1 University of Amsterdam
Direct impact of customer characteristics and moderator impact of mortgage characteristics on mortgage fraud detector
Submitted by: Vikas Gupta (UvA id: 11421738)
Course: Executive Programme in Management Studies ( Digital Business track) Thesis Supervisor: Dr. Andreas Alexiou
2
Statement of Originality
This document is written by Student Vikas Gupta who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
This thesis was done to study the direct impact of few selected customer characteristics and moderating impact of mortgage characteristics on the mortgage fraud. The dataset used in the analysis contained 5951 data points of the mortgage cases of a bank. It was observed during this research that there is a lack of research and articles on this topic even though it is a very important area for the financial industry, especially after observing number of financial turmoil so far in the current century. The driving force of this thesis was to put some efforts in extending the limited research done earlier. There were mainly four important findings of this thesis. First, there is a direct positive impact of the age of the customer on the mortgage default. Second, there is no direct relation between the income of the customer and the mortgage default. Third, there is a direct positive impact of the job tenure of the customer on the mortgage default. Fourth, the ratio of the mortgage amount requested to the amount needed to purchase the relevant property doesn’t moderate the direct relationship between the average salary of the applicant and mortgage default.
4
Content
1 Introduction ... 5
2 Literature review ... 9
3 Methods and Results ... 26
3.1 Measures... 26 3.1.1 Independent Variables ... 26 3.1.2 Control Variables ... 27 3.1.3 Moderator variable ... 27 3.1.4 Dependent Variable ... 27 3.2 Methodology ... 28 3.3 Results ... 29 3.3.1 Correlation analysis ... 29 3.3.2 Regression Analysis ... 30 4 Discussion ... 32
4.1 Major findings and contribution of study ... 32
4.2 Limitation and Future Research ... 36
5 Acknowledgement ... 37
5
1 INTRODUCTION
The financial industry has been dealing with the various kinds of frauds since many decades in the past. Every now and then, we hear about a new kind of fraud or a method that is intended to perform fraud. Data analyst, data scientist, software programmers and various other kind of professionals in financial industry have now started working together to deal with such kind of activities by massaging and analyzing the historical data and predicting the future. According to Abbasi et al (2012), financial fraud has also contributed to the
bankruptcy of the major organizations. They also mentioned the following statistics in support of this remark:
Financial fraud has also contributed to the bankruptcy of major organizations. Above stats clearly shows the most financially impactful frauds in the area of financial world. This problem appears to have been found in many developed as well as developing countries (Albrecht, Albrecht, and Albrecht 2008). Based on the research done by various associations and also by collecting many documents related to financial frauds around the world by them, it was estimated that the total sum of the losses is in trillion dollars. Adrian (2015) argued that companies internationally has been facing major problems related to economic fraud and crime. He further believed that fraud detecting and preventing process and technologies are now much needed than before eve to maintain the desired growth in terms of productivity and
6 sustainability. Shaun Hipgrave (2013) in his article, mentioned that the UK's National Fraud Authority in 2013 reported £52bn loss in terms of the amount the economy or entire UK that included not only the identified losses but also any hidden losses due to fraud. This figure can be further broken down into a cost of various sector like in public sector of £21bn, private sector of £16bn and remaining to others. There is also the threat of possible incoming
legislation with which businesses will need to contend. So the pressure is on for companies to find effective solutions to detect and prevent fraud which can cause significant harm to businesses and society as a whole. To counter the risk of attack from increasingly sophisticated fraudsters, organizations need to tackle fraud in a smarter way. Dwane Hal Dean (2004) mentioned more about the impact of fraud in significant amount. In dollar value, he referred to the fraud related to the insurance as the second most impactful financial crime in US, that was second only the fraud related to tax (Cams and Colin, 1997). Estimation of annual $100 bn related to fraud in insurance was identified, that was close to the 10% of entire sum of not only the claims but also the losses (Foppert, 1994). The prevalence of fraud varies by insurance line. Worker's compensation and health care account for the greatest amount of fraud with auto and property and casualty lines (Brostoff, 1996). The property and casualty insurance fraud, amounts to about $20 billion annually nationwide (Cams and Colin, 1997).
Kirlidog and Asuk (2012) described that huge amount of losses have mainly been caused by the fraud in the insurance sector. They also cited Gill et al. (1994), who mentioned about incorrect and forged claims by submitting falsely obtained claim or increasing the amount of claim. They further referred to the fraud in the insurance part of the industry in Canada that was estimated to be around CAD 1.3 bn that was close to the 10% to 15% of the entire claims in Canada that were paid out (Gill, 2009). It was found in their study that the that difficulty level of the detection method of fraud is quite high that increases the cost of the overall
7 process of investigating and detecting fraud. This causes many company to not to go deep in performing such a difficult and costly process.
Ravisankar et al. (2011) talked about the wrong as well as false usage of accounting ways in procedures that also contributes toward the fraud in finance. They took a closure look into the role of auditors in major scams of Enron and also WorldCom that were mainly caused due to the incorrect disclosures. Talking about the solution, Ravisankar et al. (2011) suggested to use the data mining techniques to take some pressure off the shoulder of auditors to do perform more effective audit tasks .
Barnes and Webb (2007) also argued that big size firms have a greater degree of chances of being hit by the financial theft and also fraud. Lin et al (2015) researched using the data mining methods like neural network, regression (logistic) and decision trees to more cautiously analyze the historic data and make the more appropriate predictions. There are various types of frauds that has so far been detected and analyzed. Ngai et al (2011) used data from FBI to categorize the types of financial fraud to come up with following financial fraud categories and activities:
Following prior studies that investigated only partial relation, the main focus of this thesis is to cover most important features of a mortgage customer as well as mortgage itself and to see if those features could be used to predict whether or not the mortgage would turn out to be a default, which is a big risk for lenders and financial companies. This thesis paper is focused
8 on finding the relation between the customer characteristics and mortgage default, and also the level of moderation influence of mortgage characteristics on this direct relationship. This thesis investigates whether the customer characteristics has any positive contribution towards having a mortgage of being a default case. It also investigates whether this contribution becomes stronger with the increasing level of few mortgage characteristics. Specifically, this thesis found that the higher level of age and lower level of the job tenure of the customer could increase the chance of the mortgage default. It was also observed that the ratio of the mortgage amount requested by the customer to the amount needed to purchase the relevant property doesn’t moderate the direct relationship between the average salary of the applicant and mortgage default. Also, the direct relation between the income of the customer and the mortgage default could not be proved.
Among the various types of bank frauds, the mortgage frauds are considered as the most impactful as well as painful for the world economy. Considering the importance of mortgage fraud management in financial institutions, this thesis would try to contribute to the area of identifying the relationship between certain characteristics and the mortgage default. The research question of this study is: What is the direct impact of customer characteristics and moderating impact of mortgage characteristics on the mortgage fraud? The aim of this study is to detect and analyze this impact and then judge if this research question can be answered. The next section will be on the literature reviewed and referred for this thesis along with the description of the hypothesis tested. Subsequent sections would be on the measures used to perform this study, description of used research methodology and results observed. The last section would be on the discussion relevant to this thesis.
