Collateralized shadow banking: still at risk of fire sales

A few basic points about shadow banking ten years after the crisis:

“What shadow banking is” isn’t very complicated if banking is defined as “borrowing short to lend long”

What makes banks unstable is that their liabilities are on demand (i.e. they borrow short) while their assets pay out only over the course of years (i.e. they lend long). A principle reason that we are worried about “shadow” banks is that they have the same instability as banks, but lack the protections in the form of a strict regulatory regime and a lender of last resort. When shadow banks have this instability it is because they borrow short to lend long.

This approach makes it easy to understand the world of shadow banking, because there are only a limited number of financial instruments that are used to borrow on a short-term basis. Thus, for the most part shadow banks have to finance themselves on the commercial paper market (unsecured financing) or on the repo market (secured financing) or, especially for investment banks, via derivatives collateral (e.g. that is posted by prime brokerage clients). These are the major sources of wholesale short-term funding.

So typically when a financial product is subject to losses due to a run-prone (and therefore classified as a shadow bank), it’s because of the product’s relationship to the commercial paper market, to the repo market, and/or to the derivatives market.* The latter two, which comprise the collateralized segment of shadow banking, are the most complicated, because the run can come from many different directions: that is, lenders may stop lending (e.g. Lehman Bros), borrowers who post collateral may stop posting collateral (e.g. novation at Bear Stearns), and for derivatives contracts conditions may shift so that suddenly collateral posting requirements increase (e.g. AIG).

Collateralized shadow banking is governed by ISDA protocols and contracts, not the traditional law governing debt

While repos have been around for centuries, a “repo market” in which anyone can participate and where collateral other than government debt is posted is a relatively new phenomenon. Similarly derivatives contracts have been subject to margin requirements for more than a century, but in the past these contracts were exchange-traded and exchanges set the rules both for margin and for eligibility to trade on the exchange.

Thus, what made repo and derivatives financially innovative in the 1980s and 1990s was that suddenly there were unregulated over the counter (OTC) markets in them. What “unregulated” really meant, however, was that the big banks wrote the rules for this market themselves in the form of International Swaps and Derivatives Association (ISDA) protocols and contracts.

In the early days of repo and derivatives it was far from clear that they wouldn’t fall under the existing regulatory regime as securities (regulated by the SEC), or as commodities and/or futures (regulated by the CFTC). (The legal definitions of the SEC’s and the CFTC’s jurisdiction was deliberately made very broad in the implementing legislation, so an intuitive understanding of these terms will not coincide with their legal definitions.) Similarly, it was far from clear that the collateral posted in these OTC contracts would not be subject to the standard terms in the bankruptcy code governing collateralized debt. (Kettering who describes repos in this era as too big to fail products is great on this.)

Thus, one of the ISDA’s first projects was lobbying in the US for exceptions to the existing regulatory regime. Progress was incremental, but a long series of legislative amendments to the financial regulatory regime starting in 1982 and culminating in the bankruptcy reform act of 2005 effectively placed the whole system of repo and margin collateral outside the financial regulatory regime that had been set up in the 1930s and 1940s (for details see here, or ungated). These reforms also exempted these contracts from the bankruptcy code’s protections for debtors (see here or ungated).

Where the US led others followed. Gabor (2016) documents how Germany and Britain came to adopt the US model of collateralized lending, despite the central banks’ serious reservations about the system’s implications for financial stability. The world economy entered into 2008 with repo and derivatives markets effectively subject only to the private “regulation” of ISDA protocols and contracts.

Despite reforms, the instability at the heart of the collateralized shadow banking system has yet to be addressed

We saw in 2008 how the collateralized shadow banking system relies extremely heavily on the central bank for stability. (Federal Reserve programs to support the repo market included the TSLF and the PDCF.  Data released by the Fed indicates that at the peak of the crisis it accepted substantial amounts of very risky collateral.)

Indeed the International Capital Markets Association has put it quite bluntly that it considers the systemic risk associated with fire sales in repo and derivatives markets to be a problem that “the authorities” are expected to step in and address.

“The question is how to mitigate such systemic liquidity risk. We believe that systemic risks require systemic responses. In this case, the authorities can be expected to intervene as lenders of last resort to ensure the liquidity of the system as a whole. For their part, market users should be expected to remain creditworthy and to have liquidity buffers sufficient to sustain themselves until official intervention restores sufficient liquidity to obviate the need for fire sales.”

In short, the collateralized shadow banking system is constructed on the expectation of a “Fed put”. Instead of attempting to build a robust infrastructure of debt, shadow banking embraces the risk of fire sales and expects the governments that don’t make the shadow banking rules to bail it out.

The only sure-fire way to eliminate the risk of fire sales is to reduce the financial system’s reliance on repo- and margin-type contracts that allow a decline in the value of collateral to be a trigger for demanding additional funds. Based on financial market history this would almost certainly require an increase in the use of unsecured interbank debt markets. However, not much progress has been made on this front, especially since the EU’s proposed Financial Transactions Tax stalled in 2015.

