THREE ESSAYS IN MICHIGAN PROPERTY ASSESSMENT By Yunni Deng A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Economics - Doctor of Philosophy 2021 ABSTRACT THREE ESSAYS IN MICHIGAN PROPERTY ASSESSMENT By Yunni Deng This dissertation consists of three essays examining and analyzing Michigan property assessment practice s . The first essay estimates the property as sessment expenditure function for Michigan local government (townships and cities) using panel data with 1,310 governmental units for years 2015 through 2017. Results show that assessment expenditure is related to district size and government type, but not related to assessor type or the required assessors certification level. The second essay measures and compares the performance of different types of Michigan property assessors. By analyzing over five million properties in Michigan from 2008 through 2016, I find that neither private nor public property assessors are always superior to the other . B oth approaches have relative strengths depending on community characteristics. My analysis and subsequent comparison between the existing and a more optimal arrangement of assessor types suggests a potential misallocation in Michigan. Scenario forecasts suggest that merging small assessment jurisdiction s coupled with moving toward a more optimal arrangement would improve assessment performance by as much as 32 percent. The third essay uses detailed parcel level housing price and property tax data during and following the financial crisis for five counties that comprise the Detroit Metropolitan Area in Michigan to evaluate the relationship between changing housi ng prices and property assessments. My findings are generally consistent with previous research in that there is about a three - year lag between the housing price changes and changes in property assessments for tax purposes. However, I also find that before the crisis - hit struggling communities tended to over - state assessments to support property tax revenue streams. Assessment adjustments to market declines in these places also lagged in other fiscally healthier cities. On a positive note, the evaluation sh ows that local authorities in these struggling communities used the financial crisis period to align assessments more closely to actual market conditions as per Michigan property assessment policies. iv ACKNOWLEDGEMENTS I would like to extend my sincere and deepest gratitude toward my major advisor, mentor and philosopher , Dr. Mark Skidmore. Without his active guidance, encouragement, and ment al support , I would not have made headway in this endeavor. I am ineffably indebted to the co - chair of my committee, Dr. Leslie Papke, and my committee members, Dr. Ronald Fisher , and Dr. Peter Schmidt , whose efforts and suggestions have greatly improved the quality of this research. I am grateful to the Lincoln Institute of Land Policy for sponsoring the data used in my dissertation . I would also like to thank Robert Strauss, and David Rowley for all the valuable discussions and suggestions . T o all the friends and colleagues in the Department of Economics, Michigan State University, and fell ow economists in Missouri Valley Economic Association, and Midwest Economics Association, I thank you for listening to my ideas, your thoughtful insights, and precious feedback, and genuine camaraderie. I would like to thank my husband , Kui , my parents , Shuming and Siju, who have always supported me morally and financially. I would also like to thank my children, Jade and Declan. Your arrival has made my days extremely colorful. I love you with all my heart. v TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ........... viii LIST OF FIGURES ................................ ................................ ................................ ........... ix CHAPTER 1 : Introduction ................................ ................................ ................................ .. 1 CHAPTER 2 : Determinants of Property Assessment Expenditures in the State of Michigan ................................ ................................ ................................ ............................. 4 2.1 Background ................................ ................................ ................................ ............... 4 2.1.1 Property tax in a glance ................................ ................................ ...................... 4 2.1.2 Property Assessment Process in Michigan ................................ ......................... 6 2.1.3 Local Assessors Variation ................................ ................................ .................. 7 2.2 Literature Review ................................ ................................ ................................ ...... 9 2.3 Model Specification ................................ ................................ ................................ 11 2.4 Data ................................ ................................ ................................ ......................... 17 2.5 Results and Discussion ................................ ................................ ............................ 20 2.6 Conclusion ................................ ................................ ................................ ............... 24 APPENDICES ................................ ................................ ................................ ............... 26 APPENDIX A : Evolvement of Michigan Property Tax Statutes .............................. 27 APPENDIX B : Three - Step Michigan Property Assessment Process ........................ 30 APPENDIX C : Assessing Eligibilities and Training Requirements of Assessors with Different Certifications ................................ ................................ .............................. 34 APPENDIX D : Functions of Local Assessing Offices ................................ ............. 37 APPENDIX E : Tables ................................ ................................ ............................... 39 CHAPTER 3 : An Analysis of Public vs. Private Contracting Performance in Michigan Property Tax Assessment ................................ ................................ ................................ .. 44 3.1 Introduction ................................ ................................ ................................ ............. 44 3.2 Background ................................ ................................ ................................ ............. 47 3.2.1 Michigan Property Assessment Process ................................ ........................... 47 3.2.2 Michigan Property Assessment Performance Overview ................................ .. 48 3.2.3 Michigan Assessors Overview ................................ ................................ ......... 49 3.3 Literature Review ................................ ................................ ................................ .... 51 vi 3.4 Model Specification ................................ ................................ ................................ 54 3.5 Data ................................ ................................ ................................ ......................... 60 3.6 Distribution of Private vs. In - house vs. County Assessors ................................ ..... 63 3.7 Results and Discussion ................................ ................................ ............................ 66 3.7.1 Overall Analysis ................................ ................................ ............................... 66 3.7.2 An A nalysis for Assessor Types ................................ ................................ ....... 68 3.7.3 Scenario Forecasts ................................ ................................ ............................ 71 3.8 Conclusion ................................ ................................ ................................ ............... 72 APPENDICES ................................ ................................ ................................ ............... 75 APPENDIX A : Tables ................................ ................................ ............................... 76 APPENDIX B : Figures ................................ ................................ .............................. 82 CHAPTER 4 : A Dynamic Analysis of the Impacts of Changes in Housing Values on Property Tax Assessment During and After the 2008 Financial Crisis ............................ 87 4.1 Introduction ................................ ................................ ................................ ............. 87 4.2 Institutional Details ................................ ................................ ................................ . 90 4.2.1 Michigan Property Assessment and Tax Collection Process ........................... 90 4.2.2 Michigan Property Tax and Assessment Cap ................................ ................... 91 4.3 Indices Construction ................................ ................................ ................................ 93 4.3.1 Hedonic Price Index ................................ ................................ ......................... 93 4.3.2 SEV index ................................ ................................ ................................ ......... 97 4.3.3 Median Sales Ratio ................................ ................................ ........................... 97 4.4 Data ................................ ................................ ................................ ......................... 98 4.4.1 Data used to calculate Hedonic HPI ................................ ................................ . 98 4.4.2 Data used to calculate SEV index ................................ ................................ ... 100 4.4.3 Data used to calculate Median Sales Ratio ................................ ..................... 101 4.5 Results and Discussion ................................ ................................ .......................... 102 4.5.1 Time Dummies Hedonic Regressions ................................ ............................ 102 4.5.2 Comparison of Indices ................................ ................................ .................... 103 4.5.3 Different behavior of over - estimating governments ................................ ...... 104 4.6 Conclusion ................................ ................................ ................................ ............. 107 APPENDIX ................................ ................................ ................................ ................. 109 CHAPTER 5 : Conclusion ................................ ................................ ............................... 122 vii REFERENCES ................................ ................................ ................................ ............... 126 viii LIST OF TABLES Table 2.1 Assessment Cycle and Other Requirements of States in U.S. .......................... 40 Table 2.2 Summary of Statistics ................................ ................................ ....................... 42 Ta ble 2.3 Coefficients of Potential Determinants of Assessment Expenditure ................ 43 Table 3.1 Summary of Statistics ................................ ................................ ....................... 77 Table 3.2 Factors of Assessor Performance ................................ ................................ ...... 78 Table 3.3 Coefficients of Determinants of Assessment Performance .............................. 79 Table 3.4 Coefficients of Performance Factors for Different Types of Assessors ........... 80 Table 3.5 Scenario Analysis ................................ ................................ ............................. 81 Table 4.1 Results of Time Dummies Hedonic Regressions ................................ ........... 110 Table 4.2 Comparison between HPI and SEV index (base year = 2008) ....................... 118 ix LIST OF FIGURES Figure 3.1 Number of States by Assessment Level ................................ .......................... 83 Figure 3.2 Number of States by Assessment Frequency ................................ .................. 83 Figure 3.3 Frequency Distribution of Michigan Median Sales Ratio, 2016 ..................... 84 Figure 3.4 Frequency Distr ibution of Michigan COD, 2016 ................................ ............ 84 Figure 3.5 Michigan Assessment COD and Average Median Sales Ratio, 2008 - 2016 ... 85 Figure 3.6 Michigan Assessment Performance, 2008 - 2016 ................................ ............. 85 Figure 3.7 Michigan Assessment Expenditure , 2008 - 2016 ................................ .............. 86 Figure 4.1 Indices Comparison of Representative Counties and Cities .......................... 104 Figure 4.2 Comparison of Overestimating and Other Cities ................................ .......... 106 1 CHAPTER 1 : Introduction My objective in writing this dissertation is to evaluate several aspects of property assessment in Michigan , including the determinants of local government expenditures on property assessment, measuring assessment quality, the determinants of quality , and how assessments respond to housing price fluctuations . My dissertation consists of three essays. The first essay (Chapter 2) focuses on the expenditure side of Michigan property assessment. T his chapter estimate s the expenditure function of property assessment for Michigan local governme nt s (town ships and cities) using panel data with 1,310 government units for years 2015 through 2017 . Results show that assessment expenditures are related to district size and government type, but un related to assessor type or assessor certification level. For a lo cal unit with 1700 parcels ( an average size local unit in my dataset), a one percent increase in the number of parcel s reduce s the per - parcel assessment expenditure by 0.5 3 percent. A one percent increase in average assessed value is associated with a 0. 54 percent decrease of assessment expenditure per parcel. Interestingly, the hiring of private or county - level assessors as alternatives to in - house local government assessors does not reduce assessment expenditure . Further more , r equiring a higher level of a ssessor certifi cation does not significantly increase assessment expenditure . The second essay (Chapter 3) focuses on the performance side of Michigan property assessment. I measure the assessment performance of each Michigan local 2 government using panel data that includes most residential parcels in Michigan (over 5 million parcels) from 2008 through 2016. Follow ing the discussion on the impact of assessor type on the assessment expenditure in the first essay, this essay analyzes the impact o f assessor type on assessment performance. I attempt to identify the optimal allocation of the three types of Michigan assessors - assessors, and county assessors . My goal is to determine who performs better in what communities and then compare the existing allocation of assessor types to a more optimal arrangement to determine the degree of potential assessor misallocation. My analysis suggests that : (i) None of the three kinds of assessors is always superior or inferior to the other s , and they all have relative strengths within different types of c ommunities; and (ii ) there is a potential misallocation of assessors in Michigan. The analysis suggests that p rivate contractors perform better in communities with more parcels but ar e often employed in communities with fewer parcels. The third essay (Chapter 4) evaluates the relationship between changing housing prices and property assessments. Using housing transaction and property tax data for 120 Michigan assessing units in the fi ve - county region surrounding Detroit (Macomb, Monroe, Oakland, Washtenaw, and Wayne counties) from 2008 t hrough 2016, I find a three - year lag between housing price changes and changes in property assessments for tax purposes. However, I also find that before the 2008 crisis , struggling communities tended to over - state assessments with the likely objective of suppor ting property tax revenue streams. Assessment adjustments to market declines in these places also lagged in other fiscally healthier cities. On a positive note, the evaluation shows that local authorities in these struggling communities used the financial crisis period to align 3 assessments more closely to actual market conditions as per Michigan assessment policies . Chapters 2 through 4 include the analyses summarized above, respectively, and Chapter 5 concludes with a summary of my main findings. 4 CHAPT ER 2 : Determinants of Property Assessment E xpenditure s in the State of Michigan 2.1 Background Attempting to minimize public expenditure while maintaining quality public service s has been a significant issue in the U.S. and worldwide. As the primary revenue source for local government s in the U.S. , the property tax requires significant expenditures for managing and collecting revenue . This research seeks to identify, measure, and rate the determinants of expenditure s related to property tax administra tion . M ore specifically, I evaluate the determinants of property assessment and equalization expenditure s using panel data of more than 1000 local governments (townships and cities) in Michigan for the years 2015 t hrough 2017. 2.1.1 Property tax in a glance Property tax is applied to homes, land, farms , and other forms of real estate. In the U.S., the property tax is a state and local tax imposed by different levels of subnational government and is a principal source of local government revenue. In 201 5, state and local property tax revenue accounted for 31.1 percent of all state and local government general tax revenue, ranging from 13.3 percent in North Dakota to 65.7 percent in Massachusetts. 