THE EFFECT OF GROWTH ON FINANCIAL REPORTING AND AUDIT QUALITY: EVIDENCE FROM ECONOMIC SHOCKS TO BANKS By Sarah Barron Stuber A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Business Administration Doctor of Philosophy 2019 ABSTRACT THE EFFECT OF GROWTH ON FINANCIAL REPORTING AND AUDIT QUALITY: EVIDENCE FROM ECONOMIC SHOCKS TO BANKS By Sarah Barron Stuber economic environment represent a potential threat to financial reporting quality by presenting challenges to both the firm and its auditor. Despite the ubiquity of positive economic shocks , little is known of the extent to which these changes affect financial reporting and audit quality . Using exogenous economic shocks to local banks from oil and natural gas discovery and extraction, I find that financial reporting quality, measured by loan loss estimate quality , is lower in a period of rapid bank growt h due to - reaction to the positive economic shock. I also find that auditors with a combination of both task - specific and industry - specific expertise are more successful in mitigating the deterioration in financial reporting quality compared to auditors with general, Big 4 or industry - specific expertise a lone. The findings suggest that a combination of industry and task - specific auditor expertise is needed to combat deterioration in financial reporting quality resulting from a positive economic shock iii To Micah . Psalm 34:3; Psalm 126:2 - 3 iv ACKNOWLEDGEMENTS I would like to thank my dissertation chair , Professor Chris Hogan, who has been a truly exceptional advisor . Through her patient support and encouragement, I developed into a more thoughtful researcher as she provided me with insights and opportunities to learn while challenging me to become better . I am also deeply grateful to Professor Kathy Petroni for her guidance not only on this project, but throughout the program in both my research and teaching. She challenged me to think deeper and more carefully , and her guidance has shaped the way that I approach accounting research. I would like to thank Professor Matt Beck for being a member of my dissertation committee and a helpful so urce of guidance on this and other projects, as well as being a friend throughout the program. I am also grateful to Professor Soren Anderson for serving on my committee and providing insight on energy economics, methodological issues, and a unique perspec tive that greatly benefitted this dissertation. I also thank Andrew Acito, James Anderson, Will Demeré, Aaron Fritz, Yadav Gopalan, Andy Imdieke, Michelle Nessa, Hari Ramasubramanian, Joanna Shaw, Michael Shen, Dan Wangerin, Luke Weiler, and Mike Wilkins , and seminar participants at the University of Connecticut, Florida State University, University of Pittsburgh, University of Kansas, University of Illinois at Urbana - Champaign, Texas A&M University, and University of Iowa for their helpful comments and sug gestions . I thank Hunt Allcott and Daniel Keniston for sharing their measure of shale oil and gas endowment. The accounting PhD students at Michigan State are a n outstanding group that has become like a second family to me during the program. I especially want to thank my office mates, Will Demeré and Joanna Shaw for being a constant source of support, laughter, and enjoyment during my time in the program. I am thankful for Professor Andy Imdieke for his mentorship during my v first year in the program and f or being willing to continue to answer questions and provide insight that has been integral to my success. being there for me when I need her most. I have been deeply ble ssed by friends outside of the program that have provided constant support, listening ears, and encouragement: Hannah Barry, Stephanie Pitzer, Mary Beth Smith, Mary Pat Byrd, Laura Whittaker, and Casey Washington. I want to thank my parents, Tommy and Bet sy Barron , for insti lling in me a strong work ethic, a passion for learning and a deep foundation of faith . It is impossible to express my gratitude for the love they have provided me over the years. They have always kept me grounded, while encouraging me as I chase my dreams even the crazy ones. I also thank my father and mother - in - law, Dan and Deb Stuber for their help in many ways throughout the PhD program. Finally , I want to thank my husband , Micah , who has been my rock throughout this program. He encouraged me to pursue a PhD and has been my biggest cheerleader and supporter, all while completing his own degree and helping to support us financially. Our son Davis is our greatest blessing and was bor n during this program, forever changing our outlook and providing us perspective on what is truly important. Above all, I thank God for blessing me with this opportunity and the community of friends and family to surround me on this journey. vi TABLE OF CONTENTS LIST OF TABLES LIST OF FIGURES CHAPTER 1: INTRODUCTION CHAPTER 2: RESEARCH SETTING AND HYPOTHESIS DEVELOPMENT 2.1 The Setting 2.1.1 Background on Hydraulic Fracturing 2.1.2 The Local Effect of Hydraulic Fracturing 2.1.3 Hydraulic Fracturing Effect on Banks 2.2 The Role of the Auditor CHAPTER 3: METHODOLOGY 3.1 Primary Empirical Measures 3 .1.1 Identification of a ffected b anks 3.1.2 Allowance for l oan l osses 3.2 Research Design 3.2.1 Hypothesis 1: Effect of a positive economic shock on ALL q uality 3.2.2 Hypothesis 2: Effect of the a uditor o n ALL quality during a positive economic shock CHAPTER 4: DATA AND RESULTS 4.1 Sample 4.2 Descriptive Statistics 4.3 Results 4.3.1 Hypothesis 1: Effect of a positive economic shock on ALL quality 4.3. 1.1 Additional Analysis of ALL and Charge - Offs 4.3.1.2 Additional A nalysis of bank resource constraint and portfolio composition 4.3. 2 Hypothesis 2: Effect of the auditor on ALL quality during a positive economic shock CHAPTER 5: SUPPLEMENTAL ANALYSIS AND ROBUSTNESS 5.1 Broad Sample Analysis: Shift - s hare a pproach 5.2 Identifying a ssumptions and r obustness 5.2.1 Parallel t rends a ssumption and f alsification t est 5.2.2 Endowment measure exogeneity 5.2.3 E x ante a ssignment 5.2.4 Analysis of Loan Loss Provision v i ii x 1 8 8 8 8 12 14 19 19 19 21 24 24 27 30 30 34 35 35 37 40 43 49 49 50 50 52 54 55 vii CHAPTER 6: CONCLUDING REMARKS APPENDICES APPENDIX A Endowment Measure Details and Validation APPENDIX B Variable Definitions APPENDIX C Illustration of Interaction of Endowed , Exposed , and Post Variables APPENDIX D Incurred Loss Model vs Current Expected Credit Loss Model APPENDIX E Tables REFERENCES 57 59 60 64 68 70 73 104 viii LIST OF TABLES Table 1 Table 2 Table 3 Table 4 Table 5 Table 6 Table 7 Table 8 Table 9 Table 10: Table 11 Table 12 Table 13 Table 14 Table 15 Table 16 Table 17 Table 18 Table 19 Sample Calculation Sample Descriptive Statistics Correlation Matrix Effect of exposure to shale boom on ALL quality Effect of exposure to s h ale boom on ALL quality Not L i mited to O ver - R eserved B anks Effect of exposure to shale boom on charge - offs and ALL Effect of human resource constraints on ALL quality during shale booms Effect of portfolio composition on ALL quality during shale booms Effect of general auditor expertise on ALL quality during shale booms Effect of auditor industry expertise on ALL quality during shale booms Effect of expert auditors compared to unaudited Effect of auditor industry expertise on ALL quality during shale booms Effect of single e xpertise compared to non - experts Effect of Big 4 expertise compared to industry expertise Effect of changes in horizontal drilling on ALL quality Falsification Test Correlation of exposure measure with ex ante observable bank characteristics Effect of exposure to shale boom on ALL quality Banks Under 500 million Effect of ex ante exposure to shale boom on ALL quality 74 75 76 77 78 79 80 82 84 86 88 90 92 94 96 97 98 99 100 ix Table 20 Table 21 Table 22 Effect of exposure to shale boom on provision quality Effect of E xposure to S hale B oom on D eposit G rowth Effect of exposure to shale boom on ALL quality Alternative endowment measure 101 102 103 x LIST OF FIGURES Figure 1 Figure 2 Figure 3 Figure 4 Major U.S. Shale Play Regions Treatment and Control Counties within Shale Play States Time Series Analysis of ALL Quality Time Series Analysis of ALL and CO levels of Exposed Banks 32 33 36 39 1 CHAPTER 1: INTRODUCTION Economic shocks , either positive or negative in nature, affect firm operations, management incentives and ultimately financial reporting quality. Prior research has focused on the effect of negative economic shocks on financial reporting quality (e.g., Iannotta and Kwan 2014; Barth and Landsman 2010) , largely ignoring the possible consequ ences of positive shocks. The relative lack of evidence on the effects of positive shocks is an important omission given that such shocks are common and can occur at the macro - e conomy level (e.g., tax cuts, interest rate cuts), industry - level (e.g., government stimuli, technological development) or firm - level (e.g., patent development, large new customer, positive media attention). Positive shocks fundamentally differ from negative shocks in terms of incentives and challenges faced by firms , and while accounting standards are relatively straightforward in regards to negative shocks (e.g., impairments, curtailments), the appropriate accounting for positive economic shocks is largely left up to manageria l discretion. Due to the importance of high quality and unbiased financial reporting for stakeholders during rapid growth periods, t his study examines whether positive economic shock s negatively affect financial reporting quality and whether auditor expertise moderates the effect . Rapid firm growth and or ganizational change have significant disruptive effects on a resulting in a greater propensity to report internal control deficiencies and restatements (Ashbaugh - Skaife, Collins and Kinney 2007; Doyle, Ge and McVay 2007) . In financial statements, as managers of growth companies typically have greater discretion in operating and reporting decisions, including the development of accountin g estimates (Smith and Watts 1992) . An economic shock 2 future cash flows , and increased uncertainty can result in severe and systemati c errors in judgment (Tversky and Kahneman 1974) , ultimately threaten ing the quality of accounting estimates. 1 Changes in business risk as a result of changes in general economic conditions have been cited by the Public Company Accounting Oversight Board ( PCAOB ) as increasing the risk of material misstatement, and specifically material misstatement of estimates (AS 2110; AS 2501) . , and ultimately , audit p rocedures performed. Given that financial reporting quality is the joint product of the firm and its auditor (Gaynor , Kelton, Mercer and Yohn 2016) , I investigat e the effect of both firm exposure and auditor characteristics on financial reporting following a positive economic shock. Often , economic shocks resulting in rapid growth are systemic and widespread or arise from endogenous firm and industry characteristics , making it difficult to establish an appropriate counterfactual for analysis. In the banking industry, financial crises are often preceded by a period of rapid growth, with some suggesting crises . Such crises however, affect the entire financial services sector, making it difficult to identify control firms that were not affected by the crisis. Firm - specific shocks often arise due to characteristics of the firm itself, making the exposure to a shock endogenous. I overc ome this challenge by examining banks exposed to exogenous liquidity windfalls from oil and natural gas shale development to investigate the impact of a positive shock on financial reporting quality. The technological breakthroughs that allowed natural gas and oil recovery from underlying shale deposits were largely unexpected and 1 Consistent with the Conceptual Framework for Financial Reporting ( FASB Concepts Statement No. 8 ) , I define an estimate to be of higher quality when it is more informative regarding the prospects for future cash flows to an entity. High quality estimates are those prepared without bias (i.e., neutral). 3 resulted in a liquidity windfall for banks within highly localized region s . Due to the localized nature of shale - based shocks, a single state will contain some counties that are a ffected by the liquidity influx and economic changes from the shock , and some counties that are not. My identification strategy exploits this within - state variation using a geology - based measure of oil and natural gas endowment. I focus my investigation on the banking industry as banks are significantly affected by a positive shale - based shock, yet do not have any effect on the nature and timing of the shock itself (Gilje et al. 2016) . A second benefit of focusing on the banking industry is that, d ue to regulatory reporting requirements, financial information for both public and private banks is publicly available. Private ban ks with assets of less than $500 million are not required to be audited, which allows me to study the effect of variation in both external audit status (i.e. , audited vs unaudited) as well as auditor type on estimate quality . Finally, the quality of the al lowance for loan losses (ALL) reported by banks provides a proxy for financial reporting quality that is directly related to the effect of a shale - based boom . 2 I use difference - in - difference estimation on a sample of 11, 17 8 bank years from 2005 - 2014 , to examine the effect of exposure to the positive shock on ALL quality . Consistent with positive shocks having a negative effect on financial reporting quality, exposure to a positive economic shock results in lower estimate quality . Lower estimate quality can result from banks either over or under - reacting to a positive shock. I find that lower ALL quality following the shale boom is the result of management under - reacting to the shock by 2 McNichols ( 2000) suggests the ALL is a fruitful account for evaluating financial repor ting quality because it is material to the financial statements , subject to discretion, and tied directly to explanatory factors of interest . The ALL becomes more difficult to both develop and audit during periods of increased uncertainty surrounding futur e cash flows from loans, such as the period following a positive economic shock. I discuss the allowance for loan loss further in chapter 3. 4 failing to sufficiently adjust the ALL to acco unt for the effects of the operating environment on the loan portfolio . I then examine whether an external auditor mo derates the relation between a positive economic shock and financial reporting quality. First, I find that the financial reporting quality in audited banks is not significantly different compared to unaudited banks following a positive economic shock. The finding suggests that, on average , auditors do not mitigate financial reporting quality deteri oration due to a positive economic shock . I then examine the effect of specific auditor expertise on financial reporting quality of banks facing a positive economic shock by investigating the differential effect among auditors with 1) general expertise obt ained through access to significant audit firm resources (i.e., Big 4) , 3 2) industry expertise and 3) task - specific expertise (i.e., previous experience auditing a client exposed to a shale - based shock) when clients are exposed to a positive economic shock . I do not find evidence that general audit expertise or Big 4 expertise mitigates deterioration in financial reporting quality during these economic shock s . However, I do find that banks audited by industry expert auditors and banks audited by task - specif ic experts are associated with higher financial reporting quality compared to banks audited by non - industry or non - task - specific expert auditors . Further investigation indicates that the higher financial reporting quality is isolated to clients of auditors with both industry and task - specific expertise . These results indicate that both industry and task - specific expertise is necessary for auditors to learn how, from prior experiences , to adjust audit procedures to perating environments. Overall, this study finds that financial reporting quality varies based on economic conditions, and specifically that positive economic shocks are associated with a decline in 3 Big 4 refers to the following audit firms: Deloitte, PriceWaterhouseCoopers, Ernst and Young and KPMG. 5 accounting estimate quality due to management under - reac tion to the shock . This finding indicates that research on steady - state financial reporting quality does not capture a complete picture of the determinants of financial reporting quality, as findings may not generalize to periods of change in the economic environment. Additionally, the f indings suggest that a combination of auditor industry and task - specific expertise is more effective at mitigating the effects of positive economic shocks, compared to traditional measures of expertise (i.e., Big 4 and industry expert) alone . Consistent wi industry knowledge are not a substitute for intimate knowledge of the specific challenges facing a client in a pe riod of rapid growth. While the findings of this study clearly indicate a decline in financial reporting quality as defined by GAAP standards, the results do not speak to whether the bias of the ALL in a more conservative direction is optimal for bank stab ility. In this study , I focus strictly on financial reporting quality as defined by accounting standards , specifically that estimates should be developed without bias, regardless of the direction of the bias (AS 2501 ; SEC 2001; FASB SFAC No. 8) . 