Financial Literacy and Fraud Detection

Who is better at detecting fraud? This paper finds that more financially knowledgeable individuals have a higher propensity to detect fraud: a one standard deviation increase in financial knowledge increases fraud detection probabilities by 3 percentage points. The result is not driven by individuals' higher financial product usage and is observed to be moderated by individuals' low subjective well-being, effectively depleting skills to detect fraud. Interestingly, prudent financial behavior relating to basic money management is found to have negligible effects for detecting fraud. The findings attest to the fact that fraud tactics are increasingly complex and it is greater financial knowledge rather than basic money management skills that provides the degree of sophistication necessary to detect fraud. The paper draws policy implications for consumer education programs to go beyond cultivating money management skills, and provide advanced financial knowledge necessary for tackling fraud.

The study found that despite industry efforts to prevent identity fraud, fraudsters successfully adapted to net 1.3 million more victims in 2017, with the amount stolen rising to $16.8 billion."(Javelin Strategy & Research, 2018) With increased digitalization of financial services and use of plastic payments, recent years have seen an amplification in the volume of fraudulent activities, costing the economy billions of dollars.Particularly on the rise is banking fraud, which refers to the unauthorized access to another's bank account or payment card details to carry out fraudulent transactions.
Noteworthy is the high degree of sophistication with which banking fraud is committed, such that many fraudulent activities remain undiscovered, with victims being rarely compensated.
For instance, authorized push payment frauds, contactless card and card skimming frauds, to name a few, are emergent types of fraud that can continue unhindered for long periods, if detected at all.Banks' fraud detection and verification systems can miss illicit transactions that are designed to appear authentic; therefore, banks place emphasis on their customers to spot and report any fraudulent activities in their accounts.
In this paper, we study the importance of financial literacy -the ability to process economic information and make informed financial decisions (Lusardi & Mitchell, 2014)for fraud detection.We focus specifically on the role of financial knowledge and financial behavior.Financial knowledge can make an individual more attentive to fraud risk, reduce their relative ignorance to fraudulent scams and enable them to cultivate greater effectiveness in detecting fraud, while prudent financial behavior can reduce an individual's propensity to be a target for fraud.
Thus, the paper builds on the growing evidence that there is a strong relationship between financial literacy and economic outcomes.For example, financially literate individuals are observed to be more financially aware of financial products and services (Banerjee et al., 2018), better at engaging in day-to-day financial management activities such as retirement planning and wealth accumulation (Lusardi & Mitchell, 2007a,b, 2011;Klapper & Panos, 2011), more likely to participate in financial markets (Van Rooij et al., 2011;Yoong, 2011;Balloch et al., 2015) and better equipped to face macroeconomic shocks such as the financial crisis (Klapper & Panos, 2013).
In the context of fraud victimization, financial knowledge can improve individuals' attentiveness to fraudulent practices and empower them to deter fraud.For instance, Andreou & Philip (2018) find that financially knowledgeable students have a significantly higher propensity of declining an offer to engage in a Ponzi scheme than their peers, after being solicited.DeLiema et al. (2018) document that fraud incidences are non-negligible at older ages as a consequence of poor financial capability.They find evidence that older adults are more susceptible to fraud due to greater asset accumulation and as fraudsters may consider them easier targets due to potential cognitive impairments associated with aging.Van Wilsem (2011) observes that people with low self-control run substantially higher victimization risk from internet consumer fraud.
Besides the aforementioned literature, the paper also draws on the recent evidence of the effects of behavioral inattention in household finance and other domains (for surveys, see DellaVigna, 2009;Gabaix, 2017).Studies include Gabaix & Laibso (2006), who show that inattentive customers get exploited by firms that shroud high-priced add-ons, such as, in the case of banks, ATM usage fees, bounced check fees, and minimum balance fees; and Grubb (2015), who finds that consumers will benefit from regulation requiring firms to disclose information and alerts that substitute for attention.Using survey and transactions data, Stango & Zinman (2014) find that checking-account holders incur overdraft fees because they are inattentive to account balances.They observe that individuals are substantially less likely to pay overdraft fees after taking surveys that mention overdrafts or topics that they might associate with overdraft.
To analyze the relationship between financial literacy and fraud detection, we use information from 5, 698 US respondents to the National Financial Well-Being Survey (NFWBS), which was fielded in 2016 by the Consumer Financial Protection Bureau (CFPB) and designed to be representative of the adult US population.In addition to granular socioeconomic and demographic information, the survey asks respondents whether they have experienced fraud in the past, where someone has, without their permission, used or attempted to use any of their existing accounts, such as a credit or debit card, checking, savings, telephone, online, or insurance account.The NFWBS survey also captures respondents' financial literacy, financial inclusion and financial behavior information.
