121 The impact of the online marketplace on fraud: Evidence from Craigslist from its early adoption in 1995 to its wider expansion in 2006 Efthymia Antonoudi,1 Martin Seay,2 Han Na Lim,3 and Elizabeth Kiss4 Abstract This research addresses the influence of Craigslist’s adoption and presence on fraud arrests within metropolitan statistical areas (MSAs). Utilizing the consumer vulnerability framework (Hill & Sharma, 2020), the study used diverse data sources, including Craigslist entry data, the Uniform Crime Reporting (UCR) dataset, and the US Census Bureau Current Population Survey (CPS) data from 1995-2006. Employing differences-in-differences (DID) models, this study's primary findings indicate a reduction in fraud arrests, ranging from 11% to 23% following the introduction of Craigslist. While these results might appear counterintuitive, our findings suggest that online marketplace design and enforcement capacity may jointly influence fraud patterns. Creative Commons License This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License Recommended Citation Antonoudi, E., Seay, M., Lim, H., & Kiss, E. (2025). The impact of the online marketplace on fraud: Evidence from Craigslist from its early adoption in 1995 to its wider expansion in 2006, Financial Services Review, 33(4), 121-133. Background The internet has fundamentally altered the way the marketplace operates. Using the internet and advanced technologies as an online marketplace brought countless opportunities for consumers to increase their utility by decreasing the costs of searching for information (Kroft & Pope, 2014), increasing efficiency by matching consumers to suppliers of goods and services, and reducing the transaction costs associated with buying, selling and giving away used goods (Fremstad, 2017). 1 Corresponding author (eanton@uga.edu). University of Georgia, Athens, GA, USA. 2 Kansas State University, Manhattan, KS, USA. 3 California State University, Fullerton, CA, USA. 4 Kansas State University, Manhattan, KS, USA. There are conflicting hypotheses with respect to whether the introduction of Craigslist to the local consumer marketplaces has increased or decreased fraud. In this paper, we provide empirical evidence to test these two hypotheses. Craigslist, launched in the San Francisco Bay Area, is a classified advertising website operating since 1995 that enables multiple unrelated matching markets to interact on a single consolidated platform (Cunningham et al., 2017). Users can post advertisements about jobs, https://creativecommons.org/licenses/by-nc/4.0/ https://creativecommons.org/licenses/by-nc/4.0/ Financial Services Review 33(4) 122 housing, services, personal items, and for-sale items. In 2000, Craigslist expanded to include other major cities such as Boston, New York, and Los Angeles, and in 2001, it entered the Denver, Atlanta, and Austin markets. Gradually, Craigslist started operating in other smaller metropolitan statistical areas (MSAs) and, by 2010, covered most US cities and expanded to include multiple markets outside of the US (Cunningham et al., 2017). In 2005, Craigslist experienced remarkable growth and received more than 8 billion page views per month by the end of 2006 (Kroft & Pope, 2014). Because of the enormous growth and the widespread usage of the website after 2006, we focused our analysis on the years from Craigslist's inception in 1995 to its widespread expansion in 2006. Craigslist altered how consumers buy, sell, and dispose of secondhand goods (Fremstad, 2017). It created a new marketplace promoting online search and lowering information acquisition costs. Internet outlets increase market efficiencies by reducing search costs and lowering price dispersion (Brynjolfsson & Smith, 2000). Craigslist connects consumers in mutually beneficial transactions but the opportunity for fraudulent activity may also increase. For instance, in 2014, a man in Des Plaines, Illinois, was charged with multiple felonies after posting fraudulent apartment listings on Craigslist and collecting deposits from unsuspecting renters. Victims uncovered the scam when they attempted to move in and found the apartments already occupied or that the keys did not work (Knowles, 2014). While such cases illustrate that fraud can and does occur on digital platforms, the structured environment of Craigslist may also facilitate detection and prosecution in ways that were not possible with offline or traditional scams. This highlights how the introduction of Craigslist may not reduce all fraud but may influence how it is reported, tracked, and resolved. Building on this idea, we investigate whether the presence and adoption of Craigslist as an online marketplace affected fraud-related arrests in locations that adopted the platform compared to those that did not. This study contributes to the financial services literature by examining how online marketplaces influence consumer risk exposure, fraud-related enforcement, and financial behavior. As more financial transactions occur digitally, understanding the role of marketplace infrastructure—such as Craigslist—can inform fraud prevention strategies, regulatory frameworks, and consumer protection practices. Review of the Relevant Literature Fraud Millions of individuals fall victim to fraud around the world every year. According to the recent United States Federal Trade