By Rob Flatley & Lance Smith

Recent events have revealed that fear-driven digital wildfires and “fingertip” electronic transactions are an irrepressible force in finance.  For banks, the tradeoff for creating deposit inflows with convenient apps is that money can also move out faster than their ability to react. The series of bank failures in 2023 has caught the market and regulators unprepared for fast deposit runs.  It’s clear that the bank manager’s risk reflexes will need to be sharpened and tuned io new signals and tools to detect digital bank runs.

This paper considers some basic analytics that can be applied across a large swath of the US banking industry with the intention of illuminating when a bank may be at risk of deposit flight, and what steps can be taken to mitigate this risk. We found that useful patterns emerge when we look at public FDIC data and apply simple metrics to measure which banks are the most susceptible. In addition to preparing an individual bank, this methodology creates a useful indicator of comparative exposure to the velocity of hot money deposits.

We also introduce the concept of a deposit trajectory curve, and its slope, which we will refer to as the velocity of money.  These concepts will then enable us to consider various stress tests and their implications for flight risk.

We use publicly available data 

The FDIC has a comprehensive database of financial information available to the public and accessible via an API.  The data goes back in some cases several decades, depending upon the age of the banking institution.  For the analysis in this paper, we used the year end 2022 data, and extracted the following fields

  • Name of Bank and Holding Company
  • Assets
  • Tradeable Assets
    • Cash
    • Market Value of Securities
  • Total Deposits
  • Total Uninsured Deposits

However, there are numerous other values available, and time series for each.  We did not need these for the purposes of this paper.

We apply basic analytics 

We then defined two variables:

x = %uninsured deposits.  So x = (Total Uninsured Deposits)/(Total Deposits)

y = Liquidity Coverage Ratio = LCR = (Tradeable Assets)/(Uninsured Deposits)

The LCR measures how much of the uninsured deposits could be paid off by selling securities and using cash held.  In practice the securities could be instead used as collateral against a loan, with a haircut applied, depending upon the type of security.  We ignored this effect as we were looking across all banks, but in practice, one could drill into any bank and provide this detail.

These two variables represent a useful “coordinate system” to describe banks from the perspective of deposit flight risk.  That is, each bank is represented by a pair of coordinates (x,y), and we can then develop an x-y scatter plot of the approximate 1,000 banks in the database.
Now an LCR>100% means that the uninsured deposits are fully covered by the securities and cash held (ignoring the discount factor), so these banks are “safe” and we will remove them from the study.  We also chose to consider only those banks with assets of at least $25 billion. The scatter plot then looks like this:
Next, we highlighted eight “newsworthy” banks, that is, banks that were either seized or were under siege. (Numbered from right to left).


Introducing the Deposit Trajectory Curve

This curve describes the path taken as uninsured deposits are withdrawn.  As these deposits are withdrawn, the x-variable (Percent Uninsured) drops, but so does the LCR as Tradeable Assets are sold off.  The actual curve is a hyperbola; its derivation is provided in the Appendix.
Every bank has its own curve. The curve itself does not provide the probability of a bank run, as that will be affected by the nature of the depositors and the degree of their independence from each other. But what can be deduced is that if the bank “migrates” to the right of the curve, the risk of deposit flight increases. Conversely, if it moves to the left, the risk decreases. How can such migration occur?
This graph displays the main sensitivities.  It is important to note that if the MV of Tradeable Assets decreases, then the bank will move to the right of the curve, and therefore the risk of deposit flight increases. Let’s explore this in more detail with the example of Silicon Valley Bank. During 2022, SVB’s portfolio of HTM (held to maturity) securities suffered an unrealized loss of about $15 billion.  The following graph depicts the trajectories for SVB under the actual scenario, as well as what it would have been had their HTM portfolio been hedged (i.e., no loss). We note that we are using EOY 2022 numbers in our discussion.


This is a very instructive graph. Obviously, the upper trajectory terminates with 39.1% uninsured deposits once the tradeable assets are finished, which is better than the 59.2%.  But there’s something more interesting.

Deposits in the bank are like horses in the barn

Both curves begin at just below 90%.  Think of that as 9 horses in the barn.  However, in the lower curve, there will be 6 horses left in the barn, so that only 3 escaped.  The upper curve enables 5 horses to escape.

Now look at the slopes.  The slope of the curve represents the pressure on depositors to flee.  As more flee, the bank moves down the curve and the slope (pressure) increases.  The horses stampede.  But the upper curve has a much lower slope: 0.79 vs 1.79.  The ratio of these two slopes is 44%.  In other words, the pressure to flee in the upper scenario is less than half as much!

The slopes – velocity of money – can therefore be used to assess the relative flight riskiness: the higher the velocity, the greater the flight risk.


Stress testing can measure deposit flight risk 

In this example we considered a (name withheld) bank with coordinates (58.7%,62.6%)
Now assume that the tradeable securities have a 10-year duration (we do not have visibility into the actual securities held).  We can now stress their portfolio and see what the curve would be if 10Y rates were to increase 300 bp’s.


Note that we are not making an absolute statement about flight risk, as that would require information about the depositors.  However, we can make a relative statement: the flight risk would increase by 28%.

Rescue attempts may not work  

First Republic Bank received $30 billion in (uninsured) deposits from a consortium of large banks.  What was accomplished? Not a whole lot.  The slope decreased from 4.60 to 4.29. So there was an improvement in the deposit flight risk of about 7.0%. 


All banks can take immediate action 

The current bank crisis reinforces a lesson learned long ago but not well remembered, and not often contemplated in today’s risk models. That is, risk management without stress testing is religion without hell. Understanding the downside of your actions can save you from encountering a long purgatory ahead. We are calling on all banks to measure their flight risk now, so we can all avoid the burning flames of hell that bank runs inflict on financial markets.




The deposit trajectory curve