The first part of this series examined issues you need to consider before you approach deposit repricing modeling, e.g. what distinctions in your deposits base are important for your repricing. Those factors likely include differences in deposit types, such as interest checking negotiable orders of withdrawal (NOW) vs. money market deposit accounts (MMDAs), and may also include differences based on your depositor behavior or your competitors’ behaviors. The second part focused on the data required for a state-of-the-art repricing model: a history of your institution’s total balances, rates paid and balances retained in a given set of accounts. With the data gathered, we can move to the modeling process itself.
The Asset Liability Management (ALM) model repricing process should examine the relationship between the current rates paid and their typical reference rates as well as the expected impact of that repricing on the different categories of deposits. For example, further Federal Reserve rate increases would be expected to generate larger changes in Certificates of Deposit (CD) rates than MMDA rates, which in turn would move more than NOW rates. This process would be expected to eventually return pricing spreads to long-term equilibrium levels. But when constructing the repricing model, it is critical to be aware of factors that would impact the estimated betas and potentially generate different betas and different equilibrium spreads. This process implies a conservative approach to modeling and implementing rate changes when current conditions do not reflect typical historical values. Lags in the rate adjustment process are a common feature at all times but are even more important when other factors may have an outsized impact on market rates. Currently, $1.5 trillion in excess reserves may be the most important additional factor you should consider, but expected inflation, greater loan demand, changing competitors’ behavior and technological advances are other potentially important influences.
Given the history of deposit and reference rates, a model will estimate the relationship between the two, producing a beta coefficient for the sensitivity of a deposit rate to its reference rate. A well-structured model will incorporate a lag structure for a slow or gradual response of a deposit rate to its reference rate; it will distinguish between different deposit rates for different deposit classes and potentially based on different reference rates; and it will consider the implications for other factors that could potentially impact the relationship between a deposit rate and its reference rate.
Quantifying historic core deposit repricing can be undertaken with different levels of precision. At one extreme, one could look at the average deposit rate across all deposit classes levels and relate that to an overall market rate. Alternately, one could analyze deposit rates by deposit class and relate them to separate reference rates. More detail in the quantification, at least to a point, will increase the accuracy of repricing inputs but will also require a greater resource commitment. While the sensitivity of your rates to market rates is often the central question of deposit rate repricing, the most important factor impacting repricing decisions should be the sensitivity of your deposits to your repricing. The precision employed – and the resulting complexity of the model – should depend on your cost of acquiring and retaining deposits compared to the cost of undertaking more precise modeling. The payoff from more precise pricing efforts will depend on many factors, including the size of the institution, the variability of rates, and the sensitivity of deposits to repricing. When rates were low and near constant from 2010 through 2015, analysis of repricing would not have been as critical as it would be in the current period of rising rates. When market rates are changing, however, the cost of a pricing mistake may be orders of magnitude greater than the additional modeling costs.
Best practice in modeling does two things. First, it employs your institution’s historical data to create a model that relates your deposit rates to lags of general market rates and potentially to competitors’ rates. Second, it employs your institution’s historical data to relate both your total and retained balances to your lagged deposit rates and market rates. The inclusion of lagged variables is critical to model development. When you and your depositors are making decisions, you on rates and them on deposits, both decisions will be informed by the history of rates and deposits and implicitly on their expected future behavior. The use of lags is also critical because in virtually all cases the adjustment process takes place gradually, and the inclusion of lags makes this gradual adjustment process an integral part of the model development.
To summarize, the model should include rates for different types of deposits, total and retained balances for those deposits, and other variables that have been previously identified based on the specifics of your institution. Once this task has been completed, the model is estimated using regression for example. The estimated regressions will yield parameters for your institution detailing how sensitive your rates are to market rate changes and how sensitive your deposits are to your rate and market rate changes.