4 It is notable; however, that a conservative bias in the ALL provides management with the opportunity for earnings smoothing , which dampens the discipline of bank risk taking, which can have widespread negative consequences for the financial system (Bushman and Williams 2012) . 4 SAB 102 findings was that most of the reviewed institutions' loan loss allowances included large supplemental reserves that generally were not linked to an analysis of loss exposure or supported by evidence. The GAO n oted: "Such use of unjustified supplemental reserves can conceal critical changes in the quality of an institution's loan portfolio and undermine the credibility of financial reports." (GAO 1994; SEC 2001) . 6 This paper contributes to the literature in three ways. First, p rior studies on the effect of situations . The lack of evidence on how growth affects financial reporting quality is an important omission in the literature, because positive shocks and periods of rapid expansion occur frequently in the economy and have different implications for firm a nd auditor risks, challenges , and incentives compared to negative shocks . Thus, the findings of this study have implications for future research by suggesting the need for a better understanding of how the economic environment might aff ect evaluation of financial reporting quality. In contrast to studies examining the effect of accounting standards during expansionary periods (e.g., Barth and Landsman 2010; Laux and Leuz 20 09) , I focus on the quality of financial reporting information . To the extent that financial reporting quality declines during growth periods, the standards themselves may be of secondary concern. While prior literature relies primarily on shocks that are systemic or in cross - country settings, I leverage a localized exogenous shock, which allows for stronger identification through the availability of a reasonable counterfactual. 5 Second, I contribute to the banking literature by providing insight on how growth periods impact l oan loss provisioning practices . Bank financial reports are intended to reduce the information asymmetry (i.e. , opacity) that exists between banks and stakeholders (i.e., shareholde rs, debtholders and depositors) ; thus , if financial reporting quality declines, bank opacity is likely to increase. Bank opacity causes significant impediments in the interbank 5 I acknowledge that this study focuses on a single economic shock in a specific setting. Although this focus allows for s trong identification, the findings may not generalize to all situations of economic growth. However, disruptive challenges (e.g., internal control disruption, human capital constraints and capacity constraints) affect growth firms across all industries at various points. 7 market, which has repercussions for the economy as a whole (Acharya and Ryan 2016) . 6 Additionally, I extend prior studies examining the relation between growth cycles and portfolio risk and composition by providing evidence that can have direct implications for financial reporting quality. B ank financial reporting quality has a widespread impact on the financial system. Finally, I contribute to the auditing literature by providing evidence on how the economic environment of a clien t affects audit quality. This answers calls from Gaynor, Kelton, Mercer and Yohn (2016) and Knechel, Krishnan, Pevzner, Shefchik and Velury (2013) to examine how auditors adjust to client uniqueness and uncertainty and how different economic situations influence audit and financ ial reporting quality outcomes. Finally , I contribute to the literature on auditor expertise b y providing evidence consistent with economic theory on the benefit of expertise obtained through experiences (Arrow 1962) and the importance of a combination of an - specific and ta sk - operating environment . 6 Consistent with the banking literature, I consider higher bank opacity to exist when there is greater uncertainty of the valu 8 CHAPTER 2: RESEARCH SETTING AND HYPOTHESIS DEVELOPMENT 2.1. The Setting 2.1.1. Background on Hydraulic Fracturing sidered to be economically viable. However, new technology pioneered in the Barnett shale play 7 region combined with advances in horizontal drilling technology to decrease the cost of fracking and significantly increas e the amounts of oil and gas considere d to be economically recoverable around the country ( Gilje et al. 2016) . According to Allcott and Keniston ( 2018) , the key areas affected by this increase in economically recoverable oil were the Bakken Shale in western North Dakota, the Niobara shale in eastern Colorado, the Marcellus and Utica shales in Pennsylvania, Ohio, New York, and West Virginia, the Barnett, Granite Wash, and Eagle Ford shales in Texas, the Woodford Shale in Texas and Oklahoma, and the Haynesville shale in Texas and Louis iana. 2.1.2 The Local Effects of Hydraulic Fracturing North Dakota, good and bad. It has minted millionaires, paid off mortgages, created businesses; it has raised re nts, stressed roads, vexed planners and overwhelmed schools; it has polluted streams, spoiled fields and boosted crime...Oil has financed multimillion - dollar recreation centers and new hospital wings. It has fitted highways with passing 7 shale formations that share similar geological properties, contain significant amounts of oil and /or natural gas , and are in a similar geographic area. 9 lanes and rumble st truck (Brown 2013) News reports from regions impacted by shale resource booms leave little doubt that there is a significant impact of an oil boom, with man camps springing up overnight, unemployment plunging to below 1% and rent rates skyrocketing to levels higher than expe rienced in the largest U.S. cities (Galbraith, 2012; Healy, 2013) . The number of Americans working in the oil and gas industry increased by more than 40% between 2007 - 2012 (Hefner, 2014) . An increase in crime also has accompanied the economic growth, with western North Dakota and eastern Montana experiencing a 32% increase in crime since the boom began (Healy, 2013) , and while there is no debate as to whether resource booms lead to immediate job growth, questions linger as to the sustainability of the jobs in the long run (Mor etti, 2013) . In addition to the economic and social impacts of resource booms, the potential environmental impacts must also be considered (Urbina a nd McGinty, 2011) . Given the many competing factors, the answer to the question of whether resource booms have a positive net impact, in the long or the short - term is not readily apparent and has been the subject of extensive research. Prior economics research into the impact of shale resource booms has focused on local impacts, as resource booms, particularly shale - based booms are highly localized and there is heterogeneity in the local impacts of a boom. Using the interaction of local geology and tech nological development as an exogenous shock, Feyrer et al. ( 2017) , find that over half of fracking revenue stays within the regional economy, with every million dollars of new oil extraction producing $132,000 in royalty payments and business income, $80,000 in wage income, and 0.85 jobs within a community. Fracking has spillover effects onto other indu stries within the 10 local geographic area, as these industries may benefit from the additional money in the region, but also must compete for talent in a much tighter labor market, and may be unable to keep up with the wage increase s required to attract and retain talent. Jacobsen ( 2017) finds that local wages increase across all major occupational categories, regardless of whether those occupations experienced changes in employment. The impact of employment growth is not limited to the oil and gas industry alone, but is attributed to significant cross sectional employment growth across U.S. industries and a $3.5 trillion increase in aggregate U.S. market capitalization from 2012 - 2016 ( Gilje et al., 2016) . The crowding out of other industries, at the expense of focusing on natural resources can been examined extensively in the manufacturing sector, with some studies concluding that there is no evidence of Dutch Disease ( e.g., Allcott and Keniston 2018) , while other s have argued that the income specialization that comes along with resource booms can reduce income per capita, increase crime and reduce education al attainment in the long run ( e.g. , Haggerty et al., 2014) . In studies of the boom and bust cycles, the short - run economic benefits are clear, as there is a positive impact on local employment ; however, in the long run, res ults are mixed with studies finding both positive and negative impact s on local communities ( e.g., Jacobs en and Parker 2016; Allcott and Keniston 2018) . One potential reason for discrepancies in the findings of prior research is that there is significant heterogeneity in the local net impacts, thus while the overall impact may be positive, the benefits are not evenly distributed or experienced across various local communities (Bartik et al., 2016) . The findings of Muehlenbachs et al. ( 2015) , su pport the conclusion that the impacts of shale development are not evenly spread as they examine the fluctuations in the values of specific properties in the Pennsylvania shale region from 2002 - 2015. They find that the negative 11 impacts on property value ar e closely tied to the potential negative environmental impacts of fracking, as homes that rely on piped waters were largely unaffected, while homes dependent on groundwater suffered a significant decline in value. The heterogeneity of the effects of the sh ale boom make it an ideal setting to examine the effect of localized economic shocks. The new technology and time series variation associated with shale oil booms also creates a setting in which to test the theory of learning by doing (Arrow, 1971) , as the rapid technological chan ge and race to capitalize on the new opportunities in shale resource production resulted in a steep learning curve. Research has largely found support for the theory that companies learn from their experiences with hydraulic fracturing, as wells become mor e productive with time (Fitzgerald, 2015) . However, the learning process occurred slowly and incompletely, as Covert ( 2014) estimate s that companies are successful at capturing only 67% of potential fracking profits. The challenges associated with the new technology, along with unexpected swings in oil prices made it difficult for companies invested in fracking to be profitable initially , driving more than 120 companies into bankruptcy and resulting in losses in the billions ; however, the lesso ns learned from the early fracking laid the groundwork for future profitability (Olson, 2017) . Due to continuous learning and experimentation, as well as increases in scale, one company was able to reduce its production costs by 40 percent over an 18 month period (Hefner, 2014) . Similarly, auditors would be expected to learn from their prior experiences with re source - booms, thus the variation in timing and geography of the oil booms provide an ideal setting to test whether auditors gain experience - specific expertise through learning. 12 2.1.3 Hydraulic Fracturing E ffect on Banks Shale booms provide an opportunity to study the effects of economic shocks in general, and on banks specifically (e.g., Plosser 2013; Gilje, Loutskina and Strahan 2016) . The discovery of shale oil and gas has an immediate effect on both the supply of funds and demand for credit faced by banks in exposed areas. Royalty payments and lease paym ents are made to private landowners who deposit funds in banks, resulting in a liquidity windfall for banks and a sharp increase in supply of funds for bank investment. In the United States, landowners have rights to not only the surface of their propert y, but also to the minerals below the surface (Hefner 2014) . The market for oil and gas extraction has historically been dominated by small, independent companies, resulting in intense competition for leases, thus driving the value of mineral rights upwards (Davidson 1963) . Money is paid to mineral owners in two stages. First, energy extractors make an initial lease payment for the right roy alties based on the productivity of the wells on their properties. Royalty amounts can be substantial, with estimated royalties paid to private landowners in the U.S. exceeding $39 billion in 2014 (Brown et al. 2016) . Considering both the sizeable upfront lease payment and continuing royalties, an owner of 100 acres could conceivably receive $3 million in lease and royalty paym ents over a 20 - year period. Landowner royalties primarily accrue locally, while revenues earned by oil and gas companies often flow out of the local county or state (Fitzgerald 2014) . Royalty payments, increased wage s, and higher business income result in an inflow of cash to the localized community during a fracking boom. The cash inflow leads to an increase in both supply of liquidity to banks in the form of deposits and a decline in the cost of deposits ( Gilje et al. 2016) . The increased supply of funds allows banks to meet the increase in demand for 13 credit that accompanies fracking booms. The availability of jobs and opportunities associated with oil and gas bring people to the impacted areas, increasing housing demand and subsequently rents. As rents skyrocket, there is a rush to construct new housing units, increasing the demand for loans to finance the construction. New businesse s also open to capitalize on the increased population and wealth increasing the demand for commercial loans in the area. and credit boom periods lead to an increase in the risk of bank loan portfolios Marquez 2006; Ruckes 2004) . Empirical research supports these models by finding that bank crisis periods can be tied to relax ed lending standards and movement away from requiring audited financial information from borrowers during expansionary periods 2012; Lisowsky et al. 2017) . Such a change in underwriting standar ds alters the risk inherent in a In addition to impacting new loans made by banks, shale boom shocks also affect the a positive impact on business operations for many businesses in affected areas (Allcott and Keniston 2018; Feyrer et al. 2017) and; therefore, the boom alters the risk of local borrowers and ability to service loans, which results in a downward shift in expected charge - offs, at least in the immediate future. While decreases in charge - offs are positive for bank operations, the departure from historical loan charge - off patterns results in challenges for bank management as they attempt to accurately estimate loan losses. Thus, banks exposed to fracking booms experience organizational challenges due to changes in the risk of their existing loan portfolio, as well as changes in both the supply of funds 14 and demands for new credit. Periods of organizational change and rapid growth increase uncertainty, the likelihood of internal control deficiencies (Ashbaugh - Skaife et al. 2007; Doyle et al. 2007) (Smith and Watts 1992) . These changes present a clear challenge to financial reporting quality ; t hus, I make the following hypothesis, stated in alternate form: H 1 : Exposure to a positive economic shock will negatively affect the financial reporting quality of banks. While the challenges facing banks are clear, what is unclear is whether these challenges will result in over or under - reaction to a positive shock. reaction is critical to understanding, and ultimately mitigating, negative effects of positive shocks. 