Fraud Risk Prediction in Merchant-Bank Relationship using Regression Modeling

R E S E A R C H includes research articles that focus on the analysis and resolution of managerial and academic issues based on analytical and empirical or case research Fraud Risk Prediction in Merchant-Bank Relationship using Regression Modeling Nishant Agarwal and Meghna Sharma Executive Summary Banking industry has gone through one of the worst crisis in recent times, and is still recovering from the after-shocks. However, there were a lot of learnings that banks would have taken away from this crisis. One of them is the need for a robust risk management system. The crisis dealt a blow to the banking system, catching them off guard when it came to foreseeing the risk. Banks, in the credit card business, face financial risk in the form of both credit risk and fraud risk. Sharma and Agarwal (2013) proposed a model for predicting the credit risk from the merchants. This paper builds upon their technique to predict the fraud risk posed by the merchants to the banks. Fraud risk is an important aspect of risk management systems, particularly in the credit space. The uncertainty surrounding the receipt of paybacks calls for designing robust risk prediction models. Fraud risk is very different from credit risk because fraud risk does not follow a pattern. It happens suddenly, and may not always have a trend before it happens. This creates a need for separate model for fraud risk prediction. This paper develops a fraud risk prediction model that uses logistic regression technique, deployed using SAS. The setup of the study is the merchant-bank relationship in the credit card industry. KEY WORDS Merchant Risk Acquiring Bank Risk Management Fraud Risk Management Logistic Regression SAS The model developed in this paper triggers on a transaction level, and assigns a probability score of default (PF) to each merchant for a possible fraud risk whenever a transaction is done at the merchant. Such a score warns the management in advance of probable future losses on merchant accounts. Banks can rank order merchants based on their PF score, and instead of working on the entire merchant portfolio, they can focus on the relatively riskier set of merchants. The PF model is validated by comparing the actual defaults with those predicted by the model and a good alignment is found between the two. The results show that the model can capture 62 percent frauds in the first decile when the transactions are sorted by the probability of fraud computed by the model. VIKALPA VOLUME 39 NO 3 JULY - SEPTEMBER 2014 67

The merchant-acquirer relationship is often ignored by banks, when planning for risk management. The relationship, however, poses an invisible risk to the acquiring banks, because, when a transaction is recorded at a merchant today, the risk from this transaction arises a few months into the future. This makes the risk unforeseeable. The motivation behind this paper is to create an Early Warning System, to tackle the above-mentioned challenge posed by the merchant-acquirer relationship. This Early Warning system enables the banks to estimate the expected loss from the defaults predicted by the model, and strategize accordingly. Such an Early Warning System is possible to create because of the availability of live transactional data with the banks. This data can be leveraged by using tools like SAS and SQL Server to derive analytical insights and then create a statistical model. This model can then be run on live data. LITERATURE REVIEW Cressey (1953), a criminologist, first studied the motivation behind committing fraud. He formulated the wellknown fraud triangle, which has its three vertices as Pressure, Opportunity, and Rationalization. Cressey concluded from his research that Trust violators when they conceive of themselves as having a financial problem which is non-shareable, have knowledge or awareness that this problem can be secretly resolved by violation of the position of financial trust, and are able to apply to their own conduct in that situation verbalizations which enable them to adjust their conceptions of themselves as trusted persons with their conceptions of themselves as users of the entrusted funds or property (Cressey, 1953, p. 742). Credit card fraud risk prediction and modeling has been a research interest of many banks around the world and a number of techniques, with special emphasis on regression techniques. Molyneaux (2007) and George (1992) also discussed the ethics of banking and Clarke (1994), in his work, mentioned about the moral complexity of fraudulent behaviour. Ghosh and Reilly (1994) proposed a fraud detection system that used past fraud cases due to lost cards, stolen cards, and application frauds. Syeda (2002) used parallel granular neural networks (PGNNs) to improve the knowledge discovery process in credit card fraud detection. Fan, Prodromidis, and Stolfo (1999) suggested the application of distributed data mining in credit card fraud detection. Brause, Langsdorf, and Hepp (1999) developed an approach that involved advanced data mining techniques and neural network algorithms to obtain high fraud coverage. Chiu and Tsai (2004)proposed web services and