1 In Michigan, property tax revenue accounts for 34.5 perce nt of total state and local general tax revenue, which shows a slightly higher dependency on property tax than 1 State & Local Government Finance Data Query System. http://www.taxpolicycenter.org/statistics/property - taxes - percentage - state - and - local - taxes . The Ur ban Institute - Brookings Institution Tax Policy Center. Data from U.S. Census Bureau, Annual Survey of State and Local Government Finances, Volume 4, and Census of Governments (1977 - 2015). 2015 data. 5 the national average of 31.1 percent. 2 Michigan administers the property tax at the township, city, and village level s . Each township, city , or v illage decides its own tax rate for different property classes , and the state average rate for all properties is 41.7 mils for 2016. 3 The assessed value is equal to 50 percent of fair market value in the State of Michigan. The bas is of t he property tax in the U.S. is the appraisal of property value. Property values are usually re - assessed periodically to 1) fairly and correctly reflect depreciations or improvements for each property; 2) incorporate inflation into property valuation , and 3) correct inaccura te evaluations in previous assessments. State laws usually regulate reassessment cycles and assessment forms (whether a physical inspection is required). See Table 2. 1 in Appendix E for a detailed assessment cycle and form by state. 4 The authority of the M ichigan government to levy property taxes is contained in Article 9 of the Constitution of the State of Michigan. In 1994, Proposal A amended the Constitution to restrain the increasing reliance on property taxes. Growth on the taxable value of each parcel of property is capped to the rate of inflation or five percent, whichever is lesser. Please see Appendix A for a detailed evolvement of Michigan property tax statutes. 2 As above. 3 2016 Ad Valorem Property Tax Report. Doc No . 625 (Rev. 02 - 17). County governments and school districts also apply a tax rate. 4 Significant Features of the Property Tax. http://datatoolkits.lincolninst.edu/subcenters/significant - features - property - tax/state - by - state - property - tax - in - detail. Lincoln Institute of Land Policy and George Washington Institute of Public Policy. (State - by - State Property Tax in Detail; accessed: 6/18/2018 1:59:52 PM). 6 2.1.2 Property Assessment Process in Michigan In Michigan, property assessment is carried out annually by township s , cit ies , and village s , with county equalization and state government oversight. The process of Michigan property assessment includes three steps: 1) The local assessor (township assessor or city assessor) determines the assessed value of each property by determining the market price of the property, and produces the taxable value (TV); 2) The board of commissioners in each county equalization department equalizes, or applies an adjustment factor, to ensure that property owners in all cities, townships, villages, or Commission applies an adjustment factor to the county assessments to bring the total v aluation across counties as close to the 50 percent level as possible , and produces the s on the three - step property assessment process are provided in Appendix B . In - house assessors traditionally do the f irst step . However, in order to cut assessing department staff, many local units contract for this service with a for - profit company or a n individual contact or . Many individual private assessors contract their services either in whole or in part to multiple local units. 5 During this step, county equalization usually stays out of the local assessing process. However, townships and cities could request county equalization to assess every property when they do not have an assessor with the proper certif ication. The data show that Wayne, Oakland, and Macomb counties (Metro Detroit) regularly hire contract ors to provide assess ment services . 5 D. Rowley, MAAO, Michigan Assessor Association Past President, personal communication, May 1, 2018 7 Michigan is among the 19 states in the U.S. which require annual reassessment (see Table 2. 1 in Appendix E for the r equired assessment cycle of each state) . Michigan guidelines require physical inspection of a minimum of 20 percent of property parcels in any given assessment year, which requires extra time and effort (see Appendix B for more details). Therefore, analyzi ng and controlling administrative expenditure of assessment is a more pressing issue in Michigan than in the states requiring less frequent assessments. 2. 1.3 Local Assessors Variation Local Assessors in Michigan (township assessors and city assessors) vary in two main aspects: certification level and hiring type. There are four different levels of certification for property assessors: 1) Michigan Certified Assessing Technician (MCAT); 2) Michigan Certified Assessing Officer (MCAO); 3) Michigan Advanced Assessing Officer (MAAO); 4) Michigan Master Assessing Officer (MMAO). M ore training and experience are required to obtain a higher level of certification, which grants certification holders greater eligibility and authority on assessments. Only level two to level four certification holders are granted eligibility to be an assessor. Details on assessing eligibilit y and training requirements of different levels of certification are provided in Appendix C . The State Tax Commission is responsible for the educa tion and certification of assessing officers and publishes an annual guideline for required certification level for each township, village, city, and county based on their State Equalized Value (SEV) from the previous year. 8 Based on hiring type, Michigan local assessors can be classified into three categories: in - house assessors, private contractor assessors, and county contractor assessors. Traditionally, local government units (townships and cities) tend to hire their own assessors as government employee s , with a preference to have an assessor who is a part of the community. However, with cuts to assessing department staff, local government s have increasingly contract ed this service to for - profit compan ies or with a cont r act assessor. Currently, most townships and cities in Michigan hire contract assessors, and some may share the same contractor. C ontract assessors charge local government s by the hours they work and /or per parcel. 6 This is a way for townships and cities to share assessment costs . For local units having contract assessors, contract or costs ac count for only a part of assessment office expenditures. 7 Note that a number of townships and cities still have in - house assessors , though it is less common . A ssessors typically charge an annual salary , but the assessing office also covers other related expenditure s for assessments and related functions of the local assessment office. Suppose a local assessing district does not have an assessor with an adequate certification level and has not employed a certified assessor. In that case , the local government shall request the county equalization department to perform assessments . T he expenditure of preparing the rolls shall be charged to the local assessing distri ct. There are criticisms of county equalization performing local assessing duties because both assessment and 6 D. Ro wley, MAAO, Michigan Assessor Association Past President, personal communication, May 1, 2018 7 One of the local units of government interviewed pays their assessor around $75,000 a year. The entire department budget is approximately $195,000. The differen ce would then be for other staff and other expenses to run the department (retirement & health benefits, supplies, transportation, conferences & training, Michigan Tax Tribunal hearings, and any other miscellaneous). 9 appraisal will be conducted by the same assessors/appraisers . A conflict of interest may a rise , especially when one reviews his/her own assessment s. 8 By state statute, each township or city, regardless of whether it is using a private contract assessor, county assessor, or in - house assessor, must maintain an assessing office. Please see more details of local assessing office functions in Appendix D . 2.2 Literature Review There are limited systematic analys e s in the literature on minimizing the property assessment expenditure while maintaining the quality of assessment and collection . However, a number of studies attempt to estimate the effects of s everal economic and governmental factors on public good and service expenditure s . For example, the studies of economies scale in the provision of government goods and services date back to the early 1960s. Hirsch (1965) conducted the first studies investig ating the effect of various factors on the expenditure of municipal refuse collection services. Ahlbrandt (1973) extended the stud y of economies of scale to fire services provision, Deller et al. (1988) conducted similar research on the rural low - volume ro ad s , and Walzer (1972) studied police services. The studies of economies of scale w ere extended to property assessment in the 1990s . Sjoquist and Walker (1999) estimated property tax assessment expenditures using a translog expenditure function over a samp le of 138 county - level assessment offices in Georgia . They found that there are substantial economies of scale. They used both the number of parcels and the value of parcels as alternate measures of volume, which yielded similar results , concluding that pr operty assessment exhibits increasing return of scales. The estimated elasticity of expenditure to volume is 0.323 if based on the 8 D. Rowley, MAAO, Michigan Assessor A ssociation Past President, personal communication, May 1, 2018 10 number of parcels and 0.243 if based on property value . They estimated that consolidating all assessment offices with outputs below the median volume would reduce the total property assessment expenditure for these districts by over 20 percent. Their research, though enlightening, was limited by two factors: First, the available data set is small; there is complete information on only 138 assessment offices for a single year. Second, the research does not consider a possible quadratic relationship between assessment expenditure and size. Another informative section of literature focuses on the privatization of public goods and services. Over 100 independent studies compare the expenditure s of in - house government agencies versus the private sector for providing public goods and services. Th e typical findings range from expenditure reductions of 20 percent to 50 percent resulting from privatization. Deacon (1979) compared in - house versus intergovernmental production of all services and found intergovernmental contracting reduced costs by 14 p ercent. David (1987) surveyed local administrators to obtain compari sons of publicly provided versus and private ly contracted services , report ing expenditure savings in 98 percent of contracting efforts, with a weighted average expenditure saving of 19 percent. Ahlbrandt (1973, 1974) and Hike (1986) obtained similar results on expenditure reductio n s from privatization for the provision of fire protection . Schlesing, Dorwart, and Pulice (1986) also found savings from privatizing health servi ces; Deacon (1979) and Steven (1984) also found savings from private contracting for highway s . Turn ing more specifically to property assessment, Stocker (1973) analyzed in - house property tax assessors and private contractors in Ohio , concluding that priva te assessments were more accurate and 50 percent less expensive . There are arguments 11 against the privatization of certain public goods and services concern ing the quality of services. However, other empirical research on property assessment ha s found no such impact. For example, Mikesell (1987) analyzed the data from the state of Indiana , f inding no support for the argument that private provision produce s a technically inferior job relative to publicly provided assess ment . Finally, Stocker (1973 ) found that private contractors provided more accurate assessment results relative to in - house assessors. Our research will contribute to the current literature in four ways: First, it provides a more comprehensive analysis o f how local governments can minimize property assessment expenditure s . It attempts to build a system of best practices for property assessment by analyzing the effects of assessment size, assessor incentives, and other factors not thorough ly evaluated in literature such as asses sor training levels and government type. Second, this research utilizes an intensive data set panel data that includes 1310 local government (townships and cities) assessment offices in the State of Michigan over three years . Third, this research also co nsiders the possible quadratic relationship between size and expenditure , which adds value to the current literature on economies of scale in property assessment. Fourth, the Michigan property assessment data allows a three - way compar ison of in - house assessors , private contractors assessors from county equalization departments. In summary, my evaluations add value to the current literature on the privatization of property assessment. 2.3 Model Specification This research seeks to determine th e pattern of property assessment expenditure s in the State of Michigan, and the relationship between property assessment expenditure s 12 and various potential factors. P roperty assessment expenditure per parcel is modeled as being affected by six local govern ment characteristics. The first characteristic this research considers is the total number of property parcel s in each jurisdiction as an indicator of the assessing unit size. I t est the hypothesis that assessment expenditure per parcel is positively or negatively related to the number of property parcel s . The result will help identify whether property assessment exhibits decreasing, constant, or increasing return to scale. Also, this research tests the potential nonlinear ( qu adratic ) relationship between assessing unit size and assessment expenditure per parcel. In the case of Michigan property assessment, this analysis helps answer practical questions: Will consolidation of assessing offices reduce per parcel assessment expen diture? How big of an effect will such consolidation have on reducing property assessment expenditures? My assessment provide s answers that have important policy implications for the consolidation of assessing offices. Oftentimes, m ul ti ple local government s (townships and cities) hir e the same private contractor or county equalization director for assessment service . It might make th is analysis less intuitive because the assessor seems to be conducting the assessment o n a larger scale as he /she secures multiple contracts from several assessing offices. However, it is important to note that private contractor s do not determine the contract price unilaterally. Instead, the pricing of a contract is a negotiation process, and a contractor might be paid with a different rate per parcel or per hour by different townships and cities. In this sense, the research also helps t es t whether larger local government units have a larger or smaller power in negotiati ng with private contractors. 13 This d ifferenc e in negotiation power can also be seen as economies of scale of assessment to assessing offices . Some might argue that it is more appropriate to see assessors as units of assessing activities when testing the economies of scale. However, note that assessing offices, not assessors, are the organizers of property assessment activities . The reasons are: 1) assessing offices have complete discretion in choosing private contractors who propose different services and prices; and 2) assessor salaries or pr ivate contracts only count a p ortion of total assessment expenditure s . A significant part of expenditure s goes to staff salaries, benefits, tribunal expenditure , and so on. Hence, assessing offices are the most appropriate units of assessment for testing t he economies of scale. Furthermore, using assessing offices as units of assessment makes this research consistent with Sjoquist and Walker (1999), who used assessing offices of each county in Georgia to test the economies of scale of property assessment. T he research then looks at average State Equalized Value (SEV) as the second potential factor affecting assessment expenditure per parcel. SEV is assessed property value based on 50 percent of true cash value, which is the same as the market value, i.e. , wh at the property worth on the open market. Sjoquist and Walker (1999) consider the value of parcels as an alternative measure of assessing size in their analysis of economies of scale of property assessment. However, such analysis is less relevant in Michig an because private contractors in Michigan typically charge local governments based on parcel numbers and hours worked in offices, not parcel value. 