Runs have plagued the banking system for centuries and returned to prominence with the bank failures in early 2023. In a traditional run—such as depicted in classic photos from the Great Depression—depositors line up in front of a bank to withdraw their cash. This is not how modern bank runs occur: today, depositors move money from a risky to a safe bank through electronic payment systems. In a recently published staff report, we use data on wholesale and retail payments to understand the bank run of March 2023. Which banks were run on? How were they different from other banks? And how did they respond to the run?

Banks send most large payments through the Fedwire Funds Service (from now, Fedwire), which moves money between banks’ accounts at the Federal Reserve. When depositors run on a bank and wire large amounts of money to other banks, the run-on bank suffers large and unusual payment outflows but no compensating inflows. To check for unusual outflows in March 2023, we standardize each bank’s daily net payment flows by subtracting the mean and dividing them by their standard deviation. The chart below shows the most extreme payment flows between January 1 and March 31, 2023, by plotting the 1st, 5th, 95th, and 99th percentiles of the daily cross-section of standardized net payments for a sample of banks that are active in Fedwire. As the chart shows, the 1st percentile drops significantly on Friday, March 10—following the run on Silicon Valley Bank (SVB)—and on Monday, March 13, which means that 1 percent of sample banks suffered unusually large outflows on these two days. Even the 5th percentile of net payments notably declines on Monday, March 13. The outflows stop on March 14 with the runs stopping just as quickly as they started. In other words, the March 2023 run was very short-lived, with banks suffering highly unusual outflows over a period of only two days.

Were Run We identify run-on banks as those banks whose net payments were more than five standard deviations below normal during the four business days window from Thursday, March 9, to Tuesday, March 14. Using this method, we identify twenty-two run-on banks, five of which suffered a run on Friday and nineteen on Monday (so two suffered a run on both days); this is far above the number of banks which failed during the episode (only two). The value of outgoing wire transfers from these banks more than tripled on the run days. Furthermore, the average size of run payments was more than three times what would have been expected absent the run, confirming that the run was mainly initiated by larger depositors.

In the table below, we show the key differences in the balance sheets of run and non-run banks as of the last regulatory filing before the run (2022:Q4), highlighting the fact that bank runs have both fundamental and panic elements (Goldstein 2013). Run banks had significantly lower tier-1 capital, consistent with the idea that depositors run on a bank when they are concerned with the bank’s fundamental solvency. Run banks also had significantly lower cash holdings, consistent with the idea of liquidity-driven runs as depositors try to withdraw before the bank is out of cash. Further, run-on banks had significantly more uninsured deposits and these were significantly more concentrated, consistent with the presence of stronger panic element when there are “large players.” (Corsetti et al. 2004). Finally, run-on banks were overwhelmingly banks that are publicly traded on the stock market; this points to a role for public information that we will discuss in a subsequent blog post.

Importantly, however, the characteristics of run banks and non-run banks significantly overlap. For instance, at least 25 percent of banks in both groups have cash holdings lower than 2 percent as a share of their assets; we find similar results for all the characteristics associated with run behavior we discussed above. While we can relate the occurrence of runs to specific characteristics consistent with fundamental and panic elements, a notable unexplained component therefore remains, consistent with a “sun-spot” element that is impossible to predict.

In most bank run models, banks respond to deposit outflows by first using up their cash and then selling their assets. Asset sales make systemic bank runs damaging because many bank assets are illiquid and fire-selling them causes spillovers to other banks. This, however, is not what happened in March 2023. The figure below shows, separately for run and non-run banks, the changes in banks’ balance sheets from the Wednesday before the run to the Wednesday after, based on bank-level data.

As the figure shows, almost all run banks responded to the deposit outflows by borrowing new funds from other sources. In fact, they borrowed so much that 75 percent of run banks increased their cash balances in the week of the run, that is, they more than offset the deposit outflows.

The next figure shows the sources of the change in banks’ cash balance at the Federal Reserve over the same one-week period.

Almost all run-on banks borrowed from Federal Home Loan Banks (FHLBs), whereas only a few borrowed from the discount window (including the newly established Bank Term Funding Program); those that did borrow from the discount window, however, borrowed in large amount. This is consistent with banks using the discount window only as their lender of last resort, to be tapped only when funds from FHLBs are not available.

