The end of my three-part discussion on banking (link) ended with the argument that the circular nature of funding flows means that they do not represent a constraint on debt growth, but rather the willingness (and ability) to absorbing credit risk is the constraint. If we accept this, there is an important implication for our ability to predict recessions that are driven by a contraction of “animal spirits”. I made this argument in the first volume of my (planned) two-part texts on recessions. (I’ve only completed parts of the second volume, as I’ve hit some difficult areas that I think I need to think about more carefully.)
I only have an informal version of the argument, and it goes like this. The usual driver for business expansion is private investment, which normally has to be financed by the investing firms/households. This implies that financial firms must be willing to extend credit. This will—the “animal spirits” of corporations and financiers—is therefore the engine of the cycle. So far, that’s pretty much what Keynes wrote. The catch, which may or may not be novel, is that credit markets are actually markets. If we could predict credit market trends, a lot of that involves the ability to time the markets, which implies that forecasting recession comes up against some version of the efficient markets hypothesis.
There is a lot of academic controversy over the efficient markets hypothesis, which I don’t care about. I’m referring to a simplistic version of this: if you think you can reliably predict the direction of the markets, why are you publishing academic papers and not on a yacht somewhere? My belief is that this is close to the intent of post-Keynesian views of the business cycle, but they don’t frame it that way due to their allergy to “market efficiency”, and they use therefore more confusing terminology to explain the concepts I am discussing here.
Not all recessions are credit driven
Although most historic recessions have been associated with slowdowns in debt growth, it is possible that the economy could contract for other reasons – and such a possibility is a major risk today. The figure above shows the growth of non-financial debt in the US, along with shading from recessions. We can see that credit growth has historically coincided with slowing debt growth. In recent decades, the savings and loan crisis has been associated with the 1990 recession, the 2000 recession has been associated with technology companies losing access to credit markets, and the 2008 recession was notoriously a debt crisis. These are the types of recessions that I consider difficult to predict.
However, we can also experience recessions caused by other causes, which requires us to anticipate these other factors. We may have a policy-induced recession (e.g. euro periphery), where we have to predict the policy. The recent recession was caused by a pandemic (and the response to the pandemic), which is a problem of medical forecasting. We can also have downturns triggered by things like loss of energy imports.
Currently, the most pressing problems seem to be of this type: the new confinement in China and the effects of the restriction of energy (and food) supplies. A credit crunch could be triggered by these events, but it’s hard to see credit as the driver of any downturn. As such, the comments here are not directed to the current situation.
Problem for macro templates
Credit risk being the constraint for debt growth is a problem for most macroeconomic models—agent-based models being the notable exception. It is difficult for a mathematical model to capture the uncertainty associated with credit risk and the trend in the evolution of risk appetite.
As a starting point, consider the simplest macro model structure: consistent stock-flow (SFC) models. (I have a book on building SFC models in Python.) An SFC model has two categories of equations:
The equations that define the accounting structure of the model (“accounting identities”), and
Equations that determine the behavior of sectors (which are constrained by the previous type of equations).
The arguments in my article on banking can be rephrased as follows: accounting identities do not limit the growth of debt, but it is entirely driven by behavior.
We could imagine a model where some model parameter acts as a “credit growth” factor. That is, the model will settle at a certain rate of private sector debt growth once other variables enter steady growth states. To then model a credit-induced recession, we could jump the parameter(s) to new values that reduce debt growth, which then has a ripple effect on growth.
Although this technique creates a “teaching model” that helps explain why recessions occur, the problem is that it does not help predict. We need to forecast the parameter change that leads to weaker credit growth, which is no easier than forecasting a recession in the first place.
We could assume that the “credit growth parameters” are determined by variables within the model. We might then be able to predict parameter changes. However, it does imply that we could use this technique to predict market developments (and thus be on the right path to becoming a yacht owner).
DSGE models have notorious difficulties with credit. The premise of the models is that entities (households, firms) have a starting amount of tradable goods and services, and then trade these endowments among themselves to achieve an optimal outcome.
“Currency” in these models is government money, created by politics, and therefore political rules limit its growth. Since these templates are usually primarily used to discuss policy, that’s not a problem.
The addition of private credit creates inherent problems in the optimization strategy of these models.
The models are built around the futures prices of everything which are traded on the futures markets. If companies can buy and sell everything forward, there is no fundamental uncertainty that can lead to defaults.
Since private credit ‘comes out of nowhere’, there is no initial ‘endowment’ for trade. Trying to impose a repayment constraint (“no Ponzi condition”) runs into the following problem: if every firm can issue credit, there is no reason for nominal income growth to translate into a limit finite if nominal interest rates are limited. All private sector entities can grow their balance sheets in concert, creating nominal income that allows debt service.
The models are based on representative agents who represent a sector or part of a sector. If they are indeed representative, all the firms represented would fail at the same time.
The methodology is not suitable for managing credit risk, and so the best one can hope for is to add heuristic behavior that purportedly captures credit dynamics. We find ourselves in the same position as SFC models: we have to predict changes in the parameters of the model.
Agent-based models are better positioned to capture credit dynamics because they have entities that face the same information limitations that lenders and borrowers face in the real world. Thus, it should be possible to create models that generate credit-driven business cycles.
The problem with such models is trying to fit them to observed data, because they contain an extremely large number of state variables. Although there is research pointing in this direction, the adjustment problem will always be difficult to solve.
Mathematical economic models struggle to model credit dynamics because they are driven by herd psychology. They might be able to explain why the economy shrinks if loans crash, but predicting such a crisis is just as difficult as predicting the direction of the markets.
Editor’s note: The summary bullet points for this article were chosen by the Seeking Alpha editors.