On the other hand, significant reforms have been made since 2008 (Please let me know if I’ve left out anything important.) :

  • Collateral has shifted mostly to sovereign debt. This helps stabilize the market, but perhaps only temporarily as a broad range of collateral is still officially acceptable (so deterioration of the quality of collateral can creep in).
  • Approximately 50% of derivatives now are held with central counterparties. (The estimate is based on a 2015 BIS report.) This reduces the risk that the failure of a small market participant sets off a chain of failures that results in a fire sale. There is some concern however that fire sale risk has been transformed into the risk of a failure of a central counterparty.
  • Derivatives are now officially regulated by either the CFTC or the SEC and and there has been an effort to harmonize OTC margining requirements internationally.
  • Under pressure from regulators a voluntary stay protocol has been developed by the ISDA that is designed to work with the regulators’ special resolution regimes and to limit the right to terminate a contract due the default of a related entity. In the US systemically important banks are required to include this protocol in their OTC derivatives contracts.
  • Bank liquidity regulations have been adopted that limit the degree to which regulated banks are exposed to significant risk in these markets.

Notice that these new regulations embrace the basic framework of collateralized shadow banking: much of the focus is on making sure that enough collateral is being used. Special rules are designed to protect the largest banks and the banking system more generally. But aside from protecting the banks, it’s not clear that significant measures have been taken to eliminate the risk of fire sales that originate outside the banking system. Assuming that these regulations are effective at protecting the banks, this raises the question: Who bears the fire sale risk in this new environment?

Thanks to @kiffmeister for requesting that I write up this blogpost.

* While one can usually figure this out after the run has occurred, current regulation does not necessarily make the relevant information available before a run has occurred. Mutual funds are a case in point: the vast majority of them have so little exposure to repo and derivatives markets that it can be ignored, but the few that take on significant risk may have disclosures that are hard to distinguish ex ante from the ones that don’t (e.g. Oppenheimer Core Bond Fund in 2008).

Advertisements

Integrating finance and macro: the problem of modeling debt

So I have finally read Mian and Sufi’s House of Debt. They do an excellent job of setting forth an argument that has met with quite a bit of resistance within the economics profession: the growth of household debt before the crisis and the failure to reduce it after the crisis explains to a large degree the severity of the crisis. (House of Debt was written in 2014, so if you’re thinking: “But wait, that argument is mainstream now” you would be correct.) I actually read the whole book which can be taken as approval of both its structure and the quality of the writing. (On the “life is short” principle I typically don’t get through a book is poorly structured or poorly written.) The book is widely cited and almost universally acknowledged as one of the foremost expressions of the household balance sheet view of the 2007-09 financial crisis. Thus, I am going to take the book’s many excellent qualities as given and focus on the most important flaw that underlies the book, because that flaw also underlies most economic analysis of the way financial factors played a role in the crisis.

While it is wonderful that Mian and Sufi are talking about debt, the way they are talking about debt and in particular their underlying model of debt is very problematic. Furthermore, the errors in their underlying model of debt are so ubiquitous in economic theory that these errors function as a constraint preventing the development of models that can accurately represent the relationship between finance and the real economy. In short, while this post will focus on a critique of Mian and Sufi (2014), this book is really just standing in for all the economic works that make the same assumptions, some of which I will reference below.

Holmstrom (2014) presents the standard economists’ model of debt, which underlies Mian and Sufi’s discussion too, using this diagram:

Holmstrom debt

Debt is modeled as a promise to make a fixed payment that will only be met if the borrower has enough money at the time payment is due. This diagram treats the value of the borrower’s collateral as equal to her entire wealth, assumes that the value of the collateral may take on values ranging linearly from 0 to something well in excess of the amount to be repaid on the debt, and assumes that the lender can take the collateral if the debt is not paid. Thus, the lender’s payoff increases linearly until the value of the collateral exceeds the amount due on the debt at which point the payoff to the lender is fixed.

There is nothing wrong with this model as a first pass at modeling debt. It is widely used for good reason. But the basic model also dates back to the 1980s (I connect it with a paper by Hal Cole that I can’t locate, but am not entirely sure of its origins) and it is remarkable that the model has not in ensuing decades been amended to allow for the much greater complexity of real world debt. Treating this model as if it represents the general category of “debt” and not the specific simple case that is easiest to model is a huge mistake that permeates the economics profession.

So what’s wrong with this model?

1)   It is used to treat “debt” as homogeneous

The model assumes that all debt takes a single specific contractual form modeled on a mortgage. In fact, debt is broad term that encompasses a huge range of different contractual provisions. Debt can be structured to favor the borrower or it can be structured to favor the lender. A debt contract can be designed so that it is hardly distinguishable from equity or so that the lender bears virtually no risk of loss. Economists need to stop talking about “debt” as a homogeneous product and start talking about the specific kinds of debt they mean to address.

For much of the discussion in Mian and Sufi, the standard model is appropriate, because their main focus of inquiry is mortgages, and this is a reasonable model of mortgage debt. On the other hand, this model leads them to make generalizations about debt itself that are simply nonsense, e.g. “This is a fundamental feature of debt: it imposes enormous losses on exactly the households that have the least” (p. 23). If they simply replaced the term “debt” with the phrase “the current US mortgage system” there would be nothing wrong with this sentence. When, however, they generalize from the problems with US mortgages to “debt” itself, they misfire badly. As I note above, this problem is not in any way restricted to Mian and Sufi, this is a general problem that permeates and degrades much of the economic discussion of debt.