2.2 The R ole of the A uditor As f inancial reports are a joint product of management and the auditor (Gaynor et al. 2016; Francis 2011; DeFond and Zhang 2014) auditor characteristics may affect the relation between a positive economic shock and financial reporting quality . Auditing standards require auditors to gain an understanding of the business practices of their clients and the economic environment in which their clients operate ( AS 2110) . Some of the most serious allegations made against audit firms by the SEC and clearest instances of audi t failure result from the auditor (Hall and Renner 1988) . A case study by Erickson et al. (2000) finds that the audit failure of Lincoln 15 nderstand business practices and the economic environment in which its client operated. Specifically, the auditor did not evaluate environment, leading to inappropriate accounting treatment of the transactions. This shortcoming resulted in the failure of the client, and litigation for the audit firm, ultimately leading to a settlement of more than $135 million. The need for auditors to adjust to changes in the economic en vironment is highlighted by t disruptive events, including economic shifts and technological changes (Amble, Gallagher, Joseph, and Peters 2015) . Accounting estimates are an are a where an understanding of changing business practices is particularly important, the auditing standards explicitly require that auditors . Furthermore, p rior literature has shown that auditors are able to influence accounting estimate accuracy (AS 2501 ; Petroni and Beasley 1996) . Dynamic changes in a client s economic environment increase the information - the underlying nature of the changing business practices. One way auditors gain a greater u nderstanding of business practices is through expertise (e.g. , Balsam, Krishnan and Yang 2003; Gul, Fung and Jaggi 2009; Lim and Tan 2008; Reichelt and Wang 2010) . Expert ise can be general in nature, arising from knowledge of financial reporting and auditing procedures. General expertise can be obtained through access to significant firm - level resources (e.g., human resources, talent, large professional standards groups, a ccess to a broad geographic network) , as evidenced by the higher quality of Big 4 audits identified in prior research (e.g., DeFond and 16 Zhang 2014) . Expertise can also arise from extensive experience within a particular indust ry , and industry specialist auditors are more likely to detect financial misstatements because of their ability to identify a pattern in multiple, seemingly innocuous, cues (Hammersley 2006 ) . Expertise is of particular importance in evaluating complex estimates, because effective auditing of estimates requires auditors to appropriately incorporate a wide range of information, and expert auditors are better equipped to identify relevant in formation and contradictory evidence to incorporate into their analysis of the reasonableness of accounting estimates (Griffith, Hammersley, Kadous and Young 2015) . I n the specific case of the allowance for loan loss es , industry - expert auditors assess risk differently than non - specialist auditors by properly identifying the valuation assertion as being the most imp ortant for the loan loss allowance (Taylor 2000) . In addition to industry expertise, auditors draw on specific prior experience s to develop task - specific expertise. Bonner and Lewis (1990) find that genera l experience explains less than ten percent of variance in the performance scores of experimental subjects. As l earning is the product of experience and occurs during the performance of a task (Arrow 1962) , it is specific experience that is critical, as auditor judgment is impacted by their previous experience . As auditors gain experience, their knowledge of the set of potential financial statement errors becomes more complete, and they are better able to understand the interaction of financial statement errors with the transaction cycle (Libby and Frederick 1990) . 8 In eval uating the assumptions underlying the calculation of accounting estimates, auditors with less experience are more likely to rely on management - provided 8 Further supporting the importance of task - specific expertise, Shepardson (2018) finds that individual audit committee - specific expertise affect financial reporti ng outcomes in a manner consistent with conservatism. 17 information, rather than exercising professional skepticism and are less likely to deflect attempts by m anagement to persuade auditors of the appropriateness of accounting numbers (Kaplan, O'Donnell and Arel 2008) . During periods of economic change , it becomes more difficult for an auditor to evaluate the reasonableness of management assertions, thereby increasing the importance of auditor expertise. S tudies of expert judgment find that experts mu st be able to identify, organize, measure, weigh , and combine cues in coming to a judgment (Slovic 1969; Einhorn 1974) exp (Einhorn 1974) ; however, the greater the noise in the background, the more difficult the job of the expert becomes and it is unclear whether an expert would retain an advantage. C omparability across clients audited by an expe rt also declines d uring a period of rapid growth , which can have a negative effect on a nalytical audit procedures which rely on the relative comparison of a bank to its peers . T hus , if a bank becomes less comparable to other ortfolio, it becomes difficult to identify appropriate benchmarks. 9 W ith the declining re cognize and adjust to the changing economic environment facing their clients could exacerbate these challenges. Furthermore, in situations where decision makers are constrained, evidence points to a decline in the advantage of experts compared to non - exper ts (Hoff man et al., 2003) , and not all types of experts exhibit superior performance compared to non - experts (Moroney, 2007) . 9 The benefit of having appropriate benchmarks in auditing is related to the theory of relative performance evaluation often cited in compensation literature (Lazear and Rosen 1981; Nalebuff and Sti glitz 1983) . 18 Finally, it is unclear whether the positive effects of auditor expertise documented in prior literature would apply uniformly across growth, decline and steady - state periods. While both decline and growth periods p resent challenges to an auditor in terms of changing economic environment and increased uncertainty in developing accounting estimates, the risks facing an auditor differ. One example is l itigation risk , which is a significant risk for audit firms. 10 Prior PCAOB inspection findings (Christensen et al. 2018) and that higher litigation risk results in greater audit effort, as suggested by higher audit fees (Choi et al. 2008; Seetharaman et al. 2002) . Because auditor legal disputes are typically triggered by an event resulting in material loss, such as a client bankruptcy, intuitively, the risk facing an auditor is lessened during a high - growth period when overall risk of client failure or bankruptcy is reduced (Maksymov et al. 2018) . In contrast, in a decline period, auditor litigation risk may be heightened, leading to greater auditor effort in which expertise would be most beneficial. Thus, it is ex ante unclear whether auditor expertise would have a similar effect following a positive economic shock resulting in lowere d litigation risk, compared to steady state periods or periods following a negative economic shock. Given the competing arguments of how a dynamic growth environment might affect the efficacy of auditors, I make the following hypothesis, stated in null form: H 2 a : The change in financial reporting quality following a positive economic shock will not be affected by general auditor expertise . 10 A 2008 report from the U.S. Department of the Treasury indicates that litigation - related costs represent 6.6 percent of the audit revenues of the six largest U.S. audit firm (U.S. Department of the Treasury 2008) . 19 CHAPTER 3: METHODOLOGY 3.1 Primary Empirical Measures 3.1.1 Identification of affected banks Oil and gas production decisions are influenced by several factors such as local infrastructure and economic conditions. These factors could be correlated with the quality of a anks in any given region. In contrast, ex ante oil and gas endowment is based solely on geological properties of underlying shale formations, which are unlikely to be correlated with bank financial reporting quality. 11 For this reason, I use the variation i n ex ante oil and gas endowment as the basis for identification in this study. To measure ex ante oil and natural gas endowment, I use a measure constructed by Allcott and Keniston (2018) . This measure of oil and gas endowment expands beyond oil and gas production to include both proven and undiscovered oil and gas res erves . U ndiscovered reserves are based on estimates by the U.S. Geological survey, which defines exist outside of known accumulations (Schmoke r and Klett 2013) . The expansion of the production is critical to the exogeneity of the measure , as it is unlikely that a bank chooses branch locations ba sed on the geological properties of the location, especially absent proven oil and natural gas reserves . APPENDIX A contains further details of the calculation of the measure of oil and natural gas endowment, as well as the procedures I perform to validate the measure. 11 I define endowment to be the total amount of oil and natural gas existing within shale formations before the commencement of any production activity. Endowment encompasses all oil and gas that has been produced and estimates of all oil and gas that ever w ill be produced within a given county. 20 Another key component contributing to the exogeneity of the identification measure is the unexpected nature of the technological breakthroughs allowing the extraction of resources from shale formations. The rapid development of shale - based o il and gas production surprised even the experts in the field, making it unlikely that banks had either the necessary knowledge or the time to adjust their strategy or governance structures in anticipation of the fracking boom. Resource endowment density varies within shale plays ; thus , I identify high versus low shale play. I consider any county that is above the median for endowment density within a given shale play to be highly endowed ( Endow ed = 1) ; while I consider a ll counties located within a state over a shale play (i.e., play state) and are below the median in endowment density as low - endowment counties ( Endowed = 0) . This allows me to use counties that are geographically close, but have lower or no oil and gas endowment as a control group for the analysis. 12 I use the FDIC Summary of Deposits data from 2005 - 2014 to identify bank branch locations. I construct a continuous county - bank - year measure of exposure ( Exposure ) to the fracking boo ms by determining the proportion located in a county where Endowed = 1 . The measure ranges from zero for banks with no branches in endowed counties to one for banks with all branches in endowed counties. 13 As a secondary test, I also construct a dichotomous measure to identify exposed banks using the indicator variable Exposed , which is 12 APPENDIX C provides further details on the identification of treatment and c ontrol counties and Figure 2 presents a map of treatment and control counties. In a robustness test described in chapter 5, I use the conti nuous measure of endowment and document qualitatively similar results. 13 W shock can vary over time based on the locations of its branches. I use the t ime - varying exposure measure for the main analysis, as it more accurately captures the time - The results are robust to the use of an ex ante (as of 200 7 ) bank exposure measure, which mitigates bank sel f - selection concerns. The analysis is discussed in greater detail in chapter 5. 21 equal to one if the majority of are located in a county where Endowed = 1 in year t , and zero otherwise. 3.1.2 Allowance for loan losses I use allowance for loan loss es (ALL) quality as a proxy for financial reporting quality in this study. When using specific accrual accounts to assess financial reporting quality it is important to identify an account that is material, subject to discretion, and that can be tied directly to explanatory factors of interest (McNichol s 2000) . The ALL is typically the largest , and overall systemic risk (Beatty and Liao 2014; Bushman and Williams 2012; Iannotta and Kwan 2014) . It is subject to a high level of management discretion , making it susceptible both to management bias and manipulation (Beatty and Liao 2014) , and p rior literature has found that bank management uses the ALL to manage earnings (e.g., Beatty et al. 2002) and capital (Ahmed et al. 1999) . The ALL is directly impacted by loan portfolio composition and loan growth ; therefore, I can link the disruptions that occur in a time of rapid growth directly to the ALL estimate . Some p rior literature examines the loan loss provision (LLP), which is a measure of the change in the ALL, net of charge - offs and recoveries. In this study I focus on the ALL, which is the estimate of expected losses inherent in the portfolio as of the balance sheet date. Although anal ysis of the LLP may be appropriate in a static environment, in a dynamic environment , such as following an economic shock, the analysis of the ALL is more appropriate , because t he LLP is influenced by inaccuracies in the beginning balance of the ALL and is 22 after the necessary allowance balance is calculated. Furthermore , the quality of the ALL is the focus of both auditors and regulators. 14 The ALL at period - end is the estimate of expected losses inherent in the existing loan po rtfolio as of the balance sheet date . Standards require losses be recognized when it is considered unlikely that the bank will collect all contractual payments of principal and interest. The appropriate loan loss allowance is determined based on a two - part calculation. A reserve for non - performing loans is determined on a loan - by - loan basis using methodology outlined following ASC 310 - 10 (FAS 114) and loans t hat are performing are pooled and evaluated base d on loan type following ASC 450 (FAS 5) . The requir ed reserve rate for pooled loans is for qualitative risk factors (e.g., economic factors) as determined by the bank. Once a loan is deemed uncollectible, it is charged - off and r emoved from the loan portfolio and the allowance. The quality of the ALL can be evaluated ex post based on subsequent charge - offs , (SAB) 102 outlines the procedures that should be used to validate loan loss accounting methodology is considered valid when it accurately estimates the amount of loss contained in the portfolio. Thus, loan loss estimation methods to reduce differe nces between estimated losses and actual subsequent charge - offs (SEC 2001) of subsequent charge - offs to the curren t ALL should be 1 (i.e . , there is no difference between the ALL and 14 The results of this study are robust to conducting the analysis using the LLP and following the model of Altamuro and Beatty (2010) , which examines the association between current period LLP and future period charge - offs. This model is discussed in detail in chapter 5. 23 subsequent charge - offs). 15 I consider one year to be the appropriate time period over which to examine subsequent charge - offs banks - year coverage period is generally considered appropriate because the probable loss on any given loan in a pool should ordinarily become app arent in that time frame (OCC 2012) Exa mining charge - offs over the subsequent one - year period is also consistent with prior literature (e.g., Bushman and Williams 2012; Nicoletti 2018; Altamuro and Beatty 2010) . Based on the relationship between loan charge - offs (CO) and the ALL, I u se the ratio of charge - offs in t+1 to ALL in t (CO/ALL ratio) to evaluate the quality of the ALL estimate. A portion of this study focuses on the influence of an expert auditor on financial reporting quality; thus, it is important that the financial repor ting quality proxy chosen is appropriate for use in evaluating the effect of auditor expertise. Prior literature has shown that experience matters most when the task being performed is complex (Hamilton and Wright 1982; Abdolmohammadi and Wright 1987) , and proper evaluation of auditor expertise necessitates a task to require cue selection and weighting (Bonner 1990; Dawes 1979; Einhorn 1974) . Simon's (1960) model separates tasks into structured, semi - structured, and unstructured, with the semi - structured and unstructured tasks being the most complex . The audit of the ALL can be considered a semi - structured task (Wright 2001) . The calculation of the ALL requires extensive analysis and understand ing of the importance of many factors across loan type and structure, as 15 The discussion of the timing of loan charge - offs and the relationship to the ALL in this paper is based on the incurred loss model according to ASC 310 - 10 (FAS 114) and ASC 450 (FAS 5), because these are t he standards in place as of the sample period. In June 2016, the FASB issued ASU No. 2016 - 13, which introduced the Current Expected Credit Loss (CECL) methodology which will replace the incurred loss model. See APPENDIX D for further discussion of the Incu rred Loss Model vs CECL model. 