Eli Levine, Chavez Cheong, Ridge Ren, Emily Leung, Sam Wolter
Bordering Mexico, California, Nevada, Utah, and New Mexico, Arizona has a population greater than 7.5 million and is often a hotspot for migration. Between 2006 and 2010, the population grew from 6.19 million to 6.39 million, which contributed to a steep increase in home prices.
Prior to 2005, Arizona’s house price index closely tracked the national average, but as Figure 1 indicates, home prices in Arizona shot up dramatically in 2006 and came down just as precipitously when the bubble burst in 2007. Arizona’s experience pre-and post-crisis exemplifies the speculative mania that also prevailed in several other Sun Belt states.
Figure 1: House Price Index, Arizona compared to the national average. Source: Federal Reserve Bank of San Francisco
In Arizona, mortgages and mortgage-related laws are currently regulated by the Department of Insurance and Financial Institutions (DIFI). However, before July 1, 2020, Arizona’s mortgage legislation was enforced by the Arizona Department of Financial Institutions (AZDFI). AZDFI succeeded the State Banking Department, which operated from 1968 to 2004. AZDFI’s authority and powers were aggregated in the Arizona Revised Statutes (A.R.S.), and its authorizing statute mandates that the Superintendent have experience in the business of financial institutions. The most significant mortgage-related provisions are found in Title Six of the A.R.S., which covers Banks and Financial Institutions. Provisions covering requirements and regulations of mortgage brokers, mortgage bankers, and loan originators are found in chapter nine of that title.
In the run-up to the financial crisis, applicants had to meet several requirements before receiving a license to operate as an individual mortgage broker or a sole proprietorship mortgage broker license. Per A.R.S. § 6-903, an applicant had to have three years of experience as a broker within five years of submitting the application, had to complete a course on mortgages during the three years preceding the application, and had to demonstrate integrity, among other administrative requirements. From an analysis of the annotated A.R.S, the state legislature made no significant changes to mortgage broker licensing regulations between 1997 and 2006.
Unlike mortgage enforcement records available in most other states, Arizona only has consistent data available from 2005 to 2010. While data limitations prevent analysis of how penalized mortgage-related misconduct evolved in the years before 2005, we can still look to the periods immediately before and after the housing bubble burst to get a sense of how Arizona regulators responded in the moment and the types of abuses they prioritized.
Our dataset contains 86 observations, each containing:
There was a steady increase in the number of enforcement actions issued between 2005 and 2007, with the peak occurring in 2008 (coinciding with the financial crisis) a slight decrease in 2009, and a dramatic decline in 2010. Overall, there was a notably low frequency of enforcement actions issued in Arizona relative to population size and housing market activity, as compared to other states that we examined, since the state had over 6 million residents and one of the most expensive housing markets in the country.
Figure 2: Number of Mortgage Enforcement Actions issued by year. Source: DIFI
Our exploratory data analysis has focused on the types of entities most frequently sanctioned by AZDFI for committing mortgage-related offenses. Over half of such entities were mortgage brokers, and just under a quarter were mortgage bankers.
Figure 3: Arizona MEAs, 2005-2010, by Entity Type. Source: DIFI
The plurality of enforcement actions involved revocation of license, and approximately 23% were cease and desist orders. Examples of issues cited in the cease-and-desist letters include demands that firms without a license not issue any loans, or that firms stop deliberately lying about the immigration status of employees. The latter action raises the possibility that mortgage regulators viewed the policing of immigration status as a priority and inadvertently stumbled upon cases of further unlawful activity.
Figure 4: Arizona MEAs, 2005-2010, by Enforcement Type. Source: DIFI
We leveraged natural language processing (NLP) to enhance our analysis by extracting relevant text from each MEA. We decided to concentrate our NLP work through topic modeling (Figure 5). This was predominantly done through the R package “Quanteda,” as opposed to other states we analyzed, which used SpaCy and Gensim. Quanteda is a more computationally efficient tool compared to the other two, at a slight cost of accuracy.
Our analysis focused on trends between topics. Thus, using Quentada, we clustered documents, and then undertook a manual analysis of samples from a given cluster, which helped determine what number of MEAs corresponded to which offense. Despite the algorithm recognizing ten topics, we found that many of them overlapped, so we were able to reduce the number of topics down to six. This resulted from using a more efficient but less powerful analytical tool.