It is highly unlikely that the economics can make progress in its efforts to study the relationship between finance and the real economy so long as the profession’s vocabulary for discussing something as fundamental to finance as “debt” is so utterly impoverished.

2)   Failure to model uncollateralized debt

Uncollateralized debt has very different properties from collateralized debt. In economic theory models debt is almost always modeled to be collateralized and is therefore backward looking (see, e.g. Holmstrom 2014 or Gertler and Gilchrist 2018). An agent must already own something pledgeable in order to borrow. This ensures that wealthy agents can borrow more and grow more wealthy, whereas poor agents are likely to be constrained forever. This framing of debt is closely related to the inequality dynamics described by Mian and Sufi.

By contrast, when debt is uncollateralized, it can be forward looking. If I can convince a bank that after investing the proceeds of a loan of $50,000 today, my business will give me revenues of $100,000 in a year, the bank can fund the loan with nothing more than my personal promise to pay it back (and the knowledge that our legal and social system will impose significant costs on me for a failure to pay, e.g. a public judgment against me, and a defective credit report). As long as I am expected to have the funds to pay back the loan when the debt is due, there’s no reason at all for the loan to depend on my ownership of more than $50,000 in assets to be used as collateral. For relatively small amounts and short periods of time this type of unsecured lending is very common in practice and has been very common for centuries.

Effectively the habits of thought that economists adopt when they think about debt are unreasonably constraining their ability to model the relationship between finance and the real economy. And these same habits of thought tend to rule out by assumption the possibility of inequality-reducing debt.

3)   Inaccurate assumptions about the legal framework governing debt

“Debt leads to bubbles in part because it gives lenders a sense of security that they will be unaffected if the bubble bursts” (p. 114).

This is simply not a property of “debt.” In the event of a bubble that bursts there will be a rash of bankruptcies and the basic rule in bankruptcy in this situation is cramdown: the borrowers’ debt is written down to the post-crash value of the collateral. In short, the standard legal procedure governing debt addresses precisely the macroeconomic problem in question here. A lender who lends into a bubble is at risk of loss. As a general statement, Mian and Sufi’s claim is simply incorrect. It is, however, (1) an accurate description of the model of debt that they are working with and more importantly (2) an accurate description of the law governing US mortgages on first homes, because of the explicit exception for these loans in the bankruptcy code. (Interestingly enough, the rules for the treatment of second homes in bankruptcy do allow cramdown.)

Thus, when Mian and Sufi write “Our main argument is that a more even distribution of losses between debtors and creditors is not only fair, but makes more sense from a macroeconomic perspective” (p. 150), what they are missing is an acknowledgement that “debt” as a general category is usually subject to treatment in bankruptcy that addresses their macroeconomic concerns. “The inflexibility of debt contracts” (p. 168) about which Mian and Sufi complain exists in their model and in US mortgage markets, but is not in fact a property of “debt contracts” themselves under the current legal regime in the US.

What should economic models of private debt do?

Economic models that seek to integrate finance and macro need to be very conscious of the different kinds of private debt and make deliberate decisions about why a specific form of debt is being modeled. To assist in this project, I present here a simple hierarchy of different types of debt that are likely to have very different macroeconomic consequences and thus should be modeled differently. (It’s possible and even likely that I have omitted an important type of debt, so this hierarchy is open to revision.)

The types are ranked from those that are most favorable to the lender to those that are most favorable to the borrower. (Note (i) I use “mortgage” as a general term for a collateralized loan, and (ii) the listed term of the loan should be understood as typical and not as claim that these types of debt are restricted to this term.)

  • Repurchase agreement or margin loan: ultra-short-term, overcollateralization, marketable collateral, immediate right to seize the collateral if it falls in value (and isn’t increased).
    Comment 1: A vigilant lender cannot lose money on a repo.
    Comment 2: There has been significant work on repo since the crisis, e.g. Brunnermeier and Pedersen 2008, but as far as know there is no “workhorse” model comparable to the model of debt above. (Please correct me if I’m wrong.) My impression is that much of the work on repo has been empirical (e.g. Adrian and Shin 2010).
  • Mortgage with recourse: long-term, overcollateralization, a right to seize the collateral only after the borrower defaults, right to be paid in full if the collateral value at the time of default is deficient.
  • Mortgage without recourse: long-term, overcollateralization, a right to seize the collateral only after the borrower defaults, no right to further payment. (This is the type of debt that corresponds to the “standard” model of debt discussed above.)
  • Mortgage with cramdown: long-term, overcollateralization, a right to seize the collateral only after the borrower defaults, but subject to cramdown if the borrower declares bankruptcy.
  • Unsecured debt with bank guarantee (e.g. commercial paper): short-term, no collateral, lender relies on bank guarantee.
  • General unsecured debt: short-term or long-term, no collateral. Enforcement must be via long-term incentives (reputation) and/or penalties imposed by the legal system for failure to pay. Corporate bonds fall under this heading.

On Modeling Money, Banks and Markets

Every good model is designed to emphasize certain empirical regularities that characterize the real world and by doing so to explain certain aspects of how the real world functions. Thus, the first question when discussing how to model money and banking is: What are the empirical regularities that a model of money and banking should capture?