Once the parameters of your model have been estimated, the process switches to using the model to forecast deposit rates and balances under alternate rate scenarios. In current market conditions, it would be prudent to consider a base case scenario with no rate changes assumed plus alternate scenarios projecting the implications with a potential market rate increase. Exhibits 6 and 7 present base case forecasts that assume no rate changes as well as three possible rate shocks, +100 basis points (bps), +200 bps and +300 bps. Exhibit 6 considers the changes for NOW accounts and retained balances while Exhibit 7 presents the same information for high-tier MMDAs. The forecasts in both Exhibits are based on an immediate shock to make the implications clearer in graphical form. You should also consider the implications of a gradual rate change, and for even greater realism may want to include a customized set of market rate change forecasts.
For NOW accounts, a +100 bps shock leads to an almost imperceptible increase for the first few months and only a 2 bps increase in the NOW rate even after 24 months. The adjustment process continues beyond that point leading to a 4 bps increase in the NOW rate in the long run. While this beta appears tiny and may be criticized as unrealistic, two question should put this in context. First, how much have you changed your NOW rate in the past three years as the 1-year Treasury rate has increased by about 240 bps? For most institutions, the answer is “very little, if at all.” Second, how sensitive are your checking deposit balances to rates? If they are relatively insensitive to rate changes, you would have little incentive to change rates.
Exhibit 6 also indicates that there will be an increase in the NOW rate even in the base case scenario; even with no market rate increase the historical data indicates that this institution would typically increase its rate to bring it back to the historical spread. If management views that result as inconsistent with their likely behavior, then a management override would be in order, with the base a flat line and the other rates adjusted appropriately.
The second panel in Exhibit 6 indicates that retention of balances is not substantially impacted by increasing rates. That +100 bps rate change is forecast to reduce retained balances by only 1% after one year and by 1.4% after two years. Depending on the demand for loans and the rest of the deposit supply balance sheet, this relative insensitivity of NOW accounts may provide justification for the lack of any change in NOW rates even as market rates increase.
The retained balance component of Exhibit 6 also indicates an occasional major dip. This institution experiences a major but temporary drop in checking balances in September. The model incorporates other events so that the impact of rates on deposits can be distinguished from other factors, and we can see the impact of a September rate increase as distinct from this institution’s annual September dip.
Exhibit 7 focuses on high-tier MMDAs. The results here differ dramatically from those for NOW accounts and indicate the importance of distinguishing the impact of market rate changes across different types of deposits. The top of Exhibit 7 indicates that a +100 bps shock will increase the MMDA rate by about 25 bps after one year, 40 bps after two years, and 75 bps in the long run. The adjustment process again takes a prolonged period to be complete, and again the base case projects a rise in the MMDA rate, since the current spread between the 1-year Treasury and the institution’s MMDA rate is well below the historical spread.
For MMDAs, the forecast indicates that retained balances run off more rapidly that NOW balances. After two years, about 92% of NOW balances are forecast to remain in the institution in the absence of any rate changes. In contrast, for MMDA balances only 85% are forecast to remain in the institution. Equally important, the rate sensitivity is much greater for MMDAs. While a +100 rate change would lead to a 1% decline in NOW balances after one year and a 1.4% decline after two years, for MMDAs the runoff would be 7.2% after one year and 8.3% after two years. Once again, for MMDAs there is some seasonality in the data.
None of these results should be surprising, but they do form the basic inputs for core deposit pricing decisions. In addition, they present strategy suggestions that potentially have major impacts on your repricing behavior and ultimately on your earnings. With this modeling approach, you can ask if your repricing strategy would change if your one-year runoff was only half of that estimated here. Or if you repriced more aggressively, would your runoff rate drop? Or is there some non-rate feature that can be altered to reduce the runoff even in the face of a rate increase? The modeling approach as presented focuses on the rate change question. Non-rate change factors can also be included in the analysis to the extent that they are relevant. A model may not lead you to answers to all those questions, but it should lead you to ask better questions.
In the last installment, we will discuss how to use the results of your forecasting model.
Thank you for choosing the Situs Newswatch. If you want to see your company here or have an idea for coverage, please respond to this email or email firstname.lastname@example.org for more information.