Acquisition Patterns in the Financial Sector
The Dutch case
1. Introduction
1.1 Motive
In a competitive world it is important for banks to attract and retain customers. Holding in mind the harmonization and institutional changes in the European banking environment and the rapid developments in technology (Internet), it is important for banks to develop strategies to continuously sell new products. Banks, as many other businesses, have in fact two options: acquisition of new customers and retaining and deepening relationships with existing
customers. Acquisition of new customers is often accompanied by high costs and effort. Customers and banks have to pay information and search costs. In contrast, retaining existing customers is ‘relatively easy’. Both agents already have frequent contact and hence are able to learn about the behavior of the other agent. Customers are able to inform themselves about the services and products the bank offers. On the other hand, banks can, if they truly understand their customer needs, better adapt to these customer needs. Intuitively, the latter strategy is more efficient than acquisition of new customers, because retaining customers is relatively cheap. Nevertheless, managers continuously focus on acquisition of new customers. A possible explanation of this behavior might lead to direct gains from acquisition clearly
visible in monthly or yearly sales compared to the returns associated with retaining customers. Retention of customers is achieved through taking care of a high level of customer
satisfaction. This satisfaction is a derivative of the question whether demand of customers still match with their product ownership. In general, as customers get older, increased incomes and saving considerations leads to higher wealth. As wealth increases, more sophisticated
financial needs demand different and more complex financial products. Banks adapt to this behavior by supplying a range of financial products different from each other in the level of complexity. Customers acquire these products in a logically increasing order (Stafford et al., 1982).
needs develop. Indications for the time that these needs occur might be a general rule of nature that appears for groups of customers. Based on knowledge of current customers’ characteristics and ownership of financial products, banks might be able to predict the needs that originate.
Information on customer characteristics is nowadays directly available for banks. This information varies from personal characteristics as age, dwelling ownership to monetary value, last contact moments, purchase time of the last product sold, etcetera. In this paper I try to investigate whether it is possible for banks to use information about product ownership to detect possible improvements for cross-selling strategies. Data from an anonymous affiliate of a Dutch bank serve as input for the empirical part of this thesis. I conclude that use of this type of information can help bank to identify cross selling opportunities.
1.2 Methodology
The first step in this thesis is to shed light on the methodology of this type of research. After that in Subsection 1.3 I will explain what underlying microeconomic theory I use. Then the main and sub questions are developed in Subsection 1.4. Section 2 is dedicated to other papers in the field of interest and summarizes methods and results. Next, I evaluate some models. The following step is to determine and explain the model used (Section 3.2). Once the model is chosen, the null hypotheses and alternative hypotheses for the sub-questions are derived. Data then enables me to test the hypotheses for validity. A description of the data can be found in section 3.3.
It is important to shed light on the constraints that I face. The target of this study is to exemplify the ‘world’ of the acquisition of financial products through households. Models limit the world in the attempt to illustrate the things we observe in the real world. That is why I need to indicate the boundaries of this study, because it is impossible to review all possible elements and effects that play a role in the acquisition process of financial products by households. First of all, I confine the research to a single bank. This Dutch bank is for
this method of one-point-in-time seems somewhat contradictionary to the issue of acquisition patterns during time I follow in this respect the methodology of Paroush (1965) and others. Another restriction is the number of products evaluated. Banks offer a huge range of different sorts of products and groups of products. For simplicity I focus on just six different product groups. These product groups are Saving accounts, Checking accounts, Mortgages, Insurance, Investment and Credit. Section 3.4 discusses more of the various problems involved with the data. Subsequently, Section 4 presents the results of the outcomes of the model. Finally, Section 5 concludes and offers recommendations either for managerial purposes as for further research.
1.3 Microeconomics
Standard microeconomic theory suggests that the implicit behavior of each household is aimed at maximizing the present value of future instantaneous utilities. Households are constrained by their income in the search for this maximum utility. This implies that households must decide how to spend their income on products so that maximum utility is derived. In other words, households must rank products to be obtained in order of utility, since simultaneous purchase is assumed impossible due to budget constraints.
Suppose consumers differ from each other in the respect in which they demand products. As people move along the lifetime cycle different needs develop. In other words, utility curves are dependent of time (life cycle) and hence a continuous spectrum of utility curves arises. This idea is for example used in the overlapping generation’s version of the Solow growth model (Heijdra and v.d. Ploeg, 2002). Utility could also be described as the ‘need’ for a specific product. Suppose that this need for products can only be fulfilled in a discrete manner. As a consumer shifts along the life cycle she can attain more products. In other words, in period t only one product can be purchased. In period t+1 the household is able to purchase the next product; in period t+2 the third product is purchased, etc. This theory has already been applied in marketing orientated research and is here copied to the banking industry (Paroush, 1965, Lusch et al., 1978, Stafford et al., 1982).
the first product as time passes on, income accumulates, which can be assigned for spending on the next product. The consumer, enabled through the higher income, develops additional needs. Once the household enters this next stage, this is sometimes described as the latent class in which a household is situated. A latent class method is a method to identify certain groups in different classes, in which the classes are formed by the conditional probabilities of certain discrete variables (Feick, 1987).