Drawing on my knowledge not only of the history of money and banking, but also of the structure of modern money markets, I have strong views on the empirical regularities that a model of money and banking should capture. Depending on the purpose of the model, there can be good reasons for focusing on getting either the asset or the liability side of banking right, so I will set forth the relevant empirical regularities separately for the two sides of the bank balance sheet. (Obviously there are also benefits to putting both into the same model, but frequently with formal modelling it is useful to start with something simple.) In both cases, first I state the key features that model should have and then I follow up with a brief discussion of some of the objections that I expect to hear to the approach I am describing.

Banks as issuers of money

When modelling the liability side of banking, there are two key features:

(1) Bank liabilities circulate as money. This means that bank liabilities are generally accepted, or, in other words, that the bank is trusted by everybody in normal times; and

(2) Any constraints on bank borrowing should be clearly explained, and should not imply that the individual members of the public are imposing borrowing constraints on banks. Thus, Diamond and Dybvig appropriately explains a run as a coordination problem, which is not at heart an individual action. And there can clearly be a constraint imposed by an outside authority like a regulator or central bank. But the idea that the individual members of the public refuse to lend to the bank past a certain amount should be viewed as contradicting the basic fact that bank liabilities circulate as money because banks are trusted by the public.

Discussion

Sometimes the claim is made that non-bank liabilities can also circulate as money. While it is true that there are historical examples of private non-bank liabilities circulating as money, these are almost always very localized affairs and thus don’t actually represent examples of generally acceptable means of exchange. These examples are not only lacking in geographic breadth, they are also typically short-lived, of very limited scope, and rare. In short, historical examples of circulating private non-bank liabilities are essentially measure zero events in the history of money. While certain historical events may be worth modeling in order to understand the event in question, these episodes are of far too little importance to be incorporated into a model that is trying to understand the general principles of money and banking.

The basic implication of the approach that I am advocating is that banks are not just a little more trustworthy than other economic entities. When modelling banks (in normal times), banks sit at the extreme of a spectrum of trustworthiness. Thus, models that purport to treat the trustworthiness of banks as only incrementally distinguished from other agents should not be considered as logically consistent with the statement that banks are issuers of money.

Banks as lenders

When modelling the asset side of banking — and especially when modelling how bank lending compares to market-based lending — the essential empirical regularities are:

(1) Banks, with their easy access to liquidity via the issue of monetary liabilities, are the economy’s short-term lenders.

(2) If there is going to be market-based short-term lending that competes directly with banks, then the banks’ role in “wrapping” (or guaranteeing) the short-term debt to make it saleable should be modeled. The reason for this is that in practice bank lending is frequently indirect and takes the form of a backup promise to pay in case the original borrower defaults; the use of these bank guarantees is so common that money market assets are in practice not marketable without bank support. (For a lengthier discussion of this issue, see here.) Note that for simplicity, both market-based short term lending and the bank guarantees that support it can be omitted from most models. It is, however, a clear error to include market-based short term lending without modelling the bank guarantees that support it.

(3) The market-based lending that takes place without bank support is long-term lending, such as 5-30 year bonds. Banks don’t have a comparative advantage here, because their ability to issue monetary liabilities is as likely to get them into trouble as to help them when the loan is long term. (They can easily like the S&Ls or Diamond-Dybvig run into financing problems.)

Thus, a key issue that a model seeking to address both bank lending and market-based lending is: What is the term of the lending in the model? Many models have both bank lending and market-based lending for the same term of the loan. I would argue that all models with this characteristic are effectively assuming long-term lending. Thus, when they find that markets can in many circumstances lend just as well as banks, they reach this conclusion by looking at the type of lending in which banks do not have a comparative advantage. A better way to model bank lending together with market-based lending is to model banks as lending short-term, e.g. working capital, while market-based lending is long-term (with or without banks competing in long-term lending).

Discussion

Many economic theory papers that purport to study money and banking effectively assume that markets in debt can exist in the absence of banks. One might almost say that these papers take markets as the fundamental economic unit and are trying to place banks within that context.

At least from my perspective, this presumption is precisely what heterodox theory seeks to challenge. My read of the history is that, while markets certainly existed before banks became important, neoclassical markets where there is something akin to a single price for a good could only be imagined in a world where banks were providing liquidity so that the typical trader was not liquidity constrained.

That is, “markets” in the sense of common usage have of course always been around, but this is a completely different concept from what an economist means when speaking of markets where every homogeneous good has a single price. Historically it is true that every community has, for example, weekly markets where people get together to trade. Prices in those markets are, however, typically based on individual bargaining and are very variable depending on who you are. People who have traveled broadly may have visited this kind of market, where a local friend is likely to tell you “Just let me know what you want to buy and then go away. I’ll handle the negotiations.” The neoclassical economic model is not designed to capture this kind of market.

The kind of markets that are made possible by banks are neoclassical-like markets. Based on sources like Adam Smith it appears that this type of market only started to grow up in Britain in the late 18th century. Suddenly people had access to enough liquidity that differential liquidity constraints stopped being the determining factor in prices, as is the case in traditional markets. And as Larry Neal explains in The Rise of Financial Capitalism (1990: 35) it was around the same time that published price lists expanded dramatically and began to take on “an increasingly official character.”