data mining techniques to establish a collaborative scheme for fraud detection in the banking industry. Not much literature is available on addressing the risk faced by a bank from a merchant. This study aims to fill this existing gap. MERCHANT-ACQUIRER RELATIONSHIP Business Framework Merchants, who accept credit cards, get a Point of Sale (POS) device installed at their outlet. This POS device is installed through a bank. And this bank is known as the acquiring bank, or simply the acquirer. When a transaction is recorded by a merchant, a copy of the receipt signed by the customer is sent to the bank by the merchant for claiming the amount of transaction. The acquiring bank settles this within a stipulated number of days, after deducting a surcharge, known as discount rate. Figure 1 shows the flow of events once a transaction is done by a customer at a merchant. The merchant sends a copy of the signed receipt to the acquiring bank. The bank, in turn, settles the dues to the merchant, and also communicates the transaction details to the customer bank, via the Credit Card Association. Once this communication is received by the customer bank, it adds this transaction to the customer s statement of account. Figure 1: A Typical Credit Card Business Model Customer Bank Credit Card Association (Visa and Mastercard) Acquiring Bank Customer Merchant 68 FRAUD RISK PREDICTION IN MERCHANT-BANK RELATIONSHIP USING REGRESSION MODELING

Fraud Risk Fraud Risk is the possibility that a counter party engages in an activity which entails earning revenue through the means that act as a breach of contract with the signatory party. The authors discuss the cause of fraud risk in the merchant-acquirer relationship. Fraud risk is a challenge to the acquiring banks because the merchants always have a motivation to earn money through fraudulent ways, primarily because merchants look upon the discount rate charged by the banks as a cost, and they engage in activities in order to compensate for this perceived cost. These activities invariably breach the contract between the merchant and the bank, and are termed as fraud. To understand this, an example of a merchant selling high value product, with irregular business cycle is considered. Such a merchant could have a small customer base. This concept is termed as card member concentration, which means that most of the transactions at the merchant are initiated by a small group of customers and therefore the transactions are concentrated around a few customers. In such a scenario, the possibility of fraud increases tremendously, because the merchant and customer can tie up together for committing the fraud. Another scenario where fraud is possible is when a merchant uses his personal credit card, issued by the same bank as the acquiring bank, and swipes the card on his POS device. In this way, the merchant has not sold anything. He has just swiped his personal card, and submits the slip generated by the POS device to the acquiring bank for claiming the money. Typically, this money is credited in two to three days, after deducting 2-3 percent as the surcharge, also known as the discount rate. This amount will appear on the statement of the personal card of the merchant, which would need to be paid after 50 days approximately. Upon looking at this transaction from another perspective, it was realized that this amounted to taking a loan from the bank for 50 days, at a rate of 2-3 percent. This is a breach of contract by the merchant, and should be termed as fraud. The last scenario that is considered for fraud is the possibility of a merchant who has a low business volume under usual circumstances, but has some big spikes in the volume in the recent past. In such a scenario, the mathematical slope of the spike in volume is calculated, and if that slope is more than a certain cutoff value, then that merchant is tagged as a possible fraud case. FRAUD RISK MODELING IN MERCHANT- ACQUIRER RELATIONSHIP The Need An acquiring bank can have millions of merchants in its global portfolio. While all merchants would be risky at some level, all of them do not pose the same credit risk. The management of the bank cannot build risk management strategy for the entire merchant portfolio. Hence, it would be helpful for the bank if they can assign a probability of fraud (PF) score to the merchants based on the level of risk posed by them, and then sort the list of merchants in decreasing order of this probability score. The authors propose to build a statistical model, which would help the acquiring bank in the following ways: Generate PF scores for all merchants in the portfolio of the bank, which would help the bank in identifying the high risk merchants, and plan accordingly Act as an Early Warning System, which would give the acquiring bank some visibility into the potentially invisible fraud risk in the future The above-mentioned points enable the banks to create sufficient reserves in advance, in view of the fraud risk they foresee using the model. Also, the banks can do further analysis on the kind of industries, geographies, etc. that are posing more risk, and plan accordingly. Methodology As discussed earlier, the number of merchants in the global portfolio of the bank can be in millions. This creates a need for a statistical model. To make an effective model, the authors propose to first create segments of merchants, based on their perceived risk profile, and then create the required models for each of the segments. The methodology is described in Figure 2. Create Merchant Segments Figure 3 shows the Merchant Segments based on two attributes: Ability to pay and Willingness to pay. Merchants who are high on both these attributes are least risky, and a fraud risk model may not be required for them. Merchants who are low on willingness to pay are generally risky group of merchants. VIKALPA VOLUME 39 NO 3 JULY - SEPTEMBER 2014 69

