Customer Price Elasticity – The Deposit Pricing ‘Holy Grail’?

By Tom Farin

Deposit pricing is a hot topic for financial institutions. Deposit pricing software vendors are literally coming out of the woodwork making all kinds of claims. One of the most enticing claims is that they can use linear and differential equations to predict the effect of your pricing actions on customer demand. Some of these claims remind me of the divining rod days of the 1800s when dowsers moved from farm to farm claiming they could locate water on the property holding a Y shaped willow stick.

One vendor claims to be able to predict customer response to pricing actions at the customer level. Another vendor claims to be able to forecast trends in customer supply based on interest rates, rates paid, fees, and marketing events. A third vendor promoting an elasticity solution even goes as far as to attack marginal cost calculations. The attack focuses on the price/demand assumptions that are a necessary part of marginal cost calculations. The vendor’s claim is that unless you possess his price elasticity model, your marginal cost calculations will be garbage.

Am I attacking the ability of statistical and mathematical methods to predict the effect of pricing actions on demand? From a theoretical standpoint, no. From a practical standpoint, yes, for three reasons. (1) I don’t believe any vendor currently offering elasticity solutions is gathering sufficient data on the factors that affect price elasticity to accurately predict the effect of pricing actions on demand. Read on and you’ll understand why I feel that way. (2) Even if they did, the data requirements placed on their customers would be outside the scope of data currently maintained on their core and MCIF systems and as a result would be so burdensome to small to medium sized institutions as to make it economically and operationally impractical to keep the data current. (3) Because of the amount of historical data to be tracked for price/demand conclusions to be statistically valid could span multiple years, institutions purchasing these products won’t know whether they validly predict price/demand relationships until well after they buy the products.

If your response to (3) is “All we need to do is backload historical data when the software is purchased so we can find out quickly whether price/demand predictions are valid”, please re-review points (1) and (2). Most shops lack sufficient historical data to load a three, four, or five factor elasticity model.

Unfortunately, even were the data available, no three, four, or five factor elasticity model will accurately forecast the effect of pricing actions on demand – for two reasons. (1) These models only monitor a small percentage of the factors that affect how customers respond to your pricing actions, and (2) even if the number of factors was expanded to a sufficient number to consider the major factors, the data gathering burden would further overwhelm the resources of a small or medium financial institution.

Factors Affecting the Relationship between Pricing Decisions and Demand

We’ve identified six major categories of factors affecting price/demand relationships on deposit products.

Institution and product factors
Competitive factors – direct banking competitors
Competitive factors – indirect banking competitors
Competitive factors – non-banking competitors offering substitutable products
Economic factors
Customer factors

Each category has multiple individual factors that fall into that category. Appendix A lists and discusses 31 factors organized into the above categories that affect customer pricing response.

Some factors like the customer response at various points of the economic cycle, might require a three to five year time series to evaluate. Others like the changing demographics of your customer base, competitors entering or leaving your market, the introduction of disruptive products, and how key competitors react to your pricing actions, don’t cooperate by remaining constant while the long time series needed to track the effect of the economic cycle unwind.

Some factors are extremely difficult to quantify. Maybe the most important is how the culture in your lobby staff responds to your attempts to segment of your customer base. Should your CSRs and tellers be poorly trained or left out of the communication loop, they can destroy any segmentation strategy hatched by management.

A good example is a phone call I recently received from a customer. He had seen only 20% of maturing CDs automatically renew as compared to moving into specials. I had recommended checking to see whether his CSRs were calling renewing CD customers. He found they were. He instructed them to discontinue the practice. Automatically renewing CDs jumped to 40%.

A recent BAI study on customer pricing response found that actions taken on the banking floor by CSRs and tellers is one of the two or three most important factors. How will an elasticity equation model that important behavioral input? How would the model even quantify CSR actions like working renewing CD lists as inputs into the equation?