9 Yet, average SEV is still expected to be an important factor in determining assessment expenditure because of the 9 D. Rowley, MAAO, Michigan Assessor Association Past President, personal communication, May 1, 2018 14 potential larger tribunal and other legal expenditure s associated with higher valued properties. In addition , th e potential difference in legal expenditure might result in the assessor behaving differently when assessing more expensive properties . A n error made on expensive properties might bring either a larger los s to government revenue or a larger legal expenditure fighting assessment challenge s . Thus, it is worthwhile to test the relationship between property value and assessment expenditure, co ntrolling the total parcel number, to see whether more expensive properties require more care or incur larger expenditure s . The third potential factor for assessment expenditure per parcel tested is the assessor type : whether the assessor is an in - house assessor, private contractor, or county equalization director. As the previous literature on privatization indicated, it is reasonable to expect asses sors of different type s to perform differently . T he current trend of replacing in - house assessors with private contractors was originated from local expenditure s. Different local government units choose different methods during this privatization trend . S ome governments sign contra cts with private assessing firms or with individuals with differing assessing capabilities . S ome local governments request their county equalization departments to help with assessment , whereas o thers retain in - house assessors. This research analyzes whether the replacement of in - house assessors reduced the outlays . The fourth potential characteristic affecting assessment expenditure per parcel is the required assessor certification level, as a measurement of the quality of assessors. Assessors ' quality may affect assessment expenditure in different directions . On the one hand, assessors with better education and training charge higher salaries or contract 15 prices . O n the other hand, assessing staff with better education and stronger skills might assess more efficien tly and reduce expenditure s. Assuming all assessors attempt to minimize assessment expenditure s without sacrificing basic assessment quality requirements, analyzing the relationship between assessmen t expenditure per parcel and assessor certification level ha s critical policy implications . A negative or insignificant relationship implies that requiring a higher assessor certification level does not incur extra expenditure or even help reduce expenditu re , and thus is highly recommended . A positive relationship means that it should be given a second thought when requiring a local government to have higher - level certified assessors. The fifth potential factor considered in this evaluation is the government type: township or city. There might be a difference in expenditure s between townships and cities due to different government al structure s . This research tests whether township and city governments behave differently in property assessmen t administration. There might be valuable lessons to be drawn from the potential differences. The last characteristic potentially affects assessment per parcel is the property composition ( agriculture, residential , commercial, and industrial parcel value i n percent of total property value ). It is helpful to test the hypothesis that the assessors perform assessment s differently when facing different types of propert y that may require different skills and techniques. This analysis will help to explain the beh avior of assessors facing different property classifications . In this study, let the expenditure function for the i th local government in the t th year be represented by the following specification: 16 where is the log of per parcel assessment expenditure in ith township or city in the tth year; is the log of total parcel number in ith township or city in the tth year (the square term will be centered in the estimation); is the log of average SEV in ith township or city in the tth year; are the dummy variables representing assessor employment type in ith township or city; are the dummy variables representing required assessor certification l evel in ith township or city in the tth year; is the dummy variable indicating whether ith government is a township or a city; is the vector of different types of parcel s in percent of total parcel value in ith township or city in the tth year ; is the log of average SEV in ith township or city in the (t - 1)th year (lag ged SEV is added as a control because the required certification levels are decided based on prior represents a vector of time dummies added to control for the impact of inflation on assessment expenditure ; is the unobserved time - invariant individual effect; is the error term. Since and are in log terms, their coefficients will be elasticities. represent the elasticity of as sessment expenditure to parcel number together; is the elasticity of assessment expenditure to average SEV. The c oefficient of dummy variables will be the impact of the corresponding factors on expenditure per parcel in percentage terms. and will not be interpreted because lag SEV and year dummies are added as control variables in the model. 17 2.4 Data Michigan has 83 counties, within which there are 1607 townships and cities. This research utilizes a balanced panel data of 1310 local government units (including 1090 townships and 220 cities) from 2015 to 2017 to estimate the specified model. Summary statistics are provided in Table 2. 2 in Appendix E . expenditure in the Annual Financial Report (F - 65) for each township and city in the State of Michigan. The F - 65 form is submitted by primary government units (county, tow nship, city, village) every year within six months of the end of the fiscal year. In F - 65, the item the property description department, 3) the board of review, and 4) activ ity for the equalization department. Unfortunately, assess ment expenditure s for each detailed item are not available . Assessing equalization is reported with five expense types: 1) general funds, 2) all other government funds, 3) enterprise funds, 4) component units , and 5) total. For most government units, assessing equalization expenditure s is fully covered by general funds. The number used in this research is the total expenditure of assessing equalization (expense type five). Michigan Department o f Treasury provides parcel numbers annually for the state, each county, and all local government unit s (township and city) in the L - 4023 Forms. For each government unit, detailed parcel numbers for categorical properties are also reported: Agricultural R ea l p roperty, Commercial Real property, Industrial Real property, Residential Real property, Timber - Cutover Real property, Developmental Real property, Agricultural Personal p roperty, Commercial Personal property, Industrial Personal 18 property, Residential Pe rsonal property, Utility Personal property. From Table 2. 2, we see a significan t variance i n size across government units, ranging from 65 to 66403 in 2017 , for example. Over the years, the average parcel number slightly reduced from 3165 in 2015 to 3157 in 2017. By dividing total assessment expenditure s by total property parcel counts, I obtained assessment expenditure per parcel for each county and each loc al government unit (township and city) from tax year s 2015 t hrough 2017. As shown in Table 2. 2, expenditure per parcel ranges from $ 0.17 to $ 151 in 2017 , for example. Over the years, the average expenditure per parcel increased slightly from $ 15.91 in 2015 to $ 16.88 in 2017. By dividing the value of agriculture, residential, commercial, and industrial parcels by the total property parcel value , I obtained parcel composition for each county and each local government unit (township and city) f or tax year s 2015 t hrough 2017. The percentage of agriculture, residential, commercial, and industrial classifications in 2017 are 10.17, 70.05, 9.32, and 4.68 , respectively . The composition within each property classification in Michigan is stable over time , as shown in Table 2. 2, with similar means, min, and max over 2015 t hrough 2017. Michigan Department of Treasury also reports SEV (assessed property value based on 50 percent of market value) annually for the whole state, each county, and each local government unit (township and city) in the L - 4023 Forms. By dividing total SEV by the total numbers of parcel s , I obtained the average SEV for each county and each local government unit (township and city) from the tax year 2015 t hrough 2017. Average of average SEV rises from $69,519 in 2015 to $74,801 in 2017. 19 Based on Certification Level Guidelines approved by the State Tax Commission levels standard for counties and townships and cities based on their previous year SEV annually f or tax year s 2015 t hrough 2017. Three certification level dummies: MCAO, MAAO, and MMAO are created based on the standards. As shown in Table 2. 2, most (84 - 85 percent) of loc al governments are only required to have MCAO holders as assessors . O nly a small portion (3 - 4 percent) of local governments are required to have MMAO holders as assessors. For example, in 2017, 85.11 percent (or 1115 out of 1310) local governments are requ ired to have an assessor holding MCAO or higher - level certification . 11.60 percent (or 152 out of 1310) local governments are required to have an assessor holding MAAO or higher - level certification . 3.28 percent (or 43 out of 1310) local governments are re quired to have an assessor holding MMAO certification. Each Michigan county equalization department publishes the name of its equalization director, and each township/city publishes the name of the township/city assessor and the company he/she belongs to (if applicable). For those governments with no or underdeveloped official website s , I contacted their equalization departments via phone calls and emails to obtain the assessor name list. With the assessor name list, I created the assessor employment type dummy of each township and city using the following method: if a township or city assessor is the county director or specified as nt unit; if a township or city assessor is an employee of a private , professional property assessment company, I deem that this 20 is an individual whose name appears in other government unit(s), I deem that this - a county equalization director. Except in rare cases where individual private contractor assessors only have one contract from one local government unit, this method should provide a good approximate of assessor employment type for local government units. Using this approximation, the assessor employment type dum mies are time - invariant. As calculated with the dataset, 4.35 percent (or 57 out of 1310) townships and cities hav e county directors performing their property assessments . 73.13 percent (or 958 out of 1310) townships and cities hav e private contractors per forming property assessments . 22.52 percent (or 295 out of 1310) townships and cities hav e in - house assessors performing property assessments. There are 1090 townships and 220 cities in the dataset as stated above, and a dummy o indicate whether a local government unit is a city or not. This dummy variable is time - invariant because the government type does not change over time under normal circumstances. All of the data mentioned above were transformed into log terms except parc el classification , required certification level dummies, assessor employment type dummies, city dummy, and year dummies. Th us , most of the coefficients in the results represent elasticities. Detailed analysis will be given in the next section. 2.5 Results and Discussion The specification model is estimated using three methods: Random Effects (RE), Fixed Effects (FE), and correlated Random Effects (CRE) . The results are shown in Table 2. 3 in Appendix E . 21 In the CRE result column of Table 2. 3, we see the coeff icient s of time - average s of independent variables are mostly significantly different from zero. Also, an F - test yield s a P - value of 0.000, which indicates a correlation between the independent variables and the unobserved time - invariant individual effect. Thus, the r andom e ffects estimate is inappropriate. This is confirmed with an overall Hausman Test, which rejects the hypothesis that the difference in coefficients is not significant with a P - value of 0.000. Thus, I draw conclusions based on the CRE/FE re sults. First, the results indicate a quadratic relationship between assessment expenditure and the number of parcel s . Since the coefficient on the square term is negative , and the coefficient on the level term is positive, the quadratic has a parabolic sha pe. The turning point is when log of parcel count = 5.88 / ( 2*0.4 3) = 6.83, or parcel count = 925. The parabola slopes down after the turning point, which accounts for 84 percent of all data points. In other words, assessment expenditure per parcel is negatively related to property parcel counts in local government unit s for 84 percent of the data. Therefore, property assessment generally exhibits an increasing return of scale. The quadratic relationship implies that the parcel number ha s an increasing effect on assessment expenditure per parcel the elasticity of assessment expenditure to parcel number becomes higher when the parcel number becomes bigger. For example, for a local unit with 1700 parcels (or log of parcel number = 7.45, the local unit with medium size in the dataset), the elasticity is 5.88 - 2*0.4 3 *7. 45 = - 0.5 3 , which means each one percent increase in parcel number will reduce per parcel assessment expenditure by 0.5 3 percent . For a local unit with 1100 parcels (or log of pa rcel number = 7, the local unit at the first quartile of size in the dataset), the elasticity is 5.88 - 2*0.4 3 *7= - 0.1 4 , so each one 22 percent increase in parcel number will reduce per parcel assessment expenditure by 0.1 4 percent. F or a local unit with 3000 p arcels (or log of parcel number = 8, the local unit at the third quartile of size in the dataset), the elasticity is 5.88 - 2*0.4 3 *8= - 1, so each one percent increase in the number of parcel s will reduce per parcel assessment expenditure by 1 percent. This research is consistent with assessment exhibits an increasing return of scale. However, unlike Sjoquist and Walker, who found a constant elasticity of total expenditure to the number of parcel s (0.323), this research reveals an increasing effect of parcel number on assessment expenditure s as the parcel number s grow , and the elasticity is 0.5 3 for the local unit with the medium parcel number. Thus, assessment becomes more efficient when an asses sing office deal s with a larger number of p arcels , and this impact becomes more substantial when the parcel number becomes bigger. Therefore, combining two townships with similar parcel numbers could help reduce the assessment expenditure and is thus recom mend ed if the primary goal is to reduce costs . T he property tax assessment process in Michigan is highly decentralized . E ach township or city complet es its own assessment and maintain s an assessment office , which is a potential redundancy. Intuitively, merging assessing offices into adjacent offices could save a large amount of staff salary, contract expenditure , overhead expenditure , and other expenditures. This is an important argument for the consolidation of local assessment units the expenditure s could be reduced through consolidat ing assessing offices of townships or cities to form bigger assessing district s . 23 Second, this research found a statistically significant positive relationship between assessment expenditure per pa rcel and average SEV. The elasticity of assessment expenditure per parcel to average SEV is 0. 54 , which implies a 0. 54 percent increase in per parcel assessment expenditure associated with each one percent increase in average assess ed value. In addition, considering the value of parcels as an alternate measure of size as did Sjoquist and Walker (1999), a better possible explanation is a potential larger tribunal and other legal expenditure associated with more expensive properties. Or, assessors behave di fferently when facing expensive properties, as properties with higher values need to be handled with more care than other properties, for an error on their assessments could mean either a bigger los s on government revenue or a larger potential legal expend iture . Third, one of the most interesting findings is that assessment expenditure per parcel is not statistically significantly related to the assessor type. In other words, privatization through replacing in - house assessors with private contractors or cou nty equalization director fails to help local government units to reduce property assessment expenditure , although expenditure cutting is the exact intention of this privatization. This result could contradict our intuition and previous literature on priva tization. Some possible explanations of such phenomenon include, but are not limited to 1) although privatization of assessment is prevailing in Michigan, there are still a considerable number of assessment offices hiring in - house assessors, which makes th em benchmarks on how much an assessment office can/should spend for those who privatize; 2) some professional assessing firms might have a big negotiation power in the market and thus charge a high er price; 3) assessors salary and assessment contracts only expenditure a 24 part of the assessment total expenditure, and those who save by hiring a private contractor might be more generous on other aspects, such as administrative staff or regular office activities. Additional research will be needed to determine t he reason for the failure of Michigan privatization on property assessments. Fourth, the results show no statistically significant relationship between assessment expenditure per parcel and assessors certification level required by townships or cities, whi ch implies that requiring high assessor certification levels does not incur additional expenditure . This could be due to the higher efficiency of better - trained assessors, who may save more assessing expenditure , thus offse tting their higher salaries. Hence, requiring high assessor certification levels is recommended. Fifth, the results show a statistically significant positive relationship between assessment expenditure dummy is 0. 16 , which implies that city g overnments spend approximately 16 percent more per parcel than townships do. T ownship governments are much more expenditure efficient than city governments when it comes to property assessment. Finally, the compo sition of parcel classification matters . A ssessors seem to deal with different properties differently, and such difference s result in significant differences in assessment expenditure s . 