Thus, I would argue that markets as they are typically modeled in economic theory papers exist only because banks provide the liquidity that makes the efficient prices they produce feasible. For this reason, a realistic model of banks and markets will reflect the role played by bank-based liquidity in the formation of market prices. This view, as was discussed in this post, is consistent with the realities of markets today, where short-term lending is heavily dependent on banks – and of course it’s hard to imagine how capital markets could function, in the absence of these bank-dependent money markets.

To summarize, in order to capture both bank lending and market-based lending an economic model needs to have at least a three period horizon with banks offering one period debt and markets offering two period debt. Ideally the model would be able to illustrate why markets are better for long term debt and banks are better for short-term debt.

Many thanks to David Andolfatto as this blog post was generated by email correspondence with him.

Taxonomy of liquidity II: Price stable liquidity

In Taxonomy of liquidity I I found that the distinction between market-based lending and bank lending could be clearly drawn only if the term “market-based lending” was used to refer strictly to traditional capital markets, that is, to the stock and bond markets, because money markets, repo markets, derivatives markets, etc. are all very dependent on explicit and implicit commercial bank guarantees. Here I want to address a different issue: the distinction between price stable liquidity and price disclosing liquidity.

Price disclosing liquidity is fairly intuitive. It is associated with the market liquidity that is available on stock markets or long term bond markets. Even though we consider Treasury bonds or Apple stock to be extremely liquid assets, we also understand that the prices of these assets are not stable, as any intraday chart of their prices will show. Stock and bond markets are designed to give asset holders a reliable venue in which to sell, while at the same time allowing prices to move to reflect what may be very short-term shifts in supply and demand.

Money market liquidity is different from this description of capital market liquidity. Money markets are markets where people who have cash that they will need in the near future try to earn a little interest. For this reason, money market investors are notoriously averse to sustaining capital losses (Stigum and Crescenzi 2007 p. 479). Furthermore, money market instruments are by definition short-term. Thus, unlike capital market issues, every issuer on the money market is more or less continuously raising funds. For this reason, when money market investors are worried that they may incur a loss, they don’t even need to sell their holdings to cause problems for the issuer; all they need to do is to refuse to invest in the new issues and the money market will be disrupted. In addition, because money market investors expect to need the money in the near future and are thus risk-averse, many of them avoid money market instruments that have any aura of credit risk.

An example of how money market investors react to losses is the behavior of prime money market fund investors in September 2008 after one prime money market fund, the Reserve Fund, announced that it would incur a small loss. The panic was so severe that the Federal Reserve, the FDIC, and Department of Treasury all established programs to support money market funds and the commercial paper in which they invested.

Thus, it is the nature of money markets that they are expected to provide price stable liquidity (cf. Holmstrom 2015). This form of liquidity is completely different from the liquidity provided by the stock market where losses are expected on a regular basis.

One of the reasons that banks play such an important role in money markets is that bank liabilities are promises to make payment at par. Banks offer price stable liquidity. Not only are banks generally managed so that they can offer price stable liquidity, but the banking system itself – and in particular the structural support provided by the central bank – is designed to protect the system of price stable liquidity. Indeed, it is because price stable liquidity is integral to the business of banking that credit rating agencies generally demand that money market instruments receive liquidity and credit support from a bank in order to qualify for the highest credit rating.

In my previous post I explained that a discount market is an unusual kind of market, because each seller is required to endorse the bill when it is sold and thereby to guarantee payment on the bill in case of default. The importance of price stable liquidity on the money market explains this requirement, and explains the essential difference between the London Discount Market and the London Stock Exchange in the 19th century. When every seller has to guarantee the value of the bill, the incentive structure of the discount market is such that only high quality debt trades, and with every trade the credit quality of the debt increases. This is clearly a means of supporting the price stability of the instruments that trade on the discount market. On the stock exchange, there was no such requirement, because it would have obviated the purpose of the sale.

Why is price stable liquidity so important on the money market? When short term instruments can’t be relied on to hold their value, the public starts to look for better places to put their money, and there are enough reasonable somewhat risky alternatives, including other currencies, that the monetary system will break down if it doesn’t offer enough stability. For a money market to survive over the long term it needs to be in the top of its class in terms of stability.

In short, there’s another aspect of liquidity to add to our taxonomy. Capital markets offer price disclosing liquidity, whereas banks and discount markets offer price stable liquidity. More generally, money markets need to offer price stable liquidity or they will be subject to panics and may be at risk of collapse.

A Taxonomy of Liquidity I

My recent review of Andolfatto (2018) reminds me that underlying the debate between mainstream and heterodox approaches to money is a fundamental dispute over a factual question: Do financial markets and/or non-bank financial institutions provide the same services as banks?

Mainstream approaches typically claim that “clearly” financial markets and non-banks do provide the same services and that the differences are just a matter of degree. In my view, these claims are factually wrong. In this essay I am going to work through a taxonomy of liquidity that is designed to distinguish between the fundamentally different types of liquidity provided by the different types of financial contracts. In my view it is a category error to treat these different types of liquidity as if they were equivalent and interchangeable.

Preliminary question: What do banks do?