The empirical analysis uncovers a positive and economically meaningful association between financial knowledge on fraud detection: the more financially knowledgeable the respondents, the more fraud they detect in their accounts.The results corroborate that financial knowledge enhances the financial capability of individuals by being more aware of, and better at recognizing, fraud when it transpires.Further, we observe that one's propensity to detect fraud is also influenced by the degree of financial inclusion in financial services (i.e. the number of financial services they utilize).Unsurprisingly, we find that individuals participating in informal financial services, much more than formal financial services, have a greater propensity to detect fraud.The result indicates that individuals using the informal financial sector services may be targeted more with fraud possibly due to poorer governance and supervision of the informal financial sector services.
We proceed to investigate whether individuals exhibiting prudent financial behavior are better at detecting fraud.We include in our analysis a battery of information relating to financial behaviors that enable individuals to manage their finances better.These include, among others, setting and pursuing financial goals; setting and consulting a budget; whether bills are paid on time; whether statements, bills and receipts are checked for errors; and whether the credit card balance is paid off in full each month.Surprisingly, the test results indicate that prudent financial behaviors do not really matter when it comes to the ability to detect fraud.We observe marginal negative significance for the financial behavior dimensions, pursuing financial goals, staying within budget and having a savings habit, such that these prudent financial behaviors marginally reduce the propensity to detect fraud.However, overall we see that the positive effect of financial knowledge in detecting fraud remains the strong influencing factor.
The weak result for financial behavior suggests that efficient management of finances does not directly correlate with greater effectiveness in spotting fraudulent activities.This can be explained by the fact that fraud is becoming increasingly sophisticated and it is not financial behavior but financial knowledge that provides the degree of sophistication necessary to be able to detect and deter banking fraud.Financial knowledge strengthens one's attention to the existence of fraud risk and thus empowers individuals to take necessary steps in detecting fraud when it happens and ultimately preventing its occurrence.
To test that individuals' behavioral attention attribute channels the relationship between financial knowledge and fraud detection, we capture individuals' degree of attentiveness through the survey question eliciting the behavioral trait of how often they act without thinking through all the alternatives when making a decision.We find a strong correlation between financial knowledge and individuals' attentiveness.Further, although the level of attentiveness per se cannot explain an individual's propensity to detect fraud, the attention capacities derived from financial knowledge are observed to form a strong influencing factor for detecting fraud.
We further find that the relationship between financial knowledge and fraud detection can be attenuated when individuals' subjective well-being is low, in effect reducing individuals' attention capacities to detect fraud due to the high cognitive loads low well-being imposes.This is in line with previous studies that document a negative relationship between one's capacity to make sensible economic decisions and the impediments to an individual's cognitive function.For example, Mani et al. (2013), Haushofer & Fehr (2014), Deck & Jahedi (2015), Schilbach et al. (2016) show that economic decisions worsen with increases in cognitive load.
To empirically test this, we use survey information on three subjective well-being aspects of the individuals, namely, life satisfaction, optimism about the future, and the belief that works yield success.The results identify significant interactions between well-being and financial knowledge in a meaningful way.More specifically, we observe that the greater the life dissatisfaction, the higher the pessimism about the future, and the greater the disagreement that work will yield success in the future, the weaker becomes the relationship between financial knowledge and fraud detection.By contrast, at higher levels of subjective well-being, we observe that financial knowledge emerges as a significant determinant of an individual's abilities to detect fraud.The results indicate that as well-being deteriorates, the attention-increasing effects of financial knowledge regarding fraud detection tamper off.
Overall, subjective well-being of an individual plays an important moderating role in the relationship between financial knowledge and fraud detection.
Our study has crucial policy implications given the recent interest in the importance of financial literacy for general consumers and retail investors from a behavioral perspective (IOSCO and OECD, 2018).As fraud is increasingly sophisticated, policy steps should emphasize consumer education programs to enhance financial knowledge on aspects such as risk and return.If consumers understand how financial products operate, they will be better able to identity and protect themselves from "too good to be true" offers.Further, training on aspects related to achieving prudent financial behavior must go beyond cultivating money management skills to also include training to detect and deter banking fraud.