2.6 Conclusion This research has some important findings, including the quadratic relationship between the number of parcel s and assessment expenditure , the positive relationship between average assessed value and assessment expenditure . For a local unit with 1700 parcel s (the local unit with medium size in our dataset), each one percent increase in the 25 number of parcel s will reduce assessment expenditure per parcel by 0.5 3 percent. Each one percent increase of average assessed value is associated with a 0. 54 percent decrease of assessment expenditure per parcel. Interestingly, privatization, or hiring private assessors or county assessors as alternatives to in - house assessors, fails to reduce assessment expenditure as intended. Requiring higher - level certifie d assessors does not significantly increase assessment expenditure s. City governments spend 24 percent more per parcel than do township governments. Several policy implications stem from this research . First, my analysis suggests merging small assessing d istricts to reduce expenditures because assessment is an activity with econom ies of scale. Second, privatization is not an effective way to reduce assessment expenditure . However, it is too soon to conclude the effectiveness of privatization before compari ng the assessment performance of diff erent types of assessors. T his is one of the limitation s of this research because it does not consider the possible differences in assessment quality, such as fairness and precision . It will be help ful to compare both t he expenditure analysis and assessment quality analysis , which leads to my third chapter , where I analyze the assessment performance of different types of assessors. Another drawback of this research is the short time span of data. A panel data of three years contains limited information ; it will be helpful to extend this analysis with a longer series in future research. 26 APPENDICES 27 APPENDIX A Evolvement of Michigan Property Tax Statutes 28 The authority of the Michigan government to levy property taxes is contained in Article 9 of the Constitution of the State of Michigan. Section 3 of Article 9 states that d tangible personal property not exempt by law except for taxes levied for school operating purposes. The legislature shall provide for the determination of true cash value of such property; the proportion of true cash value at which such property shall be uniformly assessed, which shall not, after January 1, 1966, exceed 50 percent; and for a system of new or rate authorized by law or charter when this section is ratified, without the approval of a majority of the tax revenue growth by demanding reduction of maximum authoriz the same gross revenue from existing property, adjusted for changes in the General Price Level, as could have been collected at the existing authorized rate on the prior assessed In 1994, the voters in Michigan a pproved the Michigan education finance amendment, known as Proposal A, as a means to amend the Constitution. Proposal A replaced most school property tax with an increase in sales tax to restrain the increasing reliance on property taxes for K - 12 funding. One primary influence to property assessment is that growth on each individual parcel's taxable value is capped to the rate of 29 inflation or five percent, whichever is lesser. When a property is sold, its taxable value gets uncapped and reset to the State E qualized Value (SEV) and then recapped , subjecting the growth limitation until it sells again. 10 10 Assessed Value (AV) is 50 percent of the usual selling price or true cash value of propert y; State Equalized Value (SEV) is the assessed value as finalized by the county and state equalization process. In most municipalities the SEV and AV are the same; Taxable Value (TV) is the lesser of State Equalized Value or Capped Value; Capped Value (CV) is the prior year's Taxable Value minus losses increased by the consumer price index or five percent whichever is less, plus additions. 30 APPENDIX B Three - Step Michigan Property Assessment Process 31 Step One: The local assessor (township assessor or city assessor) determines the assessed value o f a property based on the property's condition on December 31 of the previous year by determining the market price of the property. This process produces the taxable value (TV). To be more specific, local assessors 1) list and inventory all properties loca ted within the assessment jurisdiction; 2) equitably evaluate every item of taxable property; 3) correctly calculate the taxable value of each taxable property by identifying the taxability and the owner, describing the location and physical condition of t he property, and determining the assessed value, capped value and taxable value; and 4) prepare an assessment roll with all taxable property listed in an orderly fashion by parcel id or legal description. According to State Tax Commission guidelines, all local units (cities and townships) are expected to physically inspect a minimum of 20 percent of their parcels in any given assessment year. The expectation is that the physical data on the property record car d is no more than five years old. Sometimes a c omplete re - evaluation or reassessment is needed (i.e. , 100 percent of the parcels). Step Two: The board of commissioners in each county equalization department equalizes, or applies an adjustment factor, to ensure that property owners in all cities, towns taxes. Equalization serves to bring the total valuation across assessing units as close to the 50 percent level as possible. As the General Property Tax Act 206 of 1893 s commissioners of a county shall establish and maintain a department to survey 32 assessments and assist the board of commissioners in the matter of equalization of assessments, and may employ in that department technical and cleric al personnel which in its judgment are considered necessary. The personnel of the department shall be under the direct supervision and control of a director of the tax or equalization department who may designate an employee of the department as his or he r deputy. The director of the county tax or equalization department shall be appointed by the county board of commissioners. The county board of commissioners, through the department, may furnish assistance to local assessing officers in the performance of duties imposed upon those officers by this act, including the development and maintenance of accurate property descriptions, the discovery, listing, and valuation of properties for tax purposes, and the development and use of uniform valuation standards and techniques for the The main purpose of this step is to make sure that all the local government units in the county are assessing fairly and equitably at the constitutionally required level of 50 percent. This is accom plished by appraisers in county equalization departments who conduct appraisals and studies of all the seven classes of property (Agricultural, Commercial, Industrial, Residential, Developmental, Timber Cutover , and Personal property). Some county equaliza tion departments also perform complete assessments and appraisals of every parcel in one or more local government units when requested by the unit(s). Step Three: The State Tax Commission applies an adjustment factor to the county assessments to bring the total valuation across counties as close to the 50 percent level as 33 More specifically, the commission convenes to receive recommended state equalized valuation for each county; he ars the equalization director of each county or its duly authorized representative who desires to address the commission; prepares a statement showing detailed assessment valuation of each property assessed/appraised by each level of government units, by c ounty in an aggregate amount and by county for personal property and each classification of real property; hold a formal hearing, upon conclusion of which the commission adopts, by roll call vote, the final state equalized valuation report. 34 APPENDIX C Assessing Eligibilities and Training Requirements of Assessors with Different Certifications 35 Michigan Certified Assessing Technician (MCAT) is a limited certification that provides basic education in assessment administration. MCAT can be achieved by at tending a two and half - day course and passing the STC MCAT exam with a score of 75 percent or higher. Michigan Certified Assessing Officer (MCAO) is a certification which provides individual with eligibility to act as the assessor of record for a local uni t and sign an assessment roll with a SEV as specified in the STC Certification Level Requirements. MCAO certification can be obtained by completing one of two paths: STC six - month program or Self - Study, and pass the STC MCAO exam with a score of 75 percent or higher. Michigan Advanced Assessing Officer (MAAO) is a certification that provides eligibility to act as the assessor of record for a local unit and sign the assessment roll with a SEV as specified in the STC Certification Level Requirements, which is a broader range than granted for MCAO certification holders. MCAO certification can be obtained by completing one of two paths: STC 1 - year online/lecture hybrid program or course through an approved organization and passing an exam at the conclusion of ea ch course with a 75 percent or higher score . Michigan Master Assessing Officer (MMAO) is a certification that provides eligibility to act as the assessor of record for a local unit and sign the assessment roll with a SEV as specified in the STC Certificati on Level Requirements, which is a broader range than granted for MAAO certification holders. To obtain MMAO certification, MAAO certified assessors must have two years of assessment administration experience 36 after obtaining MAAO certification, complete thr ee MMAO required courses, and pass both the Case Study and the Oral Examination with 75 percent or higher. 37 APPENDIX D Functions of Local Assessing Offices 38 Besides step one of the Michigan property assessment process, functions of local assessing offic es also include: 1) the administration of the Principal Residence Exemption, processing of Property Transfer Affidavits, processing of tax exemption applications, and processing of property splits and/or combinations; 2) the processing of divisions of land , maintaining digital mapping (GIS) of property lines and the write up of legal descriptions, establishing special assessment districts and apportioning the special assessment within that special assessment district; 3) defending assessed and taxable value s before the Michigan Tax Tribunal and assisting other departments and divisions within the city with valuation, real estate , and property; 4) assisting residents with questions regarding property assessments. 39 APPENDIX E Tables 40 Table 2.1 Assessment Cycle and Other Requirements of States in U.S. 41 Table 2 .1 (c ont d ) 42 Table 2.2 Summary of Statistics 43 Table 2.3 Coefficients of Potential Determinants of Assessment Expenditure 44 CHAPTER 3 : An Analysis of Public vs. Private Contracting Performance in Michigan Property Tax Assessment 3.1 Introduction Privatization of services traditionally provided by government agencies is a process of transferring the management of public services from government agencies to the private sector. The rationale for such transfers includes cost savings, introducing marke t competition, a lack of in - house expertise, a desire to improve service quality, etc . In recent years, privatization has been increasingly used as a means for national, state, and local governments to fulfill their public service obligations. The most common services outsourced in Michigan include attorney/legal services, engineering, solid waste and recycling, and property assessment and inspections. 11 Michigan property assessment is currently conducted by a mix of three types of assessors: in - house ass essors, private assessors, and county government assessors. Traditionally, local township and city assessing offices are more inclined to hire their own local assessors as long - term government employees, with an implicit preference towards an assessor who is a part of the community. T o make cuts to address budget shortfalls, many local assessing offices increasingly contracted assessment services to for - profit companies or stand - alone private contractors to replace the previously kept in - house local assesso rs. 12 Several assessing offices may hire the same contractor to share the assessment expenditure and achieve economies of scale. Those assessing offices 11 Leonard Gilroy (2014) , University of Michigan Survey Finds High Use, Satisfaction with Local Government Privatization https://reason.org/commentary/michigan - local - privatization - survey/ 12 D. Rowl ey, MAAO, Michigan Assessor Association Past President, personal communication, May 1, 2018 45 without in - house assessors or private contractors can request the county equalization department to perf orm assessments. Currently, hiring private contractors is most common. About 70 percent (or 985 out of 1,400) townships and cities hire private contractors, 4 percent (or 57 out of 1,400) townships and cities seek help from county assessors, and 26 percent (or 358 out of 1,400) townships and cities have in - house assessors 13 . Privatization of government services, including property assessment, has led to a long - lasting and heated discussion in the U.S. and worldwide. The focus has been on comparing the advant ages and disadvantages of private contractors relative to the in - house provision to determine whether the private sector can provide higher quality services at a lower expenditure . In other words, do private contractors perform better than government emplo yees? Should we privatize or stay with government service providers? - of assessor arrangement is always superior to another might be misguided. Is it possible that eac h arrangement performs better under some conditions, and neither is always superior or inferior to the other? Perhaps both in - house assessors and private contractors have their comparative advantages in different types of communities; a more optimal balan ce may require matching the service arrangement to the most appropriate communities. In this chapter , I measure the performance of property assessment using detailed data from most local governments in Michigan. I explore the potential determinants of asse ssment performance, such as the size of the community and average property value. I 13 Calculated from data collected in 2017. 46 will answer the following questions: Does increased assessment expenditure help to improve assessment outcomes? Under what conditions does property assessment achieve econo mies of scale? Does the composition of property parcels affect assessment performance? This research also aims to identify the comparative advantage of contractor assessors, in - house assessors, and county assessors by analyzing how the determinants of per formance impact the three types of assessors differently. This evaluation will provide insight into how different types of assessment arrangements can be optimally allocated. For example, if community size affects private assessors negatively, but affects in - house assessors positively, then assessment performance can be improved by hiring in - house assessors in larger communities. I also compare the optimal distribution of assessor types with the current distribution to analyze whether there is a misallocati on of assessors in the state. In other words, a re private contractors and in - house assessors providing services in those communities where they have a comparative advantage ? For example, if private contractors perform better in larger communities, but tend to be used in smaller communities, then there may be a misallocation of assessors. Th e remainder of this chapter is organized as follows. Section 2 provides a review of the current Michigan property assessment process. In section 3, I summarize the relevant literature. I discuss the econometric specification in Section 4 and the data in Section 5. I present the curr ent distribution of the three types of assessors in Section 6 and then discuss the empirical results in Section 7. 47 3.2 Background 3.2.1 Michigan Property Assessment Process Michigan property assessment is a three - step process: 1) The local assessor (townsh ip assessor or city assessor) determines the assessed value of each property based on market values. The local assessor also calculates taxable values (TV), which is the tax base; 2) The board of commissioners in each county equalization department equaliz es, or applies an adjustment factor, to ensure that property taxes are paid in a fair and equitable manner across all cities, townships, villages, and school districts; 3) the State Tax Commission (STC) applies an adjustment factor to county assessments to bring the total valuation across counties as close to the 50 percent 14 level as possible, and calculates a state equalized value (SEV) for each property , which is equal to 50% of the market value . Compared to other states in the U.S., Michigan property as sessment is highly decentralized, with local property assessment conducted at the township, city, and village levels. Michigan has 83 counties, within which there are more than 1,700 townships and cities. Each township and city is required by state statute to maintain its own assessing office and to conduct independent property assessment. Michigan is among the ten states in the U.S. that require local (city, township, or municipal level) assessment. In contrast, most states conduct assessments at the count y level, and some states even conduct assessments at the state level (See Figure 3. 1 in Appendix B ). Moreover, Michigan requires annual reassessment, the most frequent required assessment cycle a mong U.S. states. (see Figure 3. 2 in Appendix B for more details ). 14 Michigan assesses property value at 50 percent of true market value. For example, the property tax base of a house that worth $1 million should be $500,000, and the tax bill should be tax rate times $500,000. 