9
2 LITERATURE REVIEW
The previous section of this paper briefly touched upon the idea of fraud, definition, various types, specifically the mortgage fraud. In this section, the idea of fraud would be further detailed out in context of this study along with the hypothesis development. Mortgage loan or simply a mortgage is the term used when any person wants to borrow money from a lender like bank, financial institution etc. to buy a property. It is usually a very common practice for an applicant to apply for a mortgage to a lender. If the applicant’s mortgage application is accepted and processed by the lender then the lender lends money to the applicant. In return, applicants pays the periodic, most probably monthly, mortgage interest to the lender apart from the original amount repayment. The lenders make money by charging interest from the applicant, and applicants usually get tax benefits based on the amount repaid to the lender. Usually the lender screens the mortgage application of the applicants thoroughly in a series of steps like personal information verification including, address, contact details etc.; financial information verification like salary, bank deposits, other property values, expenses etc.; credit information check from central bank or credit agency etc. After completing all these steps and checks, a credit score is calculated and assigned to the applicant that decides if the applicant is worth enough to borrow money, and also the mortgage amount limit upto which he can borrow money from the lender. Lender also calculates the monthly repayment amount that consists of the principal and interest amount on the remaining unpaid amount. If
everything goes as planned and in normal way, the mortgage customer, that was originally the applicant of the mortgage, repays all the amount to the lender through monthly repayment amount or a lump sum amount. But sometimes, the mortgage applicant or the mortgage agent, who helps the applicant to apply for the mortgage, performs some undesired activities that leads to the mortgage fraud. For example, the mortgage applicant mentions in the application that mortgage is being applied to buy a primary property for residence, although
10 the actual intention was to buy a property for the investment. This is a fraud because the interest rate of primary residence property mortgage offered by the lenders are usually lower than the mortgage for the investment property, and so in this case, the applicant is presenting wrong information to get the mortgage on lower interest rates although the applicant is ineligible for it. Other kind of mortgage fraud by the applicant could be the inaccurate or higher than actual income mentioned in the mortgage application. This makes the applicant eligible for the higher mortgage amount, which actually should not happen. Such applicants use the extra amount for some higher return investment illegally, which was not the actual purpose of borrowing money from the lender. This also gives them higher tax benefit. Other kind of mortgage fraud is when the applicant incorrectly mentions his higher position than actual in the mortgage application, which could make him eligible for the higher mortgage amount. Some applicants also don’t mention their liabilities and expenses correctly, such as information on existing other mortgage or credit card. This also helps them to make
themselves eligible to borrow higher mortgage amount, which is definitely a fraud.
Sometimes the price of the property is intentionally manipulated or increased to apply for a higher mortgage amount. This can be done in hidden consensus between buyer, seller and the property agent. In the same way, the price of the property sometimes is also intentionally decreased to get lower mortgage interest. Another instance is when more than one mortgage applications are submitted and processed for the same property to get the much higher mortgage amount than the actual price of the property. This poses a great amount of risk to the lenders as there is a high chance that lender won’t be able to receive back all the lent amount. One more example in the same context could be when an applicant claims to be a different person by illegally forging some documents and submitting them along with
mortgage application. This is done mainly to become eligible for the higher mortgage amount or to deliberately cause a problem or inconvenience to another person. In such cases, the
11 applicant doesn’t repay any amount after borrowing the mortgage amount. Such kind of mortgage fraud causes huge loss to the financial institutions, and has been one of the biggest contributors in the recent and historical financial crises all over the world. It is not uncommon to see the participations of the employees of the financial institutions in such frauds. Jiang et al (2014) argued that a big reason of the fraud in mortgage sector of financial world is caused by the inefficient and flawed of checking the eligibility of the mortgage applicants. The documents and information submitted by the mortgage applicants like monthly earnings, expenses and other liabilities are not properly checked and verified by the lenders during the screening process. Srivastava and Gopalkrishnan (2015) stressed upon the concept of using the big data techniques in scanning and analyzing the huge stock of historical data that banks and financial firms usually keep in their archived and back up storage. This historical data usually contains lot of unprocessed information that can be synthesized using big data tools and techniques to extract and conclude the meaningful tasks and action points for the future course of actions. Wilhelm (2004) came up with following approximate statistics on the losses due to fraud on various industries.
He researched on fraud management and argued that the fraud Management Lifecycle is fast moving and changing very rapidly, and its associated eight stages are: “Deterrence,
Prevention, Detection, Mitigation, Analysis, Policy, Investigation, and Prosecution”. Wilhelm believed that the best way of managing and handling fraud needs lot of effort in creating a right amount of balance between various factors that not only include the internal factors but
12 also the external factors. He mentioned this scenario by giving an example of the situation as: “While fraud does exist in retail originations, it is typically related to a particular loan officer and is more often than not quickly discovered. The employee is usually terminated from his [or her] position and moves on to a new company until the same thing happens all over” (Prieston and Dreyer, 2001).
Among all these aforementioned categories of frauds, the main focus of this thesis would be on the mortgage fraud. Ngai et al (2011) mentioned mortgage fraud as “material
misstatement, misrepresentation, or omission relating to the property or potential mortgage relied on by an underwriter or lender to fund, purchase or insure a loan”.
In order to detect mortgage fraud, many banks and relevant financial companies have now started adopting various technologies including data analytics and predictive modelling to detect the fraudulent cases in much earlier stage.
Smith (2010) tried to find the main contributing factors in the fraud cases in mortgage sector. He highlighted three associated factors. First, due to the rise in the usage of new technologies, there is usually a huge physical distance between the mortgage provider and mortgage
applicant, and also between the mortgage provider and the associated asset or property. Second, there is usually a huge physical separation and the lack of contact and meetings between the mortgage provider and applicants. Most of the mortgages are applied, screened and approved without the mortgage provider and the mortgage applicants ever had a chance to meet personally, and it all happens mainly due to the extended usage and presence of brokers, third party intermediates or companies etc. Third is the mix of factors and general decisions taken by the lender and the mortgage applicants that varies a lot.
Nguyen and Pontell (2010) conducted a study with its aim to find out how the fraud in mortgage sectors is causing a number of financial crises of huge volume. The study
13 conducted by them tried to probe the role of internal employees or insiders in the financial companies as a major contributing factors to the fraud in mortgage sector.
As per report from PwC on Altruistic fraud (2017), the forensics practitioners and auditors frequently refer to the so-called “fraud triangle” framework — generally, the preconditions necessary for an individual to commit fraud. Those three factors are pressure (or incentive); opportunity; and rationalization. These three factors are usually found to be the underlying reason of the fraud related activities. If rationalization is a precondition for most people to commit fraud, then they must have a privately credible excuse for doing so. In other words, they must have convinced themselves that what they were doing was somehow right, or at least justifiable. Collusion, and the ability of management to override controls, means that no system can fully prevent fraud. (For example, senior executives — those most likely to commit large-scale accounting fraud — are often well placed to collude with others and override such controls). Sometimes, the reason of committing an activity related to fraud is also motivated by the rising internal pressure of some kind, that make people to perform undesired act leading to fraud.