Data sample
We use data from the National Financial Well-Being Survey (NFWBS), fielded by the Consumer Financial Protection Bureau (CFPB) in 2016.The data were weighted to represent the U.S. adult population and key subpopulations.6, 394 respondents participated in the survey and forms a representative sample of the adult population from all 50 US states.With the intention of measuring individual-level financial well-being and its determinants, the cross-sectional survey records a rich set of individual and household characteristics at a high level of granularity.The survey captures individual attributes including socio-demographic attributes, namely age, gender, civil-status, ethnicity; socio-economic attributes, namely education, income; and spatial attributes, namely census region and urban-rural linkages.
Along with these individual-level attributes, the survey includes important information on financial fraud detection, level of financial knowledge, financial inclusion in formal and informal financial services, as well as respondents' financial attitudes and behavioral traits.After excluding respondents who did not reveal their information on the various questions that we study in the paper, we are left with a final sample of 5, 698 individuals for our empirical investigation.

Variable constructions and descriptive analysis
In order to elicit information on whether the respondent detected attempted or actual fraud, the following survey question was asked: In the past 5 years, has someone without your permission used or attempted to use an existing account of yours, such as a credit or debit card, checking, savings, telephone, online, or insurance account?
Respondents were offered four choices of answers, namely, "Yes", "No", "I don't know", and "Refused".As additional follow-on questions on the specificity and the frequency of the fraud were not asked, we are unable to clearly distinguish the case where the respondent is experiencing more (or less) fraud.However, we can safely interpret the responses as the respondent's ability to detect fraudulent activities in their accounts.Table 1 reports the number of responses to the various response categories.We observe that around 26% report that fraud has been detected in their accounts, 65% do not consider that they have been subject to fraud, and a minority 9% of respondents report to be either unaware of fraud or refused to respond to the question.In the empirical analysis, we exclude those respondents who either choose "I don't know" or have refused to disclose.
Following Knoll & Houts (2012), we measure financial knowledge of respondents using nine survey questions, eliciting their understanding of financial concepts such as long-term returns on investments, stocks vs. bonds vs. savings volatility, benefits of diversification, and the relationship bond prices and interest rates.The actual wordings of the questions and responses choices of the nine questions are reported in Appendix A.1.For each individual a composite score representing their level of financial knowledge is derived from their responses to the nine questions using item response theory (for methodological details, see Knoll & Houts, 2012).Panel A of Table 2 provides a descriptive analysis of the relationship between various levels of financial knowledge and fraud detection.We observe that the proportion of respondents detecting fraud increases with their financial knowledge.For instance, at the lowest financial knowledge score of −2.053, no respondents detect fraud; however, this increases to 36.61% for the case of respondents with the highest financial knowledge score of 1.267.
To measure the financial behavior of respondents, we make use of information from ten questions that capture the financial behaviors relating to their savings habits, their frugality, and how they plan and manage their budgets.The actual wordings of the questions and the Likert response choices are reported in Appendix A.2.The questions ask the respondents to rate their financial behaviors in a variety of dimensions such as setting and pursuing financial goals; setting and consulting your budget; whether bills are paid on time; whether statements, bills, and receipts are checked for errors; and whether the credit card balance is paid off in full each month.For each individual, we create a composite score of financial behavior by summing the response choices (which are first mapped to integers) from all ten behavior questions.Panel B of Table 2 reports the descriptive analysis of the relation between the level of financial behavior and fraud detection.Splitting individuals into terciles according to their financial behavior, we observe that, unlike financial knowledge, fraud detection across the various financial behavior groups are strikingly similar.
We measure individuals' degree of inclusion in formal (informal) financial services as the number of formal (informal) financial services they utilize.Formal financial services refers to savings accounts, life insurance, health insurance, retirement accounts, pensions, non-retirement investments, education savings account, and student or education loans.Informal financial services refers to payday or cash advance loans, pawn or auto title loans, reloadable cards that are not linked with checking or savings account, or using non-banks for international money transfers or for check cashing or purchasing a money order.Panels C and D of Table 2 report fraud detection rates for different levels of inclusion in formal and informal financial services, respectively.We observe that fraud detection increases with the degree of financial inclusion: 14.67% of respondents utilizing none of the formal financial services report detecting fraud, in contrast to 57.14% of respondents utilizing all the eight formal financial services reporting fraud detection.We see a similar trend for the case of inclusion levels in informal financial services.
Overall, the descriptive analysis in Table 2 indicates that both financial knowledge and financial inclusion have a positive relationship with fraud detection.To further investigate the relationship between these two confounding characteristics for fraud detection, Figure 1 provides a visual representation of the percentage of respondents detecting fraud at various levels of financial knowledge, for the case of low and high financial inclusion.services.For both cases, low and high inclusion is defined as using fewer or more financial services, respectively, than the median number of financial services utilized in the sample.
Two notable observations emerge.First, we find that the financial knowledge and fraud detection relationship is similar for both low and high financial inclusion, and thus the relationship is not driven by level of financial inclusion.Second, for formal financial services, we see that individuals with high inclusion detect slightly more fraud when they possess higher levels of financial knowledge.In contrast, for informal financial services, it is the low financial knowledge individuals that detect more fraud when they are also highly utilizing the informal financial services.
Next, we discuss the summary statistics of the individual-and household-level characteristics in our sample.Table 3 reports the distribution of the respondents' age groups, gender, marital status and ethnicity, in addition to their household's income, metropolitan residency status and census region.We observe that our sample is composed of individuals aged 18 to 75 years and older, with variability across all age groups, reflecting the representation of the survey.The distributions of gender, marital status and ethnicity show that the largest groups are male, married and white, respectively.Respondents are heterogeneous with respect to education: only a small fraction (6.14%) report less than high school education, and the remaining with high school (24.83%), some college/associate's degree (30.10%), bachelor's degree (21.02%) or graduate/professional degree (17.90%).Household income exhibits variability; however, the preponderance of households (57.65%) report earnings of $60,000 or more per year.The majority of respondents live in metropolitan areas, approximately evenly spread across the US census regions.