Another factor that doesn’t remain constant over time is changes in how much customers value liquidity (checking, savings, MMDA) over time deposits. During times of economic downturns the population generally has a stronger liquidity preference. Changes in your customer demographics can affect elasticity preference as some demographic groups have a much stronger elasticity preference than others.

In recent years, institutions have begun to use increasingly sophisticated segmentation strategies – CD specials, premium money market accounts, Rewards Checking, geographic pricing, geographic target marketing, tiered barriers to entry, channel barriers to entry, transactional barriers to entry, new money promotions, relationship pricing – the list goes on and on. Some of these products can be very disruptive to demand for existing products. All of these tactics affect new money raised as well as cannibalization from existing accounts. Institutions continue to experiment and change their product mixes. They develop products aimed at specific constituencies like Hispanics, young affluent adults, seniors, people who live from paycheck to paycheck, etc. None of the historical data will predict how introduction of a new sophisticated segmentation strategy will affect demand.

The bottom line is that customer response to your pricing action is driven by a large number of related and unrelated factors. Data collection measuring these inputs is often non-existent. Where the inputs may exist, gathering the data is often problematic and resource intensive. The interaction between many of these factors is extremely difficult to monitor and synthesize into equations as changes in too many of these factors are happening simultaneously. Without effective monitoring, it will continue to be impossible to discover and quantify how they relate.

Dealing With Complex Interactions Between Factors

A well known problem associated with making forecasts with highly complex interactions between data inputs is weather forecasting. Most of us are aware that current weather forecasting systems do a reasonably accurate job of forecasting the weather for the next 24 hours - most of the time. But extend that forecasting horizon to a week or a month, and the forecasts become significantly less accurate.

Edward Lorenz, a professor at the MIT, was working on a complex weather forecasting model in the 1961. He noticed that a very small change in inputs (use of rounded vs. unrounded numbers) on what he thought were trivial factors caused a very significant change in results. Out of this experimentation came the “Butterfly Effect” analogy. A butterfly flapping its wings in the air currents in Brazil could through a complex set of interactions with air currents cause a tornado in Texas. Of course, butterflys don’t cause tornados. But Lorenz’s studies point out that factors causing significant variances in outputs can be caused by changes in inputs that are not well understood.

Lorenz’s publications on the subject lead to the popularization of Chaos Theory. There are a number of definitions of Chaos Theory. The one I like the best is, "The theory that some systems, such as weather, are ultimately unpredictable because of the effects of small scale events that can't be included in the prediction equations." Gee, could he have been thinking about how a tiny little DeNovo bank entering your market by starting a rate war can really screw up the market?

A good layman’s book on the subject is Ian Stewart’s, “Does God Play Dice?” Stewart’s definition is also useful. “Chaos occurs when a deterministic (that is, non-random) system behaves in an apparently random manner.” Hmmm, could he be talking about what 9/11 or the NASDAQ crash did to balances in MMDA, checking and savings accounts?? “I’ve been dropping deposits rates like crazy, but the money keeps running in the door.”

Of course, the motion picture industry has further popularized the theory with a number of films, the best known of which have been the recently released film, “Chaos Theory”, and the movie “Jurassic Park.” Do you remember the wise guy mathematician who predicted that humans who had never seen a living dinosaur couldn’t effectively design containment systems for dinosaurs hatched from dinosaur DNA code patched with frog DNA? Do you remember whether he was right or wrong? One of his most memorable quotes was, “Life finds a way.” He could as easily have said, “CSRs will find a way.”

Chaos theory has been successfully applied to a number of physical sciences and social sciences. With customer response to pricing actions, we have a combination of physical and social science. With the number of inputs (small scale events) involved, chaos theory is much more applicable to measuring customer pricing response than any of the currently available linear and differential equation based elasticity tools or data systems.

