The Inertia-On-Inertia Paradox
Updated: May 26, 2022
How inertia impacts credit modeling methods.
Like general credit modeling techniques, Machine Learning updates credit loss understanding from more data. In supervised learning, the data teaches:
To improve and adapt the ability of the independent variables to predict a dependent variable (feature engineering) and
The appropriate underlying functional relationship best describes the independent and dependent variables. (ML Methods like Architecture and Estimators) [i]
In the following narrative and as a metaphor, I will connect physics to describe how inertia impacts Machine Learning and related modeling techniques.
In the calm environment (matter phase state = solid), best described by normal statistical distributions (kurtosis = 3) and feature independence (low inertia), a particular ML method will be chosen. This will work fine for many years, that is until it doesn't work. Once the uncertainty of a dynamic environment is added (like when heat is added to melt ice and the matter phase state = liquid), kurtosis and inertia will increase dramatically. This invalidates the current ML method. The problem is that the new phase state will not have enough new data to validate its existence, including updating both the prediction AND the ML method until the damage has been done. (in effect, the ML method itself maintains inertia) This is not a "speed of computing" problem; it is a phase state "transition time to complete" problem, aka, The Inertia on Inertia Paradox. This may also create a damaging self-reinforcing cycle, especially if the cause of the uncertainty relates to the modeling methods suffering from inertia. The model can't update, so the issue generating uncertainty continues because the uncertainty signal is not available to the existing ML model. See below for an example.
It takes data to validate a model, it takes time to realize and collect data. The problem is that the current predictor continues to operate until a new predictor is determined. In a fast-moving and chaotic environment, the current predictor is very likely wrong and inappropriate. The question becomes: Why does a storm suddenly and unexpectedly change a long period of calm? I do not know. But my experience tells me a quick and unexpected transition from calm to turbulence is to be expected. One just does not know when the transition will occur. As Mathematician and Physicist Freeman Dyson said about natural history:
“Time loves to sit quietly for millions of years, and then to pounce suddenly in a single hour of fury. Nothing is permanent, but the illusion of permanence can last a long time.” [ii]
The same can be said about credit risk, except the quiet periods are measured in much shorter business cycles.
The Storm Framework - a metaphor for uncertainty and risk:
This Storm Framework is intended to help build context for the phase change environment, like going from calm to stormy, and then back to calm. The Inertia-On-Inertia paradox is primarily found in the transition from the “calm before the storm” to the “storm” segments. (From 1 to 2) It is this transition point where great care should be afforded to existing statistical-based models. The world is transitioning to less transparent Machine Learning techniques, such as neural networks and unsupervised learning. This reality creates even more risk at this transition point. As we describe in our Simulation-based Credit Analytics: Managing tail risk and uncertainty article, this storm framework transition point is the time at which statistical-based models are subordinated to simulation-based models. Waiting until the point of transition is likely too late. A best practice is to utilize the "calm" time as an investment to prepare for the potential "storm."
An Inertia-On-Inertia Paradox example: Prior to the 2007-2008 financial crisis, there was an inappropriate practice of using past mortgage product loss models (dominated by data from lower-risk fixed-rate mortgages) to predict losses on higher-risk negative amortization mortgage products. No one believed a mortgage product could have loss rates in excess of credit cards - that is until they did. The self-reinforcing negative cycle occurred because the true credit loss signal associated with higher-risk loans was unavailable to the models. It was not until the housing market collapsed that the credit risk was revealed.
Perhaps, the message is, that the credit policy should be stable. Over long periods of time, we will have long periods of calm. This will be a time to maintain a more conservative credit policy as a credit capital savings mechanism. Like “saving for a rainy day.” In those briefer but intense storms, the credit policy stays fixed, thus releasing the savings in terms of credit availability. This approach is known as a countercyclical policy.
In a similar vein, credit policy should be used very sparingly, if at all, as a tool to expand homeownership. I have heard it said, “You can not lend your way out of an affordable housing problem.” Policymakers should focus substantially on housing supply, with credit policies only adapting to available demand and based on consistent credit standards. This case is made in our article The affordable housing paradox.
[i] Breeden, Survey of Machine Learning in Credit Risk, 2020, Available at SSRN: https://ssrn.com/abstract=3616342 or http://dx.doi.org/10.2139/ssrn.3616342
[ii] Dyson, Frogs and Birds: Selected papers if Freeman Dyson, 1990-2014, 2015