I’m going to take it as given that we can agree that banks create money by issuing monetary liabilities. Given this, what I think a lot of modern scholars miss is that those monetary liabilities can be either on balance sheet or off balance sheet. There is a tendency to focus, as Andolfatto (2018) does, on banks’ on balance sheet lending, where the banks issue money in order to fund loans. In fact, however, banks’ contingent, off balance sheet liabilities have for the past few centuries played a crucial role in the monetary system – and they still do today.

When a bank earns fee income by selling the issuer of an asset a credit line that will be used to repay the asset’s owner in the event of a default, the bank is monetizing that asset. Effectively by taking on the tail risk of the asset, the bank turns the asset into the equivalent of a bank liability, even though the bank’s liability is contingent. These contingent bank liabilities are extremely common and may go under the name of acceptance, letter of credit, standby facility, bank credit line, etc.

Because the focus of the mainstream literature on banking is on balance sheet banking, mainstream scholars typically distinguish banks, where debt is held on balance sheet, from markets, where debt is traded. But this framing elides the fact that very often debt is tradable only because of an off balance sheet bank guarantee. As a result, in using this framing mainstream scholars often draw a distinction between banks and markets that is fundamentally misguided.

More recently banks have taken on another role in markets. Morgan Guaranty Trust, which later became JPMorgan Chase, played a crucial role in the development of the modern repo market by market making in repo on the balance sheet of the depository institution so that repo regularly accounted for 10% or more of the depository institution’s assets and of the depository institution’s liabilities from the late 1990s on. Of course, JPMorgan also became a tri-party clearing bank for the repo market. Now that JPMorgan has pulled out of the repo market, the Federal Reserve itself stands ready to lend on the market through its Reverse Repo Program.

Similarly, banks like JPMorgan Chase have been dealers in the derivatives markets since their earliest development, and even today JPMorgan’s depository institution accounts for more than 20% of the US derivatives market (see Table 3 of the OCC’s latest derivatives report). So nowadays we have depository institutions that are not only supporting financial markets via the guarantees they provided to the assets traded on them – as depository institutions have always done – but that also are the key market makers in markets that are viewed as essential to so-called “market-based” lending.

In short, drawing a bright-line distinction between financial markets and banks is a mistake.

Even so, the traditional equity and bond markets continue to operate with relatively minor connections to depository institutions (at least as far as I am aware). These financial markets can properly be viewed as “market-based” lending that is distinct from banks. Thus, while it may be correct to draw a clear distinction between traditional capital markets and banks, it’s also essential to recognize that markets in most other assets, including commercial paper, securitizations, repo, derivatives, etc., rely heavily on the explicit and implicit support of depository institutions for their basic functioning.

This understanding of the nature of financial markets motivates the following taxonomy of liquidity. Taxonomy 1

In addition, to distinguishing between the market, hybrid and bank liquidity that can be provided to an asset, this taxonomy makes another point: different types of liquidity provide very different services to the asset owner.

Market liquidity is the first entry, as it is the archetype that provides the most common mental reference point when one discusses liquidity. Market liquidity refers to the ability to sell an asset without suffering much loss in terms of price. Implicit in the concept is that there is a “true” sale for accounting purposes and that the seller of the asset successfully transfers all of the risk of the asset to the new owner. Thus the balance sheet of the seller of the asset increases by the value of the asset which is received in cash and decreases by the removal from the balance sheet of the risk of the asset (both credit and liquidity).

Nowadays one sometimes hears repos referred to as a kind of market liquidity. This diagram is designed to point out the limitations of repo-based liquidity. As the chart indicates in the row titled “Overnight reverse repo”, repo allows the asset owner to have access to cash without transferring any of the risk of the asset away. This is a very important distinction between market liquidity and repo-based liquidity. Arguably the latter should be called funding and the term liquidity should not be associated with repos at all. Certainly the two concepts are very, very different.

There are two other entries under Hybrid liquidity. The discount market is a historical phenomenon that was very important in 19th century Britain. Bills could trade easily on the discount market as long as they had been “accepted” (i.e. guaranteed) by a bank. A discount market sale was not, however, like a capital market sale: in order to sell a bill the owner had to endorse it, and the endorsement obligated the owner to pay up in the event that the bill went into default. Thus a discount market sale is an effective transfer of the liquidity risk of the bill, without transferring the credit risk of the bill.

A credit default swap is designed to protect the buyer against the credit risk of the asset. Effectively an asset owner can pay the equivalent of an insurance premium in exchange for a promise of payment if the asset goes into default. Note that in this case, the asset owner continues to hold the asset unencumbered on her balance sheet and thus receives no cash upfront from the seller of credit default swap protection. This explains the zero in the “Principal value of asset” column. (Note also that I have depicted credit default swaps here as if they are an effective way to transfer the credit risk of an asset. In fact, these markets are very complicated and there is some concern recently regarding how successful credit default swaps are at transferring the credit risk of an asset.)

There are two entries under “Bank-based liquidity”. The first is a “bank credit/liquidity facility”: this represents the case where for a fee a bank guarantees payment on an asset. As in the case of a credit default swap, this functions effectively as insurance for the holder of the asset, there is no transfer of the asset to the bank, and of course the asset owner receives no payment for the value of the asset from the bank. (On the other hand, the fact that the asset is accompanied by a bank guarantee typically makes it easy for the asset owner to transfer the asset to a third-party in exchange for goods or cash, for example on money markets like commercial paper or discount markets.)