Barnes and Webb (2007) mentioned in their article that the participation of high level top employees in the fraud related activities may largely be motivated by their stakes or
investment in that particular public company. They mentioned that the way of working and the policies of the company also either motivates or demotivates the employees of the company to commit fraud. For example, if the policy of the company against fraud is very strict or if the level of punishment is too high then it can make employees to not to involve in fraudster activities. They referred to Cullinan and Sutton (2002) and assumed that the
position or level of the employee in the company to perform fraud also impacts the
14 board members high up in the ladder of corporate hierarchy have greater access to the
resources, that can be exploited to commit fraud or do some activities that leads to fraud. The other factors Cullinan and Sutton (2002) mentioned in similar line could be the number of people involved in this fraudster act or the type of fund or asset that might be on the stake to get advantage of fraud by the people who perform the act of fraud.
For example, Nguyen and Pontell (2010) further argued that it is very complex to judge or differentiate between the very narrow difference between the act of fraud or the activities that are unethical, which may lead to fraud further down the timeline. According to the Mortgage Bankers Association (2007), there are some unethical act of mortgage application, screening or approval that are considered as an act of fraud. For example, some conditions can be crafted that could lead a mortgage applicant to go for the mortgage products that have very high interest mortgage rate or mortgage application plus processing fee. Other example Mortgage Bankers Association (2007) presented was that sometimes the mortgage provider itself try to attract the mortgage applicants by misleading them in the form of incorrect mortgage product publicity or features that lead to getting the applicants enrolled and subscribed for the undesired mortgage registration.
Schloemer, Ernst, and Keest (2006) mentioned that sometimes aforementioned practices of incorrect mortgage selling or purchasing is hard to predict, detect and prove. They mentioned an example that, sometimes the broker or the agent of loan application perform fraud
activities on both lender and mortgage borrower side. That agent can either mentions the fake or incorrect terms to the mortgage applicant or he can forge incorrect income or liability or asset information of the mortgage applicant to get the benefit and get qualified for low interest mortgage.
Abdallah et al (2016) highlighted the usage of various technologies that could detect the fraud activities as soon as someone tries to perform it in the system. They mentioned that the
15 detection could be done using the big data or data mining technologies that can be plugged with other data analysis and decision making tools to make the correct judgement on fraud. West et al (2016) describes the act of financial fraud as an event that could lead to a huge impact on social and financial world. They stressed that the fraud can affect the life of million of customers by having serious impact on the daily and financial way of living. They
mentioned that the conventional way of detecting fraud using inspection and audit won’t work in present era of typical frauds. They suggested using analytical and latest technical tools using high computational power to detect the fraud as soon as possible.
16 Humpherys at el (2011) mentioned fraud as form of an act that aims to produce or present wrong or virtual information that actually doesn’t exist. This false information can produce such a scenario that could lead to fraud. Error can also be one of the reason of producing incorrect information when the intention of the person or employee of the organization is actually not to commit any fraud. If the impact of the fraud is too high due to production or reporting of incorrect or fabricated large scale accounting or profit/loss information, then it is termed as the fraud at management level.
Ravisankar et al. (2011) also highlighted the importance of accounting reporting or periodic reporting done by almost all the organizations. They mentioned that the management level
17 fraud manipulates the high level organization reports to create a wrong sense of performance or profit/loss image of the organization in front of investors and stakeholders, which creates incorrect market sentiments towards the company. They further mentioned that even the company’s financial status and periodic reports are used by financial institutions or lenders to provide loan to the organizations, and by manipulating these periodic reporting numbers, organization commit fraud.
Even a study was done to attempt to find mortgage default using the search queries entered by consumers on Google. Chauvet et al (2016) used search input data on Google engine to develop a model of risk of default in mortgage. They mentioned that such data was used an input in predicting the various factors and indicators that could be used to conclude whether or not the customer has any tendency or probability to default the mortgage. Kirlidog and Asuk (2012) stressed on the importance of using the data mining tools in fraud detection. They mentioned that by analyzing the data of the pattern of the claims of insurance to cover health issues, it can be predicted to conclude whether or not the insurance covered customer is likely to be a defaulted customer or not. They suggested to do it by analyzing the data with an aim to find particular patterns that can be generalized to the whole population of customer to make the correct judgement.
Glancy and Yadav (2011) highlighted in their article that many US firms had been struggling to find the effective way of detecting fraud rather than relying on traditional way of auditing and inspecting the processes and documents. They argued that the computational and
analytical techniques using automated detection of fraud and anomaly should be an effective way to handle the fraudulent cases and issues. They also presented an automated model in their article that could be used to design the automated method of detecting and screening the cases of fraud.
18 Humpherys at el (2011) also stressed on using analytical techniques using big data probe, and also other new methods of mining the data and text to detect and eradicate fraud. Referring to healthcare industry fraud, for example, Raghupathi et al (2014) suggested measure for finding and detecting fraud in the reported and submitted receipts and claims. They also suggested to use the combination of mobile app and analytical capabilities to probe and detect the claim related fraud much earlier in the whole process to avoid the loss in huge volume and also to cover their own risks. Further explaining the process of the detection of default and probe, they mentioned that such process takes huge amount of data as an input to do the calculations and show the output in numerical format, mostly binary to show whether or not the case is of fraud type.
This thesis was focused on the relationship between this fraud detector value and the corresponding inputs, which in this case are the certain characteristics or factors of the customers and mortgage. Seiler (2016) conducted a study on mortgage default. He explained that there are basically two types of default with mortgages, one is when the mortgage
customer himself decides not to pay the monthly installment to the mortgage provider / lender when his intention is to keep on staying in the property without spending money on rent, which is an illegal or unethical thing to do. Seiler mentioned that the second type is when the mortgage customer himself is not in good position to payback the installment to the mortgage provider or lender any more. Seiler mentioned that former type of default in mortgage sector is also very significant because of the success rate of such defaulters in eventually ending the case in their favor due to the inefficient or incompletely trained legal department of the mortgage providers.
Adrian (2015) mentioned in his study that there could be various types of data analysis techniques, that can broadly be defined in 2 ways. One is the analysis of strategic type, and
19 other is of operational type. Adrian concluded and segregated all the analysis type in
following categories:
Adelino (2016) in his study tried to understand the reasons of financial turmoil that happened in the latter half of the first decade of this millennium. He mentioned that the technology used to originate mortgages back then was not so fool-proof. So it helped many under-qualified mortgage applicants to get the approval of mortgage, that skyrocketed the property price, and then eventually the whole mortgage sector crashed due to lack of balancing power or
processes. Their conclusion was major research done by Mian and Sufi (2009) that there was a reverse correlation between the increase in mortgage amount disbursal and the earning growth at per capita level, that raised the questions on the judgement criteria of mortgage companies to approve the mortgage applications of applicants with insufficient mortgage qualification. On the flip side, Adelino (2016) added some more important points in his research. In contrary to general believe that the lower income mortgage customers were delinquent that triggered the financial crises due to the collapse of mortgage sector, he concluded that most of the delinquent mortgages near the period of financial crises were also related to the mortgage obtained by high and medium income class applicants. Adelino also found that there was an overall increase the mortgage applications and disbursal to all the income level classes including high, medium and low income.
