search trigger icon
search close button

Pro Tips for Modeling ALM Buzzwords

Joe Rezac
Jun 24, 2021

Professional woman reading ALM Model Metrics on her tablet

Over the past year, the banking industry has seen many unprecedented events that have resulted in both margin- and capital-related challenges. The flat, ultra-low yield curve coupled with the bloated growth of non-maturity deposit balances have many banks considering how to best deal with the current reality. To forecast strategic and economic impacts on key metrics, banks continue to rely heavily on financial modeling using various tools available. However, the adage “garbage in, garbage out” rings true now more than ever when it comes to a model’s accuracy. With the industry’s heightened level of uncertainty, now is a crucial time to ensure model assumptions and methods continue to be reasonable.

The 2011 Interagency bulletin on Supervisory Guidance on Model Risk Management does an effective job of identifying the need for sound modeling practices, as well as how to go about doing this conceptually. But as effective as the bulletin is, it is intended to only provide a high-level overview. As a result, specific details on how to achieve these items need further clarity.

In the supervisory bulletin, many regulatory buzzwords are mentioned. However, these buzzwords are often ambiguous and can be interpreted in several ways. To provide some additional clarity from an Asset Liability Management (ALM) modeler’s perspective, the most common model-related buzzwords from the bulletin are discussed below. Although these buzzwords can apply to any financial model, the discussion below will focus on how to properly apply them to your ALM model. Here are the five most common regulatory buzzwords related to ALM modeling:


Backtesting is a critical part of the modeling process. It is done to gauge how well the model predicted future results based on the assumptions being used.

Of all the regulatory buzzwords heard, backtesting may be one of the most ambiguous since any assumptive input could theoretically be backtested. This includes deposit runoff, loan prepayments, pricing, forecast growth, new volume assumptions, and balance sheet composition. If a request is made to backtest an ALM model without a specific set of assumptions implied, a common backtest performed is on the bank’s budget-based forecast. Focusing on the parts of the balance sheet that determine the bank’s net interest income, this type of backtest comprises three key assumptions:

  • Future rates on loan, investments, deposits, and borrowings
  • Volume growth expectations over the next 12 months
  • The composition of earning assets and funding (also known as the mix)

Done properly, a model backtest will show how accurate the model’s assumptions were, as well as the specific areas where the assumptions could be fine-tuned to potentially increase future modeling precision.


Benchmarking involves evaluating your current model’s inputs and outputs compared to an alternative model or data source. Big banks often do this by building two independent models and running them parallel to each other. However, this is not practical for community banks since it is both cost and time prohibitive. As an alternative, community banks can focus on pertinent information provided by the regulators and other vetted sources.

Banks are well equipped to do this type of benchmarking with ratios that are reported quarterly via regulatory channels. However, benchmarking assumptive inputs isn’t as clear. A common question asked by our ALM Consulting clients is, “How do our assumptions compare to other banks that you work with?” This is the type of question that benchmarking is intended to answer.

A good example of an interest rate risk (IRR) benchmarking source can be found in the Spring 2021 OCC Interest Rate Risk Statistics Report. This report is full of timely IRR data. It is broken down by bank asset size and includes current IRR volatility exposure, risk limits, and non-maturity deposit average life/pricing data. Using this as a high-level IRR benchmark will indeed help answer the question of where your IRR assumptions (inputs) and exposure (outputs) fall within the range of your banking peers.

Model Validation

Although a model validation is one part of overall model risk management, it is often the first line of defense in the process. The purpose of a model validation is to ensure model settings, assumptions, and outputs are sound. It also identifies the limitations of the model setup to assess the potential impact to result in precision. A model validation should not be confused with a model certification, which verifies that the underlying model itself is sound to perform as intended.

There are many useful ways to approach a model validation. Since different vendors have different areas of focus, the best practice recommendation is to get a fresh perspective periodically to ensure the most complete analysis of your model. For example, if you have an ALM model validation performed by your bond accounting vendor, expect their focus and recommendations to be primarily on your investment portfolio. There is nothing wrong with this type of model validation if you follow it up with a more comprehensive model validation on the next one.

One effective type of model validation is a Technical Model Validation where both the model settings and assumptive inputs are reviewed for reasonableness. This is the type of model validation performed by the ProfitStars group. There are many vendor options here. The one caveat that regulators have is that the model’s validator should be independent of the model’s primary user. This will help ensure objectivity and prevent bias.

Sensitivity Analysis

A sensitivity analysis intends to show the change in model results for an incremental change in an assumptive input. Since there are many assumptions built into a financial model, the secret to performing a successful sensitivity analysis is to isolate a single assumption and show its impact on model results. An effective sensitivity analysis will help show the modeler which assumptions have the greatest potential impact on results, particularly if the current assumption being used changes materially in the future. This should help identify which assumptions need the greatest level of scrutiny and should be regularly stress tested and fine-tuned.

A good example of an ALM sensitivity analysis would show that for every year a loan portfolio gets extended, fair value results erode by $X. This can be accomplished by extending the loan portfolio out to its maximum contractual duration and comparing fair value results in this extreme scenario to a baseline scenario that includes the bank’s current prepayment assumptions.

Stress Testing

Of all the regulatory buzzwords discussed in this article, model stress testing is the item that offers the most diversity. As the name implies, stress testing is designed to see whether a bank can withstand an economic crisis. As with other model techniques, it is recommended to start simple and build from there by running different levels of stress tests.

A good example of a multi-level ALM stress test would be to show how the bank’s liquidity position erodes under varying economic stress events. These events can be bank-specific or macro-economic. A typical three-part liquidity stress test could look like this:

Moderate (Stress Test #1):

  • Unplanned X% runoff of money market balances over the next three months.
  • CD balances above the $250K FDIC insurance limit not replaced after they mature for the next 12 months.

Severe (Stress Test #2):

  • Moderate Stress Test scenario +
  • Haircut of investment values of $Y. This devaluation is the estimated market value decline in a +300 bps rate environment.
  • 50% reduction in borrowing line availability.

Critical (Stress Test #3):

  • Severe Stress Test events +
  • Unplanned money market balance runoff over the next three months increases to 2X%.
  • Loan prepayments stop.

Since the critical stress test is often a perfect storm of conditions gone wrong, it often helps to identify when a bank hits its tipping point for a specific metric. This occurs when the metric redlines or falls outside of the bank’s policy targets. Using the liquidity example above, this is a great way to prove your defined liquidity risk limits since you can confidently state, “Our primary funding coverage becomes a deficit when the following four stress events occur…” This type of analysis is powerful since it translates actual economic events into numerical changes. And it shows your regulators that you have thought about the types of conditions that it would take to put you into a short-term liquidity pinch.

Based on the above evidence, a proper model review requires quite a bit of critical thinking time and takes quite a few different report iterations. Rather than being overwhelmed by “analysis paralysis,” a best practice recommendation is to put a plan together that tests a couple of the buzzwords listed above each quarter. This will ensure all items are covered on a regular basis and help best prepare for your next exam. In addition, most of these analyses will also provide insight that can also be used for strategic planning.

subscribe to our blog

Stay up to date with the latest people-inspired innovation at Jack Henry.

blog subscription image
floating background gradient

contact us

Learn more about people-inspired innovation at Jack Henry.