Figure 2: Fraud Risk Modeling Credit Risk Modelling Figure 3: Merchant Segmentation Framework Ability to Pay Low High 1 2 3 4 5 Willingness to Pay High Low Lowest Risk Lower Risk Create Merchant Segments Identify the type of statistical model to be built Identify the variables suitable for the model for each segment Test the statistical significance of each variable Build models of each segment and validate the results using out of time-out of sample data High Risk Highest Risk While this classification is subjective in nature to some extent, it is not critical to get this classification absolutely right. This framework is more of a guiding tool, and even if some merchants do get misclassified, it does not impact the model because in the end, all merchants are risky, and models can be built for all the four segments in Figure 3. This paper has built the fraud risk model on merchants who have high ability and low willingness to pay. A model for the Highest Risk category needs to be created separately, but since the column of merchants in the High Risk category is the highest, it was decided to create a model for this segment. Identify the type of statistical model to be built In the fraud risk model built here, the dependent variable is whether a merchant will go into a fraudulent activity or not. This is a binary variable, with the value 1 representing the possibility that the merchant will do a fraud, and the value 0 representing the possibility that the merchant will not do a fraud. Hence, logistic regression was used to model the risk. Another choice that was made was based on the nature of model required. Broadly speaking, there are two types of models: Transaction level and Account level. The transaction level model gets triggered when transaction is recorded at a merchant and transaction variables are used. The drawback of this model is that if a merchant does not have too many transactions (As per industry standards, the merchant should have at least 100 transactions per day on an average in the last one year), or has seasonal transactions, then this model may not be useful because there will not be enough transactions to get a reliable PF score. To overcome this, one can use account level model which uses variables which are merchant specific, irrespective of whether a transaction is being recorded at the merchant or not. However, the drawback of this model is that the variables are not updated because they are based on merchant history rather than on fresh transactions. This paper builds a transaction level model to demonstrate its usefulness. Modeling Dataset This includes the time period for historical data that will be used in developing the model. For validation of the model, it is necessary to create an out of time-out of sample dataset. This paper uses the following timeframes: Model Development Dataset: It is based on the time period January 2011 - December 2011. The time period chosen is one year assuming that most banks do not store data which is older than one year because of the enormity of data, and costs associated with storage. Model Validation Dataset: It begins from January 2012 and ends in December 2012, on a different sample of merchants. 70 FRAUD RISK PREDICTION IN MERCHANT-BANK RELATIONSHIP USING REGRESSION MODELING

Identify Suitable Variables Suitable variables for each segment need to be identified by combining business experience of the management, supported by statistical significance of the variable itself. Another thing that needs to be taken care of is the sign of the variable. The variable needs to have predictive power, but in the right direction Credit Risk PF Model Development A credit risk PF (Probability of Fraud) model was proposed to assess the risk associated with the merchantacquirer relationship, from an acquirer s perspective. As discussed, the logistic regression technique was adopted to build the model. Subsequently, the model was tested on out of time-out of sample data. Table 1 gives the summary of the model prerequisites that were used. Sample of Transaction level Data Tables 2a and 2b are only a snapshot of transaction level data. The exhaustive list of variables, which usually are hundreds in number, is not shown; only the variables finally used in the model are considered. Table 1: Summary of Modeling Framework Regression Technique Logistic Regression Timeframe Development Dataset January 2011 December 2011Validation Dataset January 2012 December 2012 Modeling Platform SAS, SQL Server Model Type Transaction level