Indeterminate Term Deposits in FTP: Quantified Solutions at Last! William J. McGuire, Ph D. 1. INTRODUCTION Perhaps the most contentious input in Funds Transfer Pricing (FTP) applications is the term of the crediting rate for indeterminate maturity deposits. Treasury and Retail alike are frustrated by the lack of quantified evidence on effective term and often settle for best guesses or crude approximations. No one is happy, however, and it often shows. This is a situation that no longer needs to happen. Advanced statistical approaches are now capable of assigning accurate effective terms and defining other behaviors for all types of indeterminate maturity deposits, based on comprehensive examinations of historic data. Forecasts underlying predicted terms and other values have been extensively back tested, in some case over very long periods of time, with results that confirm their accuracy. It is anticipated that solving the indeterminate maturity deposits term input impasse will bridge much of the chasm that often separates Treasury and Retail with regards to FTP. An extension of the statistical methodology provides a way to bring Treasury and Retail even closer together. Simple add-on analyses to the statistical analyses define a basis for assigning account level credits for attributes shown to be correlated with expected term. This allows Treasury to "micro-tune" indeterminate maturity deposit crediting rates, fulfilling a long-standing Retail wish. 14
This article is organized as follows. A brief overview of indeterminate maturity deposit behavior theory is presented first. This is followed by an overview of the statistical methodology that can quantitatively assess their term and other behaviors. A discussion of several types of specific adjustments to crediting rate term follows that. 2. WHY INDETERMINATE MATURITY DEPOSITS HISTORICALLY POSED A PROBLEM Contract terms for indeterminate maturity deposits normally bear little if any relationship with the actual behaviors of these funds. While contractually they appear to be short term and variable rate (if interest bearing), effectively they often behave like long term, semi-fixed rate funding. This divergence of behaviors is a source of much misunderstanding with regards to these deposits and the depositors they represent. In fact, the behaviors of indeterminate maturity deposits reflect rational decision making on the part of depositors 1. Two depositor influences drive indeterminate maturity deposit behaviors, as illustrated in Exhibit 1. The first is the finance influence, traditionally the only focus of financial managers. This influence works directly through own rate paid, competitor rates paid and fees, and indirectly through barriers to exit that raise the cost of closing an account (e.g. ACH transfers, PC bill paying, successful cross sell, etc.). Higher own rate paid relative to competitor rates and more consequential barriers to exit create longer term indeterminate maturity deposits. 1. For more discussion of this behavior model, see Maximizing Performance and Value of the Core-Deposit Franchise, by William J. McGuire, Bank Accounting and Finance, Winter 2001-2002. INDETERMINATE TERM DEPOSITS IN FTP: QUANTIFIED SOLUTIONS AT LAST! 15
But a second, and often more potent, influence also affects depositor behavior. This is the ambiguous influence. The influence works through service, convenience, products, and other dimensions of the depositor's overall relationship with the institution. The empirical record (see below) suggests that higher levels of service and convenience, products that are fine tuned to depositors needs, a less mobile customer base, a strong economic base, and an institutional commitment to a culture of deposit retention are associated with longer term indeterminate maturity deposits The ambiguous influence is the answer to a long asked question about indeterminate term deposits: Why do balances remain on deposit for long periods of time even though the financial advantage is negative? The answer is that depositors are taking some or all of their compensation in the form of service, convenience, or other non-financial benefits. Where such non-financial motivators are strong, for example in transaction-oriented categories or simple savings-oriented categories, the ambiguous influence can, and in fact and often does, dominate. This dominance (i.e. the relative weakness of financially-driven motivators) creates the long effective terms often observed for indeterminate maturity deposits despite low (or zero) rates paid and typically very limited (or no) repricing as interest rates change over time. The problem, however, is that the comparative strengths of financial versus ambiguous influences cannot be inferred from the deposit contract. Nor are they readily ascertained by simply observing current depositors, who are operating in point in time economic and interest rate environments. But a solution exists. Historic data on balances and rates paid by category provide insights into depositor motivations because they are a record of the "revealed preferences" of depositors. That is, changes in 16