Financial knowledge and fraud detection
We test the relationship between financial knowledge and fraud detection in a multivariate setting, accounting for a wide range of socio-economic and demographic attributes of the individuals.More specifically, we estimate the probit regression, for i = 1, . . ., N respondents.The dependent variable is the latent fraud detection propensity related to the fraud detection indicator variable (taking the value one if the respondent has detected fraud, and zero otherwise) via the probit link function.ε i is a i.i.d.standard normal error term.f inKnow denotes financial knowledge score (transformed to z-scores) and is the key explanatory variable of interest.To control for the level of financial inclusion, we include IncF ormal and IncInf ormal, which denote degree of inclusion in formal and informal financial services, respectively.Additionally, to account for the heterogeneity in the individual and household characteristics, we include a large set of control variables (denoted X) capturing age, gender, civil status, ethnicity, education, household income, metro residency status and census region.The definitions of all the control variables is provided in Appendix A.5.
Table 4 reports the estimation results.Column (1) presents the baseline results without accounting for degree of financial inclusion, while Columns (2) and (3) additionally include the levels of inclusion in formal and informal financial services.We observe that the effect of financial knowledge is stable and strongly significant at the 1% level across all the specifications.After controlling for inclusion levels in formal and informal services, the standardized beta coefficient on financial knowledge is 0.104.The estimation results show that financial knowledge plays an important role for detecting fraud and the finding holds even after controlling for the individual's level of financial inclusion.Further, the coefficients on degree of inclusion in the formal and informal financial services show significance at the 5% and 1% levels, respectively.The coefficient magnitude for the degree of inclusion in informal financial services is observed to be greater than that of the formal financial services.Although the exact mechanism remains unexplored due to data unavailability, the results support the conjecture that higher degree of inclusion in informal financial services exposes individuals to more fraud incidences.
Among the demographic attributes, we observe that education and income play an important role in detecting fraud.Considering the civil-status of respondents, the individuals who are single detect relatively less financial fraud, as compared to the base case of married individuals.None of the spatial characteristics of respondents have a significant impact on the ability to detect financial fraud, suggesting that geographic locations are unimportant in the digital era.

Financial behavior and fraud detection
In the section, we explore whether the financial behavior of individuals plays an important role in detecting banking fraud.To this end, we make use of a battery of survey questions that capture prudent financial behaviors, including active budgeting, setting and pursuing financial goals, paying bills on time, staying within budget, paying of credit card balance in full each month, checking accounts for errors, frugal spending and saving habits.Additionally, we create a composite financial behavior score for each individual by summing up the response choices from all the financial behavior questions.See Appendix A.2 for variables construction and definitions.To empirically test the relationship between financial behavior and fraud detection, we include the financial behavior variables (denoted f inBehav) as additional regressors in the probit regression, where the dependent variable is the latent fraud detection propensity related to the fraud detection indicator variable (taking the value one if the respondent has detected fraud, and zero otherwise) via the probit link function.Further, we include all the individual-and household-level control variables as in Equation (1).matter when it comes to the ability to detect fraud. 1 As found previously, the positive effect of financial knowledge in detecting fraud remains a strong influencing factor.Also as previously, the degree of inclusion in formal and informal financial services is significant in all regression specifications at the 5% and 1% levels, respectively.
Overall, the results suggest that efficient management of money through good financial behavior is inadequate when it comes to spotting fraud.The result alludes to the fact that banking fraud is complex, and its incidences are unexpected.Thus, a certain degree 1 It is conceivable that individuals with prudent financial behavior are less susceptible to fraud, and therefore detecting less fraud simply because they are targeted less.We test this conjecture by examining whether individuals with low financial behavior scores (i.e.those in the lowest tercile) are able to detect fraud due to perhaps experiencing higher incidences of fraud.We find that the coefficient on the low financial behavior score that explains the propensity of fraud detection turns positive, but remains insignificant.Additionally, when examining high financial behavior individuals (i.e.those in the highest tercile), we observe again an insignificant (and negative) relationship between high financial behavior and fraud detection.The results strengthen the conclusion that there is no clear relationship between financial behavior and the ability to detect fraud.
of financial sophistication through financial knowledge is necessary to detect fraud when it occurs.

