，并可能导致银行的风险行为或银行间贷款市场的混乱。
在使用这些工具时，应始终平衡潜在的不良副作用与减少风险对社会的好处，并且总会存在一些风险。我们的观点是，随着新技术的进步，这种平衡可能不得不改变。
中央银行数字货币
使用电子支付取代现金日益增多，导致越来越多的中央银行考虑引入央行数字货币(CBDC)。CBDC将允许家庭和公司将其在商业银行的存款转换为CBDC账户的存款，即在中央银行的存款。
在大多数情况下，在中央银行的存款被认为是非常安全的。根据其设计方式，CBDC可能会改变银行挤兑的动态，将其从商业银行之间的挤兑转变为从商业银行到中央银行资产负债表的挤兑。如果在中央银行的账户没有限制，被认为有风险的商业银行的客户可能会选择将所有资金转移到中央银行。令人担忧的是，这一选择本身可能会增加或加剧银行挤兑的风险。值得注意的是，由于这个原因，一些考虑引入CBDC的中央银行正在考虑限制家庭或公司可以在CBDC账户中存入多少资金。
然而，如果未来的银行技术大大提高了潜在的存款提取速度，那么银行挤兑的速度可能不会受到无限制CBDC的影响。
我们确实有应对银行挤兑风险的工具，但必须承认，没有灵丹妙药。
每个工具都有自己的优点和缺点。然而，鉴于技术突破的不可预测性及其在金融市场中的应用，密切关注事态发展并考虑如何最好地调整工具包是很重要的。以往确保金融体系安全的措施，明天可能就不够用了。
译者：西南科技大学法学院讲师 景欣