48 3.2.2 Michigan Property Assessment Performance Overview There are two key dimensions when it comes to property assessment quality (or performance): level and uniformity (IAAO, 1990). Assessment level measures the overall ratio at which propertie s are evaluated in relation to true market value, while uniformity measures the degree to which properties are equally and fairly appraised in relation to their true market value. A generally preferred measure of overall assessment level in an assessing di strict is the median sales ratio (assessed value/sale price) 15 of each parcel sold in a particular year in that district. The closer the median sales ratio is to 1, the higher the assessment quality is in an assessing district, as the assessed value of each property in a perfect assessment should be precisely the same as the true market value. Uniformity is typically measured by the coefficient of dispersion (COD) 16 , which reflects t he average percentage deviation of the ratios from the median ratio. For example, a COD of 10 indicates that individual sales ratios within a district vary from the median sales ratio by plus or minus 10 percent on average. A smaller COD means a higher qua lity assessment, as the sales ratio for each property in a fair assessment should be close to the median sales ratio. According to the International Association of Assessing Officers (IAAO), good assessments should have a COD of less than 30 (IAAO, 1990). The ideal assessment would have a median sales ratio of 1 and a COD of 0, implying that the assessed value is equal to the market price for each parcel. According to these guidelines, property assessment performance in Michigan is generally unsatisfactory in terms of both level and uniformity. Figure 3. 3 in Appendix B 15 Michigan median sales ratio is calculated as 2 times assessed value/sale price of each parcel, as Michigan assesses property value at 50 percent of true market value . 16 The calculation of COD will be discussed in later sections. 49 shows the distribution of the median sales ratio of Michigan local assessing districts (townships and cities) in 2016, 17 which ranges from 0.02 to 4, with an average of 0.94. Only 39% of assessing districts have a median sales ratio falling between 0.9 and 1. The wide range of median SR shows a high variation in assessment quality across Michigan. While many assessing units perform well in terms of level, some over - or under - estimate signi ficantly. Figure 3. 4 in Appendix B shows the distribution of COD in Michigan in 2016. 18 Only 16% of Michigan assessing districts achieved the 30 IAAO criteria. Most assessing districts have a COD much higher than 30, and some have a COD exceeding 100. Figur e 3. 5 in Appendix B shows the average median sales ratio and COD of Michigan local assessing districts from 2008 to 2016. The average COD is relatively stable over time, ranging from 75 to about 90. The average median sales ratio declined from 1.5 in 2009 to around 0.9 in 2016, indicating an apparent systematic overestimation in Michigan during the financial crisis. In summary, the primary challenges in Michigan property assessment include (i) lack of uniformity in assessment as indicated by high COD; (ii) high variation of assessment throughout Michigan as evidenced by a wide range of median sales ratio; (iii) a tendency of overestimation in the wake of the financial crisis; and (iv) potential over - spending caused by high decentralization of assessment. 3. 2.3 Michigan Assessors Overview As mentioned in an earlier section, local assessors in Michigan can be classified into three categories: in - house assessors, private assessors, and county government 17 - length, open - market transfers are included in the distribution. 18 - length, open - market transfers are included in the distribution. Assessing units with COD bigger than 300 are not shown in the graph. 50 assessors. In - house assessors are usually long - term govern ment employees who only work for one specific local assessing office. Private assessors can be either certified individual assessors or employees of private assessing companies. Most private assessors have appointments with more than one local assessing of fice. The contract terms are usually negotiated between local assessing offices and assessors, and are usually between one to five years. County assessors typically also serve as the director of the county equalization department. 19 It might be possible fo r the three types of assessors to behave differently due to different incentives. Thus, it is worthwhile to examine whether the performance of each type of assessor responds differently to various determinants of assessment performance. For example, privat e contractor performance could be more sensitive to community size, as they might think it is important to secure contracts with larger jurisdictions. To address this question, I examine the responsiveness of the three assessor types to community character istics. In this research, I will analyze potential factors affecting Michigan property assessment performance with given assessment expenditure s, including assessing district size, average property value, effective tax rate, property composition (the value of agriculture, commercial, residential parcels as in percentage of total parcels) and the number of transactions. I also test whether these factors have different impacts on in - house assessors, private assessors, and county assessors. I then consider whe ther there is a misallocation of assessors by comparing the current distribution and a more optimal distribution indicated by the analysis. 19 I conclude this base on the namelist of local assessors collected for this research. 51 3.3 Literature Review There is extensive existing literature comparing the quality and cost/efficiency of public and private delivery of public services, the results of which are mixed. Hilke (1993) compiled over 100 independent studies comparing the cost/efficiency of in - house government agencies versus private , p rofessional providers in various areas of public goods and services, including airline operation and airplane maintenance, property tax assessments, banks, electric utilities, fire protection, debt services, and so on. Most studies show cost reduction resu lting from privatization in his compil ation, and the typical reductions range from 20 percent to 50 percent. S ome studies show little difference in efficiency/cost between in - house and contract provision of services, or even higher cost for private service providers. The empirical results comparing the service quality of private and public service providers are also mixed. Mikesell (1987) analyzed the data of private and local assessors in Indiana and found that better quality assessment is likely to emerge from private appraisal firms working under contracts as opposed to elected assessors. Stocker (1973) compared in - house property assessors and private contractors in Ohio and found that private assessments are more accurate. Lowery (1982) analyzed Michigan private and in - house assessors and found that contracting leads to substantially lower levels of assessment. With inconsistent empirical findings, some research supports the argument that both public and private sectors may have certain relative strengths , and neither sector is inherently better or worse when it comes to delivering public services. Bendick (1989), Osborne and Gaebler (1992) conclude that private contractors tend to perform better on straightforward or specialized services such as processin g payments and computer 52 systems design, but worse on complex, long - range, and subjective services such as ensuring equity, continuity, and stability of services, social cohesion, etc. However, this does not explain the inconsistency in empirical results in a specific area of public service. For example , as mentioned above, Mikesell (1987), Stocker (1973) , and Lowery (1982) found opposite results, even though they all analyzed the same area of public service: property assessments. Another example is a case s tudy conducted by Hatry (1989) comparing matched pairs of public and privately administered prisons, which found mixed results for the cost reduction of private services provision. A potential explanation for such inconsistency is that both public and private sectors may have certain relative strengths even in the same area of public services, and neither sector is always superior or inferior . This research tests the hypothesis that private assessors, in - house assessors, and county assessors in Michigan have different comparative advantages, using the panel data from 2008 through 2016 for most Michigan property parcels (over 5 million parcels) over 1000 local governments. This researc h follows the framework advocated by Mehta & Giertz (1996), who proposed innovative methods for calculating technical efficiency of property assessment within the stochastic frontier framework. Their methodology incorporates both output (COD) and input (as sessment expenditure per parcel) of property assessments to compare and rank assessing district efficiency. They found that the factors that played a role in determining efficiency include the number of parcel s , tax rate, parcel composition, and legal stru ctures of each county. They suggest a possible optimal district size in property tax administration, which advocates further examin ing the policy recommendation to consolidate assessing districts. 53 The present research includes several technical improvement s to Mehta & Giertz (1996). First, i nstead of using COD as the sole measurement of property assessment output as Mehta & Giertz did, I take both assessment level and assessment uniformity into consideration. Specifically, I create a new performance measure , combining COD, the measurement of assessment uniformity, and median sales ratio, the measurement of assessment level. The inclusion of assessment level measurement is particularly relevant in Michigan, as property assessment practices are more diverse th an in other states. Michigan not only contains jurisdictions where governments systematically underestimate property values to avoid sudden tax increases and/or aversion, even appeals from taxpayers 20 , but also contains jurisdictions such as Detroit where the government systematically overestimates real property value (Hodge, McMillen, Sands and Skidmore, 2017). Furthermore, while Mehta & Giertz used a single - year cross - section data of 88 counties in I llinois, I use panel data from 2008 to 2016 for over 1000 local governments in Michigan. The substantial improvement in both the volume and quality of data increases the reliability of this research. To exploit this extensive panel data set, I use the Stoc hastic Frontier Model developed by Schmidt & Sickles (1984), which is essentially the same model used by Mehta & Giertz (Stochastic Frontier Model by Schmidt & Lovell (1979)) but adapted for panel data. The Stochastic Frontier Econometric model was initial ly developed by Aigner, Lovell, and Schmidt (197 7 ) and van den Broeck (1977). Its purpose was to practically /organizations to produce 20 Any citizen in Michigan who are not satisfied by the property assessment decision made by the local Board of Review can file an appeal of the decision with the Michigan Tax Tribunal by filing a petition. 54 below their maximum attainable output, and those due to random disturbances such as bad luck or systematic errors. The model was then extended by Battese and Corra (1977), Battese and Coelli (1988), Lee and Tyler (1978), Pitt and Lee (1981), Androw et al. (1982), Kalirajan and Flinn (1983), Bagi and Huang (1983), Schmidt and Sickles (1984), and Waldman (1984), etc. to allow adaption for various data complications. It is now one of the dominant models used in empirical industrial pr oduction studies and has been used to study public production too. 3.4 Model Specification The Stochastic Frontier Framework models the production frontier to be the set of technical efficient producing institutes, which produce the maximum feasible outpu ts with given inputs. The actual production is bounded within the production frontier, indicating that the actual output is always smaller than or equal to the maximum feasible outputs with given inputs. The Stochastic Frontier Econometric Model structures the output frontier as a function of inputs, controls, and a symmetric error as : (1) where is the jth input or control variable of unit i at time t; is the maxi mum feasible output given the inputs and controls ; is the coefficient of each input and control variable; and is the symmetric error term. The model structures the gap between actual output and maximum feasible output as a one - sided error , which comes from the concept that actual output is always smaller than or equal to maximum feasible output. The a ctual output function is the output frontier plus the one - sided error: 55 (2) where is the actual output of unit i at time t, and is the one - sided error term. Formula (1) and (2) clearly show that the actual output of unit i is always within the frontier function as the error term . There are two error terms in the output function : the symmetric error, and the one - sided error. This compound error comes from the economic logic that the production process is su bject to two different random disturbances: the one - sided error reflecting the deviation from the best output due to technical or economic inefficiencies, and under the dom factors such as luck, climate, topography, and can be favorable or unfavorable. My research will focus on , not the , as I am more interested i n learning about the factors that affect assessment output rather than how far away a specific unit is from the frontier. In the context of property assessment, the output is the quality (or performance) of assessments. As mentioned above, there are primarily two dimensions for property assessment quality: level and uniformity (IAAO, 1990). Assessment le vel measures the degree to which goals or certain legal requirements are met, while uniformity measures the degree to which properties are appraised in relation to true market value. A generally preferred measure of overall appraisal level in an assessing district is the median sales ratio (assessed value/sale price of each parcel transacted in that year) in that district . U niformity is usually measured by the coefficient of dispersion (COD), which reflects the average percentage deviation of the ratios from the median ratio: (3) 56 where is the COD of assessing district i in year t, is the sales ratio of parcel h (belonging to district i and transacted in year t) in year t, is the median sales ratio of district i in year t, and is the number of parcels transacted in district i in year t. Both median sales and COD are annual and based on the parcels sold in the relevant year. The closer the median sales ratio is to 1, and the closer COD is to 0, the higher the assessment quality is in one assessing distri ct. The ideal assessment would have a median sales ratio of 1 and COD of 0, implying assessed value equals the market price for each parcel. Because a one - dimension measurement to evaluate assessment quality is needed for the evaluation, I create a measure Performance Score and uniformity. With this measurement, it is easier to compare and rank the assessment quality of each Michigan local assessing district. The created measurement Performance score is calculated as: (4) where is the performance score of assessing district i in year t. This performance measure is very similar to COD, only replacing the median sales ratio with the ideal sales ratio 1. By analogy, performance measures the average percentage deviation of the ratios from the ideal sales ratio of 1. For example, a performance score of 10 indicates that individual sales ratios within the assessing district vary from 1 (ideal sales ratio) by plus or minus 10 percen t on average. A lower performance score indicates a better assessment. I then use the calculated performance score as the output measurement (left - side variable) in the stochastic frontier framework. I use assessment expenditure per parcel or 57 assessment expenditure as a percent of total property tax collected as the assessment inpu t of each assessing district. Since a lower performance score means a better assessment, the production frontier is the set of technical efficient producing institutes that gene rate the minimum feasible performance score with given inputs and controls. The actual performance function is always beyond the production frontier, indicating that the actual performance score is always larger than or equal to the optimal performance sco re with given inputs. To specify the performance frontier as a function of assessment inputs and controls, we have: (5) I specify the actual performance function as the performance fr ontier plus a positive disturbance, reflecting the concept that the actual performance score is always larger than or equal to the optimal performance score with everything else equal. (6) where is assessment expenditure of assessing district i in year t; represents the control for the difficulty (or the workload) of the assessment of assessing district i in year t; is the symmetric error term; and is t he po sitive error term. My research will focus on estimation of . In other words, in this analysis I am more interested i n learning about what affects assessment output than how far away a specific unit is from the frontier. In the property assessment literature, assessment expenditure is usually measured by either assessment expenditure per parcel or assessment expenditure as a percentage of tax revenue. The two measurements both have advantages and disadvantages. 58 E xpenditure as a percentage of tax revenue is highly dependent on the property tax rate, which is left to the discretion of each local government and ranges from 9.3 mil to 96.6 mil in Michigan. 