Commission (FTC) data book, individuals reported losing $8.8 billion to scams in 2022, an increase of $2.6 billion yearly (Rayo, 2023). While the dollar amount continues to increase, the number of individuals affected has remained staggering; about 35.6 million were fraud victims in 2004 (Anderson, 2004). In 2006, the FTC received over 670,000 Consumer Sentinel Complaints amounting to over $1.1 billion in fraud losses, of which 36% were identity theft complaints, and 64% were other types of fraud (Federal Trade Commission, 2006). Fraud can also have enormous nonmonetary costs, such as emotional stress and psychological trauma. These nonmonetary costs are hard to quantify and likely greater than the financial losses (Lee & Soberon‐Ferrer, 1997). Financial crimes are perpetrated in various ways, with the most common types of fraud related to false representation and identity theft. The top consumer fraud issues identified by the 2004 FTC survey are advance-fee loan scams, being billed for membership without having agreed to it, credit card insurance and credit repair services, paying money for a purchase without receiving the promised prize, being billed for internet services without agreeing, and purchasing a membership in a pyramid scheme (Anderson, 2004). Not much has changed 20 years later. In the United States, investment scams were the costliest form of fraud in 2022, with reported losses to American consumers of $3.8 billion, followed closely by impersonator scams with reported losses of $2.6 billion (Rayo, 2023). Antonoudi et al. 123 Technology and Fraud Current scholarly work on online fraud has examined a variety of types and methods of these crimes. Online consumer fraud victimization has concentrated on the sophisticated nature of fraudsters (Garg & Nilizadeh, 2013; Lee, 2021a, 2021b; Van Wilsem, 2013). Park et al. (2014) analyzed the prevalence of advance-fee fraud paying a fee upfront in anticipation of receiving something of greater value- on Craigslist and showed that ten groups of scammers were responsible for nearly half of the total scam attempts. Other scams thriving on Craigslist that have been the subject of scholarly work were automobile-related scams targeting mostly educated white males (Garg & Nilizadeh, 2013). Theory and Conceptual Framework Consumer vulnerability occurs when individuals face a heightened risk of harm due to limited access to resources or reduced control over their use in the marketplace (Hill & Sharma, 2020). In online marketplaces like Craigslist, such vulnerability can be shaped by structural features—such as platform availability—that may alter exposure to fraud. In our framework, the key independent variable is the presence or absence of Craigslist in a given metropolitan statistical area (MSA). This structural condition creates different contexts in which consumers operate, potentially influencing fraud risk. While individual and demographic factors such as income, employment, education, and race also play a role, our analysis focuses on how platform availability interacts with these factors to affect fraud-related outcomes. According to the consumer vulnerability framework, one might expect Craigslist’s entry into a market to heighten exposure to fraud by expanding opportunities for deceptive interactions and reducing face-to-face accountability. This would suggest a positive relationship between platform adoption and fraud arrests. However, several mechanisms may produce the opposite effect. Craigslist’s peer-to- peer design may increase transparency, foster community moderation through user flagging of suspicious listings, and generate digital trails that deter criminal behavior. Additionally, if Craigslist displaced aggressive or misleading advertising from traditional media, it may have reduced consumer exposure to certain fraud schemes. These competing mechanisms make the relationship between Craigslist and fraud an empirical question. Research Question Did the introduction and presence of the Craigslist online marketplace impact fraud arrest occurrences? Hypothesis Given the competing theoretical arguments, we adopt a two-tailed hypothesis structure: H₀: Craigslist’s entry has no effect on fraud arrest rates. H₁: Craigslist’s entry is associated with changes in fraud arrest rates. Methods Data and Sample We use a combination of three different data sources to determine Craigslist entry, fraud arrests, and county-level characteristics. Upon request, the authors of Kroft & Pope (2014) provided data associated with Craigslist entry for every MSA from 1995 to 2006. Craigslist was first introduced in San Francisco (1995), then in other large MSAs like New York and Boston (2000), followed by mid-sized cities in the early 2000s. Kroft and Pope (2014) noted that early adopters tended to be cities with higher incomes, higher education levels, and larger populations. To conserve space, the timeline and average demographic characteristics of early versus late adopters are available upon request. We used the US Census Bureau county-level data (U.S. Census Bureau, 1995-2006) to extract the study's control variables, including median income, unemployment percentage, poverty rate, education, and racial background of each county. MSAs are defined in terms of entire counties, consisting of at least one urbanized area with a population of 50,000 or more, along