The preceding model was originally designed for durable goods, but also holds for financial products (or services) (Kamakura et al., 1991). Financial products enable households to smooth their consumption over time. Households must decide which products they acquire first, since it is easy to imagine that financial products are not acquired at the same time, because not all needs arise at the same time. In short, the household constructs an order in which products are wanted most, the priority pattern. This priority structure corresponds with the utility function. So, for maximizing objectives, the household will eventually purchase all goods following this priority pattern, which is the order of acquisition. Let’s aggregate over households; the summation of all these orders of acquisition implies a general acquisition order in the economy as a whole. Since all households generally face the same kind of
constraints, and perhaps societal pressures exist (see Stafford et al., 1982, p. 411) the order of acquisition of groups of goods, are more or less the same for the entire economy (first obtain shelter, than convenience, than luxury etc). This priority pattern and corresponding
acquisition order is expected to hold for financial products as well. This acquisition pattern is most easily understood when it is analyzed as a matrix where the entries represent ownership or non-ownership of a particular product and in which P1, P2, P3, .. Pn represent products and S1, S2, S3, .. Sn represent subjects. A one corresponds with ownership, and a zero corresponds with non-ownership. Notice that S1 does not own any product, so he/she is at the beginning of the acquisition order. In contrast, Sn owns all products and fully completed the whole product range. Notice further that in this example the subjects do not differ from the order P1, P2, P3, .. Pn in acquisition behavior. This example is also known as the Guttman perfect scale (Paroush, 1965). It is perfect, since no deviation occurs from the order P1, P2, P3, P4 such as for
example P1, P3, P2, P4. Guttman scaling is also known as a deterministic approach, since it implies a deterministically order of product purchase.
(
)
∫
=
tZ
P
Q
H
A
Y
x
,
,
,
,
,
where x is the acquisition order of financial products. The order is explained with the explanatory variables Y,A,P,H,Q,P and Z. Y is income, A is age, H is house ownership, Q is competition among different suppliers, P is price. Probably some other (external) effects play a role in the decision making process: Z.
deterministic acquisition order. The more errors, the more likely it is that no acquisition order exists.
The discussion above only relate to the demand side of the market. On the supply side, competition is a factor of interest in the market of financial products. The Hotelling location model suggest that if two banks are fighting for the same consumer, the bank that is located at the shortest distance to the consumer will supply this consumer, ceteris paribus (Osborne and Pitchik, 1987). The reason is that households choose to visit the supplier located closest by, since this minimizes time and effort spent on each trip (also known as shoe leather costs). Although this theoretical explanation might loose some importance in a world characterized by boundaries that are fading away as a result of technological progress (Internet), this is an important fact if one considers banking competition. Logically, household product acquisition is not necessarily restricted to only one supplier. Since in the Netherlands several banks operate on the consumer market, it is interesting to see whether the influence of competitor presence is observable in acquisition patterns of households of an affiliate of one particular bank. It is easy to understand that if more competitors are fighting for the same pool of households, the fewer products a bank sells, ceteris paribus. Intuitively, loyalty is stronger in areas with fewer suppliers, simply because of a lesser amount of opportunities available.
1.4 Objectives
The main research question is:
What are the determinants of the success of cross- selling strategies of banks considering historical purchase information of customers?
The sub questions are:
1) What is the most logical order in which consumers acquire financial products according to existing literature?
2) Is this pattern found in the data for bank X corresponding with the assumed theoretical product purchase pattern?
3) If differences arise between theory and practice, what are candidate explanations for this observed behavior?
impact on the financial decisions in later life. Housing ownership is another element which has presumed predicting power in the acquisition order.
4) Do differences occur in the acquisition patterns between heterogeneous groups, segmented by age, and house ownership?
As a control variable competition is introduced in the model. This variable is used to check its impact on the acquisition pattern for the bank’s products. The degree of competition is
expected to influence the hypothetical order of acquisition.
If in a particular residences several banks have affiliates, this implies increasing competition opposed to a residence were only one bank is located. Households in the first have more options to acquire products from. For example, a household now buys its first product from one bank and the next one from another bank. Competition runs through the ideal pattern, since in the case without competition this particular household would have bought its
products at the same bank. It is therefore expected that in residence with more competitors the ideal type pattern is different than in less competitive environments.
5) The acquisition pattern deviates more from the hypothetical acquisition in a highly competitive environment opposed to an environment lacking any kind of competition.
2. Review of the literature
2.1 Single acquisition pattern
In the 1960’s Paroush was the first in economics who used a tool from the field of
psychometrics, borrowed from the work of Guttman (1950). Guttman used scale analysis in the field of psychometrics. The Guttman scale is a matrix where columns represent items and rows represent groups of consumers. The matrix shows for every item ownership, represented by one, and non-ownership (zero). Although Paroush’ order of acquisition was in his words a ‘temporal concept’ it is possible to describe this phenomenon with data gathered from one specific moment in time. The intuition behind this is that a ‘scale’ in this acquisition pattern exists. Paroush explained the existence of acquisition patterns from a microeconomic perspective. Each consumer has a number of indifference curves. As income increases the consumer is able to attain a higher level or in other words, achieve a higher indifference curve. Although income might increase continuously, consumption is altered only in discrete steps. This leads to the assumption that a path of steps is to be taken by the consumer ranked to the preferences, or in other words utility. Paroush did not pay attention to prices; he just wants to describe the actual purchases order in time of a particular consumer.
Next, suppose that all individuals have similar indifference curves and increasing income. It follows that in total economy the same specific order of acquisition is to be found. Of course, this ideal situation will not occur. However, it is possible to find a pattern that almost all consumers will follow. Paroush called the order of acquisition for the economy as a whole a “scale structure”. His next assumption was that the population is heterogeneous, i.e. there are different groups of consumers at every stage of life. This latter assumption enables to draw conclusions about the behavior of groups of individuals over time at one specific point in time. The perfect scale implies that all consumers have the same order of acquisition. But as the number of items (k) in the order of acquisition increases, it is easily understood that in practice agents do not have this perfect scale. To measure the number of deviations from the hypothetical order of acquisition the minimum number of changes from k that exists in the population is count. This number is defined as Si. Also, the number of deviations is thus
i I i iS
∑
∈ ,Coefficient of Reproducibility (Re) and it measures the “quality of prediction” and the “possibility of explanation” (Paroush, 1965). It is defined as
∑
∑
∈ ∈∗
−
≡
I i i I i i ik
S
1
Re
A practical cut-off is according to Paroush approximately 90%.