24 (Wright 200 1) . Based on these criteria, auditing of the ALL represents an appropriate task to evaluate the benefit of auditor expertise, as it is both semi - structured and requires the identification and weighting of numerous cues. 3.2. Research Design 3.2.1. Hyp othesis 1: Effect of a positive economic shock on ALL quality As with any event study, examining the impact of the fracking boom on ALL quality presents the challenge of disentangling the impact of rapid growth from other changes in economic conditions during the period. The exogenous nature of exposure to a fracking shock mitigates concerns of ex ante correlation between exposure and bank financ ial reporting quality; however, the analysis could be confounded by other local economic trends over the sample period. T o address this concern, I use banks with less or no exposure within the same state as a control group in a difference - in - differences re search design. The use of difference - in - differences estimation mitigates concerns that time - trends or contemporaneous events may influence the results (Roberts and Whited 2013) . The test of H ypothesis 1 estimates the impact of exposure to a fracking shock on ALL quality using the following model: ALL Quality = 1 Post it 2 Exposure it + 3 Post *Exposure i t 4 Size it 5 NPA it 6 Consumer it 7 C&I it s + t + e i s t (1) where subscripts refer to bank ( i ), state (s) and year ( t ), and variables are defined as follows: 25 ALL Quality it the ratio of charge - offs in year t+1 to ALL in year t 16 Post it indicator variable equal to 1 if significant fracking activity has commenced in a state where bank i has branch locations , 0 otherwise Exposure it proportion of bank i counties where Endowed =1 in year t Size it natural log of total assets at the beginning of year t NPA it non - performing assets at the end of year t scaled by beginning total assets Consumer it amount of consumer loans at the end of year t , scaled by total loans. C&I i t amount of commercial and industrial loans at the end of year t , scaled by total loans. t ) to control for time effects across all observations, and I include state fixed effects ( s ) to control for time - constant state charac teristics. In the population , 99 percent of banks have a CO/ALL Ratio of less than one, indicating that in nearly all cases, banks are over - reserved based on relevant standards for loan loss estimation as previously discussed . T his indicates that , in general , the ALL estimate is biased in a conservative direction; however, such a bias is not in line with SEC and PCAOB guidance that estimates should be developed without bias, regardless of the direction of the bias ( AS 2501 ; SEC 2001; FASB SFAC No. 8 ) . 16 The ratio of charge - offs in year t+1 to ALL in t ( CO/ALL ratio ) is slightly right skewed (skewness = 2.62). However, in the sample, the mean value of CO/ALL is 0.184. Given this small value, interpretation of the coefficien t of interest ( 3 ) is challenging using the log specification of the dependent variable (ln[1+CO/ALL]) as 100* 3 approximates the change in percentage change in CO/ALL for a one unit change in Post*Exposure only when CO/ALL is large. Thus, for the main analyses I use CO/ALL as the dependent variable of interest. To mitigate concerns regarding the non - normality of ALL Quality, I estimate model (1) using ln(1 + CO/ALL) and derive the interpretation 3 . I find results consistent with the primary analysis . 26 Given that nearly all banks are over - reserved, for the analysis, I remove all under - reserved bank - years for ease of interpret ation of results . 17 T hus, a higher CO/ALL ratio is indicative of less conservative bias (i.e., more neutral) and greater ALL quality . In this study, I am interested in how exposure to the shale boom impacts ALL quality . I use a difference - in - differences approach where the first difference ( 2 ) represents the relationship between ALL quality and the proportion of bank branches in high - endowment counties prior to the onset , and the difference - in - difference s estimato r ( 3 ) represents the difference in the relationship between ALL quality and Exposure after onset. A negative coefficient on 3 indicates that the ratio of charge - offs in t+1 to ALL in t is lower in the post - boom period when a are in an endowed county . I include the log of bank assets ( Size ) to control for variation in ALL quality based on the loan portfolio can influence the quality of the ALL. Consistent with prior literature, I control for the with the level of non - performing assets ( NPA ), which includes loans that are troubled debt restructures, greater than ninety days past due, or for which interest revenue is not being currently recorded as well as foreclosed real estate (Beatty and Liao 2014) . To control for the effect of loan portfolio composition on ALL quality , I include controls for the proportion of consumer loans and commercial and industrial loans in the portfolio ( Consumer and C&I ) . 18 17 In an additional analysis , I include both over and under - reserved banks and estimate model (1) using the absolute value of one minus CO/ALL ratio (i.e., an estimate of ALL error) rather than ALL Quality as t he dependent variable. The results are qualitatively similar to the results in the main analysis . 18 Consistent with Beck and Narayanamoorthy (2013) , I do not include economic control variables, because the ALL estimation should consider economic conditions. Furthermore, the primary interest of this study is regarding how changes in economic conditions affect ALL quality; thus, including variables to control for differences in economic conditions is inappropriate. 27 The use of the continuous variable Exposure provides information regarding how ALL quality varies with the level of Exposure ; however, the most significant effect of the shale boom would be expected when comparing banks with no exposure (i.e., Exposure = 0) to banks with any level of exposure (i.e., Exposure is greater than zero). To examine this relationship , I also estimate model (1) and replace Exposure with Exposed , an indicator variable equal to 1 if any of bank i are in a county with oil and gas endowment above the median within a shale play in year t , zero otherwise . The onset of boom is considered to be the first year of significant public drilling activity in a shale play and is identified consistent with Bartik et al. (2016) as fo llows: Permian (2005), Marcellus (2007), Bakken (2007), Andarko (2008), Eagle Ford (2008), Haynesville (2008), Niobrara (2010), Utica (2011). While these dates are consistent with information for the U.S. Energy Information Administration ( EIA ) and prior l iterature, I acknowledge that the onset of the boom does not occur at a single point in time. I conduct additional analysis that does not require a precise definition of pre - and post - periods and find consistent results. The additional analysis is described in detail in chapter 5. 3.2. 2 . Hypothesis 2: Effect of the auditor on ALL quality during a positive economic shock To test H2 , I estimate the effect of audit or expertise on ALL quality during a shale boom using four measures of auditor expertise. First, I use audit status to capture the effect of general financial reporting expertise provided by an auditor, compared to non - audited banks . Audited is an indicator variable equal to 1 if the bank is audited in year t and zero otherwise. Second, I use the presence of a Big 4 auditor to capture the effect of general audit expertise obtained through access to greater firm resou rces. I define Big4 as an indicator variable equal to 1 if the bank was 28 audited by a Big 4 auditor in year t and ze ro otherwise. Third , I define an industry expert auditor as an auditor in the top 10 percent of auditors in total count of audit clients within the banking industry. 19 I define IndExpert as an indicator variable equal to 1 if the bank was audited by a n indu stry expert auditor in year t and zero otherwise. Finally, I define a task - specific expert auditor as an auditor with previous experience auditing a bank exposed to the shale boom. The experience of these auditors provides them with specific knowledge of t he challenges faced by clients exposed to the boom, which may be beneficial in the planning and performance of audit procedures. I define SpecificExpert as an indicator variable equal to 1 if the bank was audited in year t by an auditor with task - specific expertise and zero otherwise. I estimate model (1) separately for subsamples of banks based on the types of auditor expertise and compare the coefficient on Post*Exposure across the separate regression s . 20 Specifically, I compare the following four subgroups: 1) Audited vs. u naudited ; 2 ) Big 4 vs. non - Big 4 ; 3 ) i ndustry e xpert vs. non - i ndustry e xpert ; 4) task - s pecific e xpert vs . non - task - s pecific e xpert . For the comparison of audited and unaudited banks , I limit the sample to banks with less than $ 500 million in assets. All bank s with assets greater than $500 million are required to be audited; thus, the audit choice estimation is limited to those banks below the $500 million 19 I proxy for banking industry expertise by c alculating the expert measure based on all banks within my sample. Given that all banks are required to file regulatory Call Reports, regardless of size and issuer status, the sample constitutes a reasonable proxy for the entire banking industry. Due to da ta limitations, I am unable to use audit fee - based measures of expertise. 20 In an untabulated test of hypothesis 2 , I add an interaction of the auditor characteristic of interest to model (1) and estimate the following model : ALL Quality 1 Post i,s,t 2 Exposure i,s,t 3 Post*Exposure ,s,t + 4 Auditor i,s,t + 5 Post*Auditor ,s,t 6 Exposure*Auditor ,s,t 7 Post*Exposure*Auditor ,s,t i,s,t . w here Auditor is the audit expertise measure of interest. I estimate the model first using Audited , then using Big4 , IndExpert, and SpecificExpert . The coefficient of interest ( Post*Exposure*Auditor ) is consistent with the results of the analysis of separate subsamples. I consider the use of subsample analysis to be a stronger analysis as it simulates a fully interacted model; thus, I rely on the subsample comparison for my main analysis. 29 threshold. 21 For the Big 4, industry and specific expertise analyses, I limit the sample to audited banks to isolate the incremental benefit of auditor expertise compared to auditors without such expertise. 21 The legislation governing bank audit requirements is the Federal Deposit Insurance Corporation Improvement Act of 1991 (FDICIA) . By limiting the sample to only those banks below the threshold for a mandatory audit, I can isolate the effect of the audit choice. Furthermore, limiting my sample based on size allows for greater comparability of the banks within the sample, as one of t he primary determinants of audit choice is size. 30 CHAPTER 4: DATA AND RESULTS 4.1. Sample Table 1 details the sample selection process. I begin with all banks filing at least one Call Report from 2004 to 2015 , and merge the database of Call Reports with the regulatory reports for - 9C) to obtain the auditor identity. 22 I limit the sample to year - end regulatory reports. The sample period begins in 2005, as auditor identity is not publicly available on the FR Y - 9C prior to that date. The analysis is conducted at the bank level , and c onsistent with prior literature ( e.g., Nicoletti 2018) , I do not restrict the analysis to stand alone banks (i .e., those not held by a holding company) as doing so would result in the loss of most banks of economic and practical interest as t he majority of U.S. banks are held by holding companies (Avraham et al. 2012) . I use 2004 Call Report data to obtain necessary lagged variables for the analysis, which allows me to retain 2005 in the sample. The dependent variable of the main model requires leading charge - offs, limiting the sample period to 2005 to Deposit information to determine bank branch locations for each year of the sample period. I identify eight major U.S. shale regions based on information available through the Energy Information Administration (EIA). 23 However, for several reasons, I exclude the Texas shale 22 All U.S. Banks are required to file quarterly Call Reports with bank regulators. The auditor is identified in the year end holding company report (FR Y - 9C) for all bank holding companies. Call Report and FR Y - 9C data is publicly available through the FDIC . 23 I used the county - level shale production data from eia.gov to identify the specific counties identified as being part of a shale region. 31 regions from my analyses. 24 Figure 1 shows a map of each shale region alon g with the year that significant fracking activity commenced within each region. As depicted in Figure 2, I identify the following 11 shale play states: MT, ND, WY, NE, CO, OK, AR, LA, OH, PA, and WV. 25 Using the Allcott and Keniston measure of oil and natu ral gas endowment as discussed in s ection 3 .1.1 and APPENDIX A, I calculate both a continuous and dichotomous measure of a I calculate a continuous measure of bank exposure ( Exposure s located in highly endowed counties. For the dichotomous measure ( Exposed ) , banks with any branches in high - endowment counties are identified by the indicator variable Exposed . For the analysis using Exposed , the banks with branches in play states, but with out branches located in counties with high levels (i.e., above the median ) of oil and natural gas endowment are used as control bank s . Figure 2 depicts the treatment and control counties used in this analysis . I remove all observations with missing data for key variables, as well as observations with zero or negative assets. This results in a n unbalanced panel of 11, 17 8 bank years for the analysis of H1. Table 1, Panel B shows the calculation of the sample for the test of H2 . F or a portion of 24 Significant fracking activity began in the Permian region in 2005 (West Texas), and experimentation in the region began before that date. Due to limitations on the availability of auditor information prior to 2005, I do not have sufficient observations prior to the onset of the boom period to conduct a difference - in - differences analysis. Additionally, there are two different dates for the commencement of significant fracking in shale plays within Texas and there was also significant conventional oil activity in Texas prior to the onset of the fracking boom . This comp licates the establishment of pre and post dates for the control counties. Thus , I exclude all banks with the majority of their branches located in Texas from my sample. See s ection 5 for detailed discussion of an additional analysis completed using a metho dology to allow the inclusion of Texas , which suggests that my results are robust to the inclusion of Texas . I consider the difference - in - difference methodology to provide the strongest identification in my setting, thus the primary analysis is conducted u sing the approach outlined in s ection 3 and excluding Texas. 25 New York is located within the Marcellus Region according to the EIA ; however, I exclude New York from the sample for two primary reasons, 1) The state passed a ban on hydraulic fracking, which limited shale development in the state and 2) the concentration of significant national banking activity in New York City makes banks in non - Endowed New York Counties a n inappropriate control group for the purposes of the analysis. The primary results are robust to inclusion of New York in the analysis. 32 Figure 1 Major U.S. Shale Play Regions This figure depicts the eight major U.S. shale play regions with significant fracking activity during 2005 - 201 4 . Play counties are identified by the Energy Information Administration (EIA). Years indicate the first year of major fracking activity in the region. The date of frackin g activity commencement was established based on prior research (Bartik et al. 2016; Erik P. Gilje et al. 2016) and validated based on review of production and leasing activity. Note: Texas, New York, and New Mexico shale play counties are excluded from the analysis. de Bakken 2007 Niobrara 20 10 Utica 2011 Marcellus 2007 Permian 2005 Andarko 2008 Eagle Ford 2008 Haynesville 2008 33 Figure 2 Treatment and Control Counties within Shale Play States This figure depicts the treatment and control counties used to identify Exposed banks ( Exposed = 1) for the primary analysis. Light grey counties are below the median in Endowment for the play region. Dark grey counties ar e above the median in of Endowment . Banks with the majority of branches in dark grey counties are identified as exposed banks, while banks with the majority of their branches in the light grey counties are identified as control banks. Exposure is the propo rtion 34 the analysis comparing audited to unaudited banks, I limit the sample to banks with less than $500 million in assets, resulting in a sample of 9, 296 bank years. For the analysis of the effect of auditor expertise, I limit the sample to audited banks, resulting in a total of 5,8 37 bank years. 4.2. Descriptive Statistics Table 2 , panel A presents descriptive statistics for the sample of banks used to test H1 . The sample includes 11 , 17 8 bank years with a mean (median) loan portfolio size of $2. 7 billion ($0.9 3 billion) and deposits of $3.3 billion ($1. 