with adjacent counties with a high degree of economic and social integration with the urbanized core. We used historical delineation files from the US Census Bureau webpage to aggregate CPS data at Financial Services Review 33(4) 124 the MSA level (U.S. Census Bureau, 2023). We converted the data to the most recent definitions. Police agencies report county-level fraud arrest data to the Federal Bureau of Investigation’s Uniform Crime Reporting (UCR) program each year, providing base data for this analysis. We used information from the Quarterly Census of Employment and Wages (QCEW) to aggregate county-level data at the MSA level to match the Craigslist entry data. For the study's primary analysis, the period from 1995 to 2006 was investigated using data from 1991 to test the parallel pre-trends assumption. Unfortunately, the UCR data aggregates fraud types and does not allow us to isolate fraud categories most directly associated with Craigslist transactions. During this analysis period, the sample, on average, included 384 MSAs. We analyzed MSAs over the 15 years between 1991 and 2006, except 1993. That year was excluded because the FBI crime data were not available. The available data provided an initial sample size of 5,760 MSAs, reduced by 61 MSAs that did not report crime data to a sample of 5,699 MSAs. As Anderson (2014) noted, there are problems with the fraud reports. Unusually large fraud reports have arrest counts that are a compilation of police agency reports, and not all agencies submit reports consistently every year (Anderson, 2014). Following Anderson (2014) and Fone et al. (2023), we dropped from the primary analysis MSAs that reported unusually high fraud arrests defined as more than two standard deviations from the mean.5 As such, we removed 377 MSAs, providing a final analytic sample of 5,322 observations. Of this final sample of 5,322 observations, 399 MSAs reported zero fraud arrests during the analysis period in a given year, with the remaining MSAs reporting arrests above zero. To further safeguard against this issue, we control for the number of agencies reporting within an MSA for any given year. To do that, we added a specification that controls for the number 5 MSAs with fraud arrest rates more than two standard deviations above the mean were excluded. These MSAs tended to have smaller populations and more volatile reporting patterns. 6 One possible explanation is that Craigslist’s introduction shifted local fraud patterns by of agencies that report arrest data (Anderson, 2014) and includes only those MSAs with 90% coverage or above that provided a sample of 3,624 observations. Coverage is the percentage of police and other enforcement agencies that reported arrests. Variables Dependent Variable The study’s dependent variable is fraud arrests per 10,000 people in each MSA. UCR data defines fraud as the intentional perversion of the truth to induce another person or other entity in reliance upon it to part with something of value or to surrender a legal right (Federal Bureau of Investigation, 2004). This definition includes fraudulent conversion and obtaining money or property by false pretenses. Confidence games and bad checks, leaving a full-service gas station without paying, credit card/automatic teller machine fraud, impersonation, welfare fraud, and wire fraud were all included in this definition of fraud (Federal Bureau of Investigation, 2004). Forgery and counterfeiting are excluded from this variable. While this definition includes many types of consumer fraud, such as wire fraud, impersonation and false pretenses, we note that some broader categories are also included. The UCR data does not allow us to isolate only consumer fraud, which we acknowledge as a limitation in interpreting results.6 Independent Variable The primary predictor variable is a binary variable of whether or not Craigslist was available in the metropolitan statistical area (MSA) in a given period. The Craigslist variable is constructed as a binary variable coded as 0 if Craigslist was not available and coded as 1 if Craigslist was available in that MSA for each year during the analysis period. Control Variables influencing enforcement focus or offender tactics. For example, fraudsters may have moved away from traditional schemes (e.g., bad checks or benefit fraud) toward online impersonation or listing scams, which could still affect the overall fraud arrest rate. Antonoudi et al. 125 After aggregating available individual and household-level data, we measured control variables at the MSA level. Control variables included household median income, poverty rate, and racial composition at the respondent level. In addition, we aggregated the MSA-level employment status and education status collected at the respondents' level; a summary of all variables, their definitions, and data sources is provided in Appendix Table A. Statistical Analyses A difference-in-differences (DID) approach compares the change in fraud arrests over time between MSAs that adopted Craigslist and those that did not. This statistical method allows us to isolate the effect of Craigslist's introduction by controlling for broader time trends and regional fixed characteristics. Equation The empirical model is specified as equation (1) below. Ym,t = β0 + β1Craigslistm * time t + β2Xt + ym + nt + ε,m,t where Ym,t represents