Model Merchant Segment High Risk (High Willingness and High Ability to Pay) The description of the variables is presented in Table 2b. Number of customers in the last five days as a variable was also tested for significance, and while it was found to be significant, it had high multi-collinearity with other variables, indicating that its presence in the model could inflate the error term in the regression. Hence, this variable was not included in the model. At the same time, this variable could be included in the model, but with the caveat that any bank including this variable in the model should test it across a much larger sample of data, to ensure that the influence of this variable is not exaggerated. After the identification of variables, logistic regression was applied on the framework discussed above. A typical logistic regression model is represented as: Table 2a: Variables used in PF Model Merchant_ID Trans_Dt Sub_Dt Trans_Amt Slope_Spike No_of_Cust_Last 5 Days CM_Conc 9100011001 21/2/2011 24/2/2011 167000 5 9525 5 9100011002 13/3/2011 13/2/2011 234234 4 3567 3 9100011003 12/5/2011 13/5/2011 556435 2.4 3875 2 9100011004 24/6/2011 25/6/2011 1340136 1.33 8776 5 9100011003 25/1/2011 27/1/2011 4061244 10.9 12376 2 9100011002 27/7/2011 28/7/2011 6672323 200 9876 3 9100011001 31/8/2011 1/9/2011 3004237 1 12345 5 9100011004 10/9/2011 11/9/2011 2244267 0.8 4678 5 Table 2b: Description of Variables used in PF Model Variable Variable Description Variable Type Merchant_Id A unique identifier for the merchant Text Trans_dt Date of transaction between merchant and the customer Date Sub_dt Date on which the merchant submits the transaction slip to the acquiring bank Date Trans_amt Transaction amount Numeric CM_conc A measure of the concentration of card members at the merchant. Calculated as the ratio of Numeric number of transactions in the last 5 days at the merchant to the number of card members initiating those transactions Slope_spike The slope of the spike in current month s business volume over last month s volume Numeric No_of_cust_last 5 days Number of customers transacting at the merchant in the last 5 days Numeric VIKALPA VOLUME 39 NO 3 JULY - SEPTEMBER 2014 71

z = β 0 + β 0 x 1 + β 1 x 2 + β 2 x 3 + β 3 x 4 5.1 p = 1/(1 + e z ) 5.2 where, z is the linear function of the dependent variables. p is the probability of default, that is modelled. The SAS output of our model run is as shown in Table 2c. The full blown output is not shown here. It is clear from this output that: x 1 x 2 x 3 x 4 = Trans_amt = CM_conc = Slope_spike = No._of_cust_last5days AIC - This is the Akaike Information Criterion. It is calculated as AIC = -2 Log L + 2((k-1) + s), where k is the number of levels of the dependent variable and s is the number of predictors in the model. AIC is used for the comparison of non-nested models on the same sample. Ultimately, the model with the smallest AIC is considered the best, although the AIC value itself is not meaningful. SC - This is the Schwarz Criterion. It is defined as - 2 Log L + ((k-1) + s)*log ( fi), where fi s are the frequency values of the ith observation, and k and s were defined previously. Like AIC, SC penalizes for the number of predictors in the model and the smallest SC is most desirable and the value itself is not meaningful. -2 Log L - This is negative two times the log-likelihood. The -2 Log L is used in hypothesis tests for nested models and the value in itself is not meaningful. Likelihood Ratio - This is the Likelihood Ratio (LR) Chi- Square test that at least one of the predictors regression coefficients is not equal to zero in the model. The LR Chi- Square statistic can be calculated by -2 Log L (null model) - 2 Log L (fitted model) = 231.289-160.236 = 71.05, where L(null model) refers to the Intercept Only model and L(fitted model) refers to the Intercept and Covariates model. Score - This is the Score Chi-Square Test that at least one of the predictors regression coefficients is not equal to zero in the model. Wald - This is the Wald Chi-Square Test that at least one of the predictors regression coefficient is not equal to zero in the model. McFadden s pseudo R 2 - This measure compares a random model with the model suggested in the paper. This R square is used to compare Table 2c: SAS Output of the Model Model Fit Statistics Criterion Intercept Only Intercept and Covariates AIC 224.982 165.214 SC 263.785 179.340-2 Log L 235.289 156.234 Testing Global Null Hypothesis: BETA=0 Test Chi Square DF Pr>Chi Sq Likelihood ratio 71.0525 4 <.0001 Score 58.6092 4 <.0001 Wald 39.8751 4 <.0001 Analysis of Maximum Likelihood Estimates Parameter DF Estimate Standard Error Wald Chi-Square Pr>Chi Sq Intercept 1-5.1055 0.9226 30.6231 <.0001 Trans_amt 1 0.027547 0.8607 10.2431 <.0001 CM_conc 1 0.0300 0.0159 3.5838 <.0001 Slope_spike 1 0.0532 0.0149 12.7778 <.0001 No_of_cust_last 5 days 1-0.4824 0.2785 3.0004 <.0001 72 FRAUD RISK PREDICTION IN MERCHANT-BANK RELATIONSHIP USING REGRESSION MODELING