20 Mian and Sufi (2009) (henceforth MS09) also found in their research that there was a
significant rise in the mortgage applications in the area in US where majority of the people living were in low income category. They added that there was also a strong interest from the mortgage providers or lenders to approve mortgage applications of such applicants who were denied the mortgages earlier, So rather than having a general idea that there was an
improvement in the economic conditions of the low income people, the actual condition was that there was a massive level of mortgage fraud going on with the rise in the number of applications for mortgage when the income level and qualification to obtain the mortgage was on decline.
Now to move towards the main topic of this thesis, let’s first see the conceptual model that has been referred in the entire analysis:
This thesis has considered few customer and mortgage characteristics based on the mortgage data collected from a Dutch bank. Using this data, following are the variables:
21 1. Independent variables: Age (of primary mortgage applicant), Average salary of the primary and secondary mortgage applicants, Job tenure (of primary mortgage applicant in months). 2. Control variables: Gender (of primary mortgage applicant) and Nationality (of primary mortgage applicant).
3. Moderator variables: Categorical variable of the ratio of the mortgage amount requested to the amount needed to purchase the relevant property.
4. Dependent variable: Mortgage fraud (Yes/No).
Taking these characteristics as input values, the direct impact of customer characteristics and moderator impact of mortgage characteristics on mortgage fraud detector is depicted in this thesis. In order to formulate the impact, following hypothesis would be tested:
H1: Age of the primary customer positively relates to the mortgage fraud.
The first hypothesis mentioned above is to test the direct impact of one of the customer characteristics, which in this case is age of the customer, on the mortgage fraud. Jiang et al (2014) analyzed the data of thousands of loans disbursed by a US bank between 2004 and 2008. In their study, they highlighted two problems underlying the mortgage crisis. First was the reliance on mortgage brokers who tend to originate lower quality loans, and second was the prevalence of low-documentation loans that result in borrower information falsification. This kind of feature or characteristics of borrower is termed as customer characteristics. Some examples of the customer characteristics are age, gender, employment status, type of customer (first time customer or existing) etc. In addition to this, they also concluded based on the loan data that that third-party-originated loans are more than 50% more likely to be delinquent than bank-originated loans. This kind of feature or characteristics of loan is termed as loan or mortgage characteristics. Some examples of such characteristics are loan type
22 (individual or corporate), LTV (Loan-to-value ratio), AddLTV (LTV of additional loan), prepayment penalty, tenure of loan etc. Mayer et al (2009) tried to find the contribution of product type to mortgage defaults. They also blamed poor underwriting practice as one of the reasons of mortgage default. They argued that the inefficient underwriting practice allowed many undeserved mortgage applications to get the approval and got the mortgage amount disbursed, which also allowed the mortgage customers to not to pay the advance payments for the property that further increased the chances of default. Mayer et al (2009) mentioned that the increase in the level of early loan repayment default as well as the insufficient documentation for the mortgage application laid a perfect conditions for the financial crises, especially in the mortgage sector with number of customers intentionally defaulted as
customers thought that it was not worth enough to repay the mortgage amount to the lenders. Their study also mentioned that the mortgage product features including the delinquency interest rates, repayment frequency of the mortgage amount etc. also contributed in creating the favorable conditions for the default in mortgages. In order to recommend financial organization to strengthen their underwriting process and mortgage application evaluation process, this thesis has tried to check the impact of few specific customer characteristics on the mortgage default, and assumed that higher age of the customer can lead to greater chance of the mortgage default. Hence H1 has been hypothesized to focus on the age factor of the customer characteristics.
H2: Average salary of the primary and secondary mortgage applicants positively relates to
the mortgage fraud.
The second hypothesis mentioned above is to test the direct impact of customer’s income on the mortgage fraud. This hypothesis was based on some previous studies. Mian and Sufi (2017) demonstrated that one of the key reasons of fraud in mortgage sector is also due to the
23 incorrect or more than real earning and income declared or reported by the mortgage
customers in the form they fill to apply for a mortgage. They mentioned about a case of an area in a country where the growth rate of the mortgages was quite drastic that coincides with the declaration of high income in the mortgage application by the applicants although that that arear actually didn’t have such real growth. This kind of situation is a clear signal of massive scaled fraud. Blackburn and Vermilyea (2012) also found few similar cases in their research that concluded to the increasing rate of declaration of higher earnings and income was in line with the increasing rate of mortgage applications. But at the same time, the actual income of such mortgage applicants was in fact decreasing as per the official records. They found that such a situation was a good indication of the correlation of the factors between increasing declaration of higher income to the default or delinquent customer rate. Blackburn and Vermilyea (2012) also mentioned that many relevant earlier studies pointed and hinted to the scenario that when there were more than one factors that must have contributed to higher degree and rate of default in the mortgage sector, especially like the incorrect model used for the mortgage processing or even the incorrect implementation of the models. They also used data used by local government agencies to check and prove the correlation of the factors as mentioned earlier. Considering the importance of the income in the decision making process of the mortgage, this thesis has used salary as an component to test the hypothesis H2.
H3: Job tenure of the primary customer negatively relates to the mortgage fraud. The third hypothesis mentioned above is to test the direct impact of another customer characteristics, which in this case is the job tenure of the customer, on the mortgage fraud. This hypothesis was based on some previous studies. Mortgage fraud is a generic term for many types of fraud involving misrepresentations (T. Dietrich Hill, Columbia Law Review Vol. 113, 2013) to obtain a mortgage. It could either be done by a mortgage applier who
24 presents incorrect facts about his job type, job tenure, salary or debts, or by a broker by
performing activities to visualize higher value of the property to get additional mortgage. Goodman and Smith (2010) also tried to study the impact of customer characteristics (FICO score) and some mortgage characteristics on loan default. In their study, they used a variety of data arranged from the domestic regional government entities, which they used to shortlist and use few important personal factors like the age, job tenure etc. of the mortgage
applicants, and they also considered the model used in the mortgage processing as
hierarchical model which contributed to the unethical or unexpected practice in mortgage approval and business. Their research results suggested that due to the very expensive nature of early closure of the mortgage due to high amount charges, there was a decline or very lower rate of the pre closure rate of the mortgages. There was not enough motivation factor or incentives in terms of worth for the mortgage customers to pay back all the remaining
mortgage amount and close the mortgage much earlier before the agreed tenure. They further suggested that the mortgage application and approval process should introduce such
encouraging factors into the mortgage lifeline to encourage a mortgage customer to see a benefit in paying back all the money to the mortgage provider and also realize such benefits. Smith (2010) conducted a study on Mortgage fraud and argued that the default in the
mortgage sector is driven by the combination of various factors or also by some individual high impact factors which leads to an intent or motivation of committing an unethical or illogical action that leads to the default, especially in the mortgages that are applied with an intent to buy a house or other property.