fraud/10,000 people and the subscript m is the MSA with t representing the year. The parameter β0 is the intercept and β1 is the coefficient for the Craigslist dummy variable. The coefficient y is the MSA fixed effects and n is the period fixed effects. Finally, ε,m,t is the error term. The key terms in our model are explained in Table 1 below. Table 1. Explanation of Terms Used in the Difference-in-Differences (DID) Model Term or Symbol Description How It Is Measured or Applied Y (m,t) Dependent variable: fraud arrests per 10,000 people in MSA m at time t Calculated from UCR data, adjusted by MSA population β₀ Intercept Constant term in the model Craigslist (m) × Time (t) Interaction term: treatment effect (DID estimator) Binary = 1 if Craigslist is present in MSA m during year t; 0 otherwise β₁ Coefficient of interest (treatment effect) Measures Craigslist's effect on fraud arrests X (t) Control variables Income, unemployment, education, race; aggregated to MSA level β₂ Coefficients on control variables Estimated effects of each covariate MSA Fixed Effects (yₘ) Controls for time-invariant MSA characteristics E.g., local policy or geography that does not vary over time Year Fixed Effects (nₜ) Controls for time-specific shocks National events or economic conditions in year t Error Term (ε m,t) Random error Unobserved factors not captured in the model Note. The DID estimate reflects: (Post-treatment − Pre-treatment in treated MSAs) − (Post-treatment − Pre-treatment in control MSAs). This structure helps isolate the effect of Craigslist from confounding time and regional factors. We used differences-in-differences models (DID) to examine associations between the UCR and Census-based variables over time and to study the differential impact of Craigslist’s presence in MSAs that adopted it versus those Financial Services Review 33(4) 126 that did not. The approach compares the change in outcomes over time between a group exposed to a treatment (MSAs with Craigslist) and a control group (MSAs without Craigslist). This design controls for unobserved, time-invariant characteristics of each MSA (via fixed effects) and trends common to all MSAs over time (year- fixed effects). In this study, the "treatment" is Craigslist's entry into the market. The DID estimate captures the difference between pre-and post-Craigslist fraud arrests in treated areas, net of any changes in untreated areas over the same time. Results Table 2 provides summary statistics for fraud- related crime arrests, financial status, and demographics based on UCR and IPUMS CPS data sources. The mean number of fraud arrests per 10,000 people was 10.488, with a standard deviation of 11.228. It should be noted that the total number of observations dropped when looking at aggregated CPS data by MSA, as not all MSAs had reported data. Table 2. Summary Statistics for Craigslist with UCR and IPUMS CPS Data Variable Mean Standard Deviation N Craigslist 0.071 0.258 5,322 Fraud Arrests/10Kp 10.488 11.228 5,322 Median Household Income 47,363.474 9,715.377 4,292 Poverty Rate 12.805 4.521 4,292 Unemployment Rate 0.028 0.011 3,019 Income under 50K 0.525 0.132 3,022 50K-100K income 0.191 0.072 3,020 % African/American 0.114 0.107 2,902 Less than high-school 0.175 0.085 3,022 High-school 0.234 0.075 3,022 Some College 0.209 0.043 3,022 Note. Summary statistics reflect averages across Metropolitan Statistical Areas (MSAs) from 1995–2006. Craigslist entry data were obtained from Kroft & Pope (2014). Fraud arrest data are from the FBI’s Uniform Crime Reporting (UCR) program, aggregated to the MSA level. Demographic and socioeconomic variables were derived from IPUMS CPS data. Observations vary across variables due to missing data in some MSAs or years 127 The relatively high standard deviation of fraud arrests per 10,000 residents (11.228) compared to the mean (10.488) reflects the expected right- skewness in crime data. To further address distributional concerns, we conducted additional robustness checks using alternative specifications, including the log transformation and standardization of the dependent variable. These yielded consistent treatment effects, suggesting that the skewness in fraud arrest data does not substantively affect the main results. To conserve space, results from the alternative specifications are available upon request. We also conducted robustness checks by excluding extreme outliers with alternative model specifications. Table 3 presents the results of the difference-in- differences analysis estimating the effect of Craigslist entry on fraud arrests.7 Table 3 presents the results using the full sample (column 1), a coverage of 90% or more (column 2), the full sample using the MSA population as weight (column 3), and a coverage of 90% or more using the MSA population as the weight to account for the different size of each MSA (column 4). Some MSAs may have a population of a few hundred thousand people, while others can have a population in the millions. Overall, results in Table 3 support the rejection of the null hypothesis of no effect in favor of the alternative hypothesis that fraud was reduced by the introduction of Craigslist, based on the significant and negative coefficient on the Craigslist variable. Given that the mean fraud arrests per 10,000 people was 10.488, the estimate in column (1) indicated that after