2 billion). The mean (median) CO/ALL Ratio ( ALL Quality ) is 0. 184 (0.11 1 percent ) . The relatively low CO/ALL ratio is consistent with banks being over - reserved, as discussed in chapter 3.2.1 . Regarding auditor characteristics , 52. 2 percent of bank years are audited, 5.9 percent are audited by Big 4 auditors ( Big4 ), and 10.8 percent are audited by an industry expert auditor ( IndExpert ) . 26 Table 2, panel B presents descriptive statistics for subsamples of exposed and non - exposed banks. Within the subsample of exposed banks, the mean (median) exposure is 0.6 5 (0.75). On average , the exposed banks are larger compared to non - exposed banks due to larger banks generally having a larger number of branches, which increases the likelihood of a bank being in a high endowment county . The size discrepancy between the exposed and non - expos ed banks presents a potential concern given that large banks may fundamentally differ from small banks. I address this concern in robustness tests by 1) excluding large banks , and 2) utilizing an alt ernative identification strategy . 27 Table 3 presents a correlation matrix showing the pairwise 26 In the subsample of audited banks , 11.2 percent are audited by Big 4 auditors, 42.0 percent are audited by auditors w ith previous experience, and 20.7 percent are audited by industry experts. 27 Chapter 5 describes the procedure to address concerns regarding bank size differences between exposed and non - exposed banks. Specifically, I limit the sample to only those banks u nder $500 million in assets and document qualitatively similar results. Also see chapter 5 for details of the broad sample analysis using an alternative identification strategy and a shift - share approach . 35 correlation of key variables. As expected, there is a significantly negative correlation between ALL Quality and post - boom Exposure ( - 0.03 0 ). 4.3. Results 4.3.1 Hypothesis 1: Effect of a positive economic shock on ALL quality Table 4 presents the results of estimating model (1) to test H1 . The results using a continuous measure of exposure are presented in column (1). The primary coefficient of interest is t he coefficient on Post*Exposure , which measures the marginal impact of exposure to the shale boom on ALL quality . I find a negative and significant coefficient on Post* Exposure ( 3 = - 0.0 32 , p - value < 0.05). This indicates that ALL quality is 11.2 percent lower for a bank with a mean CO/ALL ratio (0.184) at the mean level of exposure for an exposed bank (0.645) compared to a non - exposed bank following the onset of the shale boom. 28 The coefficient on Exposure is insignificant (p - value > 0.10 ), indicating that ALL quality is not significantly associated with Exposure prior to th e onset of the boom, which supports the appropriateness of the control sample. Figure 3, panel A presents a graphical time - series representation of the effect of Exposu re relative to non - exposed banks during the sample period. There is no significant difference in ALL quality prior to the onset of the boom (i.e., t - 3 to t - 1 ); however, after the onset of the boom, ALL quality is negatively related to Exposure , supporting the conclusion that the onset of the shale boom had a negative effect on ALL quality. Table 4, c olumn (2) presents the results of estimating model (1) using the dichotomous measure of Exposed to identify banks with any branches in counties where Endowed = 1. Consistent with the results in column (1), I find a negative and significant coefficient on 28 ( - 0.032*0.645)/(0.184). 36 Figure 3 Time Series Analysis of ALL Quality Panel A presents a graph of the coefficients on Exposure*Time for a regression of ALL Quality on indicator variables for the time relative to onset of the shale boom and an interaction of the indicator variables with Exposure. The black bars represent the effect of Exposure on ALL quality for exposed banks relative to non - exposed banks during the sample period. The 90% confidence intervals are shown in grey. Panel B presents a graph of a the coefficient on a time indicator variable, defined relative to the onset of the boom (t), in estimating a regression of ALL quality on time indicators and contr ol variables consistent with model (1) separately for Exposed banks (black bars) and non - Exposed banks (grey bars). In both panels, t is the first year that bank i is impacted by significant shale boom activity (i.e., t = 1 in the first year Post = 1). ALL Quality in t - 4 is used as the base period for comparison. Coefficients are estimated using control variables consistent with model (1). Panel A: Effect of Exposure on ALL Quality for Exposed Banks relative to non - Exposed Banks Panel B : Time Series comparison of ALL Quality for Exposed and Non - Exposed Banks 37 Post*Exposed ( 3 = - 0.02 4 , p - val ue < 0.05), indicating that ALL quality for exposed banks is 13.6 percent lower compared to non - exposed banks followi ng the onset of the shale boom , for banks at the mean level of CO/ALL ratio and Exposure . Figure 3 , panel B presents a time series comparison of ALL quality separately for exposed and non - exposed banks , where t is the first year of shale boom activity (i.e., first year Post = 1) . There is no significant difference in ALL quality between the base period ( t - 4 ) and periods t - 3 , t - 2 , and t - 1 for either exposed or non - exposed banks. However , beginning in year t , ALL quality for exposed banks is s ignificantly lower compared to the base period. T here is no significant difference in ALL quality compared to the base period for non - exposed banks in any year during the sample period . As discussed in chapter 3, I exclude all observations where the bank was not over - reserved (i.e., CO t+1 /ALL t > 0). As a robustness test, I estimate model (1) without excluding the over - reserved banks. Table 5 presents the results of this analysis and shows the coefficient on Post*Exposure is negative and significant (p - value < 0.05) for both the continuous and d ichotomous measures of exposure, which is consistent with the primary findings in Table 4. 4 .3. 1.1 Additional A nalysis of ALL and C harge - O ffs The l ower ratio of charge - offs in t+1 to ALL in t identified in banks with higher exposure compared to less exposed banks can result from banks 1) experiencing a change in charge - offs and failing to change the ALL estimate, 2) changing the ALL estimate and not experiencing a change in charge - offs, or 3) changing the ALL estimate in a manner that does not corresp ond with the change in charge - offs experienced. To further explore these possibilities, I estimate model (1) by replacing the dependent variable of CO/ALL first with the log of one plus charge - offs ( CO Level ), second with the log of one plus ALL ( ALL Level ) . Table 6 , column (1) presents 38 the results of estimating model (1) using CO Level as the dependent variable and show s the coefficient on Post*Exposure is negative and significant ( 3 = - 0. 30 6 , p - value < 0.01), indicating that following the onset of the shale boom, a n exposed bank with the mean level of Exposure experiences 19.8 percent less charge - offs compared to non - exposed banks . 29 The lower level of charge - offs is consistent with an overall improvement in economic environment that accompanies a shale boom having a positive effect on both the ability of borrowers to service loans as well o n the value of loan collateral. Column (2) show s t he coefficient on Post*Exposure when estimating a model with ALL Level as the dependent variable is insignificant (p - value > 0.10), indicating that the ALL level do es boom. Together, these findings indicate that banks with higher exposure to shale booms experience decreases in charge - offs, but do not ad just the ALL estimate to appropriately capture these decreases , indicating an under - reaction to the shale boom. Figure 4 shows the time trend in ALL levels and CO levels relative to the onset of the shale boom for exposed banks , relative to non - e xposed b anks . In the three years prior to the onset of the shale boom both CO and ALL levels are higher compared to the base period ( t - 4 ); however, in years t , t+1 and t+2 , CO levels are significantly lower compared to the base period, while ALL levels remain similar . These findings are consistent with exposed banks experiencing a reduction in charge - offs following the onset of the shale boom, and under - reacting to the shock by failing to adjust the ALL to account for the change in the economic environment. Th ere are two non - mutually exclusive explanations for t he failure to appropriately adjust the ALL for the change in charge - off levels . The first is an over - reliance on historical charge - off 29 ( - 0.307*0.645 ) = - 0.19 8 39 Figure 4 Time Series Analysis of ALL and CO levels of Exposed Bank s This figure presents a graph of the coefficients on Exposed* Time for regressions of ALL Level and CO Level on indicator variables for the time relative to onset of the shale boom and an interaction of the time indicator variables with Exposed . T is the first year that bank i is impacted by significant shale boom activity (i.e., t = 1 in the first year Post = 1). ALL and CO levels in year t - 4 are used as the base period for comparison. Coefficients are estimated using control variables consistent w ith model (1). -0.05 -0.03 -0.01 0.01 0.03 0.05 t-3 t-2 t-1 t t+1 t+2 t+3 % Difference from level in t - 4 Time relative to onset of shale boom ( t ) Time series comparison of ALL and CO to based period for Exposed Banks only CO ALL 40 information without making necessary qualitative adjustments for the change in the economic environment compared to the historical period. This explanation is consistent with Tversky and Kahneman ( 1974) gesting insufficient adjustment leading to bias under uncertainty . during economic boom times that will allow for earnings smoothing in times of economic downturn . I further explore these explanations in chapter 4.3.1.2. Overall, t he results of the test of H1 indicate that exposure to a positive economic shock results in lower ALL quality relative to banks not exposed to the shock. These finding s are consistent wit h the uncertainty associated with periods of rapid growth increasing the difficult y of developing accounting estimates , and with management under - reacting to positive economic shocks . 4.3.1.2 Additional A nalysis of bank resource constraint and portfolio composition The identified lower financial reporting quality following exposure to the shale boom can result from two, non - mutually exclusive channels: 1) higher difficulty in developing ALL estimate s due to increased uncertainty, and 2) intentional manag ement bias. Conversations with both auditors and bank management exposed to the shale boom indicate that a combination of the two is the most likely explanation. As discussed in chapter 2, there is an increased uncertainty surrounding future cash flows fro m loans. There is also a significant change in the historical charge - off information. The result is an increase d need for appropriate qualitative adjustments to historical charge - off numbers in the development of the ALL. Development of the qualitative factors is inherently difficult and relies he avily on management judgement. This increased reliance on management judgement provides bank management with the op portunity to 41 - loan loss. The excess reserves can be released at a future time, inflating earnings in a future period. This results in earnings smoothing, consis tent with the pattern identified by Liu and Ryan ( 2006) . In additional analysis , I attempt to determine whether o ne of the two channels dominates. If the primary channel is the challenge of compiling and analyzing the information necessary to develop a higher quality ALL, banks with greater resource constraints in terms of human resource capacity would be expected to be more significantly affected compared to banks with lesser HR constraints . I use the ratio of bank full - time employees to total assets as a proxy for human resource availability. I define the variable HR_Constrain as an indicator variable equal to 1 if a bank is below the sample median in the ratio of full - time employees to total assets, and zero otherwise. I separately estimate model (1) for banks with above and below median ratio of full - time employees to assets . 30 Table 7, panel A, column (1) shows a n egative and significant (p - value < 0.05) coefficient on Post*Exposure for the subsample where HR_Constrain = 0, while the coefficient is negative and insignificant (p - value > 0.10) for the subsample where HR_Constrain = 1. However, panel B shows that in a chi - squared test, there is no significant difference in the coefficient across the two subsamples. Thus , the analysis provides weak or no evidence that the effect of the shale boom is concentrated in those banks with less resource constraint s. The finding is not consistent with bank resource constraint resulting in more significant impact of the shale boom . 31 30 The findings are unchanged if total employees is scaled by gross loans rather than total assets. 31 I acknowledge that the proxy for resource constraint is noisy, given that it is based on total employee headcount. However, due to data limitations, the count of employees associated specifically with the loan function is not available, thus the total emp loyee to asset count is the best available proxy. 42 Loan portfolios that are more heterogeneous present a greater challenge to bank management in terms of developing the ALL, as a broader knowledge of the effect of an economic shock on different types of loans is required. Thus, i f the difficulty of developing the ALL is the primary reason for lower ALL quality following exposure to the shale boom, I predict that those banks with a more heterogeneous or higher - risk loan portfolio prior to the onset of the boom would exhibit significantly greater deterioration in ALL quality compared to the more homogenous or lower - risk loan portfolios. I measure portfolio risk by multiplying the average charge - offs by lo an types for all U.S. banks from 1992 - 2016 by the proportion of loans of a bank based on historical charge - offs across all banks. I define High_Risk as an indicator variable Consistent with Acharya et al. (2006) , I capture portfolio concentration using the Hirschman Herfindahl in dex (HHI) measure. 32 I define Diverse_Port as an indicator variable equal to 1 if a bank is below the sample median in portfolio diversification and zero otherwise. In order to have a clean measure of ex ante risk and concentration prior to the onset of the boom, I measure both loan portfolio risk and concentration as of 2007 . Consistent with the test of the effect of human resource availability, I split the sample at the median of loan risk and loan concentr ation and estimate model (1) for each subsample. Table 8, panel A shows a negative and marginally significant (p - value < 0.10) coefficient on Post*Exposure for the lower risk banks in column (1) and for banks with a more homogenous loan portfolio in colu mn (3). In contrast, the coefficient s on Post*Exposure for the higher risk 32 (Acharya et al. 2006) focus on loan concentration based on industry, using a unique single - country (Italy) dataset. Due to limitations on availability of industry - level loan da ta for the sample, I focus instead on loan type based on FDIC Call Reporting Codes. I consider this to be a reasonable level of analysis for loan concentration as it mirrors the loan type breakdown evaluated by regulators. 43 (column 2) and more heterogeneous loan portfolios (column 4) are negative and insignificant (p - value > 0.10). However; panel B shows no significant difference in the coefficients a cross the either of the loan portfolio composition splits. Thus, overall this analysis does not provide evidence of a significant difference in the effect of the boom based on loan portfolio composition. Based on these additional analys e s, it is not possible to definitively say that the effect of the shale boom is primarily driven by either the difficulty of developing the loan loss estimate or by management bias. While this could be due to the challenge of identifying a noise - free proxy for portfolio risk or resource constraint, it also points to the explanation that a combination of both channels is responsible for the results of the primary analysis. 4. 