Inline to the hypothesis H1, this thesis has tried to check the impact of few specific customer characteristics on the mortgage default, and assumed that longer job history or tenure of the customer can lead to lesser chance of the mortgage default. Hence H2 has been hypothesized to focus on the job tenure factor of the customer characteristics.
25 H4: The positive relationship between the average salary (of the primary and secondary
mortgage applicants) and mortgage fraud is moderated by the ratio of the mortgage amount
requested to the amount needed to purchase the relevant property, so that this relationship is
stronger for higher levels of this ratio.
The fourth hypothesis mentioned above shortlisted was to use the data and check the moderating influence of the mortgage characteristics upon the direct influence or relation between customer characteristics and the mortgage fraud. Anderson and Dokko (2016) tried to find out the impact of certain loan characteristics, like due date of loan repayment, on loan delinquency. They concluded that loans facing a due date within one to three months after origination have about 2% to 6% higher first-year delinquency and default rates than loans that face a property tax due date ten to twelve months after origination. Further on the exploration of the impact of mortgage characteristics, Posey and Yavas (2001) in their research concluded that the mortgage applicant’s selection of the mortgage product type can also be used to evaluate the risk in processing and approving the customer’s mortgage
application. For example, they mentioned that there are mainly two types of mortgage interest rate pattern; one is the fixed type which means that the interest rate would not change during the entire tenure of the mortgage repayment. This type of mortgage product feature is mainly chosen by the mortgage applicants that don’t want to take any risk. Such customers are put in the bracket of marginally lower risk taking customers. They mentioned that there is another type of mortgage product in which the interest rate changes as per the interest rate decided by the mortgage provider as per the changing market conditions. Such as variable type of
interest rate mortgage product is chosen by the mortgage applicants who like to take some risk and would like to get benefit out of the varying market conditions. Such mortgage applicants are put in the bracket of the ones who like to take some risk and would like to see
26 the positive effect of the market conditions on the mortgage interest rate. Xi and Yang (2009) studied how the loan default can largely be explained by various characteristics associated with the loan, the underlying property, and the default, foreclosure, and settlement process. They specifically found that the most important factor that could determine the loan default is current loan-to-value ratio, which is one of the mortgage characteristics.
These mortgage related factors represents another set of key elements that seems to have found aforementioned studies as contributors to mortgage default. It motivated this thesis study to also focus on the mortgage characteristics and check its moderation effect on the direct relationship between customer characteristics and mortgage default. Assuming mortgage amount and the actual cost of the property as the key mortgage customer characteristics, the aforementioned hypothesis H4 has been hypothesized.
3 METHODS AND RESULTS
This chapter explains the research design and approach of the thesis. The first section focuses on the description of independent, control, moderator and dependent variable used in the thesis. Then, the following section describes the steps taken and method followed for the data analysis. The last section describes the results observed during the data analysis.
3.1 MEASURES
3.1.1
Independent Variables
This section contains description of variables. First independent variable is the age of the primary applicant. This is a continuous numeric variable of scale type. The age of the
secondary applicant is not included in the variable set as this data was not available for the all the mortgage records. Focus of research is mainly on the characteristics of primary applicant unless it has financial factors. Second independent variable is the average salary of the
27 primary and secondary applicant. As salary is the most important financial factor available in the dataset, so it was decided to include the average salary of both primary and secondary applicant. Further, this average salary is recoded into a categorized variable of ordinal type based on the salary range to make the data more meaningful. The last independent variable is the job tenure of the primary applicant only. This variable is also recoded into a categorized variable of ordinal type based on the range of the job tenure in months. The reason of not including the job tenure of the secondary applicant in the variable set is same as
aforementioned for the age.
3.1.2
Control Variables
The variable set include two control variables. One is Gender of the primary applicant. This is recoded into a nominal variable with 0 for male and 1 for female. Other control variable is nationality of the primary applicant. The variable is also recoded into a nominal variable with 0 for Non-Dutch and 1 for Dutch applicant.
3.1.3
Moderator variable
The moderator is the ratio of the mortgage amount requested to the amount needed to purchase the relevant property. Both amounts in the original dataset received are in Euro. This ratio is further recoded into a nominal variable with 1 for the cases for which mortgage amount requested is less than or equal to the needed to purchase the relevant property, and 2 for the remaining cases. This was done to divide the applicants in the 2 groups based on whether applicant requested more or less than the actual amount needed. This grouping is playing a role in hypothesis H4.
28 The dependent variable is the nominal type variable, called mortgage fraud, with 0 for Non-default cases and 1 for Non-default cases.
3.2 METHODOLOGY
The dataset used in the analysis contains 5951 data points of the mortgage cases in a bank. It contained wide variety of cases with varying ages (18 to 75 years), salary, job tenure etc. SPSS was used for the whole data analysis for this assignment. The first step done was the missing value check. Surprisingly, there was no missing value found. In the frequency check, some negative values were found in the job tenure column. So as part of data cleaning, those negative values were corrected into their absolute values as the job tenure cannot be negative. Some skewness was also noticed in the dataset during skewness check. In order to deal with it using normalization techniques, some recoding of the variables was done to create
categorical variables, as mentioned in above section, for variables like Gender, average salary, job tenure and ratio of amounts. A new variable for age was also computed as the square root of the original age variable. The remaining skewness, that was still noticed after these measures taken for the average salary and ratio, can be explained by the fact that most of the applicant fell under the lower to mid category of the salaries, and only some percentage of them were at higher end of the categories. Same applies to the ratio as most of the
applicants either request less than or equal to the actual amount needed to buy the property. So the skewness in the dataset is explainable. Also, the check of reliability and counter indicative items was not performed as the dataset used is a secondary dataset, so these checks were found to be of much relevance. In order to test the moderation effect, a new interaction variable was computed as the multiplication of the categorized variables of the average salary and the ratio. After this, the next step performed was the descriptive analysis and bivariate correlation analysis. This was done among the independent, moderator and dependent
29 variables. The result will be explained in the next section. The last step of hypothesis testing was done using regression analysis to check the direct effect of the independent variable on the dependent variable, as well as the moderating effect of the moderator. The result is also explained in the following section.
3.3 RESULTS
This sub-chapter explains the result of data analysis. The first subsection covers the result of correlation analysis. The next subsection covers the results of regression analysis.
3.3.1 Correlation analysis
Table 1 below shows the matrix of stats of descriptive and bivariate correlation analysis of the dataset. As per the SPSS results extracted during the analysis, the Pearson product-moment correlation cofficient is the measure used in this paper to measure the linear correlation.
