Assumption Overrides: A Cautionary Tale
A few years ago, I was assigned as the analyst and modeler for a new DCG Advisory client. They had been performing their IRR modeling in-house for as long as anyone could remember but decided, for a variety of reasons, to switch to outsourcing. As we started the transition to the new process, we noticed they had been modeling very short average life assumptions (one year) for several core non-maturity deposit accounts. When we asked the CFO for the rationale or support behind such conservative assumptions, he seemed surprised by their existence and unsure why they were in use. After digging around, he discovered these model settings were due to an assumption override made seven years prior!
How could this have happened? Well, seven years prior, the institution had acquired another bank. To account for potential flight risk in the wake of the merger, the modeler decided to assume very conservative average lives for the acquired deposits, which were broken into their own model categories. Unfortunately, no one had documented this override, why it was done, who was responsible, how it would be monitored, or when it would be reversed. Over the course of the next seven years, personnel left, the model changed hands, and this one assumption was completely forgotten but still active in the model. The deposit categories were still in use, but now comprised of growing balances from loyal customers whose relationships had stood the test of time. Even though the flight risk for these deposits had long since passed, the model setting had been continually rolled over without it ever being questioned.
The real tragedy is this bank had been out of policy in its EVE limits for some time, largely because of this outdated assumption, and management had decided to take on expensive long-term funding to ensure they remained within their limits. They had been spending tens of thousands of dollars every year to hedge against a risk that was no longer real!
Don’t let this happen to you! The events of 2020 have thrown many model assumptions into disarray, historical trends have broken down, and overrides like these abound. Given so much uncertainty, it is crucial to up your game right now when it comes to managing assumption overrides.
If you decide to implement overrides for established assumptions, the first step is to ensure all key questions are answered:
What? Clearly define what assumptions are being over-written and which model accounts are impacted. What was the prior assumption, what is the override, and what is the impact on the risk profile?
Why? The catalyst for the override often gets lost or is never clearly articulated. Why are the established assumptions no longer reasonable? Why is the override more reasonable?
Who? Is it just the modeler who can make these judgement calls, or is dual authorization required? Does the ALCO or board need to approve the override before it is implemented in the core model? Who is ultimately responsible for reviewing and approving any assumption overrides? Is this chain of authority and responsibility clearly defined in the Interest Rate Risk or Model Risk Management policies?
When? Often, a timeline or expiration date is never even considered. When is the override effective? Is it intended to be temporary or could it become more permanent?
How? The importance of ongoing monitoring increases when overrides are in play. How will this assumption override be monitored, refreshed, sensitivity-tested, and either discontinued or converted into a core assumption?
Answering these questions is only half the battle. As important, if not more so, is to ensure these answers are communicated and documented for all key stakeholders.
ALCO and Board Documents: Are there certain graphs or tables in the ALCO or board reports where footnoting the override makes sense? Effective dates and timelines may be worth citing, particularly for period-to-period comparisons where risk profile changes may be heavily influenced by assumption overrides. Don’t forget to ensure overrides are fully communicated and documented in the corresponding meeting minutes, especially if authorization or approval is required by policy.
Change Control Logs: Regarding model risk management, change control logs can be helpful in tracking why specific model settings were updated or overwritten, serving as a corporate memory that extends beyond any individual personnel. Logs can also provide an added layer of documentation and security to the modeling process if dual authorization is required for changes. If you don’t have a change log in place, there is no time like the present to consider adding this layer of control to your process! If you do, are you confident that your log contains sufficient detail for any assumption overrides?
Modeling Procedures: The modeling process is always evolving; procedural documentation should be updated regularly to reflect this evolution. Do your procedures include explanations of all assumption overrides? Does this include processes for testing and monitoring these overrides, including any timelines in place? Ensuring procedures are both current and robust is especially vital if your staffing is limited or you are single threaded in your modeling role. These are the details that get missed in hand-offs when personnel leaves or responsibilities shift.
For that new client, we ensured their new model was set up with reasonable, documented, supported, and current average life assumptions for all deposit accounts, and ran a simulation comparison to illustrate the impact of the assumption change. At our first ALCO meeting together, we also presented alternative NII simulations to quantify how much interest expense could be saved by unwinding the funding extension, alongside an alternative EVE simulation to illustrate they would remain comfortably within their policy limits while doing so. This was the first of many profitable strategies the bank has executed since working with us.
While the basic tenants of model risk management hold true for all assumption changes, they are especially important when a decision is made to override existing support for a model assumption. Ensure that all assumption overrides are justified, tested, communicated to all stakeholders, and monitored to ensure they remain realistic. Do this now while they’re fresh in your mind! After we emerge from the chaos of 2020, your future self – and bottom line – may thank you!
Learn more about Model Risk Management.
© 2021 Darling Consulting Group, Inc.