In a recently issued staff report, we use payments data to study the March 2023 bank run. We find that the run was concentrated on only two days and was driven mainly by a relatively small number of large depositors. However, a large set of banks were run on, far in excess of the banks that ultimately failed. Although run-on banks had on average worse fundamentals, there are large overlaps between the balance sheet characteristics of run and non-run banks. Banks react to their run by increasing their borrowing, mainly from FHLBs and only as a last resort from the discount window.

MicroStrategy Inc. bought Bitcoin for a 12th consecutive week and unveiled details for the sale of perpetual preferred stock to help finance additional purchases of the cryptocurrency.

The Tysons Corner, Virginia-based enterprise software company turned leveraged Bitcoin proxy acquired $1.1 billion of the digital currency from from Jan. 21 through Jan. 26, according to a filing with the US Securities and Exchange Commission on Monday. It owns about $47 billion of Bitcoin, or over 2% of all the tokens that will ever exist.

MicroStrategy is offering an 8% fixed coupon for its $250 million of perpetual strike preferred stock, according to people familiar with the matter. The stock is being offered with a $1,000 conversion price, the people said, asking not to be identified as the information isn’t public. A representative for MicroStrategy didn’t immediately respond to requests for comment.

Co-founder and Chairman Michael Saylor has been ramping up MicroStrategy’s purchases of the cryptocurrency since the election of US President Donald Trump, whom Saylor called “The 1st Bitcoin President,” in a post on X.

The company has been using at-the-market stock sales and convertible debt offerings to fund Bitcoin purchases with the aim of raising $42 billion of capital through 2027.

Hedge funds have been helping to drive the demand as they seek out MicroStrategy for convertible arbitrage strategies by buying the bonds and selling the shares short, essentially betting on the underlying stock’s volatility.

MicroStrategy shares have risen around 600% in the past year. The stock fell about 1.4% to $348.65 as of 9:50 a.m. in New York. Bitcoin fell around 2.5% to $101,953.

Over $1 billion of those convertible notes will now be redeemed earlier than expected. MicroStrategy announced on Friday that its outstanding 0% Convertible Senior Notes due in 2027 will be redeemed on Feb. 24. Notes will reflect a conversion price of $142.38 per Class A common share, according to a statement. The shares closed at $353.67 on Friday.

“Taking maturities out and lengthening that runway will allow investors to focus on what the company is doing rather than on potential impediments to the execution of its strategy,” said Benchmark analyst Mark Palmer, who has a “buy” rating on the stock.

Redeeming these notes early could also allow the company to continue to raise more capital.

“Taking one convert out with a near-term maturity does at least theoretically position the company to be in position to issue another convert with a longer dated maturity, but there are other options that the company has with regard to raising capital, including perpetual preferred stock,” Palmer said.

MicroStrategy plans to offer 2.5 million shares of Series A perpetual strike preferred stock at a liquidation preference of $100 per share. The preferred stock will be senior to Class A common stock and offer a regular quarterly dividend beginning on March 31, according to a SEC filing. The firm announced earlier this month that it planned to raise as much as $2 billion from perpetual preferred offerings in the first quarter.

The company may also opt for more equity offerings after shareholders approved increasing the number of Class A shares 30-fold. The results of the vote on Tuesday allows the company to increase the number of authorized shares of its Class A common stock from 330 million to 10.3 billion and increase the number of authorized shares of preferred stock from 5 million to 1 billion.

MicroStrategy currently has $4.35 billion left of equity offerings under its $42 billion capital plan. Increasing share sales could allow the company to continue to fund additional Bitcoin purchases and cover expenses without having to sell tokens.

“We’ve had and expect to continue to have ample access to liquidity through our capital markets activities and cash flows from operations,” Saylor said in the shareholder vote meeting. “We haven’t sold and don’t intend to sell our Bitcoin to satisfy our interest obligations as they become due.”

Asian traders braced for a turbulent Tuesday, following a bruising session on Wall Street as cracks appeared in the artificial intelligence sector powering the bull market.

A selloff was triggered after a cheap AI model from Chinese startup DeepSeek climbed to the top of Apple’s appstore, sparking concerns that valuations of the technology may be tough to justify. The S&P 500 dropped 1.5% and the Nasdaq 100 sank 3%.

The dollar gained against its Group-of-10 peers as US President Donald Trump said he’ll soon put tariffs on foreign-produced semiconductors, pharmaceuticals and some metals in order to compel producers to manufacture in the country. Scott Bessent, whom the Financial Times said backed gradual universal levies, was confirmed as the next Treasury Secretary.