depositor behaviors when interest rates and rates paid change reveal the historical effects of the finance influence. The strength of the ambiguous influence can be quantitatively inferred. 3. A MEASUREMENT SOLUTION OF INDETERMINATE MATURITY DEPOSIT TERMS To extract behavior information from the historic record, category level historic data on balances and rates paid must be statistically analyzed at a high level of sophistication. A methodology 2 developed over the last decade holds the answer. This approach applies advanced simultaneous equations time series econometric tools to quantitatively capture category and cross-category behavior determinants present in the historic data. Forecasts produced from institution-specific applications of this advanced methodology are comprehensive (e.g., behaviors are projected by time period and interest rate scenario at the category level) and they have proven to be accurate predictors of actual future term and repricing related behaviors (see below). Terms for indeterminate maturity deposits derived from retention forecasts produced by advanced statistical methodology assessments are often longer than typical rule of thumb values or regulatory inputs. The estimated values, however, accurately represent recent institution history and thus they are the appropriate quantitative basis for the term (or terms) that set indeterminate maturity deposit crediting rates. 2. See A New Approach to Evaluating Core Deposit Behavior and Value, by William J. McGuire, Bank Asset/Liability Management, June, 1996 for an early exposition of the methodology. A more recent review is presented in US patent application 20030069147. This is available from www.uspto.gove/patft. Click on Publication Number Search and type in the application number. INDETERMINATE TERM DEPOSITS IN FTP: QUANTIFIED SOLUTIONS AT LAST! 17
Exhibit 2 provides an example of the term related information that is available from advanced methodology statistical assessments of indeterminate maturity deposit data. The statistically estimated equation for retention in this case is forecasting the run off of existing balances over approximately a 17.5 year period. The average life (run off balance in each period weighted by the period mid-point) at Base Case (current interest rates) is 5.49 years. This is the appropriate crediting rate term for this deposit category. Note, however, that the retention/run off pattern is estimated to vary as interest rates change; average lives are 4.35 and 7.08 years in the +100 bp and -100 bp rate shock scenarios, respectively. This is an example of embedded option effects in indeterminate maturity deposits. The value of this option should in theory be charged back to Retail or otherwise accounted for in the crediting rate specifications for this deposit category. Note that the exact dollar run off balances by period are available to support tranche or period specific crediting rate determination, and that the repricing behaviors of the category (period specific changes in rate paid per 100 bp interest rate shock change) are also forecast. Interest expense effects for the category can thus be quantitatively incorporated into FTP treatments also. Finally, an effective duration measure can be calculated if desired. This combines term and repricing effects into one crediting metric. Duration values, even if not utilized to set crediting term, are useful for their easy comparability to asset side characteristics (Base Case duration in the example is 4.15). Advanced statistical analyses of historic indeterminate maturity deposit retention produce forecasts that generally back test very well against subsequent retention behaviors. This accuracy has been verified in a number 18
of back tests over varying time periods. Forecast accuracy can be judged qualitatively by reviewing graphical presentations or relatively judged using simple tests, such as Root Mean Square Error (RMSE) 3 comparisons. Exhibit 3 provides an example back test of statistical retention forecast data. The forecast of future retention was produced as of 12/31/95, based on historic data from 12/31/88 to the forecast date. The statistically based, institution-specific retention ratio forecasts track very closely on average to actuals for almost 8 years, using the -300 bp rate shock scenario run (the closest forecast to the actual subsequent track of interest rates). The institution-specific retention ratio forecasts also outperform the types of regulatory inputs often used as term approximations in FTP applications (see the RMSE test results). With term forecast values from an advanced statistical methodology study in hand, there is no longer any "in" in indeterminate maturity deposits. Thus a key area of contention between Treasury and Retail is resolved. But there may be some adjustments required to ensure that Retail crediting properly reflects all term dimensions. 4. MACRO LEVEL ADJUSTMENTS TO INDETERMINATE MATURITY DEPOSIT TERMS At any given point in time, balances within a category of indeterminate maturity deposits are not likely to all be equal with regards to their expect 3. Root Mean Square Error (RMSE) is a standard statistic for establishing comparative forecast accuracy. It is calculated as the square root of the summed squared difference between each period s forecast value and that period s actual retention value, over the time series reviewed. Lower RMSE values indicate a closer match between forecasted and subsequent actual values, i.e. that projections from a forecast are more accurate in their description of future retention behaviors. INDETERMINATE TERM DEPOSITS IN FTP: QUANTIFIED SOLUTIONS AT LAST! 19
ed future term behaviors. Some balances may be seasonal in nature, relat ed to day of the week effects (e.g. DDA balances are usually higher on Fridays and on mid- and end of month pay dates), or have other short term orientations. Statistically estimated long terms are normally only credited to that portion of balances that indeed do represent long term, stable funds. A recent supply phenomena of indeterminate maturity deposit markets warrant special attention in this regard. Since approximately first quarter 2001, growth in many types of deposits has sharply increased compared to prior long-term trend growth rates 4. The substantial "surge balances" now accumulated on deposit are of uncertain motivation, and they should be segregated from balances that do in fact represent long term, stable funds. Surge balances should be credited as shorter term, and likely faster to reprice, balances compared to long-term balances (which are accorded statistically estimated terms). Finally, in many institutions, term crediting points for indeterminate maturity deposits are constrained by Treasury limits placed on allowable asset durations. This truncates allowable crediting, creating a serious source of frustration for Retail. One solution that can satisfy both Treasury and Retail is to extend assets to match indeterminate maturity deposit terms and hedge perceived interest rate risk exposures on or off balance sheet. Assuming that underlying deposit behaviors are statistically shown to be fundamentally stable, the hedge position can be partial (e.g. hedge less 4. See Adjusting Core De3posit Balances for Surge Growth, internal MPS research memo of 11/24/03. Copies may be requested by addressing a request to info@mpsaz.com. 20