Attentiveness and fraud detection
In this section, we study individuals' behavioral attention attribute, which is argued channels the financial knowledge influence on fraud detection.Although a causal relationship cannot be established due data unavailability, if financial knowledge enhances individuals' attention capacities, a strong correlation should be observed between the two attributes.Further, individuals with greater attention capacities, gained by being more financially knowledgeable, should be better at tasks such as detecting fraud.Conversely, individuals with attention deficiencies will find it difficult focusing on tasks and will tend to act without thinking.To this end, we measure individuals' degree of attentiveness using the 4-point Likert response statement: I often act without thinking through all the alternatives.
Respondents answer the statement by choosing one of the four responses, "completely well", "very well", "not very well", and "not at all".We measure attentiveness of an individual as the degree of disagreement with the statement.
To test the associations between financial knowledge, attentiveness and fraud detection, we specify the following two-step regression, The first step (as in Equation 3) is an ordinary least squares regression, where we regress financial knowledge on attentiveness and obtain the fitted value of attentiveness that is explained by financial knowledge.In the second step (as in Equation 4), we estimate a probit specification for the latent fraud detection propensity and include as independent variable the fitted value of attentiveness derived from the first step.
In Table 6, Columns (1a) and (2a) report the results from the first step, without and with the inclusion of individual-and household-level control variables, respectively.The corresponding second step results are reported in Columns (1b) and (2b).We find that financial knowledge loads highly significantly on attentiveness in the first step and in the second step, the fitted value of attentiveness is observed to significantly explain the propensity of fraud detection.Interestingly, using attentiveness directly as an explanatory variable for fraud detection propensities produces insignificant estimates for attentiveness (see Column ( 3)).The results indicate that the more financially knowledgeable an individual, the greater his/her attentiveness, and the attention capacities derived from financial knowledge constitute a strong influencing factor for detecting fraud.The findings lend credence to the underlying mechanism of behavioural attention that drives the relationship between financial knowledge and fraud detection.

Limits to financial knowledge and fraud detection
The findings so far establish a strong association between financial knowledge and fraud detection.Financially knowledgeable individuals are observed to better detect fraud and safeguard themselves against fraudulent schemes.This relationship however may be weaker for individuals who possess a low cognitive assessment of their life or a lower subjective wellbeing.That is, individuals with lower (higher) levels of subjective well-being will generally be less (more) attentive to their general well-being needs and have less (more) cognitive capacity to detect fraud.We test this premise in the data.To measure subjective well-being, we utilize three statements in the survey that invite Likert responses capturing respondents' subjective well-being: 1.I am satisfied with my life 2.I am optimistic about my future 3.If I work hard today, I will be more successful in the future Responses to all the statements are measured on a 7-point Likert scale.We accordingly map the responses to integers from 1 to 7, where the higher the integer, the higher degree of disagreement with the question.To estimate the effect of financial knowledge on fraud detection probabilities, given the different levels of subjective well-being, we specify a probit regression with interaction effects, for respondents i = 1, . . ., N .The dependent variable is the latent fraud detection propensity related to the fraud detection indicator variable (taking the value one if the respondent has detected fraud, and zero otherwise) via the probit link function.As independent variables, we include the interaction between financial knowledge (f inKnow) and the various levels of well-being (wellBeing), specified by the integer values for the Statements (1) to (3) shown above.We also include all individual-and household-level control variables in the regression specifications.
Table 7 reports the estimation results.Columns (1) to (3) report the results for the three subjective well-being statements.We observe that the interactions between financial knowledge and well-being levels 1 to 3 are statistically insignificant.However, for levels 4 and above, which relates to greater well-being, the financial knowledge coefficient estimates start to show significance.The results indicate that a low subjective well-being weakens the positive relationship between financial knowledge and individual propensities to detect fraud, while at higher levels of subjective well-being we observe that financial knowledge emerges as a significant determinant of an individual's abilities to detect fraud.Interestingly, however, we also observe that at very high levels of subjective well-being the relationship between financial knowledge and fraud detection breaks down.In this case, individuals appear too positive and optimistic to engage in effective fraud detection; as such, when it comes to detecting fraud it pays not being overly optimistic.Overall, subjective well-being of an individual plays an important moderating role in the relationship between financial knowledge and fraud detection.