US equity futures were steady early Tuesday. While Chinese markets are closed for the rest of this week, contracts show Tokyo equities may slump more than 1%, as Sydney stocks were little changed.

“What was shaping up to be a big week in the markets got even bigger with the disruption in the AI space,” said Chris Larkin at E*Trade from Morgan Stanley. “That could make this week’s megacap tech earnings even more critical to market sentiment.”

Monday’s AI plunge drove new fissures into a market narrative that prevailed since the re-election of Donald Trump in November, the America-first, tech-fueled uber bullishness that saw a clear upward path for risky assets spurred by deregulation, tax cuts and even government sponsorship of AI investment. Treasury yields slid sharply as haven-seeking investors laid aside concern - for today, anyway - that the new president’s policies will stoke inflation.

On Monday, Japan’s chip-related shares including Advantest Corp., SoftBank Group and Furukawa Electric Co. got caught in the global tech selloff.

The yield on 10-year Treasuries declined nine basis points to 4.53% on Monday. Australia’s 10-year yield fell six basis points in early Asia trading Tuesday. The Bloomberg Dollar Spot Index rose 0.3%, extending Monday’s gains. Bitcoin slid 2.9% to $101,481.84.

The severity of the rout in US assets was proportionate to the weightings of AI-enabled firms in the biggest stock indexes. Even after a recent paring to curb their influence, the cohort of Nvidia, Apple Inc., Microsoft Corp., Amazon.com Inc., Meta Platforms Inc. and Alphabet Inc. account for about 40% of the Nasdaq 100. It’s roughly 30% in the S&P 500, leaving both gauges significantly exposed to concerted drops in those names.

“The sudden, adverse market reaction to DeepSeek indicates that some of the key assumptions that have been driving the AI trade, and hence major indices, are getting reassessed today,” said Steve Sosnick at Interactive Brokers. “Part of today’s sudden adverse market reaction was a direct result of a wave of complacency that overtook the equity market.”

The Dow Jones Industrial Average added 0.7%. A gauge of the “Magnificent Seven” megacaps slid 2.7%. The Russell 2000 slipped 1%. Wall Street’s “fear gauge” — the VIX — soared the most since mid-December to about 18.

“I’m hoping this moment encourages everyone to look beyond tech stocks,” said Callie Cox at Ritholtz Wealth Management. “Not because the AI story is doomed, but because there are so many opportunities in unloved sectors that have been ignored for so long. The guts of the market’s foundation are still good, so it’s likely that the dip will be bought here.”

Chinese investors have much to ponder as they start their Lunar New Year celebrations that last for the rest of the week. The nation’s economic activity unexpectedly faltered to start the year, breaking the momentum of a recovery sparked by stimulus measures and underlining the need for Beijing to do more to prevent another slowdown.

China’s factory activity shrank in January after three months of expansion, with the manufacturing purchasing managers’ index falling to 49.1, the lowest since August. The non-manufacturing gauge for construction and services dropped to 50.2, just above the 50-mark that separates growth and contraction.

Most Asian markets are shut for the upcoming Lunar New Year celebrations, with Indonesia, South Korea, Taiwan and Vietnam among those closed. Bourses in Hong Kong — on track to open higher — and Singapore are due to end early.

Global traders’ focus will be on earnings announcements from the likes of Microsoft and Apple this week to restore confidence in the so-called Magnificent Seven group of companies.

Investors are heading into yet another pivotal Big Tech earnings cycle with the companies’ shares near record highs and valuations stretched. A key distinction this time: The group’s profit growth is projected to come in at the slowest pace in almost two years.

“This should be a fairly good earnings season, but the bar has been raised and they may not be able to live up to high expectations,” said Dan Taylor, chief investment officer at Man Numeric. “It will be very difficult for the group to perform the way it did last year, especially as valuations have increased.”

DeepSeek was founded in 2023 by Liang Wenfeng, the chief of AI-driven quant hedge fund High-Flyer. The company develops AI models that are open-source, meaning the developer community at large can inspect and improve the software. Its mobile app surged to the top of the iPhone download charts in the US after its release in early January.

The app distinguishes itself from other chatbots like OpenAI’s ChatGPT by articulating its reasoning before delivering a response to a prompt. The company claims its R1 release offers performance on par with OpenAI’s latest and has granted license for individuals interested in developing chatbots using the technology to build on it.

