ViewPoint: Spotlight: Interest Rate Risk: Deposit Behavior in a Changing Economic Environment

Introduction | Spotlight: Interest Rate Risk: Deposit Behavior in a Changing Economic Environment | Spotlight: Commercial and Industrial Loans: Analyzing Cash Flow | State of the District | National Banking Trends

Interest Rate Risk: Deposit Behavior in a Changing Economic Environment

As two previous articles on interest rate risk in Financial Update's "ViewPoint" (here and here) discussed, banks have seen net interest margins compress as interest rates have fallen in response to the Great Recession. Dupont analysis reveals that the drivers of this compression are largely declining asset yields, but more recently they are also the result of lower incremental benefits from reduced interest-liability costs—as indicated by declining gains on the banks' net interest position. As net interest margins have struggled to cover operating costs, banks—depending on their degree of operational efficiencies—have reduced liability costs and maturities through increased use of non-maturity deposits (such as demand deposits, negotiable order of withdrawal [NOW] accounts, automatic transfer service [ATS] and other transaction accounts, and money market deposit accounts [MMDA] and savings) while increasing asset maturities in an attempt to increase yields to raise revenues.

We now turn our attention to some of the potential factors that could, despite management's intent, drive changes in bank balance sheets, specifically paying attention to deposit activity given the sharp rise in non-maturity deposit balances over the last five years.

Deposit trends
The charts below show how the deposit composition for banks has changed over time from two different perspectives. Chart 1 shows aggregate U.S. bank deposits from 1961 to the third quarter of 2013, with large banks with large pools of deposits primarily driving this dataset. While large banks may not adequately reflect deposit performance for all banks, the long time series helps provide some perspective on how deposit mix performs in varying economic conditions. Chart 2 shows median, quarterly U.S. commercial bank deposit composition from the first quarter of 1990 to from the third quarter of 2013. Data labels in both charts show the current ratio of each deposit category to total deposits. Finally, table 1 shows the most recent quarter's average deposit mix by banks asset size.

Changing regulations and banking landscape can make comparing current deposit performance to historical deposit performance difficult, but some key trends in the data emerge that are captured in charts 1 and 2. First, demand deposits were once the primary deposit category making up over 66 percent of all U.S. bank deposits. However, over time as the banking environment changed, demand deposits now only make up 15.17 percent of total bank deposits (16.19 percent, on average). These lower balances exist across banks of all sizes, though banks whose assets are less than $100 million, on average, report slightly higher demand deposit compositions of 18.66 percent. Similarly, NOW, ATS, and other transaction accounts have consistently made up a lower percentage of deposits both in aggregate and at the individual bank level since the Depository Institutions Deregulation and Monetary Control Act of 1980 allowed banks to offer these types of deposits nationwide. Finally, where the difference between large and small bank deposit mix is most apparent in MMDA and savings and time deposits. Large banks tend to have a much larger percentage of their deposits as savings while smaller banks tend to utilize time deposits more. The reason for these differences goes beyond the scope of this article. However, although the degree of mix is varied, both large and small banks are seeing similar trends in the growth of savings deposits and decline in time deposits since 2008.

Much of the growth in non-maturity deposits over the last five years has been attributed to depositor risk aversion—byproducts of the Great Recession, including low interest rates, stubbornly high unemployment, low loan demand and consumer deleveraging, along with others have created an environment where depositors favor the safety of their deposit, and the ability to readily access those funds, in lieu of maximizing interest earned on those deposits.

Growth in these low-cost deposits is generally a good thing for banks. Traditional liquidity measures view these types of deposits as a stable source of bank funding, and their low cost improves net interest margin performance. However, given the increase in non-maturity deposits as a result of depositor economic uncertainty, there is concern about the stability, or stickiness, of these deposits once depositors begin to feel more confident in the economy.

Deposit activity and the economy
To better understand the relationship between deposit behavior and the economy, and so to better understand the potential impact of a changing economic environment on depositor behavior, we constructed a simple econometric model. This study looked at both annual, aggregate U.S. banking deposit data and quarterly, bank-specific deposit data. The aggregate deposit data cover the years 1967 to 2012 (46 periods), where the bank data covers quarters 1990Q1 to 2013Q3 (91 periods for 4,630 banks). Because deposit balances tend to grow over time and comparing deposit growth rates can often provide misleading signals, we turn our focus to the annual growth of changes in deposit mix. For our measures of the condition of the economy, we used a selection of metrics from the Comprehensive Capital Analysis and Review (CCAR) program. After accounting for cross-correlations, our model used the following seven independent macro variables:

  • Recession during the year/quarter: This indicator is a variable that identifies, with a 1 for "Yes" and 0 for "No," whether a recession occurred during the year or quarter. (This is not a CCAR variable.)
  • Cumulative years/quarters in recession: This is the total cumulative years/quarters of a recession. (This is not a CCAR variable.)
  • Real GDP growth
  • Real disposable income growth
  • Unemployment rate
  • CPI inflation rate
  • Three-month Treasury rate: The three-month Treasury was used as a proxy for all interest rates as the correlations between varying rate types (three-month, five-year, 10-year, etc.) are quite high.