Robustness analysis -instrumental variable approach
In estimating the effect of financial knowledge on the likelihood of detecting fraud, a potential source of concern might be a bias due to omitted variables influencing both financial knowledge and fraud detection.One such variable can be the respondents' digital literacy, which prior research has found to affect human capital accumulation and potentially financial literacy.For example, Servon & Kaestner (2008) find evidence of a possible link between digital literacy and financial knowledge.Also, Lee (2018) argues that digital media literacy can help people better understand "issues of misinformation and privacy, cybercrimes such as phishing and other types of fraud" (p.465).
To test that the results concerning the relationship between financial knowledge and fraud detection are not driven by unobserved factors, we employ an instrumental variable approach.We instrument financial knowledge by an indicator variable that takes the value of one if the parent who raised the respondent attended graduate school, and zero otherwise.
Graduate school degrees in the U.S. encompass master's degrees and PhD or other doctoral degrees, such as doctorates from medical and law schools.We argue that graduate parent as an instrument satisfies both the relevance and exclusion restrictions.First, individuals with graduate education, on average, earn higher incomes and acquire more wealth, subsequently investing in the acquisition of advanced financial knowledge.Over time, graduate parents will pass on their knowledge to their children, increasing also the children's financial knowledge.We thus expect a strong correlation between the respondents' parent having attended graduate school and their financial knowledge.Second, having a graduate parent will be, arguably, orthogonal to the respondents' fraud detection propensities, after controlling for various socio-economic and demographic characteristics.As the digital revolution is a recent phenomenon, the parents' graduate education will not entail the acquisition of digital literacy skills that can be readily passed onto their children.Moreover, we find in our analysis above that prudent financial behaviors, which can be effectively fostered by a parent with graduate education, are insignificant for fraud detection, thus also ruling out these indirect effects.Therefore, graduate parent, as an instrumental variable, extracts the exogenous part of residual variation in fraud detection, identifying all relevant parameters.
Given that our outcome variable of interest is the fraud detection indicator variable, the standard two-stage least squares estimator is likely to be inappropriate.We therefore estimate a probit model in which we instrument financial knowledge in a Limited Information Maximum Likelihood (LIML) framework.The regression setup is a recursive set of equations with the dependent variables fraud detection (f raudDetect) and financial knowledge (f inKnow), in which the endogenous variable, financial knowledge, appears on the right-hand-side of the fraud detection equation: where where the error terms ε i,1 and ε i,2 for fraud detection and financial knowledge equations, respectively, are related in a Seemingly-Unrelated Regression specification.More specifically, Equations ( 8) and ( 9) show that they are assumed to be jointly drawn from a multivariate normal distribution with a mean vector of zeros and a covariance matrix Σ.The standard deviation σ 1 is standardized to unity to identify the probit equation.The correlation term ρ captures possible endogeneity of financial knowledge for fraud detection, a testable quantity.
If the above identifying assumptions are satisfied, an estimate of β 1 yields the causal effect of financial knowledge on fraud detection.
Table 8 reports the estimation results.Column (1) reports the first-stage estimates relating to financial knowledge (Equation ( 7)), while Column (2) reports the second-stage estimates relating to fraud detection (Equation ( 6)).In Column (1), we observe that the graduate parent instrumental variable is significant at the 1% level, with a coefficient estimate of 0.162.The Kleibergen-Paap rk LM test indicates that we reject at the 1% level that the instrument relevance assumption is not satisfied.Overall, the use of graduate parent as an instrument provides a strong first-stage result, satisfying the instrumental variable relevance assumption.
The second-stage estimates in Column (2) show that the causal effect of financial knowledge on fraud detection is 0.564, with weak significance (p-value of 7%).Importantly, in contrast to the probit estimates in Table 4, we observe that the IV estimate of financial knowledge on fraud detection is approximately five times larger in magnitude.This is possibly a result driven by the subset of the population captured by the instrument.That is, respondents with a graduate parent may have significantly higher advanced financial knowledge than the rest of the population.Further, in addressing the initial concern of omitted variable bias, the Wald test of exogeneity finds no significant correlation between the errors of Equation ( 6) and ( 7) (H 0 : ρ = 0), indicating that we cannot reject exogeneity of financial knowledge for fraud detection.

Conclusion
We study the role of financial literacy, measured through the dimensions of financial knowledge and financial behavior, for fraud detection.Banking fraud is becoming increasingly complex such that detecting fraud requires a great deal of sophistication.As automated fraud detection systems do not always recognize fraudulent activities, banks place emphasis on their customers spotting and reporting fraudulent transactions in their accounts.However, the financial stakes are high, due to the previously documented evidence in the literature that consumers suffer from inattention.
We posit that financially literate individuals, through greater financial knowledge and prudent financial behavior, will be more attentive to fraud risk and better equipped to spot fraud incidences.Using a representative sample of US residents, we investigate this relationship and find strong evidence for the case of financial knowledge but not for that of financial behavior.More specifically, the results indicate that the more financially knowledgeable an individual, the greater is the fraud detection.This corroborates the conjecture that, with financial literacy, individuals become more attentive to detecting fraud when it transpires.In fact, additional analyses reveals that financially knowledgeable individuals possess a greater level of attentiveness and the attention derived from financial knowledge is a strong influencing factor for detecting fraud.
Contrary to financial knowledge, we observe that prudent financial behavior does not systematically matter when it comes to the ability to detect fraud.We proxy for prudence in financial behavior by measuring behaviors such as setting and pursuing financial goals; setting and consulting a budget; whether bills are paid on time; whether statements, bills and receipts are checked for errors; and whether the credit card balance is paid off in full each month.The weak result for financial behavior suggests that efficient management of finances does not directly correlate with greater effectiveness in spotting fraudulent behavior, and that it is indeed financial knowledge that provides the degree of sophistication necessary to be able to detect and deter banking fraud.
We observe that the positive effect of financial knowledge in detecting fraud remains a strong influencing factor even after accounting for the individuals' levels of financial inclusion.Notably, individuals participating in informal, much more than formal, financial sector services detect more fraud.The result suggests the potential higher incidences of fraud when using informal financial sector services.Further, we find that subjective well-being plays an important moderating role in the relationship between financial knowledge and fraud detection.Lower levels of subjective well-being can reduce attention due to greater cognitive loads, and thus attenuating the relationship between financial knowledge and fraud detection.
Our study has important policy implications considering the recent interest in the role of financial literacy for general consumers and retail investors from a behavioral perspective.
The findings suggest policy steps that emphasize consumer education programs to enhance financial knowledge to help consumers detect and deter banking fraud.