Hebden and Pickering (1974) investigated whether it is possible to detect a unique order in which (parts of) the population acquires durable goods, by observing past purchases. The authors recognize that acquisition patterns do not reveal anything about a time dimension, so it is hardly possible to determine the exact time of acquiring a specific product. Like Paroush the authors are only interested in the order in which products are acquired, not in the specific point of time. First, the construction of a matrix in which items correspond the shares of households owning a specific product combination. Secondly, a matrix with conditional probabilities of product ownership is constructed. These probabilities are the chances that a household owns a specific set of goods conditional on the owning of a specific number of goods from the total set. For example, a household owning TV, tape recorder, caravan conditional that it owns a specified number (3) of products out of the total set. The
methodology continues with the conditional probability that a household owns a particular product (TV) conditional to owning a specific number (3) of goods. After that for each product the difference is computed of the conditional probability that an item i is possessed conditional on the owning of a number of j goods minus the conditional probability that item i is possessed conditional on the owning of a number of j-1 goods. In short, this is the
probability that a household owns a TV and two other goods minus the probability that the same household owns a TV and one other good.
This last step reveals the probabilities of a position of a good in the pattern of acquisition. The expected position of good i in the order of acquisition is now rather straightforward: the sum of all probabilities calculated in the step before.
Kasulis et al. (1979) built further on this idea with the assumption that the utility a household derives from a durable good is registered by the purchase of the good. Their study is
choices which products will be purchased first. In other words, consumers must order or prioritize their purchases of (durable) goods. The order in which these goods are gathered reflect the utility structure for households and the economy. If there are j goods then j! patterns exist. If a priority structure is detectable than the order is D1, D2, D3 ….Dj. New to this study is the expansion of extra information by including an innovation item and second purchase goods. The analysis makes use of Guttman scalogram analysis to rank the order in which durables are acquired. All age cohorts are represented and thus all stages of the
acquisition process are represented. A similar explanation is that consumers first acquire less “difficult” goods before they buy more “difficult” goods. Next, Kasulis et al. split the sample into two heterogeneous groups: dwelling owners and renters. Results show that the
acquisition pattern is not the same in these two groups. An interesting feature of Kasulis’ et al. study is the description of the error terms in the acquisition process. They explain skip error as follows. Suppose the order in which a representative consumer buys goods is A, B, C, D, E, F. Consumer x has four items. If x follows the normal scale order, than it is expected that he owns A,B,C and D. Now suppose he owns A,B,C,E instead. This implies that he skipped good D in the acquisition process. Consumer x is therefore characterized as having a skip error. This could also be looked at the other way around. Suppose, that consumer z has item E in possession, but that he has a scale score of just four. This error is called an ownership error, since he does not own D. Notice that skip and ownership errors are the same error, only evaluated from a different perspective. A useful feature of the study is that innovative product purchase is cannot be explained by the acquisition order. Households at different stages of the acquisition order –being at a different scale order- have obtained the innovative product. Although these households already have accumulated some lower arranged goods it cannot be determined that an innovative product is the last product acquired. Another conclusion is that second purchase durables (second TV, second car1) do compete with other products. In other words, second purchase goods cut across established acquisition patterns (Kasulis, 1979, p. 56).
In another paper Lusch et al. (1978) also investigate the accumulation of durable goods by consumers. Again, the basic thought is that, due to financial restrictions, consumers must take decisions about the order in which they obtain durable goods. Income-consumption models do a good job in performance over the long run, but are unable to take the pattern in which
1
consumers acquire durables over time into account. The main result of this study is that strong evidence for an increasing order of ‘difficulty’ is present in the group of consumers for
household durables. With difficulty the authors mean that some products are more difficult to acquire than others when priorities needs to be fulfilled following a specific order.
In the first half of the 1980’s the discussion of acquisition patterns expands from durable goods to financial goods (Stafford et al., 1982). Similar to the studies devoted to durable goods, the basic idea is that consumers have similar priority structures for financial products. In other words, preferences seem to be more or less the same in the entire economy. The study of Stafford et al. focuses on eight financial products: ranging from checking accounts to mutual funds. Starting point of the study is that the fact that consumers must take decisions about the order in which they acquire the products. Constraints that force them to do so are prices of products and income. The authors shortly mention some possible reasons for the existence of a general acquisition patterns for financial products. These reasons could for example be necessity, culturally mandate lifestyle or conformity to group norms (Stafford et al., p401). For control reasons they split their 1500 household counting sample into different age cohorts. The era in which these households grew might have influence on the decision making of households. Young people during the Depression might have had different impact than a youth during the prosperity of the 1950’s. The acquisition patterns they found starts with a checking account and ends with mutual funds. The items between these extremes are in order of difficulty: husband’s life insurance, saving account, wife’s life insurance, stocks, bonds and trusts. Between the age cohorts the study reveals that the middle age cohorts acquire stocks before bonds opposed to the younger group. A possible explanation according to the authors is that the first group is better positioned to take the increased risks that are involved with stocks. In short, the study reveals that no important differences exist in the order of acquisition for age cohorts.
the household accumulates more wealth and hence the financial objectives change. The “pyramid of financial independence” (Kamakura et al., p331) orders the objectives in a particular increasing order. The authors use results of earlier studies for their assumption that higher order objectives, often associated with increasing risk taking, imply higher order needs. These needs can be classified, and hence can be described as latent classes. If a next, more difficult product is acquired, the consumer has fulfilled its latent need and shifts to a higher order latent class. A product sale is consequently the result of coincidence of the latent class and the product characteristics for the specific item. In the words of the authors to ”position both services and households along a “latent” difficulty/ability dimension” (Kamakura et al., p334). This ability dimension supposes a hierarchical order in which consumers obtain financial services (products). An implication of the study according to its authors is that is suitable to detect the most valuable prospects to cross-sell additional financial services.