The rapidity of deposit outflows during the March 2023 banking run highlights the important role that the Federal Reserve’s discount window should play in strengthening financial stability. A lack of borrowing, however, has plagued the discount window for decades, likely due to banks’ concerns about stigma—that is, their unwillingness to borrow at the discount window because it may be viewed as a sign of financial weakness in the eyes of regulators and market participants. The discount window has been reformed several times to alleviate this problem. Although the presence of stigma during the great financial crisis has been documented empirically, we do not know whether stigma has remained since then. In this post, based on a recent Staff Report, we fill this gap by using transaction-level data from the federal funds market to examine whether the discount window remains stigmatized today.

Empirically, stigma may be identified by the presence of a spread over the discount window primary credit rate that a bank is willing to pay to avoid borrowing at the window. However, a bank’s willingness to pay is not observable. Instead, we observe “realized stigma:” how much more a bank actually pays above the primary credit rate to borrow. Realized stigma and stigma are not equivalent. Realized stigma provides a lower bound on the extent of stigma. In particular, even if a bank does not actually pay a premium (for instance, because it borrows on the interbank market below the primary credit rate, thus implying no realized stigma), it may have been willing to do so ex-ante.

Evidence of realized stigma, nevertheless, is informative: it signifies that stigma is indeed present. In our paper, we follow the economic literature and measure realized stigma by looking at transactions in the federal funds market, the over-the-counter overnight U.S. interbank market for funds held by banks at the Fed; we interpret a bank purchasing federal funds above the primary credit rate as evidence of stigma.

The chart below plots the proportion of federal funds volume purchased above the primary credit rate between 2014 and July 2024. We find little evidence of stigma before the COVID-19 pandemic, with less than 0.1 percent of federal funds transactions purchased above the primary credit rate.

The onset of the pandemic saw a sharp increase in realized stigma. Between March 11 and March 31, 2020, 28 percent of federal funds borrowing occurred above the primary credit rate, with every type of bank showing evidence of realized stigma.

Evidence of stigma faded in the months that followed the pandemic but resurfaced several months before the 2023 banking turmoil. Between July 2022 and March 2023, an increasing share of federal funds (15 percent, on average) was purchased by domestic banks above the primary credit rate. Surprisingly, realized stigma did not surge after the failure of Silicon Valley Bank (SVB) on March 9, 2023. In fact, the post-pandemic peak in realized stigma happened in December 2022. Nevertheless, the average size of the stigma spread (the spread paid over the discount window rate) more than doubled, from 10 basis points (bps) in the pre-banking turmoil period to 22 bps in the weeks that followed the failure of SVB.

Note that realized stigma is common in troubled banks. Banks that showed signs of stigma between 2014 and 2024 were three times more likely to have failed over the same period. Moreover, out of the twenty-two banks that suffered a run during the 2023 banking turmoil, nine had experienced realized stigma in the preceding ninety-day period.

We also find that realized stigma is not due to a lack of prepositioned collateral at the discount window; that is, it is not due to banks’ lack of operational readiness. In particular, a bank that purchased federal funds above the discount window rate during the 2023 banking turmoil could have obtained almost 80 percent of those funds (on average) from the discount window, based on the amount of collateral it had prepositioned at the discount window.

To better understand the determinants of realized stigma, we conduct a statistical analysis. The results reveal a highly statistically and economically significant persistence in realized stigma: banks that borrow federal funds above the discount window rate are about 40 percent more likely to do so again the following month. Consistent with stigma, banks that do not visit the discount window are significantly more likely to show subsequent realized stigma.

Moreover, a bank is significantly more likely to display realized stigma after increasing in size, becoming financially weaker (that is, having a lower ratio of U.S. Treasury securities or a higher ratio of held-to-maturity losses), or when its share of alternative funding sources declines (lower Federal Home Loan Bank borrowing). Looking at the post-June 2022 period, when stigma became prevalent, some of these determinants changed; in particular, banks with lower cash holdings and higher uninsured deposit liabilities became more likely to exhibit realized stigma.

In this post, we present evidence of discount window stigma in the years since the global financial crisis using transaction-level data from the federal funds market. Evidence of realized stigma is particularly strong around periods of financial market turmoil, such as during the COVID-19 pandemic and the March 2023 banking turmoil. Our results suggest that discount window stigma is still an issue, potentially limiting the discount window’s ability to ameliorate financial stability in times of turmoil.