two models, and its value indicates the improvement achieved by one over the other. LL full model 156.234 R 2 = 1 = 1 = 0.3359 LL intercept 235.289 The value of McFadden s pseudo R 2 suggests that the model has a 34 percent improvement over a random model, or a model with no covariates. Logistic regression was run in SAS on the transaction level data. The SAS output showed the following values of the β coefficient. β 0 = -5.1055, β 1 = 2.7547, β 2 = 0.0300, β 3 = 0.0532, β 4 = 0.4824 In Figure 4, the graph, known as ROC (Receiver Operating Characteristics) curve shows that the model captures 63 percent defaults in the top 10 percent of the transaction data when the data is sorted in decreasing order of the PF score generated by the model. A high percentage of default capture in the top percentile of ROC curve indicates that the model is working as expected, and has been able to fit the data well. The steepness in ROC Curve indicates the amount by which using a model is better than not using a model at all. Figure 4: ROC (Receiver Operating Characteristics) Curve To understand this better, an example of the merchant having the id 9100011001 (row 1 in Table 2a) is considered. For this merchant, the z value comes out to be 0.532217 using equation 5.1. ROC Curve Using equation 5.2, the PF score of this merchant comes out to be 0.63 which means that this transaction has 63 percent probability to go into default. Similarly, this model can calculate the PF score of each transaction. Once the PF score is obtained from each transaction, the dataset is sorted in descending order of probability score and the data is then consolidated into deciles as shown in Table 3. Table 3: Percentile Distribution of Defaults captured by PF Model Percentile No. of Cumulative Percent Cumulative Defaults No. of Defaults No.of Defaults 10 689 689 62 20 201 890 80 30 67 957 86 40 46 1003 90 50 37 1040 94 60 24 1064 96 70 17 1081 97 80 14 1095 98 90 10 1105 99 100 7 1112 100 The 45 degree line in Figure 4 indicates the ROC curve when no model is used, and hence each merchant in the portfolio is assumed to have a 50 percent PF. The area between the actual ROC, originating from the use of the PF model, and the 45 degree line, indicates the incremental benefit emanating from the use of PF model. Once the probability score is obtained from each transaction, the dataset is sorted in descending order of probability score and then consolidated into deciles and the actual and predicted default rate calculated for each decile as shown in Table 4. Actual default rates are calculated as the ratio between the number of defaults in that decile and the total number of merchants in that decile. Predicted default rates are calculated in a similar way, but the numerator in this case represents the total number of merchants tagged as defaulters by the PF model. VIKALPA VOLUME 39 NO 3 JULY - SEPTEMBER 2014 73

Table 4: Comparison of Actual Default Rates and Default Rates Predicted by the PF Model Percentile No. of No. of Actual Predicted Defaults Transactions Default Default Rate (%) Rate (%) 10 689 8613 8.00 7.7 20 201 8613 2.33 2.0 30 67 8613 0.78 1.1 40 46 8613 0.53 0.5 50 37 8613 0.43 0.5 60 24 8613 0.28 0.3 70 17 8613 0.20 0.2 80 14 8613 0.16 0.2 90 10 8613 0.12 0.1 100 7 8613 0.08 0.1 Figure 5: Comparison of Actual vs Predicted Default Rate for the PF Model the merchant fraud risk. Moreover, managers would find it more efficient to strategize for a smaller set of risky merchants, rather than planning for the entire merchant portfolio. A possible obstacle in the use of this model could be the presence of millions of merchants in a bank s portfolio, thereby making the task of classification of these merchants using the Merchant Segmentation Framework very difficult. However, this can be easily overcome by creating geographical portfolios for merchants, instead of taking the entire global merchant base as a whole. As an example, a bank could have one million merchants in India, out of which a quarter of a million could be in the northern region. The bank can assess the northern portfolio first, and then the other regions one by one. Also, banks usually have regional teams which handle merchant relationships in their respective regions. These teams can help in using the Merchant Segmentation Framework. LIMITATIONS OF THE MODEL AND FUTURE RESEARCH Although this study has made some contribution towards a better understanding of the risk inherent in a merchantacquirer relationship, it is not without limitations. The PF model is sensitive to the explanatory variables chosen. It is also dependent, to an extent, on the modeling timeframe chosen, which makes it all the more important to test the model under different set of scenarios, and in different economies around the world. Figure 5 shows the comparison of actual default rate and the default rate predicted by the model for each decile and helps in understanding the model performance in real time. MANAGERIAL IMPLICATIONS This research on fraud risk arising from merchant acquirer relationship and the subsequent development of the PF model is founded on the concept of risk management. The approach employed to model the fraud risk from global credit card operations should be useful for modern day managers of a bank. With the help of the PF Model, managers should be able to strategize with the invisible fraud risk always in their mind. The PF model would enable the banks to keep in reserve