than the full balance of the asset deployment and/or remain out of the money until interest rates rise substantially). By applying a partial hedge, supported by known and monitored indeterminate maturity deposit behaviors, a net yield pick up is maintained while interest rate risk is controlled. 5. MICRO LEVEL ADJUSTMENTS TO INDETERMINATE MATURITY DEPOSIT TERMS Extensions of several recent statistical studies of indeterminate maturity deposits points the way to a new dimension of term crediting adjustments. This is to analyze deposit retention behaviors by attribute within a category and extend or shorten crediting terms as appropriate. In concert with the setting of term crediting points as discussed above, it holds the potential for fully bridging any chasm between Treasury and Retail. The process is straightforward. Advanced methodology statistical processes are used to establish overall category term. Underlying depositor data is then sorted by attribute (also known as segmentation characteristics). By comparing the retention behavior of an attribute to that of the category as a whole, a crediting adjustment is derived (stronger than category retention extends the crediting term and vice versa). A simple ratio can be defined by computing the ratio of attribute-to-category retention. Quantitatively based ratios could also be derived, by statistical estimations of subsets of the data by attribute. Exhibit 4 presents a simple example. Average life for the whole category is statistically estimated to be 5.49 years. Retention ratio patterns for ACH linked accounts versus non-ach linked accounts lie (on average) INDETERMINATE TERM DEPOSITS IN FTP: QUANTIFIED SOLUTIONS AT LAST! 21
10% above and 15% below the category overall. A simple term credit adjustment equal to the two outcomes would be applied to reflect the different marginal effects of accounts with each attribute. In this case, accounts with ACH transfer linkages would be credited at 6.04 years while accounts without any ACH transfer linkages would be credited at 4.67 years. Micro-level adjustments for indeterminate maturity deposit crediting terms are still in the exploratory stage. The early empirical record suggests that attributes such as younger depositor age, less developed market share, and larger balances are often correlated with lower retention and shorter terms. Conversely, barriers to exit (e.g. ACH, PC bill paying, overdraft protection), additional (cross sell) relationships, and time as a customer are related with greater retention and longer terms. Geography has an influence, but it varies. The advantages of micro-level (account or depositor attribute based) adjustments are that they let Treasury recognize the specificity of deposits that Retail knows is there from field experience. Both sides can be satisfied that both fundamental behaviors (e.g. term) and ancillary influences in behavior are properly acknowledged and credited. 22
6. SUMMARY AND CONCLUSION Advanced methodology statistical analyses can now quantitatively set the crediting term for indeterminate maturity deposits with high levels of accuracy and confidence. Special crediting adjustments can further account for macro (category level) and micro (account or depositor) based differentials in crediting terms. The "in" in indeterminate maturity deposits is thus no more, and the chasm between Treasury and Retail can be fully bridged. EXHIBIT 1 Dual Influences on Indeterminate Maturity Deposit Behavior Institution Rates Paid Local Reference Rates Selected Deposit Category Behavior National Reference Rates Depositor Characteristics Service/Convenience Behaviors of Others Deposit Categories Marketing Efforts Economic Conditions INDETERMINATE TERM DEPOSITS IN FTP: QUANTIFIED SOLUTIONS AT LAST! 23
EXHIBIT 2 Statistical Estimates of Term and Related Metrics Rate Scenario Average Life Rate Paid Present Value Eff Duration +100 bp 4.35 0.50% 847,503-3.76 Base Case 5.49 0.25% 880,623-4.15-100 bp 7.08 0.10% 920,667-4.54 Notes: Effective durations are based on present values; reverse sign for a liability. Base Case effective duration is an average of the +/-100 bp values All present values are expressed per $1,000,000 of existing book value. 24
EXHIBIT 3 Back Test Proof of Advanced Methodology Forecast Accuracy Data Type RMSE Value Institution Specific 0.1192 OTS -300 bp Approximation 0.6397 OCC Optimistic Approximation 0.5987 Notes: Institution specific analysis produced using advanced statistical methodology. OTS approximation and institution specific forecast are for -300 bp scenario OCC is the agency's optimistic (longest average life) approximation. INDETERMINATE TERM DEPOSITS IN FTP: QUANTIFIED SOLUTIONS AT LAST! 25
EXHIBIT 4 Determining Attribute-Based Micro-Level Term Adjustments Notes: Retention of balances in accounts with ACH transfer linkages is 10% higher. Retention of balances in accounts without ACH transfer linkages is 15% lower. Percentage comparisons above are to overall category retention levels. 26