Knott, Hayes and Neslin (2002) describe Next-Product-To-Buy (NPTB) models that predict not only what the next product to buy will be, but also the time when this occurs. The basis of this is data that reveals customer-specific information at time t and secondly information about product purchases at time t + 1. The data of period t is the independent variable, while data of time t + 1 is the dependent variable. The independent variables are classified in three groups: demographic variables, customer’s product ownership characteristics such as
frequency and monetary value and third marketing efforts. Their model is expanded with a closer look at the error term of the model. It supposes that the decision of customers whether to buy or not buy a next product is influenced by factors that Knott et al. describe as the inhibitions. Examples of these inhibitions are the inability of customers to recognize the problem they face or secondly, marketing efforts of competitors. These inhibitions are evenly distributed over the population. In short, the customer’s need must overcome the inhibition factors before he/she buys a product. The results of Knott et al. study show that their NPTB model is doing a good job in predicting the high potential customers that will buy a next product. Final remark is that current product ownership is the most crucial factor in predicting analysis.
that firms continuously acquire and lose customers to each other. Due to cross selling firms attempt to improve and tightening the relationship they maintain with their customers. The stronger the relationship, the higher are the costs of switching to a competitor for a consumer. This is due to the fact that if a consumer considers leaving the supplier he must invest in information costs. The more products (s)he buys from a bank the higher these costs will be. The study develops a new method “to predict consumption of new and current products by current customers who do not use them yet.” At last, the authors emphasize not surprisingly that the best prospects for cross selling are those customers with highest probabilities of consuming a specific services, but do not consume this services at the bank. Perhaps they obtain this service from a competitor and hence these are the best prospects for cross selling opportunities.
younger counterparts. The reason for this behavior is according to the authors the greater amount of available time for information research that older generations have. This model is an expansion of earlier models. The new element in this research is the addition of the demand maturity.
Paas and Molenaar (2005) describe the term cross-selling as “the right offer for the right customer at the right time”. Acquisition Pattern Analysis (APA) is the method for understanding the “..purchase dynamics and predicting future product acquisition by
individual customers” (Paas and Molenaar, 2005, p.88). The main goal of their investigation is the evaluation of the forecasting accuracy of APA models. Paas and Molenaar’s starting point is the same as the ones discussed before; an overview of the behavioral ideas of the acquisition order concept, that is, consumers always obtain the products with high
penetrations levels, before products with low penetration levels. -Alternatively we could say: this is the reason that products with different penetration levels exist.- The probabilistic APA approach is distinguished from the deterministic approach in the sense that it calculates the probability of acquisition opposed to actual acquisition. Important in this respect is that, APA just has relevance if the consumer preferences are ordered hierarchically. Paas and Molenaar introduce the Mokken scale analysis for non parametric scales for the acquisition patterns of products. In short, Mokken scale analysis can be described as a method to predict the
probability that a hierarchical acquisition pattern exists with help of penetration levels of different products. The acquisition rate is the number of actual purchases in the 1996 -2001 period divided by the number of potential purchases in the period. A potential purchase happens if a consumer does not own a specific product and is hence a potential buyer. Furthermore, only products that are just above the acquisition pattern scale position are counted as likely acquisitions. The others, the errors are counted as unlikely acquisitions. What can we learn from this research is that consumers should offered products right above their acquisition class. Higher ordered products are just too far away from their individual needs and hence products offered have low potential acquisition rates.
2.2 Literature summary
to monetary value. The dependent variables also change over time. At first, Paroush only examines just four consumer durables. In later studies the number and the kind of products under investigations grows. Besides that, product groups are introduced, if notice appears that for example leisure goods might have other influence on consumer acquisition behavior opposed to household utility goods (Hebden and Pickering, 1974). Later studies solely focus on financial products. Another point in which the authors differ is the distinction made between the points in time evaluated. Most papers took cross sample data as the basis of investigation. Opposing to that, Kasulis et al. (1979) had data available originating from two adjacent years. Even more ambitious is the work of Paas and Molenaar who used consumer acquisition data over five years as input for their investigation. Although the use of data gathered over a longer period of time clearly should lead to better insights in the moment of purchase, all studies mentioned are able to explain the sequence in which products are acquired.
Table 1 Overview of the literature Element P ar o u sh H eb d en & P ic k er in g L u sc h e t al . K as u li s et a l. S ta ff o rd e t al . K am ak u ra 1 e t al K n o tt e t al . K am ak u ra 2 e t al L i et a l P aa s et a l. # of products > 6 Durable goods Financial products
Cross section data
Response patterns from longer span of time.
Additional characteristics as predictors for LC* Competition
* LC = latent class, = not described in the paper, = described in the paper, - = irrelevant.
3. Methodology
3.1 Acquisition order
It is important to determine the most logical order in which consumers acquire financial products (research question 1). Financial products we consider are (in alphabetical order): Checking account, Credit, Insurance, Investment, Mortgage and Saving account. For this moment let’s analyze two scenarios: an altruistic-parents-scenario and a
non-bequest-scenario. In the first scenario parents save for their children and leave them an inheritance. In the second scenario each agent completely consumes its own wealth. Let’s first consider altruistic-parents-scenario. From birth on, children posses a saving account, since parents take financial care of them and leave them gifts. In their teen ages or shortly after, these agents enter the labour market. Since wage payments run via banks, they will need a checking account. Now, consider the second scenario. The first financial product they obtain is a checking account. If income is not spent totally on consumption soon a saving account is needed. Next, we move on with agents from both scenarios. With income (and in the altruistic-parents-scenario bequests) individuals are more and more able to fully become active economic agents, purchasing durable goods and services, for example cars or holidays, respectively. Minimizing risk behavior will induce individuals to insure themselves against bad risk: the hypothetical third acquisition is thus Insurance. At some point in time
products, namely the acquisition of investment products such as bonds, shares, funds etc., shortly named Investment.
Now, two kinds of hypothetical orders of financial product acquisition are developed. Let’s analyze what former studies have produced and summarize this in table 3. The table is in fact the answer on the first two sub questions: what is the most logical order in which consumers acquire financial products and what is the acquisition order according to the existing
literature?