3.2 Hypothesis 2 : Effect of the auditor on ALL quality during a positive economic shock I test the effect of the auditor on ALL quality following exposure to a fracking shock by firs t comparing subsamples of audited and unaudited banks. Table 9 , panel A columns (1) and (2) present the results of estimating model ( 1 ) separately for each subsample , and panel B 2 test of difference on the coefficient of interest ( Post*Exposure ) across the subsamples . I find that the coefficient on Post* Exposure is negative and insignificant (p - value > 0.10) for unaudited banks and negative and marginally significant (p - value < 0.10) for audited banks . Panel B , column (1) shows that there is no significant difference in the coefficient on Post*Exposure between audited and unaudited banks , indicating the effect of Exposure on ALL quality following the shale boom does not differ based on presence of an auditor . The lack of significant difference in coefficients across these subgroups suggests that an auditor alone does not mitigate the lower ALL quality associated with the boom compared to non - expert 44 auditors . This finding is consistent with auditors , on average, not appropriately adjusting their risk assessment and audit procedures to changes in the economic environment of their clients due to the shale boom. I then examine th e effect of general auditor expertise due to access to greater firm resource s within a subsample of audited banks. Ta b le 9 , panel A, c olumns (3) and (4) present the results of estimating model (1 ) for subsamples of banks audited by non - Big 4 and Big 4 auditors respectively , limiting the sample to audited banks . The coefficient on Post*Exposure is negative but insignificant (p - value > 0.10) for the both subsamples . Panel B shows there is no significant difference in the coefficient on Post*Exposure across the regression s , indicating that the effect of the exposure to the shale boom does not differ based on the presence of a Big 4 auditor. This result suggests that the general expertise of Big 4 auditors does not significantly mitigate the effect of the shale boom on ALL quality compared to non - Big 4 auditors . While I do not find evidence that the presence of an auditor in general is associated with higher ALL quality during a boom period compared to non - audited banks, there is an extensive literature that suggests that Big 4 and industry expert auditors perform higher - quality audits compared to non - expert auditors (DeFond and Zhang 2014) . Thus, it is possible that banks audited by a Big 4 or industry expert auditor may have higher ALL qual ity following a shale boom compared to unaudited banks. Table 11 presents an analysis of banks audited by Big 4 and industry expert auditors compared to unaudited banks, and in this analysis I do find evidence that auditors with expertise are associated with less deterioration in ALL quality following exposure to the shale boom compared to unaudited banks. This result suggests that there is variation in the effects of ALL quality based on aud itor expertise within audited banks and supports the analysis of auditor expertise within the subsample of audited banks. 45 Next, I examine the effect of industry expertise on ALL quality following exposure to a shale boom . Table 10 , panel A, c olumns ( 1 ) and ( 2 ) present the results of estimating model (1) for subsamples of banks audited by non - industry expert and industry expert auditors respectively. In column (1), the coefficient on Post*Exposure is negative and significant ( 3 = - 0. 047; p - value < 0.05) for the subsample of banks audited by non - experts, which is consistent with a negative relationship between Exposure and ALL quality following the onset of a shale boom . In contrast, column ( 2 ) shows a positive , although insignificant coefficient on Post* Exposure ( 3 = 0. 046; p - value > 0.10) for the subsample of banks audited by industry experts, indicating that Exposure is positively related to ALL quality for clients of industry experts following the onset of the shale boom. Panel B, column ( 1 ) shows the coefficient on Post*Exposure is significantly different across the regressions ( Difference = 0. 116; p - value < 0.0 5 ) , indicating that industry expertise is associated with significantly higher ALL quality for exposed banks following the shale boom relative to banks audited by non - industry experts. These results indicate that industry expertise is effective at mitigating deterioration in ALL quality following exposure to a shale boom. Next, I examine the effect of task - spec ific expertise on ALL quality following exposure to a shale boom. Table 10 , panel A, columns ( 3 ) and ( 4 ) present the results of estimating model (1) for subsamples of banks audited by non - task - specific expert and task - specific expert auditors respectively. In column ( 3 ), the coefficient on Post*Exposure is negative and significant ( 3 = - 0. 060; p - value < 0.05) for the subsample of banks audited by non - task - specific experts , and column ( 4 ) shows the coefficient on Post*Exposure is negative and in significant ( 3 = - 0. 038; p - value > 0.10) for the subsample of banks audited by task - specific experts . Panel B, column ( 2 ) shows the coefficient on Post*Exposure is significantly higher ( Difference = 0. 022; p - value < 46 0.10) for clients of task - specific experts , indicating that task - specific expertise is associated with higher ALL quality for exposed banks following the shale boom relative to banks audited by non - task - specific experts . The results provide some support for the effectiveness of task - specific expertise at mitigating deterioration in ALL quality during a shale boom . In the sample, there is overlap in the auditors that have industry and task - specific expertise. In total, 6 3 percent of auditors with industry expertise also have task - specific expe rtise and 31 percent of auditors with task - specific expertise also have industry expertise. The overlap makes it challenging to determine whether it is industry expertise, task - specific expertise, or a combination of the two that is responsible for mitigating deterioration in financial reporting quality following the shale boom. To disentangle the effects of industry and task - specific expertise, I estimate model (1) for each of the following three subsamples: 1) clients of auditors with neither industry , nor task - specific expertise, 2) clients of auditors with either industry , or task - specific expertise (i.e., single expertise), and 2) clients of auditors with both industry and task specific expertise (i.e., combination expertise). I compare the coefficient s on Post*Exposure across the subsample regressions . Table 12 , panel A, columns (1) and (2) present the results of estimating model (1) for clients of an auditor with neither industry nor task - specific expertise . C onsistent with the test of H1 , t he coefficient on Post*Exposure is negative and significant ( 3 = - 0. 053; p - value < 0.05) . Column (2) presents the results for the subsample of banks audited by an auditor with either industry or task - specific expertise (but not both) and the coefficient o n Post*Exposure is negative and statistically insignificant ( 3 = - 0. 0 70 ; p - value > 0.10) ; and the magnitude of the coefficient is greater than the coefficient on Post*Exposure for the subsample in column (1). Panel B, column (1) shows that the differences in the coefficient between the two regressions is marginally significant, suggesting greater deterioration in financial reporting 47 quality in banks audited by auditors with eit her industry or task - specific expertise (but not both) compared to banks audited by auditors with neither type of expertise. A potential explanation for this result is that auditors without the necessary industry background to understand how changes in the economic environment affect the ALL , or without the experience to understand the changes , may encourage clients to be more conservative in their ALL estimates as uncertainty increases and the auditors observe their other exposed clients moving in a more c onservative direction. In contrast, the coefficient on Post*Exposure for clients of auditors with both industry and task - specific expertise is in significant ( 3 = 0. 106; p - value > 0.10), indicating that auditors with a combination of industry and task - specific expertise effectively mitigate deterioration in financial reporting quality following a shale boom . Panel B, column (2) shows the coefficient is significantly higher for banks audited by dual - experts compared to the regression estimate for ba nks audited by an auditor with only a single expertise. Taken together, t he results indicate that industry expertise or task - specific expertise alone is insufficient to mitigate lower ALL quality resulting from exposure to shale booms . Instead, it is the c ombination of industry and specific expertise that is effective in mitigating the effect of exposure . In an additional analysis presented in T able 13, I separately examine only industry expertise and only specific expertise and find that neither expertise alone is associated with significantly higher ALL quality compared to banks audited by non - experts following the onset of the shale boom. I fi nd that specific expertise alone is associated with lower quality ALL following the onset of the fracking boom, compared to the ALL quality of banks audited by a non - expert. These findings are consistent with single expertise alone being insufficient to mi tigate the lower ALL quality following exposure to a fracking boom. These findings are consistent with behavior al research and theory (e.g. , Bonner 1990; Bonner and Lewis 1990; 48 Arrow 19 62) suggesting that learning by doing is an important component of expertise; however, the results also indicate it is necessary for auditors to have industry - level knowledge in order to appropriately learn from and adjust their audit approach based on specific experiences following a positive economic shock . In an additional analysis, I also compare the effect of a combination of industry and task - specific expertise to Big 4 expertise, as Big 4 auditors are commonly accepted as experts in prior literat ure (e.g., DeFond and Zhang 2014) . In Table 14, I find that a combination of industry and task - specific of non - Big 4 auditors is significantly more effective at mitigating deterioration in financial reporting quality compared to Big 4 expertise. This result underscores the finding that a fit between to the effectiveness of auditor expertise and suggests that high - level generic proxies may not capture auditor expertise in all circumstances. 49 CHAPTER 5: SUPPLEMENTAL ANALYSIS AND ROBUSTNESS 5.1. Broad Sample Analysis: Shift - share approach One of the potential limitations of the research design for the main analys e s of this paper is that it requires an arbitrary cutoff point for a county to be considered highly endowed (i.e., above the median), as well as the identification of the specific year when the shale boom began in each shale play. Furthermore, to establish a clear treatment and control group, I limit the sample to only banks operating in play states, as described in chapter 3.1.1 and APPENDIX A . In order to address concerns that these research design choices significantly influence the findings, I use an alt ernative empirical strategy inspired by the Bartik ( 1991) - (Allcott and Keniston 2018) . In this approach, the independent variable of interest is an interaction between a standardized measure of a BankEndowment ) based on its branch location s and national time series variation in active horizontal drilling rigs ( Rigs ). Horizontal drilling is a key first stage in the fracking process; thus, the national level of active horizontal drilling rigs is an indicator of the intensity of start - up fracking activity. Specifically, f rom 2005 to the height of shale - boom activ ity in 2014, the count of horizontal drilling rigs increased by more than 500 percent, from 222 to 1,372. The average increase year over year was 22 percent during the same period. H orizontal drilling activity covaries with BankEndowment , in that the impact of an increase in number of horizontal drilling rigs should be positively related with the Keniston endowment measure covers nearly all c ounties in the United States, I can include all bank - year observations in the sample, rather than limiting the sample to only banks with branches located in play states. Using the expanded sample, I estimate the following model: 50 ALL Quality i t = 1 Bank Endowment is t + 2 Rigs * Bank E ndowment i st + Controls is t + e is t ( 2 ) The model includes the first difference of control variables and logged percentage change in ALL Quality and is estimated with year and state fixed effects. 33 BankEndowment is a standardized and natural gas deposits , and Rigs is the log percentage change in the number of horizontal drilling rigs . 34 Table 15 presents the results of the shift - share analysis . The coefficient on Rigs* Bank Endowment is negative and significant ( 2 = - 0.067, p - value < 0.05), indicating that for each standard deviation increase in bank endowment, a one percentage increase in rig count reduces the ALL quality by 6.7 percent . This result is consistent with the results of the main analysis and provide s support for H1 using a broader sample of bank - year observations. The insignificant coefficient on Bank Endowment indicates that trends in ALL quality changes do not significantly vary based on a bank s endowment , providing further support for the strength of the identification strategy . 5.2 Identifying assumptions and robustness 5.2.1 Parallel trends assumption and f alsification t est The key assumption of the difference - in - differences methodological approach is the zero correlation assumption, commonly referred to as the parallel trends assumption (Roberts and 33 The main effect of Rigs is sub sumed by year fixed effects. 34 BankEndowment continuous measure of Endowment in county c, and summed by bank. BankEndowment is standardized by subtracting the sample mean of BankEndowment and dividing by the sample standard deviation. The standardization of the variable aids in interpretation. 51 Whited 2013) . This assumption requires that trends are the same on average for the treatment and control groups in absence of treatment . Although the assumption can not be explicitly tested, there are several analyses that can aid in mitigating concerns that different trends confound the results . I first analyze the reasonableness of the parallel trends assumption by graphically comparing the trends prior to the treatment period. Figure 3, p anel A demonstrates that there is no significant difference in the ALL quality for the years prior to exposure and there is no indication of divergent trends between the two groups. Furthermore, Figure 3, p anel B shows that there is not a significantly divergent trend in ALL quality between exposed and non - exposed banks prio r to the onset of the boom. Together these figures provide support for the parallel trends assumption. F ollowing Roberts and Whited (2013), I also conduct a falsification test. For the falsification test, I estimate model (1) by replacing the Post variable with Post _ False , an indicator equal to 1 for the three years prior to the onset of the shale boom (i.e., Post = 1 in year t - 3 ) and exclude all post - period observations (i.e., year = t and subsequent years) . Table 16 , column (1) presents the results of the falsification test and shows the coefficient on Post _ False*Exposure is negative and insignificant (p - value > 0. 10 ) . Additionally, I conduct the falsi fication test without excluding post - period observations . As shown in Table 1 6 , column (2), I continue to find that the coefficient on Post _ False*Exposure is insignificant (p - value > 0.10) . These findings are consistent with expectations for the falsification test and provides further support for the difference - in - differences findings. I acknowledge that there are likely spillover effects of a fracking boom due to the flow of money, jobs , and people across county lines. However, the use of a within - state control group biases against finding results in the presence of a significant spillove r effect. To further address 52 this concern, I also use an alternative identification methodology in the shift - share analysis discussed in section 5. 1 , which provides further assurance that the specific research design decisions in identifying the treatment and control group do no t drive the results. D ue to the likelihood of spillover effect s across county lines, it is unlikely that only the high - endowment counties were affected by fracking shocks ; however, the supplemental analyses provide s confid ence that the results are not sensitive to the precise definition of treatment and control groups . Furthermore, any spillover should bias against finding results. 