Figure 5: Distribution of Key Topics in Arizona MEAs, 2005-2010 Source: DIFI
Our topic models were only able to identify which documents were similar, as well as common words in each document group, however, it was the team’s discretion to label which group corresponded to which offense. Our category of “administrative MEAs” corresponds to enforcement actions that pertained to not submitting proper documentation to the DFI. Deception refers to enforcement actions related to misleading advertising. “Failure to conduct a background check” denotes those who received a mortgage enforcement action due to hiring former convicts or undocumented immigrants. “Fraud” refers to entities that scammed or stole from clients. “Operated without a license” and “Surety bond requirements” are self-explanatory.
As Figure 6 indicates, failure to meet the surety bond requirement did not generate any enforcement actions from 2005 until the introduction of the National Mortgage Licensing System (NMLS) in 2008. One of the requirements of the NMLS is that brokers and lenders must have a large enough surety bond in case the lender is unable to meet their financial obligations. In Arizona, this requirement likely was not enforced pre-NMLS. The prevalence of fraud and operating without a license suggests that many mortgage brokers in Arizona did not conduct themselves ethically in the years leading up to the recession. To make stronger conclusions about the prevalence of unethical conduct among mortgage brokers, we would need additional data, such as the total number of mortgage brokers operating in Arizona during that time frame, which would allow us to calculate changes in the proportion of mortgage brokers accused of unethical conduct.
The most notable limitation to our analysis is the lack of data points available, which complicates the identification of trends. In addition, we encountered a number of challenges in converting Arizona’s text to an Excel file. Issues included the addition of random numbers, awkward spacing, and frequent misspellings. This limitation cannot be remedied because Amazon Textract was designed to identify characters that look like English letters.
Lastly, two limitations emerged during our use of NLP. First, we subjectively decided which stop words were irrelevant. Thus, some of the words we omitted might be necessary to the topic modeling process in ways we do not understand. Second, our inexperience with Quanteda might have led us to overlook potential problems in the topic models. This can be remedied in the future by testing a few documents and identifying the accuracy of the topic models produced.
When analyzing the individual MEAs, a number of chapters in the Arizona Revised Statutes appeared across multiple documents. The most referenced statutes were Title 6 – “Banks and Financial Institutions” – and 41 – “Administration,” with a variety of subsections appearing.
The most cited provisions from Title 6 are §§123, 131.01, 132, 817, 903, 941, 942. These relate to the powers given to the Deputy Director, especially those assessing fines, (§132) as well as the punishment for not paying within the 30-day limit. The maximum fine was $5,000. §817 outlines the procedure for rejecting and rescinding an application. §903 outlines the requirements and expectations to be granted a license to issue mortgages in Arizona. While §941 focuses on relevant definitions, §942 outlines an exception to Arizona’s mortgage banking laws, specifically that the laws do not apply to financial entities under the jurisdiction of another state.
The Arizona statutory framework seems relatively standard for the time and does not appear to be unusually sparse or underdeveloped. This begs the question as to why there were only 86 MEAs issued in the relevant time frame. One possible reason for document scarcity is a lack of regulatory enforcement in Arizona due to under-resourcing, a permissive enforcement stance among relevant officials, or a combination of the two. Another possible reason is that due to the series of reorganizations that transferred regulatory responsibilities from one agency to another, digitized documents from earlier periods were lost in this transition.
 Despite requiring “integrity,” §6-903 does not define the term, noting only that the deputy director responsible for reviewing license applications “may require additional information on the experience, background, honesty, truthfulness, integrity and competency of the application and any responsible individual designated by the applicant.” However, cross reference to §6-905. (titled “Denial, suspension or revocation of licenses) provides insight how deputy director evaluates integrity. §6-905(A) empowers the deputy director to deny a license upon finding that an applicant: “is not a person of honesty, truthfulness, or good character;” “has been convicted in any state of a felony or any crime involving breach of trust or dishonesty;” has had a civil judgement or state/federal agency order entered against him/her “on grounds of fraud, deceit or misrepresentation and the conduct on which the judgement is based indicates that it would be contrary to the interest of the public to allow such a person to be licensed;” has “made a material misstatement or suppressed or withheld information on the application;” or refused to allow examination or provide information required by the examiner.
 We have not been able to ascertain why enforcement data in Arizona remains so sparse.