Table 2 Hypothetical financial product acquisition order
Product Altruistic- parents-scenario non-bequest-scenario Kamakura et al.(1) Stafford et al Saving account 1 2 1 2 Checking account 2 1 1 1 Insurance 3 3 4 3 Credit 4 4 2 Mortgage 5 5 3 Investment 6 6 5 4
As can be seen the earlier described hypothetical product acquisition order does not match exactly with the two studies. The reason might be twofold. One is the kind and number of products evaluated. For example, Mortgages and Credit cards are not incorporated in the Stafford et al. study. The second reason is that our research is located in the Netherlands, whereas other studies are performed in the United States, where acquisition order is perhaps slightly different. This implies that the acquisition order in literature is hard to determine; it depends on various different elements in each study. At this stage it is important to know that from here the study continues with the following hypothetical acquisition order: Saving account, Checking account, Insurance, Credit, Mortgage and Investment, where the reverse order of the first two items is considered as well.
3.2 Model
unfortunately does not unambiguously reveal monetary value (Knott et al., 2002) or income. On the other hand, demographic variables are good predictors of latent classes (Knott et al. 2002, Kamakura et al., 1991, Kamakura et al., 2002). Due to restrictions on data availability, demographics is in this research solely reduced to the variable ‘age’. The impact of not using monetary value as an explanatory factor of latent class, is estimated quite low, since although other studies do find proof for monetary value, the statistical relevance is low.
In order to find an order of acquisition I use methods that are commonly applied in the field of marketing. This type of research often uses principles that is built on hypothetical foundations which are often assumed not to be really convincing. However, I consider results can be useful as long as statistical significance is able to confirm results. I assume that research to agents’ utility structures dependent on the factor time can take advantage of lessons learned in marketing orientated research. In this field of interest preceding work was done with the help of Guttman scaling. Instead of that I will make use of a more refined research method, which makes use of ownership probabilities calculations (see Hebden and Pickering, 1974). The reason is that this is thought to be superior to the Guttman scaling method (Paroush, 1965, Lusch et al, 1979, Kasulis et al, 1979, Stafford et al, 1982) since the latter ignores the possibility of any other acquisition order than presented. Opposing to that, the method proposed by Hebden and Pickering, is more convenient, since the acquisition order is
determined on the basis of expected positions of products in the acquisition pattern. In short, the first part of the model to calculate the order of acquisition in this research is almost analog to Hebden and Pickering (1974). In the following part, the construction of the model is
explained stage by stage.
The first stage in the analysis is to determine the number of consumers that own one of the goods evaluated in any combination. For example, in the set with six products, some households own one product, others two, and so till six products (see for a full data
combination. Each entry in the column represents 0 or 1 for (non) ownership of each product respectively. Since there are 6 products, the maximum possible product combinations (or rows in the table) are 26 = 64 entries. Notice that this stage is in fact the construction of a Guttman scale (see for an example of the perfect Guttman scale Table 1 in section 1.3). If a perfect order of acquisition would have been present, the table would have had just 6 rows with data. But in a non-perfect world, other product ownership combinations exist. The totals of each combination are divided by the totals of households that own set s.
Stage one Pr
(
ownssets owns[ ]
s products)
≡n( )
s n[ ]
sWhere,
Pr = probability
s = a certain set of products (3 or 4 for example)
(s) = specific set of s (Checking account and Saving account for example) [s] = number of products in set s
n(s) = number of sample households owning set s
n[s] = number of sample households owning [s] products
Stage two builds further on the probability found in step one by calculating the probability that a household owns one specific product conditional on the fact that it owns a set s of products. Conditional on the fact that a household owns any number of products, this step calculates the probability that one of these products is a specific product. For example, conditional of owning three products, what is the probability that one of these is a Checking account. Notice the difference with step one in which the conditional probability of ownership of a combination of products is calculated. In this step the conditional probability of
ownership of just one product is calculated. Comparing different set sizes (for example three products in possession as opposed to two products in possession) shows which products are expected to be obtained early relative to other products.
Stage two
(
[ ]
)
(
[ ]
)
[ ]
s is fixed where s owns s owns products s owns i product owns s i∑
∈ ≡ Pr PrWhere i = any particular product
life cycle process that all households supposed to go through, but where not all households are positioned on the same point in that process. Each product that a household adds assumes a next small step (or should we say giant leap when just six products are considered) in the life cycle process. Differences in the conditional ownership of products between two of these steps is in fact the signal that some products are acquired earlier in the life cycle process than others. This assumption leads us to the next probability. This is the probability that a
household owns a specific product i conditional that this is the jth good acquired. In an example this is the probability that a household owns a Mortgage conditional on the fact that this is its fourth purchase. Although seemingly difficult, this can be easily calculated by using the results from the former stage. Stage three probability is the probability of owning good i conditional to the fact that the households own j goods minus the probability of owning good i conditional to the fact that the households own j-1 goods. In words, this simplifies to: the probability that a household owns Mortgage conditional of possessing four products minus the probability that this same household owns Mortgage conditional of possessing three products. Notice that with the use of cross sectional data it is only possible to compare groups that differ from each other in the number of products owned if we assume that different groups -in maturity terms- behave according to the same patterns.
Stage three Pr
(
[ ]
i ≡ j)
≡Pr(
jth product purchased isi)
Or more formally written as:[ ]
(
i j)
Pr(
owns producti owns[ ]
j products)
Pr(
owns producti owns[
j 1]
products)
Pr ≡ ≡ − −
For example:
(
Mortgage 4th product)
Pr(
ownsMortgageowns4products) (
PrownsMortgageowns3products)
Pr ≡ ≡ −
Stage four finishes the calculation of the acquisition order. The sum of the conditional probabilities constructed in the stage three for each product is the expected position in the order of acquisition. Since six products are considered, notice that if hypothetically all
consumers purchase Investment as their last product, the expected position is of Investment is therefore close to 6. Stage four:
( )
[ ]
≡∑
[ ]k= ∗(
)
j th i is purchased product j j i E 1 Prresults is explained in Section 4. In order to test this pattern of acquisition some tests for statistical significance are executed. I follow the method proposed by Hebden and Pickering (1974). The explanation of these tests can be found in the Appendix. The research continues with a discussion of some subgroups of consumers. As extension I will implement the element of competition.