Craigslist entry, the fraud arrests decreased by 0.709 or about 7% compared to the control group. The results in column (2), where only MSAs with 90% coverage or more were included, painted a similar picture, showing that fraud arrests decreased by about 10% after Craigslist entered an MSA. The last two columns, (3) and (4), where we used population weights, produced similar results. When accounting for the population weights, the coefficients showed a decrease of 11% and 23% for the full sample and the one with 90% coverage or higher, respectively. The results were generally robust to other specifications using other coverage levels. Table 3 Differences-in-Differences Analysis with All Years and Coverage Restrictions and Weights All MSAs MSAs with Coverage>90% All MSAs, Weighted MSAs with Coverage>90%, Weighted (1) (2) (3) (4) B SE B B SE B B SE B B SE B Craigslist - 0.709*** 0.294 -0.994*** 1.117 -1.127** 0.586 -2.445*** 0.792 Obs. 5,322 3,624 5,322 3,624 e (e2-a) 0.75 0.775 0.789 0.807 e (df-a) 377 370 377 370 Note: * p < 0.05; ** p < 0 .01; *** p < 0.001. All specifications include MSA-fixed effects and time-fixed effects. Robust standard errors (SE) are provided in separate columns. To validate the DID approach, we tested for parallel pre-treatment trends using event study specifications with leads and lags of Craigslist entry (Table 4). No significant differences were observed in the pre-entry periods, supporting the validity of the DID assumptions. 7 Results are robust to the exclusion of San Francisco, the first city to adopt Craigslist. The estimates remain statistically significant and consistent in magnitude. Full results are available upon request. 128 Table 4 Leads and Lags All MSAs MSAs with Coverage >90% All MSAs Weighted MSAs with Coverage>90%, Weighted (1) (2) (3) (4) B SE B B SE B B SE B B SE B 5 periods prior -0.799** 0.357 -0.865** 0.386 -0.410 0.722 0.137 1.005 4 periods prior -0.715** 0.35 -0.722** 0.436 -0.274 0.57 -0.099 0.78 3 periods prior 0.561 0.345 -0.203 0.432 0.245 0.891 0.654 1.150 2 periods prior -0.110 0.335 0.168 0.459 0.493 0.57 0.606 0.843 1 period prior -0.442 0.364 -0.507 0.488 0.508 0.68 -0.545 0.54 1 year after -0.732 0.463 -0.805* 0.469 -0.152 0.574 -1.033*** 0.61 2 years after -0.893* 0.503 -1.286* 0.704 -1.925* 1.170 -3.228** 1.517 3 years after -0.731 0.587 -1.060 0.858 -1.851 1.378 -3.187* 1.839 4 years after -1.157 0.794 -1.376 1.014 -2.931* 1.765 -4.263* 2.177 5 years or more after -0.995 0.742 -1.253 0.901 -3.262** 1.502 -4.725** 1.858 Obs. 5,322 3,624 5,322 3,624 e (e2-a) 0.749 0.774 0.793 0.812 e (df-a) 377 370 377 370 Note: * p < 0.05; ** p < 0.01; *** p < 0.001. All specifications include MSA-fixed effects and time-fixed effects. Robust standard errors are provided in separate columns. To further support the robustness of our analysis and incorporate additional control variables, we accessed CPS data spanning from 1995 to 2006 via IPUMS CPS (Flood et al., 2023). We subsequently aggregated the control variables at the MSA level by aggregating the data utilizing Bureau of Labor Statistics delineation files (U.S. Census Bureau, 2023). The empirical results were not affected by this adjustment. These additional results are available upon request. Robustness Checks In Table 5, we conducted a series of robustness checks to examine the lagged effects of Craigslist entry on fraud arrests. These checks aimed to investigate whether the introduction of Craigslist in prior years had any significant influence on subsequent fraud arrest rates. The table provides results for different time lags, ranging from one to three years, and incorporates controls for lagged outcome variables. More specifically, column (1) used fraud arrests forwarded by one year regressed on Craigslist entry using the full sample results. For example, instead of using fraud arrests in 2000, we used fraud arrests in 2001. Column (2) used fraud arrests forwarded by 2 years, and column (3) used fraud arrests forwarded by 3 years. The estimates in all three columns were negative and significant, indicating that Craigslist entry in prior years had a negative impact on fraud arrests. Columns (4), (5), and (6) added controls for the lagged outcome variable. For example, column (4) added an additional control for the fraud arrests in a prior year (i.e., if the dependent variable was fraud arrests in 2000, the lagged variable was fraud arrests in 1999). The estimates remained consistently negative and statistically significant in these columns, except Antonoudi et al. 129 for the last column, where statistical significance was not observed. Table 5. Robustness Check with Lagged Effects Post lag1 Post lag 2 Post lag 3 (1) (2) (3) (4) (5) (6) B SE B B SE B B SE B Craigslist (SE) -1.858*** (0.707) -2.594*** (0.788) -1.985** (0.788) -1.149** (0.547) -1.292** (0.553) -0.220 (0.519) Fraud Arrests (SE) 0.602* (0.037) Post lag 1 (SE) 0.701*** (0.046) Post lag 2 (SE) 0.68*** (0.04) Post lag 3 Obs. 5,321 5,320 5,319 5,321 5,320 5,319 e (e2-a) 0.665 0.573 0.518 0.737 0.729 0.715 e (df-a) 377 377 377 377 377 377 Note: * p <0.05; ** p <0.01; *** p <0.001. All specifications include MSA-fixed effects and time-fixed effects. Robust standard errors are provided in brackets in separate rows. To further assess the robustness of our findings, we conducted several additional checks. These included models interacting