Elasticity equations focus on a limited number of relationships between numbers – your rates, competitive rates, market rates and conditions, etc. No model I’ve seen to date incorporates social factors like the CSR interaction with customers, changing customer preferences for liquidity, the role of customer confidence in the economy and other social factors much more difficult to quantify, measure and monitor.

I’m not saying there is anything wrong with the algebra or calculus in these models. As Stewart puts it, “If you set up the wrong differential equations, when you model the motion of the moon, no matter how correctly you apply calculus, you’ll get nonsense.” We don’t have the monitoring systems and the understanding of the relationships needed to construct accurate mathematical models that predict the effect of pricing actions on demand. As a result, those models monitoring a limited number of factors are highly likely to produce nonsense.

Dealing With Uncertainty

So are we totally helpless when dealing with a complex system like weather or how a customer will respond to your pricing action? Actually, the best computer in existence for dealing with a wide variety of complex inputs and formulating the appropriate response is still the human mind.

Think about your personal life or your business career. I can think of a number of small scale events that have had major effects on either my personal or business life - a flirtatious glance from a member of the opposite sex, a chance demo of an Apple II computer and Visicalc, another human being in the wrong place at the wrong time in a traffic intersection, the college class I took as a lark that changed my career path. Yet as humans we assimilate all these complex inputs, small scale and large scale, and somehow successfully navigate our lives and our professional careers. Can we predict an individual’s life or career path in advance based on a limited number of inputs – DNA, IQ, parenting style, religious belief?

I also believe that a well trained pricing committee or ALCO committee can do a much better job predicting the effect of pricing actions on demand than any regression or differential equation based model. That’s because the mind can deal with all the inputs (small scale and large). The mind may be at a disadvantage relative to a computer in running accurate numbers based on known mathematical relationships. But it is much better at fuzzy logic, when all the mathematical data isn’t there and the relationships not well defined.

Beyond that, we humans can take actions to manage some of the factors that are key inputs. Remember how a manager, eliminating the use of renewing CD lists by CSRs, caused automatically renewing CDs to jump from 20% to 40%?

Education, Training, Software, and Consulting

The education issue is crucial. That’s why we offer Deposit Boot Camp and Advanced Deposit Boot Camp. In those Webinars the entire team can participate. We teach how to develop strategies. We make teams aware of the complex array of factors that affect customer response and how to manage as many of those factors as possible. We teach the importance of tracking systems and certain key analysis tools like marginal cost.

Most importantly, we teach how to modify customer pricing response with targeted products, segmentation tools, barriers to entry, lobby personnel training and management, and how to deal with uncertain response. We also provide the tracking systems, and analysis tools. Finally we facilitate the first few strategic reviews to keep the management team on point and to insure the implementation issues are discussed and resolved.

Dealing with Demand Response that is Hard to Predict

As is true with weather forecasting, it is crucial to keep planning horizons short in accurately evaluating and managing potential demand response to pricing actions. We limit our planning horizons to 90 days. The shorter the horizon, the less likely a number of the 31 factors in Appendix A that affect customer response to pricing actions will vary significantly. Demographic changes, economic cycles, competitor pricing response, customer liquidity responses, and actions taken by CSRs and tellers on the banking floor are less likely to change in 90 days than over the long historical time series needed to make elasticity models statistically valid.

But even in a 90 day period, it is difficult to predict the effect of pricing changes on demand. There are just too many butterflies flapping their wings. That’s why we have always encouraged management teams to perform sensitivity testing. If you can’t reliably predict how customers will respond to pricing actions, then test the strategy across a reasonable range of demand responses. How good do you feel about the results of the strategy across the range? We teach you how to perform sensitivity testing in boot camp and iPrice supports sensitivity testing.

Of course, it is very helpful to be able to look back over the effect of the same strategy on demand in the previous quarter or two. While elasticity models may have difficulty predicting the effect of pricing actions on demand for reasons discussed earlier, tracking past relationships puts you in a better position to predict the range of potential responses than having no historical data at all.

