Optimize Your Modeling Process
Over the years, financial institutions have gained confidence in the modeling software they rely on to guide asset-liability management, but there is still significant room for improvement. Validations of credit union ALM models from 2012 through 2015 indicate that 52 percent are still ranked as average and 24 percent of models are ranked below average.
Here are 10 areas that we have encountered in our work with financial institutions where modeling can be improved:
1. Data inputs/current contractual position
The precision of an interest-rate risk simulation model depends on the quality of the input data and the category designs that capture the data. The adage “garbage in, garbage out” applies here. Financial analysts should regularly review cash flow and other model inputs. In addition, regulatory guidance is clear about the critical importance of rigorous data quality assessment and documentation. Through regular audits, internal and external data inputs must be continually verified to be accurate, complete and consistent with the model purpose and design.
Balancing the model to the general ledger is only the first step of the review process. It is also necessary to review maturity and repricing schedules and rates for accuracy. For example, ask: Do variable or adjustable rate balances reprice appropriately?
2. Chart of accounts structure
The degree of data aggregation specified within a model’s chart of accounts represents a key component. The chart needs to be granular enough to recognize key ALM behaviors, option effects and pricing. Break out callable investments; loans by product, pricing terms and amortization method; CDs by term; non-maturity deposits by tier; and borrowings with options.
3. Model and category setup attributes
Model and category setup attributes facilitate the projection of future balances and instruct model calculation methodologies and routines. Key attributes include amortizing or bullet (nonamortizing) terms; repricing terms, including caps and floors; teaser terms; balloon intervals; and off-balance sheet instruments. It is crucial to capture all contractual terms and to model current position rate caps and floors.
Proper measurement of interest rate risk requires that assumptions underlying your credit union’s IRR exposure estimates be regularly assessed. For example, option-related risk is defined as the additional variability in earnings performance or equity at risk that arises from option behaviors embedded within the balance sheet. Behavior assumptions applied to the credit union’s non-maturity deposit balances significantly impact the accuracy of model forecast results.
4. Option-related assumption inputs
In the area of investments, preferred modeling should incorporate modeling prepayments of mortgage-backed securities and collateralized mortgage obligations and capture investment call options. Preferred practice loan prepayment assumption inputs should be scenario-specific and based on your credit union’s empirical data. However, if historical data is not available, at a minimum loan assumption inputs should encompass industry or peer inputs and be confirmed to be reasonable with back-testing. Funding assumption inputs involve modeling for variable-rate and “step-up” types of CDs as well as option-related wholesale funding sources.
5. CD early withdrawal assumption inputs
Given the longevity of low market rates, the option value has a low breakeven point in a rising-rate environment, and customer withdrawals are likely. As rates increase, credit unions can anticipate a greater percentage of withdrawals, and this forecasting should be incorporated into IRR and what-if analysis, with both sensitivity and stress testing.
6. Non-maturity deposit assumption inputs
Behavior assumptions applied to your credit union’s non-maturity deposit balances significantly impact the accuracy of model forecast results. These assumptions include repricing betas that should change across products and pricing tiers as well as repricing limits, or rate floors. Repricing limits should represent the lowest rate that can be paid and the lowest that will be paid based on your credit union’s pricing strategy.
Accurate repricing betas should encompass both historical tendencies and management expectations with respect to future conditions. Three components should be considered: historical quantified repricing, supply response to repricing such as product migration in a rising-rate environment, and the credit union’s current and future deposit market. Incorporate your board’s repricing preferences; understand your depositors by non-maturity deposit type; know your competition, including Internet companies with deposit offerings; and consider liquidity needs as interest rates rise.
Decay rates forecasting the trends of non-maturity deposits leaving the balance sheet should be based on a historical analysis of the credit union’s experience, and not on industry estimates or default vendor assumptions. Decay assumptions should also include any exposure to transient balances that have more rapid runoff.
7. New volume assumption inputs
Accurate, supportable new volume assumption inputs specific to interest rate scenarios are essential to earnings forecasting. New volume inputs should reflect your credit union’s current pricing and terms and include caps and floors if material in originations.
8. Present value assumption inputs
Equity-at-risk analyses aim to measure the potential long-term exposure of the credit union’s balance sheet. For investments, separately derived option-adjusted spread valuations are preferred over discounting from a yield curve. Loans can be discounted using a curve plus/minus rate spread that is reflective of current market pricing and appropriately captures cash-flow terms. Analysis of variable-rate loans should use a repricing index.
9. Other forecast inputs
Additional forecast inputs, such as balance sheet forecasts and non-interest income and expense inputs, are material drivers of forecast accuracy. In the balance sheet forecast, model a static balance sheet for IRR reporting and periodically model a dynamic balance sheet and measure IRR. Model non-interest income and expense based on current actual or budgeted amounts.
10. Risk reporting
A model’s reporting component translates the estimates and forecasts produced by the model into useful business information. However, model risk can result from the incorrect or misuse of model outputs and reports, leading to potential financial loss and/or poor management decisions. IRR risk reports should be generated directly from the model, and any external modifications to IRR results should be fully documented.
In sum, we recommend an analysis to determine if any or all of these practices could help improve your credit union’s model and process:
- Start with the right ALM model and appropriate internal commitment.
- Continuously review and look for potential improvements.
- Base assumption inputs on historical analyses specific to your credit union.
- Confirm model results via back tests to achieve board buy-in.
- Develop a production calendar to keep your team on track and use resources efficiently.