Craigslist’s presence with poverty and income levels, specifications using different sample restrictions, and estimators robust to staggered treatment timing, such as the Callaway and Sant’Anna (2021) approach. Because this method requires a balanced panel, we restricted the analysis to MSAs with complete data across all years. Across all tests, the results consistently supported the main conclusion: Craigslist’s entry was associated with a statistically significant decline in fraud arrests. Full results are available upon request. Discussion and Implications Discussion The findings from this study consistently show a negative relationship between Craigslist entry and fraud arrest rates. Rather than increasing fraud, Craigslist’s presence may shape detection processes, reduce certain types of fraud, or shift activity to environments where it is less likely to result in arrest. There are several possible explanations for this outcome. The platform's peer-to-peer model may enhance transparency, limit third-party intermediaries, and increase the chance that users identify and flag suspicious behavior. Even in largely anonymous settings, digital trails—such as emails or IP addresses— can deter certain types of fraud by increasing the risk of detection. In addition, Craigslist may displace more misleading or aggressive advertising that consumers might otherwise encounter offline. These results are consistent with prior studies suggesting that Craigslist and similar platforms can reduce certain types of offline risks. For instance, Gurun et al. (2016) found that Craigslist’s introduction reduced high-cost mortgage advertising, while Cunningham et al. (2024) linked its “erotic services” section to declines in violent crime. At the same time, other work, such as Heese et al. (2022), has shown that Craigslist may reduce institutional oversight in other domains, such as corporate governance. This suggests that platform effects may vary depending on the type of fraud or enforcement mechanism involved. Overall, while our study does not test mechanisms directly, the findings are consistent with the idea that platform design and user interactions can help mitigate consumer vulnerability in digital markets. These dynamics warrant further investigation in future research. Financial Services Review 33(4) 130 Conclusions Utilizing both traditional DID analysis and the staggered DID approach along with a wider range of tests for the robustness of the findings, we consistently find a statistically significant negative relationship between Craigslist entry and fraud arrests across various MSAs. This finding supports the view that the introduction of Craigslist offers potential benefits in securing consumer transactions in a region. Nevertheless, it is crucial to highlight that a decrease in fraud arrests does not automatically imply a decline in the overall instances of fraud. This research illuminates the nuanced impacts of online marketplaces like Craigslist on societal metrics such as fraud arrests. The findings underscore the need for continued exploration in this domain, ensuring that the implications of these platforms are measured, examined, and understood. Our results align with findings from other empirical studies suggesting that the introduction of Craigslist leads to a decrease in certain types of offline crimes by shifting transactions from less public spaces to more public and digital spaces. For example, the introduction of the "erotic services" section on Craigslist was linked to a reduction of violent crimes like female homicide rates, suggesting that the platform contributed to making sex work safer by moving it off the streets and allowing workers to better screen clients (Cunningham et al., 2024). While small-scale frauds, like false listings, might persist and go unreported (mainly if they have not caused substantial financial or emotional loss for the victims), introducing a public and popular platform might reduce larger-scale fraud schemes due to enhanced public scrutiny. Unlike numerous other online endeavors, Craigslist has maintained its significant popularity for over twenty years (Oravec, 2014). As such, listings receive significant attention from the public. Larger-scale fraud schemes may find it challenging to operate under such scrutiny without being flagged or reported by multiple users. In-person interactions, often recommended for exchanges of goods, can reduce the anonymity that some larger fraud schemes rely upon. Furthermore, Newmark and his colleagues at Craigslist have collaborated with law enforcement agencies throughout the United States, limiting postings that might lure people into unsafe or illegal situations (Freese, 2011). Implications and Limitations This study provides empirical evidence of a decrease in fraud arrests following the adoption of Craigslist. However, this decrease does not necessarily mean a decrease in fraud occurrence. It could imply a shift in how fraud is detected and reported, with more incidents being resolved through the platform's mechanisms or private settlements rather than official law enforcement channels. Online platforms like Craigslist can stimulate local economies by offering an avenue for people to sell and buy used and other goods using a free or low-cost medium for trading. As such, economic benefits and opportunities might