The last row in Table 1 gives us an impression of the correlation between the dependent variable and other variable. It shows that the dependent variable Mortgage fraud, with (r = -0.107, p < 0.01) is not correlated to job tenure of the primary applicant. On the other hand,
30 with stat (r = -0.028, p < 0.05), the dependent variable Mortgage fraud is highly negatively correlated to average salary of the primary and secondary applicants. After this, looking at the correlation between dependent variable and control variables, it can be interpreted that with stat (r = -0.179, p < 0.01), there is a high negative correlation between mortgage fraud and nationality of primary applicant. Also, with stat (r = 0.035, p < 0.01), it shows the absence of relation correlation between mortgage fraud and the gender of the primary applicant. The other correlation stats of the Mortage fraud with other variables including the moderator is not significant, and so doesn’t indicate any correlation.
3.3.2 Regression Analysis
This paper used hirerarichal linear modelling regression technique to perform the hypothesis testing. The Table 2 as shown below summaries the stats extracted from the regression analysis done on the dataset using SPSS. It is analyzed to interpret the direct effect of independent variables on dependent variable, as well as the moderating effect of the moderator.
31 In the regression options for missing values, it was opted to exclude cases listwise. The regression analysis examined the linear relationship between dependent variable (Mortgage fraud), independent variables (age, average salary, job tenure) and moderator (ratio of mortgage requested amount to the actual amount needed to buy the property). The above table contains the cofficient estimates and White’s heteroscedasticity-consistent standard errors of the ordinary least squares regression for the mortgage fraud. Step 1 shows a foundation model that considers only the two control variables to make sure that these variables don’t explain the whole relationship between independent and dependent variables. Step 2 includes the main effect of independent variable along with the ratio of requested amount to actual amount. Step 3 includes the variable to analyze the interaction of the moderator variable with independent variable to interpret the moderating effect. To interpret the direct relationship in step 2 shown above, the stat (β = 0.05, p < 0.001) supports the H1 that the age of the primary customer positively relates to the mortgage fraud. But with stat (β = -0.010, p > 0.05), the relationship of the average salary of the primary and secondary applicant with mortgage fraud is not significant. So it can be interpreted in relation to the hypothesis H2, that the average salary of the primary and secondary mortgage applicants positively relates to the mortgage fraud is not supported. With stat (β = -0.106, p < 0.001), there is a strong support of H3 that the job tenure of the primary applicant negatively relates to the mortgage fraud. Now coming to the last hypothesis H4, with stat (β = 0.010, p > 0.05), there is no support of the moderating effect of the ratio on mortgage fraud. So it cannot be stated that the positive relationship between the average salary (of the primary and secondary mortgage applicants) and mortgage fraud is moderated by the ratio of the mortgage amount requested to the amount needed to purchase the relevant property. So the dataset doesn’t support that direct relationship is stronger for higher levels of this ratio. The overall strength and robustness of the model measured by the R2, coefficient of determination, is relatively
32 low varying from 3.3% to 4.4%, being the highest in the final model (Step 3). Therefore, it is concluded that more than 95% of the variation in mortgage fraud remains unexplained by the model. This would be considered as the limitation of this paper, and should invite a further research on this topic. The main reason of such an unexplained variation could be given to the limited set of characteristics found in the available dataset. Perhaps further research could bring in more customer and mortgage characteristics to support the hypothesis tested, as well as test some other hypothesis. I believe that this thesis would be a further effort in
determining the importance of certain factors in fraud detection, that might be of some help to banks and other financial institutions for further trial and research. This thesis could also be considered as my effort to add one more research in the area of fraud management that has been of key importance to banks although have not received enough attention from the literature.
4 DISCUSSION
The first section of this chapter is on the discussion of the thesis is to review and explain the major findings and the contribution of this thesis to the current literature. The second section outlines the research limitations and recommendations for future research.
4.1 MAJOR FINDINGS AND CONTRIBUTION OF STUDY
The research focused on the effort to detect mortgage fraud based on its direct relation with few customer characteristics and mortgage characteristics. It was observed during thisresearch that there is lack of research and articles on this topic, although it is a very important area for the financial industry after observing huge financial turmoils so far in the current century. The driving force was to put some efforts in extending the limited research done earlier. There are manily four important findings of this thesis. The first finding is that the
33 dataset proved a direct positive impact of the age of the customer on the mortgage default. For any lender, this could also be considered as a contribution in improving the fraud
management process as they can strengthen their underwriting process to pay attention to the age of customer as an important factor that could impact the mortgage approval process. The second finding was that the direct relation between the income of the customer and the
mortgage default could not be proved. There was not enough support rom the dataset to prove this direct relationship. But even then, the income of the customer could be an area of interest to the underwriters to be extra cautious while making the judgement. Nguyen and Pontell (2010) in their study came up with some recommendations for the prevention of the fraud in mortgage sector of financial world. They highlighted the importance of first understanding the culture and activities that encourages such kind of illigal or unethical practices that lead to the fraud, and then based on such understanding, it is important to setup and build up the strategy and plan to deal with such practices to handle the problem of default and fraud. They recommend to have a such strategy that not only completely eradicate the fraud related activities but also deal with any activity that could lead to difference in practices of making a deal between mortgage seeker and provider by creating some encouragement in the form of incentives for not engaging in the wrong set of activities. They mentioned that the mortgage providers should improve their process and practice of the evaluation and initial screening of the mortgage applications in an efficient and more vigilant way, especially the screening and double checking of the documents that depicts the financial standing and position of the applicant of mortgage, which can be done by faciliating the use of unbiased or third party individual or organization that can critically and thoroughly evaluate such applications and also audit them to find some flaws as well as suggest the areas that can be effectivily improved with a well built plan. Their further recommendation was to stop selling some mortgage products that encourages or made it easier for the people, like the brokers or
34 applicants of the mortgage or even the employees of the mortgage providing organizations to engage in morally incorrect or illegal activities that leads to fraud., and also to create
awareness among such people by continuously facilitating the learning seasons or by
radiating the right set of material and information to discourage unethical practice, and also at the same time highlighting the profit or intellectual or personal gains on following the rules and correct procedures. They also proposed to have right set of security measures to provide minimum access and processing rights on the documents and information that is sensitive for the mortgage processing to the undesired people. Nguyen and Pontell (2010) conluded in the end that by following such recommendations or practices based on well formed vision and strategy, it would be easier for the mortgage providing organizations to create a healthy and ethical culture of practices, and also difficult to encourage the usage and practice of illegal activities. Jiang et al. (2014) provided some numbers and proofs to believe that the earning and financial position mentioned by the applicants of mortgage were not near to the expected figures to justify their eligibility for obtaining such mortgages. They used some methods and models to see the relationship between the higher earning or incomes and the lower rate of default in the mortgage sector, but they could not establish the similar relationship or effect between the lower earnings and the higher rate of default in mortgage, perhaps due to many other factors that plays an important part while deciding the default behavior of the applicants of mortgage or the people involved in the mortgage providing side.