21 The expenditure per parcel does not capture differences in the workload of assessing different properties. After controlling the average property value and tax rate, both approaches are good reflections of the assessment input of individual townships and cities. This res earch uses expenditure per parcel as the input measure in the core analysis , and also reports a robust analysis with expenditure as a percentage of tax revenue as the alternative input measure to confirm the results. Control variables include the number of parcels (in log), average assessed property value (in log), parcel composition (agriculture, residential, and commercial parcels in percent of total parcels), effective property tax rate (in %), number of transactions (in log), and time dummies. The number of parcels measures the size of an assessing district and is expected to affect the assessment workload. Average assessed property value measures how wealthy an assessing distr ict is, and is expected to affect tax assessment performance as well the effort involved in assessing a $100 million worth parcel is expected to be greater than that of a $100,000 worth parcel. Parcel c lassification is expected to affect assessment perform ance as the difficulty in assessment may vary from one parcel classification to another. The effective tax rate is expected to affect assessment performance as a higher tax rate might boost taxpayer awareness. The number of transactions is expected to impa ct the assessment performance because recent sales can reference similar properties in an assessment. I included time dum my variables because tax performance might be affected by the financial crisis in 2008, as the government 21 2016 data in the dataset. 59 might over - or under - estimate property values to differing degrees over the period of analysis . It is essential to point out that in equation (6) is homogeneous over time to assess any given district, reflecting the assumption that each assessing district has the same deviation from best performance due to technical inefficiencies over time. This assumption is necessary to differentiate between the symmetric error term and the one - sided error term. It also has economic justifications: the factors contributing to technic al inefficiencies, such as insufficiently trained personnel and inadequate management system, are usually stable over time. T o test the hypothesis that the three types of assessors respond differently to changes in district size, average value, etc. , I est imate the following model. This model is similar to equations (5) and (6), with additional terms interacting with each control with assessor type dummies. The performance frontier is: (7) And actual performance is: (8) and equals to 0 if not. if assessing district i has a private assessor, if assessing district i has a county assessor. Fac tors include assessment expenditure per parcel and all the controls previously discussed : number of 60 parcels (in log), average assessed property value (in log), parcel composition, effective property tax rate (in %), and number of transactions (in log). Time dummies are included but are not interacted with assessor - type dummies to simplify the model. 3.5 Data I calculate the output measurements performance score, COD, and the median sales ratio from individual parcel tax records and current sales data obtained from CoreLogic 22 . I include only parcels transacted in the current year in the calculation. For example, 1,854 out of 43,181 parcels in Lansing were involved in transactions in 2016 (after trimming out non - - length transactions, errors, and ou tliers). I calculate the 1,854 sales ratios and use them to measure performance, COD, and median sales ratio for Lansing in 2016. To ensure all sales ratios included in the calculation are correctly - length, open - market transfer, I eliminate d all individual transactions with Eliminated transaction types include non - - length transactions, construction loan/financing, and seller carrybacks. Those transactions are not standard ope n - market transactions, and their sales prices are not proper reflections of their market values. I also eliminated all individual transactions with sales prices less than $1,000 or sales ratios greater than 20 as they are most likely errors or non - - le ngth transactions. Since Michigan assesses property values at 50 percent of market value, I calculate the sales ratios as two times the assessed value divided by market value. Figure 3. 6 in Appendix B shows that the new measurement performance score succes sfully combines information from both level and uniformity. From 2008 to 2015 , 22 CoreLogic, Inc. provides financial, property and consumer information, analytics and business intelligence. The data used in this research are the tax roll records they maintain. 61 when the COD is relatively stable, the change of performance score coincides with the average median sales ratio. From 2015 to 2016, performance score picks up the increasing trend of average COD. In this research, assessment expenditure is measured by either assessment expenditure per parcel or assessment expenditure as a percentage of tax revenue. Assessment expenditure per parcel is obtained by dividing total assessment expe nditure s by total property parcel counts, and assessment expenditure as a percentage of tax revenue is self - explanatory. I obtain total assessment expenditure s of each county, l Report (F - 65). Parcel numbers of the whole state, each county, and each assessing unit (township and city) are published by the Michigan Department of Treasury in L - 4023 forms annually. Total tax revenues of each county, township and city are obtained fr om the Ad Valorem Property Tax Report published by the Department of Treasury annually, reflecting the sum of county taxes, township/city taxes, school taxes, and authority taxes if applicable. Figure 3. 7 in Appendix B shows that the average expenditure as a percent of tax revenue i ncreases throughout the years, but the expenditure per parcel is relatively stable, with some fluctuations in 2013. Following the literature, I measure the size of assessing districts by the log of total parcel numbers . I measure property compositions of each assessing unit by the percentage of total parcel valuation in the agriculture, commercial, and residential property classes. I determine the type of assessor in each township and city using the following method. Ea ch Michigan county equalization department publishes the name of its 62 equalization director, and each township/city publishes the name of the township/city assessor and the company he/she belongs to (if applicable). For those governments with no or underdev eloped official websites, I contacted their equalization departments via phone calls and emails to obtain the assessor name list. I then created the assessor employment type dummy variable from this name list. If a township or city assessor is the county I f a township or city assessor is an employee of a private , professional property assessment company, I determine that this township o I f two or more township or city hires the same person who is not county equalization director , then I determine that they also use a . A township or city uses an - ue assessor but not a county equalization director. Except in rare cases where individual private contractor assessors have only one contract from one local government unit, this method should provide a good approximate of assessor employment type. Only assessing units (townships and cities) with more than 30 correctly recorded - length, open - market transfers are included in my data, because calculating the median, COD, performance from individual sales ratio is not statistically meaningful for a sample smaller than 30. Such trimming might create bias because the number of transactions is highly correlated with the total parcel number in an assessing unit, and I may be excluding the smaller assessing units. However, it is necessary to eliminate the unrepresentative observations to assure me aningful analysis. If an assessing unit has more than 30 of such transfers in some years but not in other years, only the observations in the years with more than 30 transfers are kept. For example, if Lansing has 31 property 63 transactions in 2016 and 29 tr ansactions in 2015, then only the 2016 record would be included in the dataset. This trimming method creates an unbalanced panel, but it helps to prevent excessive or unnecessary trimming. I provide a detailed summary of the statistics in Table 3.1 in App endix A . We can see that the dataset is unbalanced. Out of the 1,607 observations, there are 762 for 2016, but only 326 are for 2008. This is due to my trimming method, as an assessing district might have more than 30 transactions in one year, but not in o it is clear that some of the townships and cities have very large COD, as the max reaches more than 1,000 in some years. T o en sure that the analysis is not affected by those extreme observations, all observations with COD larger than 500 are considered outliers. I estimate the specifications with all data, as well as after eliminating the outliers. It is also apparent in parcels (in 2008), but it is not the smallest dis trict in Michigan because the smallest one (which only contains 55 parcels) is eliminated from the dataset as it does not have a sufficient number of transactions. There is also the property transactions are less fre quent during financial crisis years (2008 - 2012) and become more frequent after the economy begins to recover. Total effective tax rates are stable throughout the years, reflecting the relatively stable tax environment. 3.6 Distribution of Private vs. In - house vs. County Assessors As mentioned earlier, hiring private contractors is the current trend: 70 percent (985 of 1,400) of townships and cities hire private contractors, 4 percent (57 of 1,400) of 64 townships and cities seek help from county assessors , and 26 percent (or 358 out of 1,400) of townships and cities keep in - house assessors 23 . To learn more about where the three types of assessors are allocated in Michigan, I estimate the following multinomial probit model, in which the probability of a loc al assessing district having a private or in - house or county assessor is a function of parcel number and average assessed p roperty value in that district. Let the outcomes be numbered 1 (in - house assessor), 2 (private assessor), and 3 (county assessor ). For assessing district , the model assumes a latent (unobserved) utility or score for each outcome : (9) Here = [ ], a row vector of explanatory variables for district = ve ctor of coefficients for outcome ; and the are independently and identically distributed as N(0,1). The probability that outcome is chosen is the probability that is the largest of and It is therefore a m ultivariate normal probability. Since the differences of the utilities depend only on the differences of the , they are not all identified, and we use the normalization that = 0. Thus and influence the effects of the explanatory variables on the prob abilities of outcomes 2 and 3 , respectively. The calculations were done in Stata using the command mprobit. The discussion above is a summary of the Stata documentation. For more theoretical detail, see Cameron and Trivedi (2005, 516 - 519), Greene(2012, 770 - 771) , or Wooldridge (2012, 648 - 649). 23 Calculated from data collected in 2017. 65 These parameters are estimated by maximum likelihood with 2017 data 24 , and the result is shown in Table 3 .2 in Appendix A . 25 The coefficients are shown in the first column, and the average partial effects (APE) are s hown in the third column. The results indicate that private contractors are more concentrated in smaller and inexpensive assessing districts, and county assessors are concentrated in larger and expensive assessing districts. To be specific, a one percent increase in district size reduces the probability of hiring a private contractor by 0.11 on average . A one percent increase in average property value reduces the probability of hiring a private contractor by 0.07. A one percent increase in district size will increase the probability of using a county assessor by 0.02 . A one percent increase in average property value will increase the probability of using a cou nty assessor by 0.04. All the relationships are statistically significant at the 0.05 level. The rationale behind such distribution of assessors seems straightforward. Smaller assessing districts usually have a smaller budget on property assessments and hi gher fixed expenditure s per parcel . They have a greater incentive to contract out for assessment services as a method of expenditure - saving. The same argument also holds for assessing districts with less average property value because they might collect le ss revenue. However, such a distribution might not be an optimum allocation of the three types of assessors, as they might not be assessing the districts they are relatively better at assessing . Th is issue will be discussed in greater detail in the followi ng section. 24 The data used for multinomial probit model estimation does not eliminate observations with transactions less than 30, or COD greater than 500. All observations will non - missing values are included. 25 The Probit analysis uses 2016 data and does not exclude the observations with transactions less than 30. 66 3.7 Results and Discussion 3.7.1 Overall Analysis I estimate equation (6) using fixed effects and present the results in Table 3.3 in Appendix A . The first two columns show the results of the core analysis with assessment expenditure per parcel as the input measurement and without eliminating the outliers. The first column shows coefficients , and the second column shows the standard errors. The third and fourth columns show the results from the robust analysis with assessment expendit ure in percent of property tax collected as the input measurement. Following this are the results from the core and robust analysis with outliers eliminated. All four analyses point to the same consistent results: (i) average assessed property value and th e number of parcels is the most important factors of performance; (ii) assessment quality tends to be better in larger communities and communities with higher average property value; (iii) parcel composition and tax rate also matter. The results are consis tent with Mehta & Giertz's findings. The overall analysis does not find a statistically significant relationship between assessment expenditure and performance score, implying that assessment quality is insensitive to expenditure changes in general. This r esult contradicts the intuitive speculation that more input on property assessment will bring better output. It seems that the assessors in Michigan are not taking advantage when the assessment expenditure increases. The p erformance score is negatively rel ated to the number of parcels. Since performance score is smaller for higher quality assessment, this result implies that assessment quality tends to be higher in bigger assessing districts with everything else 67 equal. A one percent increase in parcel numbe r brings down the performance scores by 0.223 percent , on average. This result is consistent with Mehta & Giertz's finding that property assessment is an activity with an efficiency of scale. This finding suggests that is not the best approach , considering the strong economies of scale nature of assessing equalization activities 26 . Decentralization increases expenditures by requiring more assessing offices and assessors and may substantially decrease assessment quality. There is a statistically significant negative relationship between average parcel value and assessment performance score, and the relationship is strong in terms of magnitude. A one percent increase in av erage property value is associated with a 0.311 percent decrease in performance scores on average. A possible explanation is that assessments for properties with high assessed value are completed with greater care, and such care affects assessment quality. Parcel classification also affects assessment performance. Michigan groups its properties into different classifications: agricultural, commercial, industrial, residential, developmental, and timber cutover . This research studies the three main classes: a gricultural, commercial, and residential. The performance scores are statistically significantly lower when the properties are more concentrated in commercial and residential classes. The result implies that commercial and residential parcels are easier fo r assessors to assess than other kinds of parcels. A one percent increase of commercial or residential parcels as a percent of total parcels is approximately associated with a 0.009 or 0.006 percent reduction in assessment performance score, respectively. A 26 Yunni Deng. Determinants of Property Assessment E xpenditure s in the State of Michigan. (Working Paper) 68 possible explanation is that assessors might be more experienced in handling these two classes of properties, rather than agricultural, industrial, developmental , and timber cutover properties. The result also indicates a positive relationship between th e overall property tax rate and assessment performance . The higher the tax rate is, the lower the performance score is, holding other factors constant the effective tax rate will lead formance scores , on average. This confirms the conjecture that a higher tax rate improves assessment performance , perhaps by boosting taxpayer awareness in assessing districts. Surprisingly, the result indicates a slight negative relationship between transaction frequenc y and assessment performance . Intuitively, current property transactions can provide assessors with more information on the property market, and thus enhance assessment quality. However, the evaluation suggests that Michigan assessors d id not take full advantage of the information provided by real property sales. 