indirectly contribute to reducing economic- driven types of fraud. The findings of this study can instill more confidence in consumers to transact on online platforms like Craigslist. Aware of the platform's impact, consumers might become more proactive in educating themselves about safer online transactions, utilizing platform guidelines correctly and best practices to avoid scams. This engagement can lead to economic savings and access to better deals along with a broader range of products and services. Consumers should still exercise caution and understand that fraud that goes undetected or unreported on these platforms likely still exists. Overall, consumers must understand the differences between online platforms, stay updated on recent developments, and take vigilant and proactive measures to protect themselves from potential risks. Financial services professionals need to understand emerging digital behaviors to help clients navigate online marketplaces more securely. By tailoring financial advice to reflect the growing role of peer-to-peer platforms, practitioners can support informed, risk-aware participation in digital economic activity- particularly as clients pursue side income or engage in informal trading environments. This study contributes to the fraud and consumer behavior literature and demonstrates how DID methods can be applied to evaluate marketplace innovations within financial services. As Antonoudi et al. 131 financial transactions increasingly occur online, platforms like Craigslist, Venmo, and Facebook Marketplace offer rich opportunities for further DID-based evaluations—ranging from fraud to credit access, trust, and digital financial literacy. We hope this study encourages future applications of quasi-experimental designs in digital financial research. This study has several limitations. First, the skewed distribution of fraud arrests is inherent in administrative crime data. While we employed robustness checks and sample restrictions to address this issue, future studies may benefit from alternative fraud measures, including monetary loss or victim-reported data, to capture additional dimensions of fraud. Second, the Uniform Crime Reporting (UCR) dataset does not allow isolation of consumer fraud from other fraud-related arrests. While many categories (e.g., impersonation, wire fraud, false pretenses) likely reflect crimes affecting consumers, the aggregate nature of the data introduces some ambiguity. Third, the use of arrests is an indirect measure of actual fraud occurrence that depends on several factors related to crime reporting and law enforcement action. Fourth, although our difference-in-differences (DID) framework adjusts for unobserved time-invariant differences and shared temporal shocks, Craigslist adoption may still be correlated with unobserved local characteristics. While we used robustness checks and staggered rollout to mitigate this concern, a formal probit model predicting Craigslist entry could further validate our identification strategy and remains an area for future research. Finally, we do not test the mechanisms underlying the observed reduction in arrests—such as digital trails, peer flagging, or reduced exposure to deceptive advertising. These remain promising areas for future research on how marketplace design influences consumer safety outcomes. References Anderson, D. M. (2014). In school and out of trouble? The minimum dropout age and juvenile crime. Review of Economics and Statistics, 96(2), 318-331. Anderson, K. B. (2004). Consumer fraud in the United States: An FTC survey. https://www.ftc.gov/sites/default/files/d ocuments/reports/consumer-fraud- united-states-ftc- survey/040805confraudrpt.pdf Brynjolfsson, E., & Smith, M. D. (2000). Frictionless commerce? A comparison of Internet and conventional retailers. Management Science, 46(4), 563-585. Callaway, B., & Sant’Anna, P. H. (2021). Difference-in-differences with multiple time periods. Journal of Econometrics, 225(2), 200-230. Cunningham, S., DeAngelo, G., & Tripp, J. (2017, January). Craigslist’s effect on violence against women. Paper presented at the American Economic Association Conference, Chicago, IL. Cunningham, S., DeAngelo, G., & Tripp, J. (2024). Did Craigslist’s Erotic Services Reduce Female Homicide and Rape? Journal of Human Resources, 59(1), 280-315. Federal Bureau of Investigation. (2004). Uniform Crime Reporting Handbook. U.S. Department of Justice. https://ucr.fbi.gov/additional-ucr- publications/ucr_handbook.pdf Federal Trade Commission. (2006). Consumer fraud and identity theft complaint data January - December 2006. https://www.ftc.gov/sites/default/files/d ocuments/reports_annual/sentinel-cy- 2006/sentinel-cy2006.pdf Flood, S., King, M., Rodgers, R., Ruggles, S., Robert, J. W., Backman, D., Chen, A., Cooper, G., Richards, S., Schouweiler, M., & Westberry, M. (2023). IPUMS CPS: Version 11.0 [Dataset]. https://www.ipums.org/projects/ipums- cps/d030.V11.0 Fone, Z. S., Sabia, J. J., & Cesur, R. (2023). The unintended effects of minimum wage increases on crime. Journal of Public Economics, 219, 104780. Freese, S. M. (2011). Craigslist The Company and its Founder (R. Rowell, Ed.). ABDO Publishing Company. Fremstad, A. (2017). Does craigslist reduce waste? Evidence from California and https://www.ftc.gov/sites/default/files/documents/reports/consumer-fraud-united-states-ftc-survey/040805confraudrpt.pdf