Figures
Figure 1: Rates of detected fraud given the levels of financial knowledge, split by low and high degrees of inclusion in formal and informal financial services.The lines indicate the percentage of respondents that detected fraud given their levels of financial knowledge and financial inclusion.Financial knowledge is defined as the composite score derived from nine financial knowledge questions, which elicit the respondents' understanding of various financial concepts.The financial knowledge composite score ranges from -2.053 to 1.267.Inclusion in formal (informal) financial services is the number of formal (informal) financial services respondents utilize.Low and high inclusion is defined as using fewer or more financial services, respectively, than the median number of financial services utilized in the sample.Table 1: Detection of banking fraud.This table reports the distribution (counts and percentages) of responses to the question: In the past 5 years, has someone without your permission used or attempted to use an existing account of yours, such as a credit or debit card, checking, savings, telephone, online, or insurance account?This table reports probit regression estimates of financial knowledge, financial behavior and financial inclusion, together with a host of demographic and financial control variables.The dependent variable takes the value of one if the respondent has detected banking fraud in the last five years, and zero otherwise.Financial knowledge is defined as the composite score derived from nine financial knowledge questions.The construction of the financial behavior variables used in Columns (1) to ( 11) are described in Appendix A.2.In Column ( 12), the composite financial behavior score is used, which is created by summing up the responses to the financial behavior questions.Inclusion in formal (informal) financial services is the number of formal (informal) financial services respondents utilize.Definitions of all the variables are reported in the appendix.Robust standard errors are reported in parentheses, and the stars * * *
(   (0.12) (0.12) (0.12) (0.12) (0.12) (0.12) (0.12) (0.12) (0.12) (0.12) (0.12)   3) report results for the three well-being statements, I am satisfied with my life, I am optimistic about my future and If I work hard today, I will be more successful in the future, respectively.The higher the integer value for the level of well-being, the higher the degree of agreement to the statement.Financial knowledge is defined as the composite score derived from nine financial knowledge questions.Definitions of all the variables are reported in the appendix.Robust errors are reported in parentheses, and the stars * * * , * * and * denote level of significance at 1, 5 and 10 percent, respectively. (

Age
The survey captures age of respondents in seven non-overlapping age brackets, between 18 and 74, and the eighth age bracket captures respondents older than 75.The variable "Age" for a respondent is equal to the midpoint age of the age bracket the respondents belong to.
For respondents in the eighth age bracket, the variable takes values equal to the lower limit of the age bracket.The variable is then transformed to z-score.

Female
It takes the value of one if respondent is female, and zero otherwise.

Widowed
It takes the value of one if the respondent has reported their civil status as widow, and zero otherwise.

Married
It takes the value of one if the respondent has reported their civil status as married, and zero otherwise.

Divorced
It takes the value of one if the respondent has reported their civil status as divorced, and zero otherwise.

Single
It takes the value of one if the respondent has reported their civil status as single, and zero otherwise.

Cohabiting
It takes the value of one if the respondent has reported their civil status as cohabiting, and zero otherwise.

Black
It takes the value of one if the respondent has reported their ethnicity as black, and zero otherwise.

Hispanic
It takes the value of one if the respondent has reported their ethnicity as hispanic, and zero otherwise.
Other It takes the value of one if the respondent has reported their ethnicity as other than white, black or hispanic, and zero otherwise.

Education
The survey captures the education level of respondents, classified into five categories.The variable "Education" takes values equal to the minimum number of schooling years required to attain the degree.To map the academic degrees to number of schooling years, we adapt the mapping in Fujiwara & Kawachi (2009).The variable is then transformed to z-score.

Income
The survey captures the income level of respondents, classified into nine non-overlapping income brackets and the ninth income bracket captures income of $150,000 or above.The variable "Income" for a respondent is equal to the midpoint income of the income bracket the respondents belong to.For respondents in the lower most income bracket, the variable takes values equal to the upper limit of the income bracket.Similarly, for the upper most income bracket, the variable takes values equal to the lower limit of the income bracket.The variable is then transformed to z-score.