Finally when the end of the planning horizon is reached, it is important to back test. Plug in what actually happened where previously there were just assumptions and back test what it actually cost you to raise the funds. If the actual response is significantly outside what was predicted, gather your management team and brainstorm the potential explanations for the response you saw. Then if the factor is controllable, take actions to control the factor. Because we work with our customers on those quarterly reviews, we have come to realize just how insightful a good pricing or ALCO committee can be. These are all out takes from these reviews.

“Of course raising our CD rates had no effect on demand. Our major competitor matches our rates within 24 hours.” - Maybe we need to consider using stealth products.

“Of course all of our CD customers appear to be rate sensitive. Our CSRs are calling all of our renewing CD customers and selling them into the special.” – Maybe the CSRs need training.

“No, we weren’t successful marketing our 5.01% Rewards Checking accounts to young adults. We marketed in the newspaper. The marketing research people tell us they don’t read the newspaper. They get their news on the Web.” – Maybe we need to institute Google point and click marketing and upgrade our Web site.

“Yes we are seeing a big movement from CDs into checking by some of our older customers. Wouldn’t you move if you had a choice of 5.01% on checking as opposed to 3.5% on a 12 month CD?” – Maybe our checking product needs higher barriers to entry to reduce cannibalization.

“Of course we’re seeing big inflows into CDs. Local CD rates are 100 bp above Treasuries. Our customers are moving funds from Treasury mutual funds into CDs.” – Maybe we can reduce rates to 75bp over Treasuries and still get the CD supply we’re looking for.

Put the knowledge of the pricing committee together with good price/demand trend data, and tools like marginal cost. You’ll have a much better environment for making effective pricing decisions than plugging three to five factors into an elasticity model that purports to know everything you need to know to make effective pricing decisions, and makes them for you. Sorry, I’ll take the well trained pricing committee, thank you.

Of course we can only dream of the day when a vendor announces his breakthrough ‘Chaos Theory Elasticity Model’. I can just hear the hype now. “We monitor all of Farin’s 31 inputs required to predict the effect of pricing actions on demand and tell you the optimal rate to charge.” Are you ready to order the blood pressure cuffs and brain scan devices you will need to monitor the behavior of your CSRs and provide appropriate electronic inputs to your “Chaos Theory Elasticity Model?” You’ll need them!!

I hope by then I’ll be retired, laying in the sun, and drinking lemonade.

Appendix A – Factors Affecting Customer Response to Pricing Actions

Institution and Product Factors

Relationship between rates for product being repriced and offering rates on the institution’s other products that completely substitute for the product being repriced. A good example is the relationship between an institution’s pricing on regular CDs and similar term CD specials.
Relationship between rates for product being repriced and offering rates on the institution’s other products that partially substitute for the product being repriced. In rising market rate environments, institutions tend to raise rates on CDs significantly and non-maturity deposits more slowly. As the differential between the accounts increases, customers move funds from non-maturity deposits to CDs in order to take advantage of significantly more attractive rates. When market rates fall the differential between CD and non-maturity deposit rates closes and funds flow back into non-maturity deposits.
Disruptive products introduced by the institution. Say your shop rolls out a Rewards Checking account at a 5.01% introductory rate at a time when the top CD and MMDA rate is less than 4%. That product could cause significant cannibalization from MMDAs, savings, checking, and even CDs that no price/elasticity model would have predicted, especially given the behaviors customers need to exhibit to actually earn the 5.01% rate.
Extent to which the product is marketed and how the marketing is targeted. Take the 5.01% checking account introduction in the previous bullet. Should the product be rolled out but not marketed or marketed ineffectively, there will be significant cannibalization of existing accounts as compared to new money raised.
Speed at which customers are exposed to the pricing action. With CD customers, balances are laddered by remaining terms to maturity. With non-maturity deposits, existing customers often don’t find out about rate changes until they receive their next quarterly statement.
Execution by your staff on the lobby floor. Here’s where the rubber really meets the road. Say you introduce an aggressively priced CD special, while holding down rates on regular CDs. Do you think it will affect your cannibalization/new money assumptions if your CSRs lay their hands on the renewing CD lists and call their customers saying, “You better get in here or the bank will renew your CD at a crappy rate.”
Policy guidelines relating to exception pricing. If the institution tightens its exception pricing policy, will the model looking back at results from a looser policy predict the effect of tightening on demand?
Product features that are important to customers other than rates. Fees can be looked upon as a negative rate paid. Will an increase or reduction in fees have an effect on growth of checking balances that is totally independent of the interest rate paid?
Product features designed to block entry to premium products by segments of the current customer base. Tiered pricing structures, minimum balance requirements, relationship pricing, geographic segmentation, stealth marketing, new money requirements, channel barriers, and transactional barriers are all techniques used to allow entry by new customers while blocking entry by existing customers.