Model results: Annual, aggregate deposit performance
Charts 3 through 5 provide some key output from the annual, aggregate deposit regression. The first chart graphs the t statistic (t Stat) for each macro variable for each deposit mix multiple linear regression. We highlight a t Stat of greater than or equal to 2 and those less than or equal to –2 as an indicator for the significance of the macro variables in the regression and provide the corresponding p Value in the table below (p Values less than 5 percent are highlighted). The second chart provides the coefficient of determination (R-square) for each multiple regression. Finally, to help determine the strength of relationship between deposit mix and the economy, the third chart shows the individual correlation coefficients between each deposit mix category and each macro variable.

As we can see in the first chart, the three-month Treasury has a very strong significance in the regression for all but savings deposits. Then when we look at the three-month Treasury correlation coefficients, we also see fairly strong correlations of +45 percent for time, +23 percent for NOW/ATS, and –40 percent for demand deposits. Other economic indicators appear important as well, though to varying degrees. Unemployment is a particularly strong macro variable for time (t Stat = -6.14) and savings deposits (t Stat = 5.16) with equally strong correlation coefficients. While the correlation isn't as strong, unemployment is also a significant variable in the demand deposit regression. CPI has a very strong positive correlation with time deposits but very little significance in the regression, yet it does have a strong significance with savings. While we set our regression significance level at the 5 percent level, we could add a few additional macro variables to our significance list if we used a 10 percent significance level. For example, Cumulative Years in Recession, Real GDP, and Real Disposable Income would become significant (p Values less than 10 percent), with moderate levels of correlation performance. Finally, the R-squared for savings and time are quite good at 62 percent and 61 percent, respectively. Demand deposits also has a strong R-squared (39 percent), however NOW, ATS & Other lag the other deposit regressions with an R-squared of only 14 percent.

For savings and time, their relationship to unemployment is quite strong as is the relationship between interest rates and time and demand. It may be initially surprising to see such a weak relationship between savings deposits and interest rates, however the correlation is negative, and we suspect the low t Stat may be more of a large bank phenomenon than industry-wide one, and in fact, the percentage of banks with assets of $50 billion or more that have an MMDA & Savings/3-Month Treasury Rate t Stat of +2 or more, or –2 or less, is only 18 percent compared to 35 percent for all other banks.

Model results: Quarterly, bank-specific deposit performance
As with the previous model, we ran the same t Stat regression and correlation coefficient statistics for bank-level deposit mix performance and macroeconomic variables (we were able to calculate correlation coefficients for 6,782 banks). Again, we use a t Stat of greater than or equal to 2 and less than or equal to –2 as our indicator of variable significance. Charts 6 through 9 graph the percentage of banks, by deposit category, whose bank-level multiple linear regression model macro variable t Stats are significant. Unlike the previous study where the regression helped explain how macro variables influence overall aggregate deposit performance, the results represented in these four charts provide a better understanding of the degree to which the economy influences individual bank deposit performance.

Correlation coefficient frequency distribution charts
The 14 charts below show the distribution of individual bank deposit mix correlations to each of the seven economic variables. The first chart in each grouping provides a frequency distribution of the number of banks whose correlation coefficients correspond to the number bins on the scale from –1 to 1 (–100 percent to 100 percent). The second chart provides the same detail but as the cumulative percentage of banks. A line where the majority of it is above 50 percent (y-axis) and is to the left of zero (x-axis) indicates that the majority of banks, more than 50 percent, have a negative correlation to the macroeconomic variable. Conversely, if the line is above 50 percent and is to the right of zero, the majority of banks have a correlation that is positively correlated to the macroeconomic variable.

For demand deposit, 22 percent of banks show a significant negative relationship with interest rates. Almost 20 percent (10 percent positive and 9 percent negative) show the unemployment rate as significant indicator. These results are also seen in the aggregate regression numbers. For 32.54 percent of banks, interest rates are significant for growth in MMDA and savings. This differs from the aggregate model results, because, as previously discussed, savings depositors at large banks appear less sensitive to interest rates by themselves. Unemployment was also a significant factor for savings deposits for 22.20 percent of banks, and this too was reflected in the aggregate deposit model. For NOW, ATS & Other Transaction Accounts, unemployment is a significant economic variable for 27.62 percent of banks, as are interest rates for 23.04 percent of banks (14.73 percent negative, 8.31 positive). Although the aggregate model results do not show the unemployment rate as a significant variable in the regression (though it is close), it does for interest rates. Just as they are in aggregate, interest rates are highly significant for time deposit growth at the bank level with a very large 62.09 percent of banks showing a t Stat of +2 or more. Unemployment is also a negatively significant variable for time deposits for 24.70 percent of banks.