The third finding of this thesis is that the dataset proved a direct positive impact of the job tenure of the customer on the mortgage default. Again, for any lender, this could also be considered as a contribution in improving the fraud management process as they can
strengthen their underwriting process by paying attention to the job tenure of the customer as an important factor that could impact the mortgage approval process. Garmaise (2015) found two important points during research on borrower misreporting. The first one was that the
35 mortgage providing organizations didn’t have any prior strategy or awareness about the incorrect practice being followed by the mortgage agents or even the applicant to wrongly mention their financial position while applying for the mortgage, and due to this lack of awareness, those mortgage providing organization did not have any process to increase the charges of availing such loans to cover the uncertaining or risk to become default. The second point in Garmaise (2015) research was that there was no gains or benefit noticed for the applicants of the mortgage who represented their financial asset position over and above the desired or elgible limit. It was also found in their research that some applicants were found to be real in representing those numbers, but at the same time, there were applicants who
intentionally represented incorrect numbers to prove their eligibility in the application process of loan, although in reality, they were not worthy of obtaining the requested loan, which is not a ethical practice.
This thesis considered the ratio of the mortgage amount requested to the amount needed to purchase the relevant property as an important mortgage characteristic for mortgage default detection. The fourth finding of this thesis was that this ratio doesn’t moderate the direct relationship between the average salary of the applicant and mortgage default. Generally, these factors individually are always checked before making the mortgage approval decision. The lenders can also consider the moderating effect of such important factors on default detection. Abdallah et al (2016) stated that the requirement to tackle and handle the problem of fraud is the real requirement of the current financial situation. They suggested to use the analytical strategy and methods to deal with such problems by implementing the well known data analysis and machine learning concepts with the help of tool and softwares. They further suggested that using such concepts, it is not only important to prevent the fraud in the
mortgage sector but also to detect such tendencies in much earlier stage to avoid any rise in cost for dealing with such sensitive financial situation of crises. As per Hoffmann and
36 Birnbrich (2012), it is important for the mortgage providing institutions to prevent the frauds from happening by using data analytics techniques to keep and improve their dealing and relation with their clients. Such opportunities enables the mortgage providing instituions to bring up the level of trust and faith that their clients have on such institutions (Guardian Analytics, 2011). Wilhelm (2004) proposed and advocated the use of the fraud management lifecycle to manage fraud in relevant industries. He researched and found that there is a great significance and importance of using such a lifecycle with its benefit of bringing down the overall cost of managing the activities that lead to fraud, and also using it as a framework in various companies and industries to avoid any rework in adopting such practices. He
mentioned that such savings in the cost can be utilized by these mortgage providing
companies to pass on the gains and benefits to their clients to further boost up the relation and potential growth inc current challenging economically tough situation, and also to reduce the cost of producing efficient products. He further mentioned that one of the potential factors that could contribute in an organization’s success would be to implement and thoroughly use such frameworks to create awareness about the management steps needed to detect and also prevent the fraud from happening, and also to realize the gains and benefits of doing such activities to see overall success and growth with its replicating effect on the satisfaction level of the customers.
4.2 LIMITATION AND FUTURE RESEARCH
During the data analysis, there were few points discovered that could be considered as the limitation of this thesis. First limitation is the number of mortgage data points or records used for this thesis. There were 5951 records in the dataset, which might be considered as not enough to generalize the findings. Second limitation is that the dataset used is a secondary data that is relevant only to specific customer base in just one location. The data was not
37 gathered using a survey, that could have bright a bit of extra variation and dimension to the data analysis. The third limitation was that the data of secondary applicant was not
completely available. So this thesis mostly focused on the characteristics of primary
customer. The fourth limitation found is that the dataset could not explain more than 95% of the variation in mortgage fraud. Although the unexplained variation could be explained, but the R2, coefficient of determination, was relatively low varying just from 3.3% to 4.4%.
These limitation opens plenty of space for future research, which can be carried out using even bigger and diverse dataset by approaching and collecting datasets from various lenders like banks, non-banking financial companies etc. The bigger dataset could surely bring in more realistic and generalizeable results and analysis. Perhaps further research could bring in more customer and mortgage characteristics to support the hypothesis tested, as well as test some other hypothesis. Further research could also be done to include more mortgage
characteristics like interest rates, mortgage product types, payment frequency of the mortgage etc. Some additional customer characteristics could also be considered in future research like job type (like full time, part time, self-employee etc.), number of family members, expense types of the applicant etc. Future research can also be carried out by considering the customer characteristics of both primary and secondary applicants equally. Again, the bigger dataset might be able to address such limitations.
5 ACKNOWLEDGEMENT
In this section, I want to take this opportunity to appreciate and thank my thesis supervisor Dr. Andreas Alexiou for the guidance, encouragement and helpful feedback during various stages of the research. He made my journey of performing thesis much smoother than I initially expected. Additionally, I would like to thank my colleagues, who were willing to provide me with advice and comments for this research, and for cooperating with me
38 professionally by taking some work load off my shoulder to enable me to make enough time to perform the thesis and write this report.
39
6 REFERENCES
1. Hill, T. Dietrich. “THEARITHMETICOFJUSTICE:CALCULATING
RESTITUTIONFORMORTGAGEFRAUD.” Columbia Law Review, vol. 113, no. 7, 2013, pp. 1939–1976. JSTOR, JSTOR, www.jstor.org/stable/23561383.
2. The application of data mining techniques in financial fraud detection: A
classification framework and an academic review of literature. E.W.T. Ngai, Yong Hu, Y.H. Wong, Yijun Chen, Xin Sun, 2010.
3. Loss given default of high loan-to-value residential mortgages. Journal of Banking & Finance 33 (2009) 788–799. Min Qi , Xiaolong Yang, 2009.
4. Mortgage Fraud: Strengthening Federal and State Fraud Prevention Efforts. Mortgage Bankers Association. 2007.
5. Losing Group: Foreclosures in the subprime market and their cost to homeowners. Center for Responsible Lending. Schloemer, Ellen, Keith Ernst, and Kathleen Keest. 2006.
6. LIAR’S LOAN? EFFECTS OF ORIGINATION CHANNEL AND INFORMATION FALSIFICATION ON MORTGAGE DELINQUENCY. Wei Jiang, Ashlyn Aiko Nelson, and Edward Vytlacil, 2014.
7. LIQUIDITY PROBLEMS AND EARLY PAYMENT DEFAULT AMONG SUBPRIME MORTGAGES. Nathan B. Anderson and Jane K. Dokko, 2016 8. MetaFraud: A Meta-Learning Framework for Detecting Financial Fraud. Ahmed
Abbasi, Conan Albrecht, Anthony Vance and James Hansen. Source: MIS Quarterly, Vol. 36, No. 4 (December 2012), pp. 1293-1327.
9. The prevalence and impact of misstated incomes on mortgage loan applications. ScienceDirect, Journal of Housing Economics 21 (2012) 151–168. McKinley L. Blackburn, Todd Vermilyea.
40 10. Fraud detection system: A survey. Aisha Abdallah, Mohd Aizaini Maarof, Anazida
Zainal, 2016.
11. Mortgage default risk: New evidence from internet search queries. Journal of Urban Economics 96 (2016) 91–111. Marcelle Chauvet, Stuart Gabriel, Chandler Lutz (2016).
