A previous post discussed some of the challenges associated with modeling non-maturity deposits (NMD), especially with respect to economic value of equity (EVE). Deposit modeling for earnings-at-risk (EAR) is more tractable but presents challenges of its own.
What is EAR?
EAR looks at changes to net interest income from changes in market interest rates. Unlike EVE, EAR’s starting value is known and accepted. Banks calculate NII at least quarterly, and those calculations show up on the bank’s financial and regulatory reports. In contrast, EVE represents an estimate for a mix of liquid and illiquid assets and liabilities. You may be able to benchmark an EVE for reasonableness, but the estimate still involves a lot of assumptions.
In other ways, however, EAR can be more difficult to model than EVE. Or at least to model well. EVE merely takes a snapshot of the current balance sheet and then estimates present values under different interest rate scenarios. An EAR projection needs to estimate not just runoff of assets and liabilities, but also replacement and growth. There are a lot of moving parts.
These moving parts also mean multiple approaches to EAR, each with their strengths and limitations. A dynamic approach allows balances to grow or shrink over time, depending on the scenario. A common problem with a dynamic approach is that it can assume that management will consistently make the correct choice, leading to the optimal asset and liability mix. This can too closely resemble peeking at the answer sheet. It’s not as though this sort of optimal decision making necessarily happens in practice. Overwise, we would not have seen so many banks load up on 3% fixed rate MBS in 2020.
A static approach assumes a fixed balance sheet, with like for like replacement of assets and liabilities. Assuming a static balance sheet makes the EAR analysis much simpler and makes it harder to manipulate assumptions to make a bank’s risk profile look better. Unfortunately, the static approach doesn’t account for product growth or divestment. That can also make it difficult to assess the accuracy of projections. For NMDs, that may mean capturing pure rate adjustments but not effectively capturing deposit attrition and movement between accounts.
Modeling NMDs for EAR
EAR projects interest expense for NMDs over some fixed time horizon, typically one year. Interest expense will vary by rate environment. The analysis starts with current rates on various deposit products and how those rates may change as market rates change. The projection makes assumptions on the sensitivity of deposit rates to market rates, how quickly those rates adjust, deposit runoff or attrition, new deposits, and movements between accounts.
Deposit rate sensitivity (beta) represents the percentage change in deposits rate for a given change in market interest rates. Changes to deposit rates tend to lag market rates, but start to catch up later, especially for larger changes in rates. Some banks use a multiple beta approach, assuming, say, a 50% beta for a +100bps change but a 90% beta for the next 100bps. Others assume a single beta, while acknowledging that rate sensitivity varies over time and by the size of the market rate change. That single beta may overstate rate sensitivity for the first 100 basis points but understate beta for the next 100 basis points.
Rate sensitivity alone won’t capture everything. Even accounts that pay little or no interest can be affected by higher rates since they can migrate to higher rate accounts. Banks also need to estimate runoff or attrition for specific account types and look at whether those customers move to higher yielding accounts or leave the bank entirely.
EAR analysis also needs to consider new business. New deposits may replace those deposits that run off. There will be net deposit outflows if these new deposits fall short and banks use wholesale borrowings to fill the gap. There will be net inflows if new deposits exceed the attrition of existing accounts.
The Allure of One Stop Shopping
The time and effort to build EAR projections, along with the multiple components of an EAR estimate for NMDs can make a one stop shopping approach attractive. Larger banks already build pre-provision net revenue (PPNR) models for stress testing purposes. These models are designed to forecast earnings under both baseline and adverse macroeconomic scenarios. It may make sense to leverage off these stress test models when estimating EAR. But a model needs to be fit for purpose for each of its uses. Stress test scenarios have historically focused on the impact of a recession and generally assumed stable or even falling rates. A model that makes sense for that scenario may work less well under sharply rising rates.
The other problem is that banks often build models based on a host of macroeconomic variables, which may or may not include interest rates. Those variables may also correlate with interest rates. That can make the impact of market rates on deposit outflows appear weaker than it is. For example, market interest rates may move together with the CPI. Including CPI in the model is likely to diminish the impact of interest rate variables on model output. Remember, however, that IRR measures reflect the impact of rising or falling nominal interest rates rather than real interest rates (i.e., after controlling for inflation).