The correlation coefficient distribution charts provide support for the model results. Most notable are the unemployment and three-month Treasury distribution charts. Looking at the unemployment chart, time deposits are overwhelmingly skewed to the left. In fact, 89 percent of banks have a negative correlation between time deposit mix growth and interest rates with almost 60 percent showing a correlation between –100 percent and –30 percent. For non-maturity deposits as a group, 70 percent to 80 percent have a positive correlation to unemployment rates, which corresponds to the models' positive t Stat results. We again see very strong relationships between deposit mix growth and interest rates. Whereas time is negatively correlated to unemployment, it is strongly positively correlated to interest rates, with 50 percent of banks having a correlation of 40 percent or more. Conversely, for a large majority of banks, non-maturity deposit categories are negatively correlated to interest rates. This, too, supports the t Stat regression values seen in the bank-level regressions.

At both the aggregate and bank-level regressions, the relationships between deposits and interest rates and unemployment are quite strong. The connection between unemployment and interest rates is well documented since the Federal Reserve influences short-term rates as a tool to maximize employment and stabilize prices (inflation). Thus, in layman's terms, fed funds are driven lower in an attempt to lower the unemployment rate. Then when the economy has reached a level of growth and employment, interest rates rise in attempt to offset inflation. Given these relationships, it is not surprising to see deposit behavior act one way with interest rates and the opposite way with the unemployment rate. However, the correlation between the two is not always perfect since there are periods where both interest rates and unemployment are rising, and periods where both interest rates and unemployment are falling. Because of this, we believe unemployment and interest rates each provide value in explaining deposit activity.

While there are many factors that can drive depositor behavior, we do see that depositors are influenced by economic conditions. Interest rates play a particularly important role in depositor preference with time deposit growth being highly, positively sensitive to increases in interest rates, while non-maturity deposits are negatively correlated. This seems logical to us because in a rising rate environment where the economy is stable or growing and unemployment is falling, depositors will look to maximize a return on their deposits, thus time deposits will typically be the beneficiary when interest rates rise. Conversely, the growth rate of savings deposits tends to rise when the economy is slowing. We know this relationship exists on the macro level as the personal savings rate tends to rise during a recession, so it seems appropriate that savings deposits tend to increase during periods of economic stress and high unemployment.

While growth in time and savings deposits tend to have the strongest connection with interest rates and unemployment, we see from the bank-level regression results that each of the macroeconomic measures is significant for some percentage of banks. This is important to note because while deposit growth may be influenced by a particular economic measure at one bank, it may not at another. Also, because the composition of deposits always equals 100 percent, growth in exposure to one deposit category necessarily equates to a decline in another (this is why we focus on changes in deposit mix rather than changes in deposit balances). So even if one deposit category, such as time deposits, is strongly affected by changes in interest rates, while other deposit categories are limitedly influenced by other economic indicators, time deposit performance may ultimately drive the overall change in deposit mix. Thus, understanding how the economy affects one deposit category can lead to a better understanding of the performance of the overall deposit mix.

As we have shown, the economy can drive depositor behavior in a variety of ways. Certain deposit types, such as time and savings, are especially influenced by interest rate activity and unemployment. However, other economic indicators also play a role. If the growth in non-maturity deposits over the last five years is in fact due to depositor risk aversion or economic uncertainty, than it seems reasonable that once this uncertainty dissipates, depositors will inevitably change their behavior. Given the historical relationships we have just discussed, as we ultimately move from a low-rate environment with historically high unemployment to an increasing-rate environment with normalizing unemployment, the data tell us that we can expect a shift in depositor preferences toward higher-yielding deposit types, such as time deposits, and away from lower-yielding deposits, such as savings, demand and other transaction accounts. Of course, competition for deposits can offset the impact of the economy on depositors. Additionally, the regulatory and business landscape will also often mitigate economic influence over deposit activity. As a recent example, the repeal of Regulation Q in 2010, which formerly prohibited the paying of interest on demand deposits, may change the way demand deposits perform moving forward regardless of interest rates or unemployment.

The increase in non-maturity deposits since 2008 has led to an improvement in the banks' cost of funds and a reduction in their reliance on wholesale funding. These deposits provide a low-cost, seemingly stable source of deposits that will be available to lend once loan demand returns. Just how stable these deposits will be moving forward remains to be seen. Certainly some banks will have a competitive advantage in the markets they serve and will be immune from non-maturity deposit withdrawals once interest rates rise and the economy improves. However, this may be the exception and not the norm. Because of this, as part of their asset/liability management, banks should consider scenarios where unanticipated deposit withdrawals occur and assess the potential impact to earnings, liquidity and capital. While the rise in non-maturity deposit volumes may be new to a bank and the behavior of those depositors not completely known, bank management can look to the historical relationships between the economy and depositor behavior to complement their current analysis. Even if these events fail to materialize, bank management will at least have a better understanding of some of the extraneous drivers of deposit activity.

This article was written by Dean Anderson, a senior technical expert in the Atlanta Fed's supervision and regulation division.