Another important form of bank-based liquidity is the central bank discount window. All loans at the discount window are recourse loans, and as a result in exchange for the central bank’s cash the owner of the asset is able to lay off the liquidity risk, but not the credit risk of the asset.

The point of going through this Taxonomy of Liquidity in somewhat excruciating detail is to make it clear that it is a mistake to talk about “credit” or “liquidity” as if they are simple one-dimensional concepts. Similarly, it is very difficult to draw a bright line distinction between financial markets and banks. Anyone who wants to model money needs to be aware of these issues.

 

A regression discontinuity test error

This is post 3 in my HAMP and principal reduction series. For the introductory post see here.

The series is motivated by Peter Ganong and Pascal Noel’s argument that mortgage modifications that include principal reduction have no significant effect on either default or consumption for underwater borrowers. In post 1 I explained how the framing of their paper focuses entirely on the short-run, as if the long run doesn’t matter – and characterize this as the ideology of financialization. In post 2 I explain why financialization is a problem.

In this post I am going to discuss a very technical problem with Ganong and Noel’s regression discontinuity test of the effect of principal reduction on default. The idea behind a regression discontinuity test is to use the fact that there is a variable that is used to classify people into two categories and then exploit the fact that near the boundary where the classification takes place there’s no significant difference between the characteristics of the people divided into the two groups. The test looks specifically at those who lie near the classification boundary and then compare how the groups in the two classifications differ. In this situation, the differences can be interpreted as having been caused by the classification.

Borrowers offered HAMP modifications were offered either standard HAMP or HAMP PRA which is HAMP with principal reduction. In principle those who received HAMP modifications had a net present value (NPV) of the HAMP modification in excess of the NPV of the HAMP PRA modification, and those who received a HAMP PRA modification had an NPV of HAMP PRA greater than NPV of HAMP. The relevant variable for classifying modifications is therefore ΔNPV (which is economists’ notation for the different between the two net present values). Note that in practice, the classification was not strict and there was a bias against principle reduction (see Figure 2a). This situation is addressed with a “fuzzy” regression discontinuity test.

The authors seek to measure how principal reduction affects default. They do this by first estimating the difference in the default rates for the two groups as they converge to the cutoff point ΔNPV = 0, and then estimating the difference in the rate of assignment to HAMP PRA for the two groups as they converge to the cutoff point ΔNPV = 0, and finally taking the ratio of the two (p. 12). The authors find that the difference in default rates is insignificant — and this is a key result that is actually used later in the paper (footnote 30) to assume that the effect of principle reduction can be discounted (apparently driving the results on p. 24).

My objection to this measure is that due to the structure of HAMP PRA, most of the time when ΔNPV is equal to or close to zero, that is because the principal reduction in HAMP PRA is so small that there is virtually no difference between HAMP and HAMP PRA. That is, as the ΔNPV converges to zero it is also converging to the case where there is no difference between the two programs and to the case where principal reduction is zero.

To see this consider the structure of HAMP PRA. If the loan to value (LTV) of the mortgage being modified is less than or equal to 115, then HAMP PRA does not apply and only HAMP is offered. If LTV > 115, then the principal reduction alternative must be considered. Under no circumstances will HAMP PRA reduce the LTV below 115. After the principal reduction amount has been determined for a HAMP PRA mod, the modification terms are set by putting the reduced principal loan through the standard HAMP waterfall. As a result of this process, when the LTV is near 115, a HAMP PRA is evaluated, but principal reduction will be very small and the loan will be virtually indistinguishable from a HAMP loan. In this case, HAMP and HAMP PRA have the same NPV (especially as the data was apparently reported only to one decimal point, see App. A Figure 5), and ΔNPV = 0.

While it may be the case that for a HAMP PRA modification with significant principal reduction the NPV happens to be the same as the NPV for HAMP, this will almost certainly be a rare occurrence. On the other hand, it will be very common that when the LTV is near 115, the ΔNPV = 0, which is just a reflection of the fact that the two modifications are virtually the same when LTV is near 115. Thus, the structure of the program means that there will be many results with ΔNPV = 0, and these loans will generally have LTV near 115 and very little principal modification. In short, as you converge to ΔNPV = 0 from the HAMP PRA side of the classification, you converge to a HAMP modification. Under these circumstances it would be extremely surprising to see a jump in default rates at ΔNPV = 0.

In short, there is no way to interpret the results of the test conducted by the authors as a test of the effect of principal reduction. Perhaps it should be characterized as a test of whether classification into HAMP PRA without principal reduction affects the default rate.

Note that the authors’ charts support this. In Appendix A, Figure 5(a) we see that almost 40% of the authors’ data for this test has ΔNPV = 0. On page 12 the authors indicate that they were told this was probably bad data, because it indicates that the servicer was lazy and only one NPV test was run. Thus this 40% of their data was thrown out as “bad.” Evidence that this 40% was heavily concentrated around LTV = 115 is given by Appendix A, Figure 4(d):

GanongNoel

Here we see that as the LTV drops toward 120, ΔNPV converges to zero from both sides. Presumably the explanation for why it converges to 120 and not to 115 is because almost 40% of the data was thrown out. See also Appendix A Figure 6(d), which despite the exclusion of 40% of the data shows a steep decline in principal reduction as ΔNPV converges to 0 from the HAMP PRA side.

