Financial institutions are required to have their asset liability management (ALM) process and model validated by qualified third parties. Model validations can include a wide range of procedures including policy reviews, sample testing, and assumption and results reviews. To maximize value, validations can be taken one step further by also providing a full replication of the model being used. Typically, ALM models are considered “black boxes” where the institution inputs the data and assumptions, and the model outputs the results. The outputs from ALM models are then used to drive major business decisions. Financial institutions should be able to have full confidence in their model and its ability to produce results that are reasonable for their organization. In order to do so, financial institutions should perform full replications of their model to not only better understand the fundamental aspects of a model but also to gain complete assurance in the outputs. Following is a discussion of the four main areas to focus on throughout an ALM model validation.
ALM models often require complex methodologies which in turn require full support and documentation on every step of the process. Validations generally start with an evaluation on the model’s policies, internal controls and model documentation. The validation assesses if the model governance framework is sound and regulatory requirements are covered as well as provides any recommendations for enhancements with management. As compliance requirements continue to expand, transparency around the model, process and assumptions is key.
To produce meaningful results, the data that goes into a model needs to be clean and accurate. The next step in a validation includes a full review of not only the input data but also the historical financial performance data. Understanding the historical financial performance provides insight to how the institution has performed in the past and helps the validators better understand certain aspects of the model. For instance, analyzing historical trends in various loan and/or deposit categories provides the validator with a baseline of the institution’s operational strategy. This information can then be used to compare against the model’s assumptions for projecting forward.
A key requirement of ALM is incorporating forward-looking assumption data into the calculation of cash flows over the life of each financial instrument. These assumptions help drive the short-term projections for net interest income and the long-term projections for fair value across various interest rate shock scenarios. The main drivers are voluntary and involuntary (default) prepayment rates on the loan categories, the decay, beta and effective maturities on the non-maturity deposits, and the interest rate shock methodology. Therefore, a major component of a successful validation is a complete review of all of these assumptions in addition to the historical behavior of the loan and deposit portfolios in order to backtest against the assumptions used. These reviews help to ensure the validity of the assumptions to the balance sheet, the institution’s current and historical performance as well as the expected future environment.
This is an area that is often missed in validations but it is important to confirm that the inputs are reasonable for each specific institution. For instance, a complete validation should analyze the comparison between the prepayment rates on the institution’s loan categories against its actual prepayment rates recorded during various interest rate environments. This will allow the validator to identify how prepayments changed as interest rates increased or decreased over time and compare it against how prepayment rates are projected to change throughout the various interest rate shock scenarios. This procedure can then be conducted against the deposits as well as any major investment categories.
The most important part of any validation is the testing of the model. Typically, a validation might conduct sample testing of various financial instruments. However, ALM models are becoming more and more complex, and assumptions used on certain financial instruments can differ drastically from other financial instruments. For instance, assumptions on high-balance money market accounts can vary significantly from low-balance checking accounts. Therefore, performing a full replication of the ALM model is a best practice when conducting a validation. This includes obtaining all of the data sets and assumptions used by the institution in order to produce an independent ALM profile. Once the profiles have been run, including modeling consistent interest rate shock scenarios, a category-by-category level variance analysis can be performed. Variances can occur for a variety of reasons: user error, data errors and/or methodology differences. A recent client validation uncovered a methodology error that the client wasn’t aware of. The client had thought the prepayment methodology on a specific loan category was being vectored over time based on changes in interest rates. However, when an attempt was made to replicate the prepayment methodology on that loan category, large variances were identified, and it was discovered that the prepayments used by the client were actually remaining constant over time. Without the full replication of each loan category, this error may not have been uncovered.
Overall, the full replication approach to a validation gives the user confidence that the model is estimating cash flows and providing insight into their interest rate risk in an accurate and reasonable manner.
For more information on this topic, or to learn how Baker Tilly’s Value Architects™ can assist your organization, contact our team.