Competitive factors – local banking competitors

Entrance to or exit from your market by local competitors who price aggressively.
The extent to which competitors are willing to match your pricing actions quickly. Say you decide to raise your rates on MMDAs 25 bp having seen the prediction from your elasticity model that there will be a 10% increase in demand. The next day, your major competitors match your rate increase. How much new money will you raise?
If competitors elect to match your pricing moves it makes a big difference how they match those moves. Do they match with exception pricing or stealth products that are primarily designed to hold onto existing customers? Or do they fire back in your face in the local newspaper?

Competitive factors – nearby, regional, and national banking competitors

The presence and aggressiveness of national and regional competitors that are attacking your customers over the Internet.
The general level of rates in your market relative to the national markets.
Extent to which these National competitors promote their products in your local markets.
Barriers to entry that exist in front of these high rate products. For example, products like ING’s Orange Savings account limit deposits and withdrawals to Web ACH transactions. My 87 year old mother will not find the necessity of conducting transactions on the Web to be at all appealing.
The willingness of your customers to do business with institutions outside your local market.

Competitive Factors – non-banking competitors offering substitutable products.

Relationship between rates on CDs and substitutes like US Treasuries. Yes, a 1 year Treasury is a direct risk substitute in the eyes of many customers for a 1 year FDIC or NCUA insured CD.
Relative attractiveness of insurance industry products like annuities and brokerage products like cash management accounts.
The extent to which the customer perceives risk associated with being uninsured. Economic conditions can cause these perceptions to change over time.
The relationship between returns being offered by your products and stock market returns.

Economic Factors

The presence or absence of economic events that might cause a significant flow of funds into deposit accounts like Y2K, 9/11, the 2001 stock market crash, the recent stock market fluctuations, and government use of incentives like tax credits to stimulate the economy.
The general health of the local and national economy and the level of consumer confidence and its effect on consumer and business cash flow and the willingness of consumers and businesses to spend versus save.
The extent to which the tax code encourages savings through tax exemptions related to retirement savings, health savings, college savings, etc.
The willingness of those outside the US to invest surplus funds in the US economy. This affects interest rates, particularly long-term rates in the US economy.
Where we are in the economic cycle.

Customer Factors – Aside from those listed under Economic Factors

The customer preference (at the time of the pricing changes) toward liquidity (checking, savings, money markets), as compared to short-term investments (CDs with maturities less than one year), as compared to long-term investments (CDs greater than one year).
The perceived value customers place on features like bump-rates, early withdrawal penalties, add-on options, etc. These perceived values can vary greatly based on volatility and direction of market rates.
The extent to which the customer is willing to use alternative products (like Treasury Mutual funds, and brokerage cash management accounts) and alternative delivery channels (like the Internet).
The demographic mix of your customer base today as compared to one year ago at this time and five years ago at this time.
The behavior and price sensitivity of high balances vs. low balance customers.

Tom Farin