From Knott et al. I use the idea that competitive actions or presence are a potential reason of inhibition to buy a product from a bank. In order to answer the control research question for competition, the factor competition is an element that must somehow be measured. Perhaps the most elementary measure as proposed by Hotelling (1929) is distance.
Table 3 &umber of competitors in place of residence Residence Bank α β γ δ I 1 1 0 0 II 1 0 0 0 III 1 1 0 0 IV 1 0 0 0 X 1 1 1 0 Total 5 3 1 0
Notice that α has the lowest concentration of banks, which implies the highest degree of competition. In β fewer banks have affiliates, so the degree of competition is lower. In γ the only bank with an affiliate is X, so here X is in fact ‘monopolist’. Keep in mind that distance does not prevent households to obtain financial products from banks in other towns, but this possibility is further ignored here. Although the degree of competition in α and δ is
customers for a specific good. For example, if a bank has a marketing focus on Insurance products this perhaps increases its experience and success in selling Insurance. This bank is expected to supply relatively more Insurance products in the market opposing to its
competitors. In the aggregate acquisition pattern for this bank this would become visible as the expected position of the Insurance is relatively at the beginning of the continuum. In a place of residence where no competition is present, it is to be expected that the order of acquisition does not deviate much from the hypothetical order of acquisition composed earlier is. In contrast to that, increasing competition is expected to distort this acquisition order and the order deviates from the hypothetical order. In practice for this study this implies that the H0 is stated as follows:
H0 : In γ the hypothetical acquisition order is approached, while in the distortion from that order is larger in β and the distortion in α and δ is even more pronounced.
Against the null hypothesis, the Ha is stated
Ha : no difference is detected in the order of acquisition for the different types of residences.
3.3 Data
3.3.1. Data description
The data used for the empirical research originates from December 2006 and is supplied by a commercial bank in the Netherlands. This consumer data set is retrieved from internal record system of the bank and holds information on product ownership of several product groups. Unfortunately it is not possible to compress the data into household figures. Data is only originated from consumers living in α, β, γ and δ. First of all, and most important, the data set contains information about individual’s ownership of six different product groups which are in order of appearance in the internal record system of the bank, Checking accounts (Check), Saving accounts (Save), Investment product (Invest), Credit (Cred), Mortgage (Mort) and Insurance (Ins). Ownership of one of these product groups is in the system represented with a J (yes) and non-ownership is recorded with N (no). The data contain demographic and other characteristics of each individual that will be used to classify different heterogeneous groups of customers. The demographic variables are those representing the place of residence and age. Notice that although the data contains all consumers, I only make use of individuals aged over 18, since it is not to be expected that children already own the whole product line.
indeterminable, not known, respectively. Only the J and N are taken into account, the other two options are ignored and deleted from the data set since this variable is used for
constructing heterogeneous groups. In total the data sample contains 26263 consumers.
In order to differentiate between age cohorts, the construction of three age cohorts is executed, analog the method in Stafford et al. (1982). Cohorts all approximately contain a third of all consumers. These are age below 36 (<36), age 36 until 53 (36-53) and age 54 and higher(>=54). In §3.2.2, figure 2 subtotals for each age cohort as well as number of
3.3.2 Data Composition
Figure 1 Analysis of data composition
Notice that that the number of individuals in the residences considered differs remarkably. This can be explained by the fact that each residence varies in the number of inhabitants within their borders. Another explanation could be bank X having very different market penetration grades for either one of the product groups. Intuitively, common sense favors the first explanation. Furthermore, the percentage of housing ownership is close to the national average, which is approximately slightly more than 50% (Sterken, 2006, p589). This is a signal that the data is a good reflection of the population in the Netherlands.
Analysis of the data shows that age cohorts are not evenly distributed in the four residences. Housing ownership seems to be related to age. This effect is strongest for the Elderly group. This is easily understood when one is holding in mind that usually income (and thus wealth) increases during lifetime. Wealthier households normally have higher house ownership than the less wealthy groups. Evidence for this is also found in the data, see the next table.
Table 4 House ownership
Age group House ownership
< 36 y. 25,0%
37-53 y. 65,0%
>= 54 y. 59,5%
Total 48,9%
Obviously, house ownership is concentrated in the middle and oldest age groups.
3.4 Restrictions
The major problem is that for reasons as time and budget restrictions data of product ownership of only one bank is available. Further, the construction of classes of households assumes the use of data about income. Unfortunately, this type of data is not freely accessible. Latent classes also could be constructed on data about education, occupation, house
ownership and so on (see for example Kamakura et al., 1991). Although the use of sample based information is used in some studies, this still requires effort in organization of a field study. Setting up a large investigation for collection of this type of data including the use of competitor product use has not been executed for reasons of budget and time. Another major restriction is the absence of some elements of demographics. Only age is available, while the factors education and income would enable better prediction of the latent classes.
With cross-section data it is impossible to detect variations in product ownership over time. For example, it is not possible to determine whether households only shortly before the moment of registration have shifted from one financial supplier to another. Another lack of this type of research is that it completely ignores the effect of prices on the decision making of households. It is quite obvious however that prices of various products are factors with impact on the decision taking of households. In the data set, for unknown reasons,
4. Results
First, the results of the expected order of acquisition for six financial products for the total sample are presented. After that shortly the results for the heterogeneous groups - age cohorts and by house ownership- are evaluated. Finally the results for implementation of the effect of competition are discussed.