https://www.ftc.gov/sites/default/files/documents/reports/consumer-fraud-united-states-ftc-survey/040805confraudrpt.pdf https://www.ftc.gov/sites/default/files/documents/reports/consumer-fraud-united-states-ftc-survey/040805confraudrpt.pdf https://www.ftc.gov/sites/default/files/documents/reports/consumer-fraud-united-states-ftc-survey/040805confraudrpt.pdf https://ucr.fbi.gov/additional-ucr-publications/ucr_handbook.pdf https://ucr.fbi.gov/additional-ucr-publications/ucr_handbook.pdf https://www.ftc.gov/sites/default/files/documents/reports_annual/sentinel-cy-2006/sentinel-cy2006.pdf https://www.ftc.gov/sites/default/files/documents/reports_annual/sentinel-cy-2006/sentinel-cy2006.pdf https://www.ftc.gov/sites/default/files/documents/reports_annual/sentinel-cy-2006/sentinel-cy2006.pdf Financial Services Review 33(4) 132 Florida. Ecological Economics, 132, 135-143. Garg, V., & Nilizadeh, S. (2013, May). Craigslist scams and community composition: Investigating online fraud victimization. In 2013 IEEE Symposium on Security and Privacy Workshops (SPW) (pp. 123-126). IEEE. https://doi.org/10.1109/SPW.2013.21 Gurun, U. G., Matvos, G., & Seru, A. (2016). Advertising expensive mortgages. The Journal of Finance, 71(5), 2371–2416. https://doi.org/10.1111/jofi.12423 Heese, J., Pérez-Cavazos, G., & Peter, C. D. (2022). When the local newspaper leaves town: The effects of local newspaper closures on corporate misconduct. Journal of Financial Economics, 145(2), 445–463. https://doi.org/10.1016/j.jfineco.2021.08 .015 Hill, R. P., & Sharma, E. (2020). Consumer vulnerability. Journal of Consumer Psychology, 30(3), 551-570. Knowles, J. (2014, October 29). Additional charges filed in Craigslist rental scam. ABC7 Chicago. https://abc7chicago.com/i-team- craigslist-apartment-scam-rental-des- plaines/372081/ Kroft, K., & Pope, D. G. (2014). Does online search crowd out traditional search and improve matching efficiency? Evidence from Craigslist. Journal of Labor Economics, 32(2), 259-303. Lee, C. S. (2022). How online fraud victims are targeted in China: A crime script analysis of Baidu Tieba C2C Fraud. Crime & Delinquency, 68(13–14), 2529–2553. https://doi.org/10.1177/0011128721102 9862 Lee, C. S. (2021b). Online Fraud Victimization in China: A Case Study of Baidu Tieba. Victims & Offenders, 16(3), 343-362. Lee, J., & Soberon‐Ferrer, H. (1997). Consumer vulnerability to fraud: Influencing factors. Journal of Consumer Affairs, 31(1), 70-89. Oravec, J. A. (2014). Craigslist in crisis: Issues of censorship and moral panic in the context of online communities. International Journal of the Academic Business World, 8(2), 1-11. Park, Y., Jones, J., McCoy, D., Shi, E., & Jakobsson, M. (2014). Scambaiter: Understanding targeted Nigerian scams on Craigslist. In NDSS 2014: Proceedings of the Network and Distributed System Security Symposium. Internet Society. https://elaineshi.com/docs/scambaiter.pd f Rayo, A. (2023, February). The top scams of 2022. https://consumer.ftc.gov/consumer- alerts/2023/02/top-scams-2022 U.S. Census Bureau. (1995-2006). U.S. Census Bureau county–level data [Data set]. U.S. Census Bureau. (2023, August 4). Delineation Files. https://www.census.gov/geographies/ref erence-files/time-series/demo/metro- micro/delineation-files.html Van Wilsem, J. (2013). 'Bought it, but Never Got it' Assessing Risk Factors for Online Consumer Fraud Victimization. European Sociological Review, 29(2), 168-178. https://doi.org/10.1093/esr/jcr053 https://doi.org/10.1111/jofi.12423 https://abc7chicago.com/i-team-craigslist-apartment-scam-rental-des-plaines/372081/ https://abc7chicago.com/i-team-craigslist-apartment-scam-rental-des-plaines/372081/ https://abc7chicago.com/i-team-craigslist-apartment-scam-rental-des-plaines/372081/ https://elaineshi.com/docs/scambaiter.pdf https://elaineshi.com/docs/scambaiter.pdf https://consumer.ftc.gov/consumer-alerts/2023/02/top-scams-2022 https://consumer.ftc.gov/consumer-alerts/2023/02/top-scams-2022 https://www.census.gov/geographies/reference-files/time-series/demo/metro-micro/delineation-files.html https://www.census.gov/geographies/reference-files/time-series/demo/metro-micro/delineation-files.html https://www.census.gov/geographies/reference-files/time-series/demo/metro-micro/delineation-files.html 133 Appendix Table A. Variable Definitions and Data Sources Symbol Variable Description Data Source Y (m,t) Fraud Arrests Fraud arrests per 10,000 people in MSA m at time t FBI Uniform Crime Reporting (UCR) Program Craigslist_mt Craigslist Entry Binary indicator if Craigslist is available in MSA m at time t Kroft & Pope (2014) X (t) Control Variables Income, unemployment, education, racial composition IPUMS CPS β1 Treatment Effect Coefficient estimating Craigslist's effect on fraud arrests Author’s model Ym MSA Fixed Effects Controls for time-invariant characteristics in each MSA Author’s calculations Nt Year Fixed Effects Controls for national shocks or trends at year t Author’s calculations Εmt Error Term Captures unobserved factors influencing fraud arrests Author’s model Note. All variables were aggregated or matched to the Metropolitan Statistical Area (MSA) level using US Census delineation files and the QCEW County-to-MSA crosswalk. IPUMS CPS refers to data from Flood et al. (2023). Craigslist entry data were obtained from Kroft and Pope (2014).