35 5.2.2. Endowment measure exogeneity Although the location of a bank or audit firm over a shale play is not directly chosen, it is possible that banks located in counties with high levels of oil and natural gas endowment differ from banks in counties with lower levels of endowment, even within the sa me state. If this is the case, then characteristics of banks other than the exposure to the disruptive resource boom could drive differences in audit and financial reporting quality. To address this concern, I test the correlation of the endowment measure with various characteristics of the counties that may potentially affect financial reporting quality prior to the onset of the oil boom ( Post = 0 ). The purpose of this test is to examine whether endowment can be required by a quasi - experiment, or if additional empirical strategies are needed to address initial assignment. I examine the correlation between size , deposit growth, charge - off s , non - performing loan s/asset levels a nd loa n growth and I estimate the correlation by separately regressing each 35 According to Glaeser and Guay ( 201 7) in the pre - treatment period and that all firms in the treatment group and only those firms received the treatment in The stable unit value treatme (Glaeser and Guay 2017) . The supplemental analyses performed mitigate concerns related to potential violations of these assumptions. 53 characteristic of interest on a continuous measure of share of bank branches located in highly - endowed counties in 200 7 ( Exposure 2007 i ) . Consistent with the main analyses, I estimate each relation using state and year fixed effects and cluster standard errors by bank. Table 17 presents results showing no s ignificant correlation (p - value > 0.10) between Exposure 2007 i counties and bank characteristics of interest . The results provide confidence that the Exposure measure is uncorrelated with bank characteristics that may explain the results and provides evidence of the exogeneity of the exposure measure. 36 While this analysis shows no significant correlation between size and the Exposure measure, Table 2 shows that there is a size discrepancy between the subgroups of exposed and non - exposed banks. To address concerns that size differences may influence resul ts, I reperform the primary analysis using a sample of only banks with assets less than $500 million . Consistent with the primary results, Table 18 shows a negative and significant coefficient on the interaction of Post with both the continuous and dichoto mous measures of exposure ( 3 = - 0.027, p - value <0.10; and 3 = - 0.027, p - value < 0.05). The limited sample removes the large banks which drive the size discrepancy between the two groups and mitigates concerns that size differences between exposed and non - exposed banks explain the results . Furthermore, the shift - share analysis described in section 5.1 utilizes a first - differencing estimation method, which further mitigate concerns that co rrelation between bank - level characteristics and oil and natural gas endowment explain the primary results. 36 I acknowledge that this analysis addresses only observable characteristics, while unobservable or difficult to quantify characteristics such as governance and bank cost structure are not directly addressed. Howe ver , given that there is no correlation in observable characteristics, and no theoretical reason that the geological features associated oil and natural gas endowment should be correlated with bank characteristics, I consider the assumption of ignment to be reasonable. 54 5.2. 3 Ex ante assignment The variable Exposure is calculated on a bank year basis to most accurate ly capture the exposure of a given bank to the fracking shock in a given year . However, the time - varying nature of this variable may raise concerns that banks chose to increase the number of branches in highly impacted counties during this period ; thus , there may be self - selection of banks into the exp osed counties. T o mitigate this concern, I replace Exposure with Expos ure2007 , which is the - endowment county in 200 7, the first year of significant oil boom activity in the sample . The identification of a b ank based on their branch location before the fracking shock, mitigates concerns that banks were able to self - select into greater exposure over the course of the boom activity . Table 19 , column (1) presents the results of the analysis using ex ante exposure to the shale boom. Consistent with the primary results, the coefficient on Post*Exposure2007 is negative and significant ( 3 = - 0.025, p - value <0.10), mitigating concerns that changes in bank branch locations during the sample period could explai n the findings. Another potential concern is that banks may choose to change from a non - industry expert auditor to an industry expert auditor in response to being exposed to the fracking boom. To mitigate this concern, I reperform the analysis excluding al l banks that changed auditors . Table 19 , column (2) shows that the coefficient on Post*Exposure2007 is negative and significant ( 3 = - 0.040, p - value <0.10), and is even greater in magnitude compared to the estimation using the full sample. This results mi tigates concerns that auditor changes in response to the fracking boom may explain the primary results. 55 5.2. 4 Analysis of Loan Loss Provision As discussed in Chapter 3, the loan loss provision (LLP) is often used in prior literature to assess the quality ( e.g., Altamuro and Beatty 2010; Beatty and Liao 2014) . For the purposes of this study, I consider the balance sheet component, the ALL, to be the appropria te focus of the empirical analysis, nevertheless, it is important to also understand how d future charge - offs. To examine this relationship, I test hypothesis 1 using the followi ng model adapted from (Altamuro and Beatty 2010) : ChargeOff ist+1 = 1 Post ist 2 Exposed ist 3 Post*Exposed ist 4 LLP ist 5 Post*LLP ist 6 Exposed*LLP ist + 7 Post*Exposed*LLP ist 8 Size ist 8 Size*LLP ist 9 NP A ist + 6 Consumer it 7 C&I it + s + t + e ist ( 3 ) where subscripts refer to bank ( i ), state (s) and year ( t ) . The loan loss provision ( LLP ), is defined as loan loss provision during year t scaled by beginning total assets , and charge - offs ( ChargeOff ), defined as charge - offs in year t+1 scaled by beginning total assets . All other variables are defined consistent with model (1). t , to control for time effects across all observations, and I include state fixed effects, s , to control for time - constant state characteristics. Consistent with Altamuro and Beatty (2010) , I interact Size*LLP to allow for the association between LLP to vary based on bank size. 37 37 In an untabulated sensitivity analysis , I interact variable Post , Size , and Post*Size with all explanatory variables to allow for the effect of each variable to vary across bank size and across the pre - and post - boom periods. There is no change in the inference of the results from estimating the model specification with all interactions . 56 A positive relationship between the LLP and subsequent charge - offs is considered to be indicative of higher LLP quality . For the purposes of the analysis, I am interested in how exposure to the shale boom impacts the relationship between LLP and subsequent chargeoffs. The model uses a difference - in - differences approach where the first difference ( 6 ) represents the difference in LLP quality between banks in endowed and non - endowed counties prior to the onset of the boom and the difference - in - difference esti mator ( 7 ) represents the difference in LLP provision quality between exposed and non - exposed banks after the onset of the fracking boom compared to the difference prior to the boom. 38 A negative (positive) coefficient on 7 indicates that the quality of the LLP is lower (higher) in the post - boom period for exposed banks as compared to the quality in the pre - boom period. A lower (higher) association between current year LLP and subsequent charge - offs suggests lower (higher) LLP quality . Table 20 prese nts the results of estimating model ( 3 ) to test hypothesis 1. The primary coefficient of interest is Post*Exposed*LLP , which measures the marginal impact of exposure to the fracking boom on bank LLP quality . Using the continuous measure of exposure in colu mn (1), I find a negative and significant coefficient on Post* Expos ure *LLP ( 7 = - 0.16 0 , p - value < 0.05), indicating a weaker association between the LLP accrual and subsequent charge - offs for banks located in counties with high resource endowment after the onset of the fracking boom. Similarly, I find a negative and significant coefficient on Post*Exposed*LLP ( 7 = - 0.1 26 , p - value < 0. 10 ) using a dichotomous measure of exposure. The results suggest that exposure to fracking shocks results in lower LLP quality relative to banks not exposed to the shock. Th e finding is consistent with the primary findings presented in table 4 . 38 The difference - in - difference estimator is determined as follows: where represents the mean relationship between LLP and future charge - offs. 57 CHAPTER 6: CONCLUDING REMARKS In this study , I investigate the effect of rapid growth cause d by a positive economic shock on financial reporting quality. I argue that exposure to a positive economic shock presents internal and external threats to the quality of accounting estimates. Internally, rapid growth can lead to a deterioration in internal controls , whi ch provides an opportunity for both error in operating environment increases the challenge in developing and validating the assumpt ions to predict future cash flo ws . Consistent with an economic shock resulting in both internal and external challenges for banks, I find that ALL quality is lower for banks exposed to a positive economic shock . I investigate the efficacy of auditor expertise in mitigating lower estima te quality positive economic shock. task - specific expertise, developed through prior experience auditing clients exposed to a fracking shock , combined with industry - specific expertise is associated with higher estimate quality , compared to auditors with expertise resulting from general expertise, greater resources (i.e. , Big 4 auditors) or industry - specific experience alone . These findings are consistent with the e importance of auditors gaining a proper understanding of economic environment, particularly when auditing accounting estimates. The results suggest that general and industry specific expertise, and access to extensive firm resources alone do n ot fully compensate for lack of first - hand knowledge of the external economic challenges facing an audit client. I acknowledge that , as with any natural experiment, the results may not be widely generalizable. However, this study capitalizes on a strong setting to provide evidence that a 58 quality. I identify important cross - sectional variations in auditor expertise that suggest that further research is needed to un derstand how and when different types of auditor expertise contribute to financial reporting quality. The primary contribution of this paper is in demonstrating that financial reporting quality and the effects of auditor expertise are not constant across a business cycle. The findings suggest that research examining financial reporting quality and auditor expertise should consider how the economic challenges facing firms could affect the results and conclusions of the research . Additionally, this paper cont ributes to the banking literature by examining how positive economic shocks impact ALL quality . The results have practical implications given the importance of bank financial reporting quality to the overall economy and the importance of financial reporting during expansionary periods. 59 APPENDICES 60 APPENDIX A ENDOWMENT MEASURE D ETAILS AND VALIDATION 61 Endowment Measure Calculation To measure ex ante oil and natural gas endowment, I use a measure constructed by Allcott and Keniston (2018) . The measure is calculated as follows: where subscripts represent county (c) and year (t). Production is the total production of natural gas and oil from 1960 to 2013. ProvenReserves are based on the amount of reserves that are known, but have not yet been extracted. UndiscoveredReserves are based on estimates by the U.S. geologic knowledge and theory to exist outside of known accumulations (Schmoker and Klett 2013) differences in county size. The resulting endowment density measure provides physical units of oil and gas, which are converted to dollar amounts based on the average price of oil and gas from 1960 - 2011, using real 2010 dollars. 39 The measure that I use is calculated as of 2013; however, the total amount of endowment does not change from year to year as it is based on the total amount of oil and gas reserves that were available for extraction as of 1960. The various components of th e measure (i.e., Production , ProvenReserves and UndiscoveredReserves ) may vary from year to year, for example as oil and gas is produced it moves from being a ProvenReserve to being included in the production summation, but the total endowment number remai ns constant. The reliance of this measure on geological factors and proven reserves, rather than production alone results in the consistency, and ultimately exogeneity of the measure for use in determining which banks are exposed to fracking shocks. 39 Average p rices used are $34.92 per barrel of oil and $3.20 per mm Btu of gas (Allcott and Keniston 2018) . 62 Endowm ent Measure Validation The first step in validating the appropriateness of the Allcott and Keniston measure for the setting was to replicate the deposit growth results of Gilje et al. (2016) to demonstrate that bank endowment. I follow Gilje et al. (2016) in estimating the effect of high levels of endowment using the following model: it 1 Exposure it 2 Size it - 1 3 Deposits it - 1 3 Liquidity it - 1 s + t + e (4) where the subscripts represent bank ( i ), state ( s ) and year ( t ), and represents change in deposits calculated as the natural log of the ratio of deposits in year t to deposits in year t - 1. Exposure is the standardized proportion of bank i - endowment counties in year t . Size is the natural log of total assets at the end of t - 1, Deposits is total deposits scaled by total assets in year t - 1 , and Liquidity is total liquid assets (cash and cash equivalents) scaled by total assets in year t - 1 . Consistent with the primary analysis, the sample size is limited to only banks with branches located within a play state. I limit the sample period to only post - boom observations, as this is the period when exposure would be expec ted to impact deposit growth. The model is estimated with state fixed effects ( s ) to control for time - invariant state characteristics t ) to control for time trends. The coefficient of interest is 1 which represents the relationsh growth. A positive value for 1 indicates that exposure to the fracking boom results in a higher deposit growth rate , which is consistent with the results in Gilje et al. (2016) and the expectations of the impact of a fracking shock on deposit growth. Table 21 presents the results of the measure validation and shows that the coefficient on Exposure is positive and significant, indicating that 63 exposed banks experienced higher deposit growth than non - exposed banks following the onset of the fracking boom and prov iding support for the validity of the Allcott and Keniston endowment measure. Allcott and Keniston (2018) obtained confidential access t o the U.S. Energy Information - level data for the proven reserves portion of the endowment measure. Due to the confidential nature of the data, I was not able to re - create the full measure, thus relied on the information provid ed by Hunt Allcott and Daniel Keniston. However, as a further validation of the measure, I independently construct a similar measure using information from DrillingInfo, a market research company. DrillingInfo provides an estimate of ultimate recovery of o il and natural gas (EUR) on a well - level basis. 40 This measure provides a reasonable proxy for the production and proven reserves portion of the Allcott and Keniston measure. From the well - level EUR I calculated a county - level EUR and replicated my test of H1 in this study using the EUR measure. Table 22 presents the results of this analysis and I find a ll results consistent with the findings using the Allcott and Keniston measure. I relied on the Allcott and Keniston measure data allow for stronger identification. 40 According to DrillingInfo, the EUR for each well is the technically recoverable resour ces and are calculated based on a combination of historical production data and forecasts. The EUR estimates the total amount of oil and/or gas that could be recovered from a given well. 64 APPENDIX B VARIABLE DEFINITIONS 65 66 67 68 APPENDIX C ILLUSTRATION OF INTERACTION OF ENDOWED , EXPOSED , AND POST VARIABLES 69 First Exchange Bank, Mannington, WV Marion County, WV (FIPS: 54049) State: West Virginia Play: Marcellus Allcott and Keniston 2013 Endowment: $2,923,458,000 County Square Miles: 311.518 mi 2 Endowment Density: $9,384,556 / mi 2 Marcellus Play Endowment Density Top Quartile: $9,307,775 / mi 2 First Year of Fracking Boom in State: 2007 First Year of Fracking Boom in Play: 2007 The endowment measure places Marion County in the top quartile for the play, thus Marion County is an Endowed percent) are located in Marion County ; t hus , Exposure = 0.667. F branches located in Endowed Counties; thus, Exposed = 1. Fracking activity began in the Marcellus Play in 2007, thus , for First Exchange Bank, Post = 0 for 2005 - 2007, and Post = 1 from 2008 - 2014. All banks with branches in West V irginia counties, but with oil and natural gas endowment density below the top quartile for the Marcellus play (<$9,307,775/mi 2 ) are coded as Exposed = 0 for the full sample period, Post = 0 for 2005 - 2007, and Post = 1 from 2008 - 2014. 