3. 7.2 An Analysis for Assessor Types Given that the three types of assessors may behave differently, it is helpful to analyze how assessor performance responds to the factors based on assessor type. I thus estimate equation (8) using fixed effects and present the result in Table 3.4 in Appendix A . Again, both core analysis and robust analysis are done with and without eliminating outliers. The key results from the analysis are: (i) larger parcel counts, higher average property value, and higher effective property tax rates enhance assessment qualities for all three types of assessors; (ii) in - house assessors, private contract assessors, and county 69 assessors respond very differently to changes in assessment expenditure s, average property values, assessing district sizes, and tax rates; (iii) private contractors perfo rm relatively better in communities with more parcels; (iv) c ounty assessors perform relatively worse in communities with high average property value. County assessors appear to be more expenditure - sensitive than private or in - house assessors. The perform ance scores of both private and in - house assessors are slightly positively related to either assessment expenditure per parcel or assessment expenditure in percent of property tax collected, implying that increases in assessment expenditure would not impro ve assessment quality. In contrast, there is a statistically significant negative relationship between assessment expenditure per parcel and performance scores for county assessors. A one percent increase in assessment expenditure per parcel translates int o an average (0.197 0.034 =) 0.163 percent reduction in assessment performance scores . The n umber of parcels is negatively related to performance score s for all three types of assessors, but the relationship is stronger in magnitude for private contract ors than for in - house assessors and county assessors. A one percent increase in parcel number reduces performance scores by 0.139 percent on average for in - house and county assessors . This number is (0.139 + 0.112 =) 0.251 percent for private contractors. This result indicates that the private contractors are most effective in dealing with communities with a large number of parcels and should be allocated in such communities. From this research, t he reason s for the private contractor comparative advantage in larger communities is unclear. I speculate that private contractors might emphasize contracts with larger assessing districts and thus conduct the assessments more 70 carefully to e nsure the renew al of their contrac ts when the existing contracts end. However, such speculation should be formally tested in further research. This result also indicates a misallocation of private contractors in Michigan: private contractors are best in assessing larger districts, but they are more concentrated in smaller assessing districts. The policy implications followed by this result are clear merging small assessing districts to form large assessing districts would improve performance , and large assessing districts should consider hi ring private contractors. Average property value affects property assessment performances differently for the three kinds of assessors as well. Average property values are also negatively related to performance scores for all three types of assessors, but the relationship is weaker in magnitude for county contractors than for in - house assessors and private contractors. According to the core analysis without eliminating outliers, a one percent increase in average property value reduces performance scores by 0.286 percent on average for in - house and private assessors. This number is (0.286 0.138 =) 0.148 for county assessors. Although c ounty contractors are least effective in assessing communities with higher value properties, they are most concentrated in s uch communities. The policy suggestion is that assessing districts with high average property values should consider hiring private contractors or in - house assessors instead of county assessors. A higher effective tax rate improves all three types of asses quality, and the impact is stronger in terms of magnitude for county assessors. According to the core analysis without eliminating outliers, a one mil increase in total effective tax rate reduces performance scores by 0.1 mil s (or 0.01 per cent) on average for in - house and private assessors. This number is (0.1 + 0.19 =) 0.29 mil for county assessors. 71 It seems that parcel composition affects the three types of assessors almost to the same degree. The number of transactions affects private co ntractors and county assessors more than in - house assessors. Again, the reasons for such similarities and differences are not the scope of this research and need to be explored in the future. 3. 7.3 Scenario Forecasts To more clearly demonstrate how misallo cation of assessors affects overall property assessment performance, I provide linear forecasts for average performance score under the following four scenarios: (i) each assessing district remains status quo; (ii) if all private contractors are replaced b y in - house contractors; (iii) if all in - house assessors are replaced by private contractors; (iv) if the assessors are optimally (according to this research) allocated based on community size: private contractors assess communities containing more than 50, 000 parcels, and in - house assessors assess the rest of the communities. I present the result in the first part of Table 3.5 in Appendix A . I provide forecasts for average performance scores assuming every assessing unit performs without technical inefficiency (lies on the performance frontier). It is interesting that changing to all private contractors makes things worse , whereas changing to all in - house makes things better. The average performance score would be 29.03 if all private contractors are replaced by in - house contractors, and 30.20 if all in - house assessors are replaced by private contractors. The average performance score would be 28.94 if assessors wer e allocated optimally, which indicates the highest quality assessment among the four scenarios. allocation approximately 2.3 percent. The simulation result serves a s further evidence of 72 comparative advantages among different types of assessors since neither scenario (ii) or (iii) is the optimal allocation. As this research suggested, consolidating smaller assessing units into larger ones would affect the performances since all three types of assessors performs better in larger communities. In order to estimate how consolidation would affect overall performances, I provide the linear forecasts for average performance scores under the previous four scenarios, but assumi ng smaller assessing units are consolidated into one large assessing district. I do that by limiting the observations to assessing units with more than 5,000 parcels (larger than median size communities ), as this will be a good approximation to the scenari o in which all small assessing units are consolidated into one large one. I present the result in the second part of Table 3.5 in Appendix A . The performance would be significantly improved once small assessing districts are consolidated . Table 3.5 shows t hat the overall performance score comes down to 21.48 even if we keep the assessors ' types unchanged, which accounts for a 27.5 percent improvement compared to the current situation. The performance score would further decrease to 20.17 if we were to alloc ate assessors optimally. improves performance by 32 percent compared to the current situation, and the improvement is statistically significant considering the standard deviatio n. 3.8 Conclusion In summary , the key results of this research are: (i) average assessed property value and the number of parcels are the most important factors for performance; (ii) assessment quality tends to be better in larger communities and communities with higher 73 average property value for all types of assessors; (iii) in - house assessors, private contract assessors, and county assessors respond very differentl y to changes in assessment expenditure s, average property values , and assessing district sizes; (iv) there is a misallocation of Michigan assessors: private contractors have a comparative advantage in assessing communities with more parcels, but are actual ly more heavily employed in communities with fewer parcels; c ounty assessors have a comparative advantage in assessing communities with lower average property value, but are actually not used as much in such communities; (iv) consolidating small communitie s and moving to the situation. The policy implications drawn from these findings are consistent with those of Chapter 2. First, my analys es suggest merging small assessing districts and form larger assessing areas . This consolidation would improve assessment performance as all types of assessors tend to perform better in larger communities and reduce assessment expenditure since assessment is an activity wit h econom ies of scale. Second, I recommend allocating assessor s based on their comparative advantages. Specifically, private contractors should b e hired in communities containing more than 50,000 parcels , and in - house assessors should be hired in the rest o f the communities. Such reallocation would enhance assessment performance without increasing assessment expenditure because hiring different types of assessors does not affect assessment expenditure , as shown in Chapter 2. With all the important policy implications offered , I acknowledge some limitations to this analysis . As discussed in the data section, eliminating all communities 74 with fewer than 30 parcel sales may introduce selection bias. Moreover, due to data limitations, there are some potential f actors of assessment performance not included in development in each assessing district. Furthermore comparative advantages are not within th further research. Finally , the jurisdictions might have reasons for not taking optimal assessing regime s as discussed in this chapter . For example, they might be foregoing assessment performance for more control to the asses sors to deliberately over - or under - estimates properties. I t would be interesting to extend this research with a focus on government assessing objectives. Going forward, one possibility is to extend this research with a more comprehensive analysis. One possibility is to calculate the technical efficiency of each Michigan local government and rank them based on their efficiency . Th e econometric methodology used in this research the Stochastic Frontier Model could also be used to calculate technical e fficiencies for each government by combin ing both assessment expenditure and performance . S uch an analysis would offer a more comprehensive view of Michigan property assessme nt instead of focusing on either input or output aspect s . 75 APPENDICES 76 APPENDI X A Tables 77 Table 3.1 Summary of Statistics 78 Table 3.2 Factors of Assessor Performance 79 Table 3.3 Coefficients of Determinants of Assessment Performance 80 Table 3.4 Coefficients of Performance Factors for Different Types of Assessors 81 Table 3.5 Scenario Analysis 82 APPENDIX B Figures 83 Figure 3.1 Number of States by Assessment Level Figure 3.2 Number of States by Assessment Frequency 84 Figure 3.3 Frequency Distribution of Michigan Median Sales Ratio, 2016 Figure 3.4 Frequency Distribution of Michigan COD, 2016 85 Figure 3.5 Michigan Assessment COD and Average Median Sales Ratio, 2008 - 2016 Figure 3.6 Michigan Assessment Performance, 2008 - 2016 86 Figure 3.7 Michigan Assessment Expenditure , 2008 - 2016 87 CHAPTER 4 : A Dynamic Analysis of the Impacts of Changes in Housing Values on Property Tax Assessment During and After the 2008 Financial Crisis 4.1 Introduction The property tax is a primary revenue source for state and local governments. In 2017, state and local property tax revenue accounted for 31.8 percent of total states and local general tax revenue, ranging from 17.3 percent in Alabama to 66.6 percent in Massachusetts , and in Michigan, the figure is 33.7 percent. 27 However, there is a concern that property tax revenue could experience rapid changes if housing price fluctuations were to be entirely pass ed through to property tax collection. This concern was especially heightened when housing prices declined dramatically during the 2008 Global Financi al Crisis . Changes in house prices are typically not immediately reflected in the property assessment and tax collection. 28 This phenomenon has captured many researchers' interest ; how do property assessments respond to large cyclic upturns and downturns, especially real estate fluctuations? Many studies have examined the relationship between property price fluctuations and property tax revenues since the 2008 Global Financial Crisis. Most of this research found that the housing pric e fluctuations are not fully reflected in property tax revenue. In a relatively early study, using census data at both national and local levels from the 1980s to 2008, Lutz (2008) found that it takes about 27 State & Local Government Finance Data Query System. http://www.taxpolicycenter.org/statistics/property - taxes - percentage - state - and - local - taxes . The Urban Institute - Brookings Institution Tax Policy Center. Data from U.S. Census Bureau, Annual Survey of State and Local Government Finances, Volume 4, and Census of Governments (1977 - 2017). 2017 data. 28 ement, Volume 13, Number 4, pp. 310 - 334, 2013 88 three years for house price changes to affect tax revenues ; and the elasticity of property tax revenue to house prices reaches 0.4 over that period, Chernick, Langley, and Reschovsky (2011) forecast the impact of the recession and the housing crisis on central city revenue s and expenditures between 2009 and 2013, using data from nation's largest central cities from 1997 to 2008. They estimate th at property tax collection in these cities is 3.8% lower three years following a 10% reduction in the housing price index. Using local level housing data from Zill ow® matched to property tax data from 1998 to 2012, Goodman (2018) also found that decreases in property values have an elasticity between 0.3 and 0.4 ; and that it takes three years for changes in values to be fully reflected in property tax revenues. As the economic situation improves following a downturn, governments are relieved from immediate fiscal pressures. Thus, in recent years fewer researchers have focused on the relationship between housing prices and property assessments. The most recent publis hed research in this area is Goodman (2017), which includes data for 2012, the starting point of housing market reversal after the dramatic decline . One contribution of the present work is that I extend the period of analysis to more fully explore the long er - run post - recession property assessment patterns. Another line of related research has focused on the quality and timing of property assessment during periods of significant property price fluctuations. Such studies include Paglin and Fogarty (1972), Che ng (1974), IAAO (1978), Kochin and Parks (1982), Bell (1984), Sunderman et al. (1990), Clapp (1990), Hodge, et al. (2017), among others. The general findings in this line of research indicate that assessment ratios decline with the property sale price, whi ch is a violation of the principle of vertical equity. Of most direct 89 relevance to the present study is Hodge, et al. (2017) , who examine assessment quality in the wake of the financial crisis. They found that official s in the City of Detroit, who were f acing imminent bankruptcy , did not reduce property assessments in response to dramatic real estate price reductions. The present research aims to analyze the medium - term impact of the Global Financial Crisis's on property assessment by examining how local officials changed assessments during and following the crisis. To examine this issue, I construct and compare a Hedonic Housing Price Index (HPI) and a State Equalized Value (SEV) index for 120 Michigan assessing units in the five - county region surrounding Detroit (Macomb, Monroe, Oakland, Washtenaw, and Wayne counties), using individual parcel - level data on housing sales and local government property assessment in Michigan from 2008 through 2016. This analysis offers a multi - government evaluatio n of how housing price declines and then increases were reflected in assessed values, how long it t ook for such changes to be implemented, and the potential asymmetry in responses to price declines and increases. My examination also enables the differentia l behavior across different types of local units: do assessment patterns in struggling places such as Detroit behave differently than places that have a more r o bust economic base? My analysis also demonstrates variability across local governments in prope rty assessment practices; townships and cities can either significantly over - estimate or under - estimate assessments as indicated by the wide range of median sales ratios. 29 My analysis offers an evaluation of the different assessment practices across over - and under - estimating local governments. Do local governments that overestimate assessments pass 29 90 through housing price declines more quickly to assessed values than an u nderestimating local government and vice versa? This chapter provides descriptive and an ecdotal analyses in an attempt to explain the differences in governmental assessment objectives. To do this, I compare the HPI and the SEV indices for several local governments with high and low median sales ratios. 