Wall Street is asking regulators for more time to implement a rule requiring centralized Treasury clearing as banks and funds trading U.S. government bonds face a 2026 deadline.

The Securities Industry and Financial Markets Association (SIFMA) together with other trade associations sent a letter to the SEC on Friday requesting that the implementation timeline be extended by at least one year for the cash and repo clearing deadlines. The repo market is where banks and funds exchange short-term loans backed by Treasuries.

"We believe final implementation of the Clearing Rule will provide improvements for this market," SIFMA and the other signatories of the letter said.

"However, the importance of the Treasury market to the financial system and the economy, along with the expected significant issuance of Treasury securities in the coming years, argues for an implementation timeline for the Clearing Rule that allows for a smooth transition so as not to disrupt this market," the letter said.

Other signatories include MFA, which represents hedge funds and other private funds, the Alternative Investment Management Association

"Association members are concerned that, without an extension, the success of the transition to central clearing will be seriously compromised and will inevitably lead to disruptions in the cash and repo markets in Treasury securities to the detriment of the financial system," said the letter.

Reuters reported last year that a request for a timeline extension was being considered, as crucial details on how mandatory central clearing would work had not been yet defined and market participants feared the remaining two years might not be sufficient to transition.

The rule originally said clearing houses would have until March 2025 to comply with provisions on risk management, protection of customer assets and access to clearance and settlement services.

Their members would have until December 2025 to begin central clearing of cash market Treasury transactions and until June 30, 2026, for repo transactions.

The central clearing rule is the key reform of a broader government effort to fix structural issues that regulators believe have caused market volatility and liquidity problems in the Treasury market.

U.S. President Donald Trump last week tapped Mark Uyeda, a Republican member of the SEC, to be acting chair of the agency. Trump has said he will nominate former SEC Commissioner Paul Atkins to run the agency on a permanent basis.

Uyeda replaces Gary Gensler, former President Joe Biden's hard-charging SEC chair whose ambitious agenda led him to clash with Wall Street and the crypto industry.

Operating within the High-Flyer hedge fund, Deepseek specializes in open-source large language models (LLMs) and algorithmic refinement. The company’s latest innovation, Deepseek V3, boasts an astonishing 671 billion parameters yet was developed for a mere $5.58 million—a fraction of the expenditures by rivals such as Openai and Anthropic, who have invested billions in similar projects.

This accomplishment has elevated Deepseek as a formidable contender against industry heavyweights like Openai, Meta, and Nvidia. The unveiling of its models has reverberated across various sectors. In the U.S. tech industry, companies including Nvidia, Microsoft, and Meta have experienced sharp declines in stock value as investors reassess the viability of high-cost AI ventures.

Similarly, European corporations such as ASML and Siemens Energy have witnessed notable downturns. Analysts attribute this disruption to Deepseek’s innovative software-driven optimizations and its strategic stockpiling of hardware before U.S. export restrictions came into effect, enabling it to achieve performance parity with Western AI models at a significantly reduced cost.

Bitcoin.com News conducted a brief trial of Deepseek’s latest models and found their reasoning capabilities to be particularly impressive. When compared to Openai’s o1 reasoning model, Deepseek’s product demonstrates superior speed and intuitive reasoning, offering an experience that hints at the potential of artificial general intelligence (AGI).

While the models provide a sense of AGI-level sophistication by explaining their thought processes, they remain limited in their ability to fully replicate human intellectual versatility. Beyond traditional markets, Deepseek’s influence has also extended into the cryptocurrency market. Speculation surrounding AI-driven crypto tokens has intensified, with investors predicting a surge in adoption powered by Deepseek’s efficient models.

However, this enthusiasm has not spared AI tokens, including the prominent AI agent coin spectrum, from steep declines. Bitcoin has also been subject to pronounced volatility, with analysts linking these fluctuations to broader market uncertainties sparked by Deepseek’s entry into the fray. Bitcoin, which often mirrors trends in the U.S. tech sector, slipped below the $100K mark on Monday, Jan. 27.

Deepseek’s rapid rise signals a pivotal shift in the global AI sphere, challenging U.S. dominance and redefining assumptions about the resources required for cutting-edge AI innovation. Its trajectory highlights China’s expanding role in the AI arena and provokes critical discussions about the future competitiveness of Western technology giants. As of 2 p.m. ET on Jan. 27, the Deepseek app has claimed the top spot among free apps on Apple’s App Store, outpacing both Openai’s Chatgpt and Meta’s Threads.