The results make clear that it is indeed possible to give an answer on the second sub question: Is an acquisition order found in the data for bank X corresponding with the assumed
theoretical product purchase pattern? The answer is obviously yes.
It follows that for the entire sample the acquisition order found is Checking account, Saving account, Credit, Insurance, Mortgage and Investment respectively. Table 5 Average order of acquisition, presents the average position in the acquisition order.
Table 5 Average order of acquisition
C h ec k in g a cc o u n t S a v in g a cc o u n t C re d it In su ra n ce M o rt g a g e In v es tm en t Average position1 1.54 (*) 2.21 (*) 3.00 (*) (**) 3.76 (*) 4.91 (*) 5.57 (*) 1) otice that we assume that all individuals eventually acquire all six products. So, in a perfect order the corresponding integers would have been 1,2,3,4,5,6. The Checking account is expected to be acquired at position 1.54 which is before the 2.21 of the Saving account. A single asterisk (*) implies significant distance between two adjacent average positions at the α=5% level . A double asterisk (**) implies no significant distance between the average position and its corresponding integer at the α=5% level. Single and double asterisks hence favor the hypothesis of a unique pattern.
It follows that the altruistic-parents-scenario is not viable and the non-bequest-scenario is a better representation of the actual acquisition order.
Figure 2 Probability density function of the acquisition order (total sample)reflects
probability density function for the six products with products on the x-axis, the number of purchase on the y-axis and the probability of purchase on the vertical z-axis. It is easy to see that probability of ownership increases if more products are owned.
are in the dataset. This implies that some individuals only purchase these products from bank X and obtain other products from competitive banks. This means that the probability that their first purchase is one of these products is higher than the probability that this is individual’s second purchase.
Figure 2 Probability density function of the acquisition order (total sample)
The third sub question was: If differences arise between theory and practice, what are
The fourth sub research question was: Do differences occur in the acquisition patterns between heterogeneous groups, segmented by age and house ownership? Table 6 Average positions in the acquisition order for heterogeneous groups shows the results.
Table 6 Average positions in the acquisition order for heterogeneous groups
Average positions Ch ec k in g a cc o u n t S a v in g s a cc o u n t C re d it In su ra n ce M o rt g a g e In v es tm en t < 36 y. 1.40 (*) 1.99 (**) 3.02 (**) 3.81 (*) 5.06 (*) 5.72 36-53 y. 1.54 (*) 2.43 (*) 2.80 (*) 3.71 (*) 4.92 (*) 5.61 >= 54 y. 1.67 (*) 3.28 (*) 3.13 (*) 3.73 (*) 4.75 (*) 5.43
Non house owners 1.43 (*) 2.13 (*) 2.91 (*) 3.70 (*) 5.53 (*) 5.29 House owners 1.74 (*) 2.33 (*) 3.06 (*) 3.71 (*) 4.59 (*) 5.57
otes, see table 5
The results for (non) house ownership should give low expected positions on the acquisition order for Mortgage in the group non-house ownership. This is exactly what is found2. Furthermore, non house owners obtain a Checking account earlier than house owners;
expected position of Checking account is for non house owners lower than for house owners.
Now the function of the acquisition order is expanded with Q, competition. Remember that competition is expected to be more severe in cities α and β, since here more bank affiliates are located. In residence γ no competitors are present, so here bank X is considered to be a
monopolist. Individuals in γ have no other option than to buy is from X. Table 7 Average positions in the acquisition order for four residences presents the results for the four different places of residence. In residence α all expected positions differ statistical significant from the adjacent position of the next product. All average positions deviate significant from the corresponding integers (1,2, ..6). Significant distance between average positions of adjacent positions also holds for β and γ. In δ no statistical difference can be found between Checking account and Savings account. In β and δ for three products a statistical significant closeness between the product and the corresponding integer is found.
Table 7 Average positions in the acquisition order for four residences
Average positions Ch ec k in g a cc o u n t S a v in g a cc o u n t C re d it In su ra n ce M o rt g a g e In v es tm en t α 1.50 (*) 2.30 (*) 2.93 (*) 3.80 (*) 4.88 (*) 5.59 β 1.68 (*) 2.02 (**) 3.04 (**) 3.67 (*) 5.01 (**) 5.58 γ 1.49 (*) 2.05 (**) 3.34 (*) 3.81 (*) 4.77 (*) 5.52 δ 1.89 2.03 (**) 3.17 (**) 3.63 (*) 4.86 (**) 5.42
otes, see table 5
In γ, only Saving account is statistically close to its integer. Two other things must be mentioned before going further. In γ and δ the probability that a Checking account is owned
2
5. Conclusions and recommendations
5.1 Conclusions
Before conclusions can be drawn it is important to emphasize that theory used in this research is built on marketing orientated research. Although it is acknowledged that the theoretical foundations of this type of research might seem rather intangible, it can help to better understand the issues treated here. The use of this theory makes it important that results are justified with solid statistical material.
The first conclusion is that the answer on the main research question “What are the determinants of the success of cross- selling strategies of banks considering historical
purchase information of customers?” is ambiguous. For age and house ownership significant evidence is found that expected positions of products in the acquisition order differ from each other. On the other hand, it is hard to find any significant evidence for the statement that these expected positions do not differ much from the corresponding integers in the theoretical acquisition order. Hence, I found proof for the statement that age and house ownership relate to expected positions in the acquisition order, not on different acquisition patterns. In other words, for all subgroups the same acquisition pattern is found, but the underlying
relationships between product purchases are somewhat different. At this point, some possible explanations are the issue of different needs for different age cohorts. Perhaps younger generations develop different needs that ask for different financial products. Another
explanation is that ownership of residential property obviously causes other product demand.