70 APPENDIX D INCURRED LOSS MODEL VS CURRENT EXPECTED CREDIT LOSS MODEL 71 Prior to 2018, (i.e., the entire sample period of this paper), banks determined the necessary ALL based on the incurred loss model. Under this model, losses should be recognized only when incurred and only those losses which are inherent in the loan portfolio at the time of the financial statements should be included in the ALL. Using the incurred loss model, the ALL is developed in two parts. In one part, loans that are impaired (i.e., not expecte d to collect all contractual interest and principal payments) are evaluated individually to determine the inherent losses under ASC 310 - 10 (FAS 114). The second part of the ALL is estimated for the performing loans on a pooled basis following ASC 450 (FAS 5). The required ASC 450 allowance is determined using historical losses and qualitative adjustment factors to cover economic changes in the existing loss emergence period (i.e., the time that it will take for losses that are inherent in the portfolio to b ecome evident). Long - horizon forecasts are explicitly prohibited in determining appropriate ALL levels using the incurred loss model. The requirement that losses be inherent in the portfolio before being recognized in the ALL is the basis for the reasonabl eness of the one - year period of charge - off coverage often referenced by standard setters and regulators and used as the basis for the analysis in this paper. In June 2016, the FASB issued ASU No. 2016 - Financial Instrumen Current Expected Credit Loss (CECL) methodology to evaluate pooled performing loans. In addition to consideration of historical losses and current economic conditions, the methodolo gy allows for the inclusion of reasonable and supportable forecasts in determining the appropriate ALL. Thus, rather than restricting the ALL to only the losses inherent in the portfolio as of the reporting date, CECL methodology requires banks to look for ward to incorporate future expected losses into the ALL calculation. The change to the CECL model for calculating loan losses makes 72 the understanding of the effect of an economic shock on the loan portfolio even more salient for bank management and auditor s, as these groups must now not only react to the current effect of an economic shock, but also consider reasonable forecasts of the future effects of the shocks on the collectability of loans. The change will increase not only the number, but also the unc ertainty of the assumptions underlying the development of the ALL estimate. 73 APPENDIX E TABLES 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 REFERENCES 105 REFERENCES Abdolmohammadi, M., and A. Wright. 1987. An Examination of the Effects of Experience and Task Complexity on Audit Judgments. The Accounting Review 62 (1): 1 13. Acharya, V., I. Ha san, and A. Saunders. 2006. Should Banks be Diversified? Evidence from Individual Bank Loan Portfolios. The Journal of Business 79 (3): 1355 1412. Stability. Journal of Accounting Research 54 (2): 277 340. Ahmed, A. S., C. Takeda, and S. Thomas. 1999. Bank loan loss provisions: a reexamination of capital management, earnings management and signaling effects. Journal of Accounting and Economics 28 (1): 1 25. Allcott, H., and D. Keniston. 2018. Dutch Disease or Agglomeration? The Local Economic Effects of Natural Resource Booms in Modern America. The Review of Economic Studies 85 (2): 695 731. Altamuro, J., and A. Beatty. 2010. How does internal control regula tion affect financial reporting? Journal of Accounting and Economics 49 (1 2): 58 74. Amble, J., M. Gallagher, J. Joseph, S. Peters, and P. Santarelli. 2015. Emerging issues that could affect Audits, Auditors or the PCAOB presented at the Standing Advisor y Group Meeting, November, Washington, DC. Arrow, K. J. 1962. The Economic Implications of Learning by Doing. The Review of Economic Studies 29 (3): 155 173. Ashbaugh - Skaife, H., D. W. Collins, and W. R. Kinney. 2007. The discovery and reporting of inter nal control deficiencies prior to SOX - mandated audits. Journal of Accounting and Economics 44 (1). Conference Issue on Corporate Governance: Financial Reporting, Internal Control, and Auditing: 166 192. Avraham, D., P. Selvaggi, and J. Vickery. 2012. A St ructural View of U.S. bank Holding Companies . FRBNY Economic Policy Review. New York, NY: Federal Reserve Bank of New York. Balsam, S., J. Krishnan, and J. Yang. 2003. Auditor industry specialization and earnings quality. Auditing: A Journal of Practice & Theory 22 (2): 71 97. Barth, M. E., and W. R. Landsman. 2010. How did Financial Reporting Contribute to the Financial Crisis? European Accounting Review 19 (3): 399 423. 106 Bartik, A. W., J. Currie, M. Greenstone, and C. Knittel. 2016. The Local Economic a nd Welfare Consequences of Hydraulic Fracturing. NBER Working Paper Series. Bartik, T. 1991. Who Benefits from State and Local Economic Development Policies? Kalamazoo, MI: Upjohn Insitute for Employment Research. Beatty, A. L., B. Ke, and K. R. Petroni. 2002. Earnings management to avoid earnings declines across publicly and privately held banks. The Accounting Review 77 (3): 547 570. Beatty, A., and S. Liao. 2014. Financial accounting in the banking industry: A review of the empirical literature. Journ al of Accounting & Economics 58 (2 3): 339. policies and their informativeness? Journal of Accounting and Economics 56 (2 3, Supplement 1): 42 65. Berger, P., M. Mi nnis, and A. Sutherland. 2017. Commercial Lending Concentration and Bank Expertise: Evidence from Borrower Financial Statements Chicago Booth Research Working Paper 15 - 42. Bonner, S. E. 1990. Experience Effects in Auditing: The Role of Task - Specific Knowledge. The Accounting Review 65 (1): 72 92. Bonner, S. E., and B. L. Lewis. 1990. Determinants of Auditor Expertise. Journal of Accounting Research 28: 1 20. Brown, C. 2013 . North Dakota Went Boom. The New York Times , January 31. Brown, J. P., T. Fitzgerald, and J. G. Weber. 2016. Capturing rents from natural resource abundance: Private royalties from US onshore oil & gas production. Resource and Energy Economics 46: 23 38. Bushman, R. M., and C. D. Williams. 2012. Accounting discretion, loan loss provisioning, and - taking. Journal of Accounting and Economics 54 (1): 1 18. Choi, J. - H., J. - B. Kim, X. Liu, and D. A. Simunic. 2008. Audit Pricing, Lega l Liability Regimes, and Big 4 Premiums: Theory and Cross - country Evidence*. Contemporary Accounting Research 25 (1): 55 99. Christensen, B., N. Lundstrom, and N. Newton. 2018. Does the Disclosure of PCAOB Inspection ion Exposure? Working Paper. University of Missouri. Covert, T. R. 2014. Experiential and Social Learning in Firms: The Case of Hydraulic Fracturing in the Bakken Shale. 107 Davidson, P. 1963. Public Policy Problems of the Domestic Crude Oil Industry. The American Economic Review 53 (1): 85 108. Dawes, R. M. 1979. The robust beauty of improper linear models in decision making. American Psychologist 34 (7): 571 582. DeFond, M., and J. Zhang. 2014. A review of archival auditing research. Journal of Accounti ng and Economics 58 (2 3): 275 326. from the Subprime Mortgage Market. Journal of Money, Credit and Banking 44 (2 3): 367 384. 2006. Lending Booms and Lending Standards. The Journal of Finance 61 (5): 2511 2546. Doyle, J., W. Ge, and S. McVay. 2007. Determinants of weaknesses in internal control over financial reporting. Journal of Accounting and Economics 44 (1 2): 193 223. Einhorn, H. J. 1974. Expert judgment: Some necessary conditions and an example. Journal of Applied Psychology 59 (5): 562 571. Erickson, M., B. W. Mayhew, and W. L. Felix. 2000. Why Do Audits Fail? Evidence from Lincoln Savings and Loan. Journal of Accoun ting Research 38 (1): 165 194. Feyrer, J., E. T. Mansur, and B. Sacerdote. 2017. Geographic Dispersion of Economic Shocks: Evidence from the Fracking Revolution. American Economic Review 107 (4): 1313 1334. Financial Accounting Standards Board (FASB). 201 0. Conceputal Framework Satement of Accounting Concepts No. 8. Fitzgerald, T. 2014. Importance of Mineral Rights and Royalty Interests for Rural Residents and Landowners. Choices 29 (4): 1 7. . 2015. Experiential gains with a new technology: An empiri cal investigation of hydraulic fracturing. Agricultural and Resource Economics Review 44 (2): 83 105. Francis, J. R. 2011. A Framework for Understanding and Researching Audit Quality. Auditing 30 (2): 125 152. Galbraith, K. 2012. West Texas Oil Boom Crea tes Housing Shortage and Other Issues. The New York Times , July 12, sec. U.S. Gaynor, L. M., A. S. Kelton, M. Mercer, and T. L. Yohn. 2016. Understanding the Relation between Financial Reporting Quality and Audit Quality. Auditing: A Journal of Practice & Theory 35 (4): 1 22. 108 Gilje, Erik P., E. Loutskina, and P. E. Strahan. 2016. Exporting Liquidity: Branch Banking and Financial Integration. The Journal of Finance 71 (3): 1159 1184. Gilje, E, R. Ready, and N. Roussanov. 2016. Fracking, Drilling, and Asse t Pricing: Estimating the Economic Benefits of the Shale Revolution . Working Paper. National Bureau of Economic Research. Glaeser, S., and W. Guay. 2017. Identification and generalizability in accounting research: A discussion of Christensen, Floyd, Liu, and Maffett (2017). Journal of Accounting and Economics 64: 305 312. Government Accountability Office (GAO). 1994. Depository Institutions: Divergent Loan Loss Methods Undermine Usefulness of Financial Reports . Washington, DC: GAO. Griffith, E. E., J. S. Hammersley, K. Kadous, and D. Young. 2015. Auditor Mindsets and Audits of Complex Estimates. Journal of Accounting Research 53 (1): 49 77. Gul, F. A., S. Y. K. Fung, and B. Jaggi. 2009. Earnings quality: Some evidence on the role of auditor tenure and au Journal of Accounting and Economics 47 (3): 265 287. Haggerty, J., P. H. Gude, M. Delorey, and R. Rasker. 2014. Long - term effects of income specialization in oil and gas extraction: The U.S. West, 1980 2011. Energy Economics 45 (Supplement C): 186 195. Hall, W., and A. Renner. 1988. Lessons that auditors ignore at their own risk. Journal of Accountancy 166: 50 58. Hamilton, R. E., and W. F. Wright. 1982. Internal Control Judgments and Effects of Experience: Replications and Extensions. Journal of Accounting Research 20 (2): 756 765. Hammersley, J. S. 2006. Pattern Identification and Industry - Specialist Auditors. The Accounting Review 81 (2): 309 336. Healy, J. 2013. As Oil Floods Plains Towns, Crime Pours In. The New York Times , November 30, sec. U.S. Hefner, R. A. I. 2014. The United States of Gas: Why the Shale Revolution Could Have Happened Only in America Shale and the Future of Energy. Foreign Affairs 93: 9 14. Hoffman, V. B., J. R. Joe, and D. V. Moser. 2003. The ef fect of constrained processing on Accounting, Organizations and Society 28 (7): 699 714. Iannotta, G., and S. Kwan. 2014. The Impact of Reserves Practices on Bank Opacity Working Paper 2013 - 35, The Federal Reserve Bank of San Francisc o. 109 Jacobsen, G. D. 2017. Who Wins in an Energy Boom? Evidence from Wage Rates and Housing. Upjohn Institute Working Paper 17 - 217. Jacobsen, G. D., and D. P. Parker. 2016. The Economic Aftermath of Resource Booms: Evidence from Boomtowns in the American W est. Economic Journal 126 (593): 1092 1128. the Persuasiveness of Information Provided by Management. Auditing: A Journal of Practice & Theory 27 (1): 67 83. Kne chel, W. R., G. V. Krishnan, M. Pevzner, L. B. Shefchik, and U. K. Velury. 2013. Audit Quality: Insights from the Academic Literature. Auditing: A Journal of Practice & Theory 32 (1): 385 421. Laux, C., and C. Leuz. 2009. The crisis of fair - value accounti ng: Making sense of the recent debate. Accounting, Organizations and Society 34 (6): 826 834. Lazear, E. P., and S. Rosen. 1981. Rank - Order Tournaments as Optimum Labor Contracts. Journal of Political Economy 89 (5): 841 864. Libby, R., and D. M. Frederi ck. 1990. Experience and the Ability to Explain Audit Findings. Journal of Accounting Research 28 (2): 348 367. Lim, C. - Y., and H. - T. Tan. 2008. Non - Audit Service Fees and Audit Quality: The Impact of Auditor Specialization. Journal of Accounting Research 46 (1): 199 246. Lisowsky, P., M. Minnis, and A. Sutherland. 2017. Economic Growth and Financial Statement Verification. Journal of Accounting Research 55 (4): 745 794. Liu, C. - C., and S. G. Ryan. 2006. Income Smoothing over the Business Cycle: Changes in - Offs from the Pre - 1990 Bust to the 1990s Boom. The Accounting Review 81 (2): 421 441. Lo, A. 2015. Accounting Credibility and Liquidity Constraints: Evidence from Reactions of Small Banks to Monetary Tightening. The Accounting Review 90 (3): 1079 1113. Maksymov, E. M., J. S. Pickerd, D. J. Lowe, M. E. Peecher, and A. Reffett. 2018. The Settlement Norm (and Exceptions) in Audit Legal Disputes: Insights from Prominent Attorneys. Working Paper. McNichols, M. 2000. Research design issues in earnings management studies. Journal of Accounting and Public Policy 19. 313 - 345. Moretti, E. 2013. Where the Good Jobs Are and Why. Wall Street Journal , September 17, sec. Opinion. 110 Moroney, R . 2007. Does Industry Expertise Improve the Efficiency of Audit Judgment? AUDITING: A Journal of Practice & Theory 26 (2): 69 94. Muehlenbachs, L., E. Spiller, and C. Timmins. 2015. The Housing Market Impacts of Shale Gas Development. American Economic Review 105 (12): 3633 3659. Nalebuff, B. J., and J. E. Stiglitz. 1983. Information, Competition, and Markets. The American Economic Review 73 (2): 278 283. Nicoletti, A. 2018. The Effects of Bank Regulators and External Auditors on Loan Loss Provisions. Journal of Accounting and Economics 66 (1): 244 265. Office of the Comptroller of the Currency (OCC). 2012. Allowance for Loan and Lease Losses. In . Washington, DC. Olson, B. 2017. Big Oil Firms Are Exploring a New Frontier in Shal e: Profits. Wall Street Journal , June 16, sec. Business. Auditing: A Journal of Practice & Theory 23 (1): 89. Petroni, K., and M. Beasley. 1996. Errors in Accounting Estimates and Their Relation to Audit Firm Type. Journal of Accounting Research 34 (1): 151 171. Plosser, M. C. 2013. shocks . Staff Report. Federal Reserve Bank of New York. Public Company Accounting Oversight Board (PCAOB). AS 2110: Identifying and Assessing Risks of Material Misstatement . Auditing Standards. Washington, DC: PCAOB. . AS 2501: Auditing Accounting Estimates. Washi ngton, DC: PCAOB. Reichelt, K. J., and D. Wang. 2010. National and Office - Specific Measures of Auditor Industry Expertise and Effects on Audit Quality. Journal of Accounting Research 48 (3): 647 686. Reinhart, C., and K. Rogoff. 2008. Is the 2007 U.S. Su b - Prime Financial Crisis So Different? An International Historical Comparision? American Economic Review 98 (2): 339 344. Roberts, M., and T. Whited. 2013. Endogeneity in Empirical Corporate Finance. In Handbook of Economics of Finance , 2A:493 572. Rucke s, M. 2004. Bank Competition and Credit Standards. The Review of Financial Studies 17 (4): 1073 1102. 111 Schmoker, J., and T. Klett. 2013. U.S. Geological Survey Assessment Concepts for Conventional Petroleum . DDS - 69 - D. Denver, CO: U.S. Department of the Int erior. Securities and Exchange Commission (SEC). 2001. SEC Staff Accounting Bulletin No. 102: Selected Loan Loss Allowance Methodology and Documentation Issues . Seetharaman, A., F. A. Gul, and S. G. Lynn. 2002. Litigation risk and audit fees: evidence fr om UK firms cross - listed on US markets. Journal of Accounting and Economics 33 (1): 91 115. Shepardson, M. L. 2018. Effects of individual task - specific experience in audit committee oversight of financial reporting outcomes. Accounting, Organizations and Society . Simon, H. 1960. The new science of management decision . 1st ed. Vol. 3. The Ford distinguished lectures. New York: Harper. process. Journal of Applied Psychology 53 (4): 255 263. Smith, C. W., and R. L. Watts. 1992. The investment opportunity set and corporate financing, dividend, and compensation policies. Journal of Financial Economics 32 (3): 263 292. and confidence judgements. Contemporary Accounting Research 17 (4): 693 712. Tversky, A., and D. Kahneman. 1974. Judgement under Uncertainty: Heuristics and Biases. Science 185 (4157): 1124 1131. Urbina, I., and J. C. McGinty. 2011. Drilling Down: Fighting Over Oil and Gas Well Leases. The New York Times , December 1, sec. U.S. U.S. Department of the Treasury. 2008. Advisory Committee on the Auditing Professi on . Wright, W. F. 2001. Task experience as a predictor of superior loan loss judgments. Auditing: A Journal of Practice & Theory 20 (1): 147 155.