a pool of funds in advance, to negate the losses arising from While the PF model can predict the possibility of a merchant going into default, it cannot predict the amount of default. Since this model is based on logistic regression, the dependent variable needs to be binary in nature. Therefore, to predict the amount of default, the PF model can be extended to another level, where a regression model can be built, with the dependent variable being the LGD (Loss Given Default). The model seems to have good explanatory abilities. However, it is pertinent to mention that the variables used in the PF are not the only set of variables that can be used. The choice of variables would depend upon the sample of data used, along with the nature of the data. Hence it becomes important to test this model under different conditions, and in different geographies. Also, it is important to mention that apart from disputes, other causes of 74 FRAUD RISK PREDICTION IN MERCHANT-BANK RELATIONSHIP USING REGRESSION MODELING

fraud risk could also be explored. While the PF model proposed in this paper is useful for both practitioners as well as researchers, caution needs to be exercised while generalizing the findings. In this paper, PF model was tested on sample data of a bank. Future research attempts should test the generalizability of the PF framework of different banks in different geographies. CONCLUSION The management of risk in global business scenario has taken more importance than ever. In the financial services sector, fraud risk is inherent in the nature of business. The aim of this research was to create an Early Warning System for banks to effectively manage the risk arising out of their merchant-acquirer relationship. To facilitate this, the PF (Probability of Fraud) Model was proposed. The PF model assigns a PF score to the entire merchant portfolio of the acquiring bank, which can be used by the banks to assess their merchant portfolio more effectively. In this paper, the business framework of the merchantacquirer relationship was explained, along with the ways in which fraud risk can arise from this association. The PF model was created in SAS and applied to the transactional data of a bank. It was observed that the model captured 63 percent defaults in the top decile of transactions, when sorted in decreasing order of PF scores. The PF model was also validated by comparing the actual defaults with those predicted by the model and it was found that there was a very good alignment between the two. This indicated that the model rank ordered the defaults quite close to reality. To summarize, the PF model is expected to act as a tool for acquiring banks in both analysing as well as strategizing for credit risk arising from the merchant side. REFERENCES Agarwal N., & Sharma M. (2013). Default risk modeling in credit cards: A study of merchant and aquiring bank relationship. IIMS Journal of Management Science, 4(1), 40-48. Brause, R., Langsdorf, T., & Hepp, M. (1999). Neural data mining for credit card fraud detection. Proceedings of IEEE International Conference. Tools with Artificial Intelligence, pp. 103-106. Chiu, C., & Tsai, C. (2004). A web services-based collaborative scheme for credit card fraud detection. Proceedings of IEEE International Conference. e-technology, e-commerce and e Service, pp. 177-181. Clarke, M. (1994).Fraud and the politics of morality. Business Ethics: A European Review, 3(2), 117-122. Cressey, D. R. (1953). Other people s money. Montclair, NJ: Patterson Smith. Fan, W., Prodromidis, A.L., & Stolfo, S.J. (1999). Distributed data mining in credit card fraud detection. IEEE Intelligent Systems, 14(6), pp. 67-74. George, E. (1992). Ethics in banking. Business Ethics: A European Review, 1(3), 162-171. Ghosh, S., & Reilly, D.L., (1994). Credit card fraud detection with a neural-network. 27th Hawaii International Conference on Information Systems, Vol. 3 (2003), pp. 621-630. Molyneaux, D. (2007). Two case study scenarios in banking: A commentary on The Hutton Prize for Professional Ethics, 2004 and 2005. Business Ethics: A European Review, 16(4), 372-386. Syeda, M., Zhang, Y. Q., & Pan, Y. (2002). Parallel granular networks for fast credit card fraud detection, Proceedings of IEEE International Conference on Fuzzy Systems, pp. 572-577. Nishant Agarwal is pursuing doctoral studies in the area of Accounting from Indian School of Business, Hyderabad. He has degrees in Engineering (Delhi College of Engineering) and MBA (Indian Institute of Management, Calcutta). Ha has more than 10 years of work experience spread across technology, consulting, risk management, and teaching. e-mail: nishanta2006@email.iimcal.ac.in Meghna Sharma is an academician, researcher, and a consultant with over 18 years of corporate and teaching experience. She has done her Post Graduation in Economics, PGDBM in International Business, and a Ph.D. in Economics (Microfinance). She has several national and international publications to her credit. Currently, she is a faculty with Amity International Business School, Amity University, Uttar Pradesh. e-mail: meghnaasharma@gmail.com VIKALPA VOLUME 39 NO 3 JULY - SEPTEMBER 2014 75

76 VIKALPA VOLUME 39 NO 3 JULY - SEPTEMBER 2014