12. Residential mortgage default: Theory works and so does policy. Journal of Housing Economics 19 (2010) 280–294. Allen C. Goodman, Brent C Smith (2010).
13. Adjustable and Fixed Rate Mortgages as a Screening Mechanism for Default Risk. Lisa L. Journal of Urban Economics 49, 54-79 (2001).Posey and Abdullah Yavas (2001).
14. https://en.wikipedia.org/wiki/Mortgage_fraud
15. Identification of fraudulent financial statements using linguistic credibility analysis. Decision Support Systems 50 (2011) 585–594. Sean L. Humpherys, Kevin C. Moffitt, Mary B. Burns, Judee K. Burgoon, William F. Felix.
16. Big data analytics in healthcare: promise and potential. Raghupathi and Raghupathi Health Information Science and Systems 2014, 2:3. Wullianallur Raghupathi and Viju Raghupathi.
17. Detecting and Preventing Fraud with Data Analytics. Procedia Economics and Finance 32 ( 2015 ) 1827 – 1836. Adrian Banarescu.
18. Intelligent financial fraud detection: A comprehensive review. computers & security 57 (2016) 47-66. Jarrod West, Maumita Bhattacharya.
19. A computational model for financial reporting fraud detection. Decision Support Systems 50 (2011) 595–601. Fletcher H. Glancy, Surya B. Yadav.
20. Insurance fraud: The business as a victim. In M. Gill (Ed.), Crime at work, Vol 1. (pp. 73-82). Leicester: Perpetuity Press. Gill, K. M., Woolley, K. A., & Gill, M. (1994).
41 21. How the detection of insurance fraud succeeds and fails. Psychology, Crime & Law,
12(2), 163-180. Morley, N. J., Ball, L. J., & Ormerod, T. C. (2006).
22. A fraud detection approach with data mining in health insurance. Procedia - Social and Behavioral Sciences 62 ( 2012 ) 989 – 994. Melih Kirlidog, Cuneyt Asuk. 23. Detection of financial statement fraud and feature selection using data mining
techniques. Decision Support Systems 50 (2011) 491–500. P. Ravisankar, V. Ravi, G. Raghava Rao, I. Bose.
24. How good intentions can mask rationalizations -- and fraud. Report from PwC on Altruistic fraud. Publication attributed to the American Bar Association's Criminal Justice Section White Collar Crime Committee e-Newsletter, Summer/Fall 2017 issue.
25. Impact of Big Data Analytics on Banking Sector: Learning for Indian Banks. Procedia Computer Science 50 ( 2015 ) 643 – 652. Utkarsh Srivastava, Santosh Gopalkrishnan. 26. Smarter fraud investigations with big data analytics. Network Security Volume 2013,
Issue 12, December 2013, Pages 7-9. Shaun Hipgrave, IBM.
27. Alexander, N. and Colgate, M. (2000), “Retail financial services: transaction to relationship marketing”, European Journal of Marketing, Vol. 34 No. 8, pp. 938-53. 28. Rauyruen, P. and Miller, K. (2007), “Relationship quality as a predictor of B2B
customer loyalty”, Journal of Business Research, Vol. 60 No. 1, pp. 21-31.
29. Asif, S. and Sargeant, A. (2000), “Modelling internal communications in the financial services sector”, European Journal of Marketing, Vol. 34 Nos 3/4, pp. 299-317. 30. Behram, D. (2005), “Fraud management as tool to attract new customers”, American
42 31. Guardian Analytics (2011), “2011 business banking trust study”, available at:
http://info.guardiananalytics.com/2011-TrustStudy-Download.html (accessed January 6, 2012).
32. The impact of fraud prevention on bank-customer relationships: An empirical investigation in retail banking, International Journal of Bank Marketing (2012), Vol. 30 Issue: 5, pp.390-407. Arvid O.I. Hoffmann, Cornelia Birnbrich.
33. Perceptions of the Ethicality of Consumer Insurance Claim Fraud. Journal of Business Ethics, Vol. 54, No. 1 (Sep., 2004), pp. 67-79. Dwane Hal Dean.
34. Growing Public Tolerance Increases Fraud Claims, National Underwriter, Life & Health/ Financial Services Ed., 100(23), 31. Brostoff, S.: 1996.
35. Insurance Fraud and the Industry Response, CPCU Journal, Society of Chartered Property & Casualty Underwriters, 50(2), 92-103. Carris, R. and M. A. Colin: 1997. 36. Foppert, D.: 1994, 'Waging War Against Fraud', Besfs Review 94(12), 20-27. 37. Current Trends in Fraud and its Detection. Information Security Journal: A Global
Perspective (17), pp. 2-12. Albrecht, W. S., Albrecht, C. O and Albrecht, C. C. 2008. 38. Organizational Susceptibility to Fraud and Theft, Organizational Size and the
Effectiveness of Management Controls: Some UK Evidence. Managerial and Decision Economics, Vol. 28, No. 3 (Apr., 2007), pp. 181-193. Paul Barnes and Jill Webb.
39. Defrauding the public interest: a critical examination of reengineered audit process and the likelihood of detecting fraud. Critical Perspectives on Accounting 3: 297-310. Cullinan CP, Sutton SG. 2002.
40. Mortgage Fraud, The Impact of Mortgage Fraud on Your Company’s Bottom Line, Mortgage Bankers Association of America. Prieston, Arthur J. and Dreyer, Jaqueline A., (2001)
43 41. The Fraud Management Lifecycle Theory: A Holistic Approach to Fraud
Management. Journal of Economic Crime Management, Spring 2004, Volume 2, Issue 2. Wesley Kenneth Wilhelm
42. Detecting the financial statement fraud: The analysis of the differences between data mining techniques and experts’ judgments. ScienceDirect, Knowledge-Based Systems 89 (2015) 459–470. Chi-Chen Lin, An-An Chiu, Shaio Yan Huang, David C. Yen. 43. The Structural Causes of Mortgage Fraud, 60 Syracuse L. Rev. 473 (2010). James
Charles Smith.
44. Mortgage origination fraud and the global economic crisis A criminological analysis. American Society of Criminology. Criminology & Public Policy, Volume 9, Issue 3. Tomson H. Nguyen, Henry N. Pontell (2010).
45. The perceived moral reprehensibility of strategic mortgage default. ScienceDirect, Journal of Housing Economics 32 (2016) 18–28. Michael J. Seiler
46. Loan Originations and Defaults in the Mortgage Crisis: The Role of the Middle Class. Advance Access publication March 28, 2016. Published by Oxford University Press on behalf of The Society for Financial Studies. Manuel Adelino, Antoinette Schoar and Felipe Severino.
47. The consequences of mortgage credit expansion: Evidence from the U.S. mortgage default crisis. Quarterly Journal of Economics 124:1449–96. Mian, A., and A. Sufi. 2009.
48. Fraudulent Income Overstatement on Mortgage Applications During the Credit Expansion of 2002 to 2005. The Review of Financial Studies / v 30 n 6 2017.
Published by Oxford University Press on behalf of The Society for Financial Studies. Atif Mian and Amir Sufi.