Metropolitan
It takes the value of one if the respondent resides in a metropolitan residency area, and zero otherwise.
North-east It takes the value of one if the respondent resides in the north-east census region, and zero otherwise.

Mid-west
It takes the value of one if the respondent resides in the mid-west census region, and zero otherwise.

South
It takes the value of one if the respondent resides in the south census region, and zero otherwise.

"
The 2018 Identity Fraud Study released today [February 6, 2018] by Javelin Strategy & Research, revealed that the number of identity fraud victims increased by eight percent (rising to 16.7 million U.S. consumers) in the last year, a record high since Javelin Strategy & Research began tracking identity fraud in 2003.
Figure (a) plots inclusion in formal financial services, while Figure (b) plots inclusion in informal financial

Table 5
reports the estimation results.Columns (1) to (11) present the results for the financial behaviors added individually as regressors, while Column (12) shows the results for the financial behavior composite score variable.Surprisingly, we observe that except for the financial behavior variables Stayed within budget and Saving habit, which are marginally significant at the 5% and 10% levels, respectively, all the other financial behavior variables are insignificant.The test results indicate that prudent financial behavior does not substantially

Table 2 :
Fraud detection, financial knowledge, financial behavior and financial inclusion.This table reports the counts and percentages of fraud detection given the different levels of financial knowledge (Panel A), financial behavior (Panel B), and financial inclusion (Panels C and D), respectively.Financial knowledge is defined as the composite score derived from nine financial literacy questions.The financial knowledge scores range from -2.053 to 1.267.The financial behavior score is created by summing up the Likert responses (after mapping them to integers) to the financial behavior questions and is then divided into terciles.Inclusion in formal and informal financial services is measured as the number of financial services respondents hold or use from the formal and informal financial sector, respectively.Inclusion levels in formal and informal financial services range from 0 to 8 and 0 to 4, respectively.

Table 3 :
Individual and household-level summary statistics.This table reports the sample summary statistics (counts and percentage) of the demographic attributes including age, gender, civil status, ethnicity, education, income, residency status and census region.

Table 4 :
Financial knowledge and fraud detection.This table reports probit regression estimates of financial knowledge, formal and informal financial inclusion, together with a host of demographic and financial control variables.The dependent variable takes the value of one if the respondent has detected banking fraud in the last five years, and zero otherwise.Financial knowledge is defined as the composite score derived from nine financial knowledge questions.Inclusion in formal (informal) financial services is the number of formal (informal) financial services respondents utilize.Definitions of all the variables are reported in the appendix.Robust standard errors are reported in parentheses, and the stars * * * , * * and * denote level of significance at 1, 5 and 10 percent, respectively.

Table 5 :
Financial knowledge, financial behavior and fraud detection.

Table 6 :
Attentiveness and fraud detection.This table reports the results for individuals' attentiveness and fraud detection.The dependent variable in Columns (1a) and (2a) is respondent's attentiveness measured from the Likert response statement: I often act without thinking through all the alternatives.. Attentiveness variable takes values 1-4, corresponding to the level of disagreement with the statement.In Columns (1b), (2b) and (3), the dependent variable takes the value of one if the respondent has detected banking fraud in the last five years, and zero otherwise.Columns (1a) and (2a) report results of the first step OLS regression, while Columns (1b) and (2b) report the second step probit regressions.Definitions of all the variables are reported in the appendix.Robust standard errors are reported in parenthesis, and the stars * * * , * * and * denote level of significance at 1,5 and 10 percent, respectively.

Table 7 :
Financial knowledge, subjective well-being and fraud detection.This table reports the probit regression estimates of financial knowledge interacted with levels of well-being, controlling for financial behavior and financial inclusion, in addition to a host of financial and demographic variables.The dependent variable takes the value of one if the respondent has detected banking fraud in the last five years, and zero otherwise.Columns (1) to (

Table 8 :
Financial knowledge and fraud detection: instrumental variables approach.This table reports the results for the instrumental variable (IV) probit regression.Column (1) reports the first stage of IV regression, where the dependent variable is the financial knowledge score of respondents.Column (2) reports the second stage of IV regression, where the dependent variable takes the value of one if the respondent has detected banking fraud in the last five years, and zero otherwise.We use graduate parent as an IV for financial knowledge, which takes value of one if either of respondent's parents is a graduate, and zero otherwise.Definitions of all the variables are reported in the appendix.Robust standard errors are reported in parentheses, and the stars * * * , * * and * denote level of significance at 1, 5 and 10 percent, respectively.To what extent do you agree or disagree with each of the following statements?Panel B: Please indicate how often you have engaged in the following activities in the past six months To what extent do you agree or disagree with the following statements: Wordings of attentiveness and subjective well-being questions Please indicate your level of agreement with the following statement: Please indicate the degree to which you agree or disagree with each of the following statements: Individual and household-level control variable definitions Variable name Variable definition