Knowledge of this research can help banks to better understand the actual ownership
acquisition patterns of their customers. Unfortunately the opportunities to use subcategories of households different in age diminishes as age increases. For places of residence, the degree of competitive concentration will provide banks information for maximizing cross selling
efforts. According to this research banks are able to improve their success in cross selling strategies in areas with low degrees of competitive concentration.
The fact that for all four places of residence and the subcategories of different types of groups of individuals the same hierarchical order of the acquisition pattern is found is a signal that the results are closely related to each other. Another, more serious problem is the fact that the methodology used is not completely suited to detect all effects included in the data.
Recommendations for further research are hence provided in the next section.
5.2 Recommendations
The methodology used in this research has a hard job in providing convincing evidence for the implied effect of competition on the acquisition patterns of individuals. Since the initial discussion about the choice of the correct hypothetical acquisition order, due to the order of the Checking account and Savings account it is recommended for further research to omit one or both products to see whether more pronounced results can be found. Another indication for this recommendation lies in high probability of acquiring a Checking account as the first purchase. In future research the use of a different product line could help to improve predictions about banks cross selling strategies.
The element of competition in the model was based on the distance to a bank. Obviously, price can be a factor of major importance in individual supplier decisions. In this research the effect of price is completely ignored. A suggestion for further research is to incorporate the element of price of the various products in the model, for better exploring all elements playing a part in competition. Of course, this is would be quite complicated since it is hard to
construct one single ‘price’ for products like mortgage and investment.
groups in the total population are interesting factors for further investigation. Cultural differences accompanying these ethnic groups could perhaps have influence in the
Appendix
The assumption of a unique order of acquisition implies that if indeed this order exists the expected position in the order is for each product equal to its integer (1,2,3…k). But since deviations of the perfect order probably will occur, the positions deviate from this perfect order. The uniqueness of the deviation of the expected position E([i]) from the corresponding integer can be tested with the help of a Student’s t-test for significance of deviation, in which the null hypothesis states that no deviation occurs. Since the sum of expected positions on the acquisition pattern is bound to k(k+1)/2 the deviation of one of the positions must be
compensated elsewhere to reach ultimately the k(k+1)/2 constraint But since one “mistake” of a product relative to its corresponding integer, and the sum of expectations is restricted3, this “mistake” must be balanced with another expected position. That is why a second t-test must be carried out. This test measures the difference between expectations of positions of the subsequent products in the acquisition order..
So the second test is used to check whether the position differ from each other. The expected positions are adjacent to each other. Here, also the procedure of Hebden and Pickering (1974) is followed. The test is actually a little bit to strict since it ignores the covariance element in the following formula. The reason of this omission is the expectation that the omission of the covariance term will not have much influence on the results.
[ ]
( )
( )
[ ]
[ ]
( )
( )
[ ]
( )
[ ]
( )
[ ]
− + − = j E i E Cov j E Var i E Var j E i E t ^ ^ ^ ^ ^ ^ , 2where i and j are products.
As mentioned before, the term in italic is deleted from the equation.
Since the sample is large (n > 40) the t – procedures is used on legitimate grounds (Moore, McCabe, 1993, page 402)
3
References
Chilton, R.J., A Review and Comparison of Simple Statistical Tests for Scalogram Analysis, American Sociological Review, Vol. 34, 1969, 238-245.
Feick, L.F., Latent Class Models for the Analysis of Behavioral Hierarchies, Journal of Marketing Research, Vol. 24, 1987, 174-186.
Hebden, J.J., Pickering, J.F., Patterns of Acquisition of Consumer Durables, Oxford Bulletin of Economics and Statistics, Vol. 36, 1974, 67-94.
Heijdra, B., Ploeg, van der, R., The Foundations of Modern Macroeconomics, Oxford University Press, 2002.
Kamakura, W.A., Ramaswami, S.N., and Srivastava, R.K., Applying latent trait analysis in the
evaluation of prospects for cross-selling of financial services, International Journal of Research in Marketing 8, 1991, 329-349.
Kamakura, W.A., Wedel, M., de Rosa, F., Mazzon, J.A., Cross-selling through database marketing: a mixed data factor analyzer for data augmentation and prediction, International Journal of
Research in Marketing 20, 2002, 45 – 65.
Kasulis, J.J., Lusch, R.F. and Stafford, E.F. jr., Consumer Acquisition Patterns for Durable Goods, Journal of Consumer Research, Vol. 6, 1979, 47-57.
Knott, A., Hayes, A., Neslin, S.A., Marketplace; Next-Product-To-Buy models for cross-selling applications, Journal of Interactive Marketing, Vol. 16, 2002, 59-75.
Li, S., Sun, B., and Wilcox, R.T., Cross-Selling Sequentially Ordered Products: An Application to Consumer Banking Services, Journal of Marketing Research, Vol. 42, 2005, 223-239. Lusch, R.F., Stafford, E.F. jr. and Kasulis, J.J., Durable accumulation: an Examination of Priority
Patterns, Advances in Consumer Research; Vol. 5, 1978, 119-125.
Moore, D.S., and McCabe, G.P. Statistiek in de Praktijk, 2nd edition, W.H. Freeman and Company, 1993.
Osborne, M.J. and Pitchik, C., Equilibrium in Hotelling’s model of spatial competition, Econometrica, Vol. 55, 1987, 911-922.
Paas, L.J., Molenaar, I.W., (2005), Analysis of acquisition patterns: a theoretical and empirical evaluation of alternative methods, International Journal of Research in Marketing, Vol. 22, 2005, 87-100.
Paroush, J., The Order of Acquisition of Consumer Durables, Econometrica, Vol. 33, 1965, 225-235. Rindskopf, D., A General Framework for Using Latent Class Analysis to Test Hierarchical and
Nonhierarchical Learning Models, Psychometrika, Vol. 48, 1983, 85-97.
Stafford, E.F. jr., Kasulis, J.J. and Lusch, R.F., Consumer Behavior in Accumulating Household Financial Assets, Journal of Business Research, Vol. 10, 1982, 397-417.
