Blog

One of the more common findings in an ALM validation is a disconnect between the replacement rates assumed in the model and the rates the institution is actually posting. The model continues to run. The outputs look reasonable. But the assumptions have not kept pace with current pricing, and the income simulations no longer reflect reality.

This happens most often in internally managed models. Pricing decisions get made, but the model does not get updated. Responsibility for maintaining assumptions is informal. Over time, the gap grows.

The fix is straightforward. Replacement rate assumptions should be reviewed and updated on a defined schedule. Someone should own that process. A simple cross-check comparing model inputs to current posted rates will catch most problems before they affect the results.

These steps are not complicated. They are just easy to skip when no one has been assigned to do them. Model accuracy depends less on how the model was built and more on whether the inputs reflect current conditions. Even well-constructed models produce unreliable results when assumptions are stale.

ALM models are built on assumptions. Betas, deposit retention, and prepayment speeds all shape the results. Assumption sensitivity analysis is one of the most effective tools for understanding how dependent those results are on the inputs behind them.

The concept is straightforward. After the model is run using standard assumptions, it is re-run using different and less favorable ones. The results are then compared. This is not a single test. It is a series of runs, each isolating a specific assumption. Deposit betas may be increased to reflect more aggressive member pricing. Separately, non-maturity deposit retention may be shortened to reflect faster runoff or prepayment speeds may be adjusted to slow in rising rate environments and accelerate in falling ones.

Each run answers a different question. How sensitive is net interest income to higher funding costs? How does economic value shift when deposit retention assumptions are changed? How exposed is the institution to changes in loan cash flow timing? Viewed separately, the results reveal which inputs are most impactful.

The value of this analysis is not in producing a single stressed outcome. It is in identifying which assumptions matter most. In many cases, one or two inputs drive a disproportionate share of the risk profile. That insight allows ALCO to focus attention where it is most needed. Results should also be interpreted carefully. Assumption sensitivity analysis uses deliberately conservative or even unrealistic inputs. The goal is not to predict a likely outcome. A scenario that produces results outside policy limits under extreme assumptions does not necessarily indicate a need for corrective action.

The analysis is best used as a learning tool. It helps management understand how the risk profile would change if real world conditions were very different from the standard model run, which assumptions are the most impactful, and where additional assumption support may be warranted.

It is surprising how many ALM models still rely almost entirely on parallel rate shocks as part of their standard reporting. While parallel scenarios provide a useful baseline, they do not reflect how interest rates actually move. The yield curve rarely shifts in a uniform fashion. It steepens, flattens, and twists, often in ways that create very different outcomes for earnings and value.

This is not a new expectation. Regulatory guidance has been clear for over a decade. The Advisory on Interest Rate Risk Management issued in January 2010 states that institutions should assess exposures beyond typical conventions, including larger rate changes and movements across different tenors to reflect changes in the slope and shape of the yield curve. Despite this, non-parallel testing is still frequently absent or treated as optional.

The impact of this gap can be significant. A balance sheet that appears well positioned under a parallel shock may respond very differently under a steepening or flattening scenario. For example, funding costs tied to short-term rates may reprice quickly, while longer-term assets remain fixed. In a flattening environment, this dynamic can compress margins more than a simple parallel shock would suggest. Without non-parallel testing, these risks remain hidden.

The issue is not model capability. Most modern ALM systems can run these scenarios. The issue is whether they are incorporated into regular analysis and decision-making.

The takeaway is straightforward. Parallel shocks are a starting point, not a complete assessment. A credible IRR framework requires a broader view of how rates move and how the balance sheet responds when they do.

As a credit union grows in size or complexity, examiner expectations tend to evolve alongside it. One common request is for formal back testing of the ALM model. At its core, back testing compares what the model projected for net interest income to what actually occurred over a defined period, typically one year. Done properly, this is a valuable diagnostic tool. Done poorly, it becomes little more than a compliance exercise.

The most important step is aligning the comparison to the actual rate environment. If market rates increased by 100 basis points over the period, the appropriate benchmark is not the model’s base case projection. It is the model’s +100 basis point net interest income simulation. This ensures that the comparison reflects the conditions the balance sheet actually experienced.

Once aligned, the variance between projected and actual results begins to tell a story. In most cases, differences fall into two categories. The first is volume. Changes in loan growth, deposit mix, or balance sheet composition can materially affect earnings, particularly if those shifts were not anticipated in the original projection. The second is rate-related. If actual yields or funding costs diverge from modeled assumptions, it points to issues with betas, pricing assumptions, or repricing dynamics.

The value of back testing lies in how these insights are used. Persistent volume-related variances may indicate a need for more robust growth forecasting or more frequent updates to model inputs. Rate-related differences may suggest that key assumptions require refinement to better reflect observed behavior.

Back testing analysis is a forward-looking tool. When used effectively, it improves the accuracy of future projections, strengthens confidence in the model, and enhances the quality of strategic decision-making.

ALCO meeting minutes sometimes amount to a copy and paste exercise, with little evidence of discussion, no reference to economic or rate outlooks, and no questions or follow up items. This pattern is more common than it should be, and it often appears as a repeat examiner finding.

What makes the situation notable is that the underlying risk management is often not weak. The CFO may have a strong grasp of interest rate risk, ALCO meetings are held regularly, and reporting to the board is consistent. The gap is not in practice, but in documentation.

From an examiner’s perspective, sparse minutes lead to a straightforward conclusion. Either the discussion is not happening, or the participants are not engaged. In either case, the governance process appears ineffective, regardless of the actual level of expertise within the institution.

The fix is to align documentation with reality. Effective minutes include key discussion points, economic context, assumption considerations, and documented questions with follow up where appropriate. AI based recording tools can capture meetings and generate a draft set of minutes, but the output should be treated as a starting point. It requires review, refinement, and the application of judgment by the CFO or designated owner.

Expanding the question and answer portion of meetings is also helpful. In some cases, prompting discussion is necessary to ensure that critical topics are clearly reflected in the record.

The takeaway is simple. If it is not documented, it did not happen, at least in the eyes of an examiner.

Across a number of ALM validations, a recurring pattern appears in ALM policies that is worth addressing. Many policies define ALCO membership using general language such as “members will be designated by the CFO” or “comprised of the senior leadership team.” While this may seem flexible, it does little to clearly assign responsibility or ensure consistent participation.

From a governance standpoint, this approach creates ambiguity. If roles are not explicitly defined, accountability becomes informal. Over time, this can lead to inconsistent attendance, uneven engagement, and a lack of clarity around who is responsible for key decisions and oversight. Examiners often cite this as a requirement.

A stronger approach is straightforward. ALCO membership should be defined by title within the policy. Roles such as CFO, CEO, Head of Accounting, VP of Lending, and other key leaders should be clearly identified. This establishes accountability and ensures that the appropriate perspectives are consistently represented at the table.

In many cases, it is also beneficial to include a Board representative. While not required, this can strengthen the connection between management-level discussions and Board-level oversight.

Clear structure leads to better communication. Better communication leads to better decisions.