I think this is mostly a lesson that details matter and economics is hard. It is also important, however, to set the record straight: running a regression discontinuity test on HAMP data cannot tell us about the relationship between mortgage principal reductions and default.

What’s the problem with financialization?

This is post 2 in my HAMP and principal reduction series. For the introductory post see here.

The series is motivated by Peter Ganong and Pascal Noel’s argument that mortgage modifications that include principal reduction have no significant effect on either default or consumption for underwater borrowers. In post 1 I explained how the framing of their paper focuses entirely on the short-run, as if the long run doesn’t matter – and even uses language that indicates that people who take their long-run financial condition into account are behaving improperly. I call this exclusive focus on the short-run the ideology of financialization. I note at the end of post 1 that this ideology appears to have influenced both Geithner’s views and the structure of HAMP.

So this raises the question: What’s the problem with the ideology of financialization?

The short answer is that it appears to be designed to trap as many people into a state of debt peonage as possible. Debt peonage, by preventing people who are trapped in debt from realizing their full potential, is harmful to economic performance more generally.

Here’s the long answer.

By focusing attention on short-term payments and how sustainable they are today, while at the same time heaping heavy debt obligations into the future, modern finance has had devastating effects at both the individual and the aggregate levels. Heavy long-term debt burdens are guaranteed to be a problem for a subset of individual borrowers, such as those who are unexpectedly disabled or who see their income decline over time for other reasons. Mortgages with payments that balloon at some date in the future (such as those studied in Ganong and Noel’s paper) are by definition a gamble on future financial circumstances. This makes them entirely appropriate products for the small subset of borrowers who have the financial resources to deal with the worst case scenario, but the financial equivalent of Russian roulette for the majority of borrowers who don’t have financial backup in the worst case scenario. (Remember the probabilities are in your favor in Russian roulette, too.)

Gary Gorton once described the subprime mortgage model as one where the borrower is forced to refinance after a few years and this gives the bank the option every few years of whether or not to foreclose on the home. Because the mortgage borrower is in the position of having sold an option, the borrower’s position is closer to that of a renter than of homeowner. Mortgages that are structured to have payment increases a few years into the loan – which is the case for virtually all of the modifications offered to borrowers during the crisis – similarly tend to put the borrower into a situation more like that of a renter than a homeowner.

The ideology of financialization thus perverts the whole concept of debt. A debt contract is not a zero-sum transaction. Debt contracts exist because they are mutually beneficial and they should be designed to give benefits to both lenders and borrowers. Loans like subprime mortgages are literally designed to set the borrower up so the borrower will be forced into a renegotiation where the borrower can be held to his or her reservation value. That is, they are designed to shift the bargaining power in contracting in favor of the lender. HAMP modifications for underwater borrowers set up a similar situation.

Ganong and Noel treat this distorted bargaining situation as if it is normal in section 6 of their paper, where they purport to characterize “efficient modification design.” The first step in their analysis is to hold the borrowers who need modifications to their reservation values (p. 27).[1] Having done this, they then describe an “efficient frontier” that minimizes costs to lenders and taxpayers. A few decades ago when I studied Pareto efficiency, the characterization of the efficient frontier required shifting the planner’s weights on all members of the economy. What the authors have in fact presented is the constrained efficient frontier where the borrowers are held to their reservation values. Standard economic analysis indicates that starting from any point on this constrained efficient frontier, direct transfers from the lenders to the borrowers up until the point that the lenders are held to their reservation value should also be considered part of the efficient frontier.

In short, Ganong and Noel’s analysis is best viewed as a description of how the financial industry views and treats underwater borrowers, not as a description of policies that are objectively “efficient.” Indeed, when they “rank modification steps by their cost-effectiveness” they come very close to reproducing the HAMP waterfall (p. 31): the only difference is that maturity extension takes place before a temporary interest rate reduction. Perhaps the authors are providing valuable insight into how the HAMP waterfall was developed.

The unbalanced bargaining situation over contract terms that is presented in this paper should be viewed as a problem for the economy as a whole. As everybody realized post-crisis the macroeconomics of debt has not been fully explored by the economics profession and the profession is still in the early stages of addressing this lacuna. Thus, it is not surprising that this paper touches only very briefly on the macroeconomics of mortgage modification.

In my view the ideology of financialization with its short term focus has contributed significantly to growth of a heavily indebted economy. This burden of debt tends to reduce the bargaining power of the debtors and to interfere with their ability to realize their full potential in the economy. Arguably this heavily indebted economy is losing the capacity to grow because it is in a permanent balance sheet recession. At the same time, the ideology underlying financialization appears to be effectively a gamble that it’s okay to shift the debt off into the future, because we will grow out of it so it will not weigh heavily on the future. The risk is that, by taking it as given that g > r over the long run, this ideology may well be creating a situation of permanent balance sheet recession where g is necessarily less than r, even given optimal monetary policy.

[1] The authors justify this because they have “shown” that principal reductions for underwater borrowers do not reduce defaults or increase consumption. Of course, they have shown no such thing because they have only evaluated 5-10% of the life of the mortgage – and even that analysis is flawed.