Model developers may be tempted to not only use a single beta across rate scenarios but also to have the beta reflect some combination of rate sensitivity, movement between accounts, and attrition. Providing this sort of top of the house measure can provide a clear, if simplified picture of how the bank’s deposits and overall funding costs respond to changes in market interest rates. Unfortunately, this “beta” measure can become a black box and make it difficult to determine what is driving rate sensitivity.
The role of strategy
Rate sensitivity of NMDs can vary from product to product and from bank to bank. In addition, the rate sensitivity of deposits at the same bank can vary over time. Management’s short- and long-term strategies can play a key role. Banks looking to grow will tend to price their deposits more aggressively to meet growth targets. Conversely, those looking to shrink or even maintain current balances might price more defensively. While historical experience and model inputs should provide the starting point for rate sensitivity estimates, it may make sense to adjust those estimates upward or downward depending on growth plans. These adjustments need to follow rigorous documentation and challenge processes, whether or not they meet the bank’s definition of a “model.”
Scenario Selection for EAR
Banks base their IRR measures on specific rate scenarios. The analysis usually starts with current rates and increases and decreases of 100 basis points. Rate changes can include both parallel and nonparallel shifts in the yield curve. A parallel shift means that both short- and long-term rates change by the same amount. A nonparallel shift may include a steepening of the curve where long-term rates rise more. A flattening or inversion means short-term rates rise more. Parallel shifts are easier to calculate and may be sufficient for EVE purposes. Nonparallel rate shifts become much more critical for EAR. A bank’s NII is often much more vulnerable to a steepening or flattening of the yield curve than to an across-the-board change in rates. Short-term rates have also seen much larger swings than long-term rates. The graph below shows short-term (Federal Funds) and long-term (10-year Treasury) rates over the past 25 years.
Regulatory guidance on IRR recommends assessment of exposure under larger (300 or 400 bps) rate shocks. Most big banks only disclose ±100bps exposures in their public filings. Internal reporting usually includes larger rate shocks, but the focus remains primarily on ±100bps scenarios. Limiting the focus to moderate rate changes can lead to some serious gaps in EAR analysis. The rate sensitivity of NMDs is often not a constant but can increase for larger rate changes. In addition, short-term rates can change by well over 100 basis points over short periods. For example, the Federal Funds rate rose by 425 basis points between May 2004 and August 2006 and by 525 basis points between February 2022 and August 2023.[1]
Taking a Sensible Approach to Segmentation
The rate sensitivity of NMDs can vary from product to product and specific products can become more (or less) prominent over time. That may make it necessary to divide up, or segment NMDs by customer or product type. More segmentation is not necessarily better but should reflect differences in price sensitivity. For example, higher rate deposits, larger deposits, or products that the bank is seeking to grow can all be expected to experience more rate-sensitivity.
Outcomes Analysis and EAR
The process for making and evaluating earnings forecasts is a mature one for many banks. However, banks still struggle with identifying earnings drivers and how those drivers might change if rates go up or down. As with EVE, a model that performs well during benign periods can fall apart under more stressful conditions. With EVE, performance under benign periods matters a lot less than under stress. That’s much less clear with EAR. Benign or even baseline earnings forecasts matter a lot, as do projections under more stressful scenarios. There can be a real tradeoff between a model’s performance for both small and large changes in rates.
Models miss the mark on occasion, but the size of the miss isn’t the only concern. You also need to consider the direction of the miss and whether model forecasts systematically over- or undershoot actual results. If, for example, deposit attrition is a little higher than the forecast one month but a little lower the next, the overall estimate might be quite accurate. On the other hand, if attrition is consistently higher (or lower) than forecast and that gap grows over time, you may have more of a problem. This situation should prompt an investigation by model developers and risk managers to identify the source of the error.
Concluding Thoughts
Assumptions around NMDs play a critical role in both EVE and EAR assessments. Modeling for EAR is somewhat more tractable because EAR ties back to a widely accepted baseline (net interest income) rather than a mark to model valuation (EVE). EAR estimates still have a lot of moving parts and modeling NMDs presents a significant challenge. There is no magic elixir to make those challenges go away, but a better understanding of those limitations can lead to better informed risk assessments.
[1] The most extreme Fed Funds move was between June and January 1981, when it increased by 1005 basis points.
Leave a Reply