30 This research also contributes to the l iterature by building an HPI at a much less aggregated level. All the above - cited studies use house price indices aggregated to a larger scale, such as county - , metropolitan - , or state - levels, which is much more aggregated than the level of property assess ment and tax collection. Hence, a housing price index that is calculated at a more aggregated level does not always map well with local - level property tax assessments. The remainder of this chapter is organized as follows. Section 2 provides a review of t he current Michigan property assessment processes and guidelines. I discuss the construction of the indices i n Section 3 and the data in Section 4. I present and discuss the empirical results in Section 5 and conclude in Section 6. 4.2 Institutional Detail s 4. 2.1 Michigan Property Assessment and Tax Collection Process In Michigan, property assessment is carried out annually in each local assessing district (townships, cities , and villages), with county equalization and state government oversight. 30 Sales ratio is assessed value/sale price. Median sales ratio in one assessing district is the median sales ratio of each parcel sold in a particular year in that district, which is a widely used measurement of assessment precision. 91 The process of Michigan property assessment contains three steps: i) the local assessor (township, city, or village assessor) determines the assessed value (AV) of property, which is 50 percent of what the assessor determines to be the market price. The lo cal assessors evaluate the market price of a property based on its condition on December 31 of the previous year, including: the advantages and disadvantages of location; quality of soil; zoning; existing use; present economic income of structures; present economic income of land if the land is being farmed or used for the production of income; quantity and value of standing timber; water power and privileges, etc.; ii) The board of commissioners in each county equalization department equalizes to ensure th at property owners in all local assessing districts in the county pay their fair share of that unit's taxes; 3) the State Tax Commission (STC) applies an adjustment factor to the county assessments to bring the total valuation across counties as close to t he 50 percent level as possible. This process produces the property's state equalized value (SEV). 31 4. 2.2 Michigan Property Tax and Assessment Cap Michigan has two different limitations on how much property tax revenue can grow each year. The first limit i s on property tax growth for each local government taxing unit (cities, villages, counties, and townships). According to the Headlee Amendment, an amendment to the Constitution enacted in 1978, if the total assessed value of a local tax unit's all existing properties, not including new properties, grows faster than the inflation rate, the maximum property tax rate must be rolled back. S o property taxes on existing 31 92 properties will not increase by more than inflation unless the local unit asks the voters for an override of the automatic rollback. The second limit is on the annual growth in the assessed value of each parcel of property. In 1994, voters approved Michigan Education Finance Amendment, known as Proposal A, which restrained Michigan's increasing rel iance on property taxes for K - 12 funding. Proposal A introduced the term " taxable value (TV) ," and since then, each parcel has three values, a capped value, an assessed value (SEV) , and a taxable value. 32 Proposal A capped the taxable value to be the 1994 a ssessed value. Since 1994 the annual increase of the taxable value for each parcel of property may not exceed five percent or the rate of inflation, whichever is less. Unless a property is sold, its TV gets uncapped and reset to the SEV and then recapped, subjecting the growth limitation until it sells again. To elaborate , the relationship of the three values, assessed value (AV) is 50 percent of its determined true cash value; state equalized value (SEV) is the assessed value as finalized by the county and state equalization process. In most municipalities, the SEV and AV are the same; taxable value (TV) is the lesser of SEV or capped value; capped value (CV) is the prior year's TV minus losses increased by the consumer price index or five percent, whicheve r is less, plus additions. It is worth noting that TV is the tax base of property tax, i.e., what is used to calculate property taxes, even though SEV is the final result of the assessment and equalization process. In this research, I use assessed value ( or state equalized value) as the measurement of property assessed value, because it is not capped and not affected by 32 State Tax Commission, Guide to Basic Assessing, May 2018. 93 Proposal A, and only related to the assessor - determined true market value of the properties. 4.3 Indices Construction 4. 3.1 Hedonic Price Index In this section , I describe how I construct a Hedonic Housing Price Index for the 120 Michigan assessing units in the five - county region surrounding Detroit (Macomb, Monroe, Oakland, Washtenaw, and Wayne counties) , using the parcel - level data on sale prices and the quality and characteristic attributes of the residential parcels sold. Hedonic pricing is a method that takes into account the inherent characteristics of the products. It is often used to calculate pri ce indices for heterogeneous goods, because traditional measures of changes in the average price of such goods reflect not only changes in the price level but also changes in the quality - mix of goods transacted. The hedonic pricing index approach, on the o ther hand, helps to eliminate price differences that arise from changes in characteristics or quality of goods and focus on the pricing trend. The hedonic pricing method is advantageous when calculating the price index of the property market, where the pri ce of a specific parcel is affected significantly by its own characteristics, such as square feet, appearance, age of structures, and condition, as well as characteristics of the surrounding environment, such as the distance to the nearest highway, crime r ates, school districts, the pollution level, etc . The hedonic pricing method uses a regression model to estimate the influence that each characteristic has on the price of a good, and interprets the resulting estimated coefficients on the characteristic v ariables to be the marginal price of a particular characteristic. The estimated marginal prices of various characteristics are then used to 94 extract the price difference arising from the overall pricing trend from the compound price changes. The methods of constructing a hedonic price index can be divided into two groups: direct and indirect. Direct methods can be further divided into the Time Dummy Variable method and the Characteristic method. I use the Time Dummy Variable (TD) method in this paper for the following reasons: i) Among all hedonic pricing methods, TD is most relevant to the property market because properties are sold less frequently than other products and have their unique characteristics, structural or locational 33 ; ii) TD method is relative ly easy to apply to panel datasets, as it only requires the inclusion of time dummy variables into the data set, and does not require matching of properties. The starting point is to construct a hedonic regression for all the property transactions in the f ive - county area from 2008 to 2016, as the TD method suggests pooling together transactions of different assessing districts to analyze the marginal prices of housing characteristics. This chapter uses a log - linear functional form of the hedonic regression. I have the prices of an individual property on the left - hand side (LHS) as the dependent variable and their characteristics on the right - hand side (RHS) as the explanatory variables. The hedonic regressions are estimated for the base year 2008 (t=0) and e - linear functional form for period t is given by: (1) is the selling price of property i in period t . is the intercept term that captures the time effect of the base period. is a vector of dummy variables that take 33 95 on the value of 1 if the transaction takes place in period t , and 0 otherwise, is a vector of 7 constants. Thus, the time effect of period t is represented by . is a vector of K price - determining characteristics for property i in period t . are the coefficients from the hedonic equation, and can be interpreted as the marginal price of the k - th characteristic. is the error term. The price - determining characteristics in the hedonic housing price function usually include property attributes, such as building size, building age, physical condition, location influence (positive or negative aspects associated with the parcel's location such as waterfront, airport, etc.). Considering the data constraints, I used the following attributes in this analysis: land area (square feet), number of bathrooms, garage area (square feet), the month of the transaction, and five - digit zip code. I would like to include other variables, suc h as building size , the total number of rooms, and building age, but data limitations pre vent the inclusion of these factors . However, based on the regression results presented later, the variables included in my analysis explain a large proportion of pric e variation. I offer a more detailed discussion regarding variables selection and the hedonic regression results in later sections. One assumption of the TD hedonic method is that the coefficients of the hedonic regression are constant over all time p eriods: for all t . This restriction assumes that the prices for each price - determining characteristic remain unchanged throughout the sample period, and among the whole state. This assumption is unlikely to hold precisely but has its adva ntages for i) making the model much more straightforward; and ii) making estimation of equation (1) possible even with a limited number of transactions e ach year in each assessing unit. Although in this particular 96 analysis, the assessing units with less th an 30 transactions are eliminated in the final analysis for their small sample sizes. Equation (1) illustrated the regression model: (2) where , , , and are the estimator of , , , and . I then evaluate the TD hedonic price index for each assessing district based on its average characteristics and the estimated parameters in equation (2). The computation is shown in equation (3), and details of the index construc tion process can be found in de Haan (2004): where = (3) where is the estimator of the quality - adjusted price index for assessing district h in period t , means parcel i is in assessing district h transacted in period t , is the selling price of property i in period t, is the number of parcel s sold in district h in period t . Thus, is the geometric averag e of selling prices within district h in period t. is the estimator of , which is the price effect of the k - th characteristic. = is the average of the jth characteristic in period t, and thus reflects the housing price changes due to the average changes of all characteristics from period 0 to period t. According to de Haan (2004), approximately equals to , which makes the interpretation very intuitive, as is the estimator of the difference between the time effect of period t ( ) and that of period 0 ( ). The price index can therefore be readily derived from the estimated 97 coefficients of the time dummy v ariable . The price index is 100* for each period t=1,..,T, with respect to base period t=0, which equals 100. 4. 3.2 SEV index I also create the SEV index, which reflects the fluctuation of average property assessed value and is comparable to the HPI index constructed in section 3.1. Similar to the HPI index , the SEV index is 100 for the base period t=0, which is 2008. And for each period within t=1,..,T, the SEV index is calculated as dividing the SEV per parcel in period t by the SEV per parcel in the base period, or (4) is the calculated SEV index for assessing district h in period t. is the total assessed value, or state equalized value for assessing district h in period t , is the total SEV for assessing district h in the base period 2008. is the total number of parcels in assessing district h in period t , is the total number of parcels in assessing district h in the base period 2008. 4. 3.3 Median Sales Ratio The median sales ratio is a widely used measurement of assessment precision, or assessment level in this profession (IAAO, 1990). It measures the degree to which goals or specific legal requirements are met. The median sales ratio in assessing district h in period t is the median assessed value over the sale price ratio of each parcel transacted in district h in period t . The media n sales ratio is annual and is based on the parcels sold in a given year. The closer the median sales ratio is to 1, the higher the assessment quality is in the assessing district. The ideal assessment would have a median sales ratio of 1, implying 98 that th e properties in the district are neither systematically ove r - assessed nor under - assessed. Please note that this only provides a general measure of assessment quality, other methods such as those used in Hodge, et al (2017) are beyond the scope of the prese nt analysis. 4.4 Data 4. 4.1 Data used to calculate Hedonic HPI I calculated the Hedonic Housing Price Index with the base year of 2008 ; in the 120 Michigan assessing units in the five - county region surrounding Detroit (Macomb, Monroe, Oakland, Washtenaw, a nd Wayne counties). The selection of this five - county region is consistent with Skidmore, Reese, and Kang (2012) in their analysis of tax competition effects of Michigan property tax, and is adopted for this research for two reasons: 1) This five - county re gion is the most densely populated area in Michigan and covers many of the most populated cities in the state, including Detroit, Ann Arbor, Sterling Heights, and Warren; This dense concentration of people and property parcels gives us confidence that we can estimate a n HPI for each local government jurisdictions; 2) this region includes Detroit, the city that witnessed one of the most drastic housing price changes du ring the 2008 financial crisis. The dataset used in the calculation of the Housing Price I ndex is individual parcel data for all single - family residential units obtained from CoreLogic, LLC., which covers nearly all the property parcel information in the five - county region. For each single - family residential parcel, there are records on the lat est transaction information, including sale price, and sale date, location information, including county, and local government belong to, address, census track, and zip code, and over 50 property attributes, including 99 building size, land size, number of ro oms, number of bathrooms, building age, building condition, number of stories, number of furnaces, etc. Out of the many property attributes, I include the following i n my analysis: land area (square feet), number of bathrooms, garage area (square feet), th e month of the transaction, and five - digit zip code. The inclusion of these variables is based on th e need to balance the data quality/availability and the number of observations included in my regression. I would prefer to add building size, the total num ber of rooms, and building age, but adding them would require omitting many observations because of low - quality and/or data unavailability. Robustness analyses using smaller samples with these additional variables suggest that inclusion of these other con trol variables may be unnecessary given the purpose of this paper because the number of bathrooms is closely related to building size and the total number of rooms , to some degree. Thus, the reduced variable specification, though potentially misspecified , generates relatively high explanatory power. Recall that the goal is to generate an HPI and not necessarily focus on coefficient estimates o f specific property characteristics . The specification used in my analysis generates an R - square of about 0.5 depend ing on the local unit for the hedonic regressions, thus accounting for a large portion of price variability over time for each local unit. Out of over 1,900,000 parcels in the five - county region, around 316,000 parcels are used in the analysis. The select ion of properties is based on the following criteria: 1) the land usage of the property is " single house family" since it is most appropriate to construct a housing pricing index; 2) the property has a recorded sales transactions between 2008 and 2016; 3) there are no missing values or er rors in the variables specified included the specification; 4) the transaction type is either "resale" or "sale of new 100 construction" (in order to eliminate of transactions where the sale price does not reflect true market v alue, such as refinancing); 5) the property with sales price larger than $1,000, and sales ratio, which is assessed value divided by sales price, smaller than 20 (to eliminate non - arm's length transactions). 34 The number of observations used in the hedonic regression and the number of total parcels of each county and assessing unit are shown in Table 4. 1 in the Appendix. The table shows the parcels used in the hedonic regression are around 10 to 20 percent of th e total parcels in the majority of assessing units. However, some regressions used a considerably smaller proportion of total parcels. For example, Ray township in Macomb county has 2267 parcels, but only 52 parcels, or 2.3 percent of total parcels , are in cluded in the regression. The reason for the low number of observations is due to the low data quality of these local governments . More specifically, a large proportion of the parcel level data in these governments are missing one or more attributes. Those governments using more than 10 percent of total parcels generally have high - quality data. In such governments, most parcels trimmed out are for the property type and lack of sales between 2008 to 2016, not for missing data. 4. 4.2 Data used to calculate SE V index I calculated a SEV per parcel index with the base year of 2008 for each of the 120 assessing districts and the five counties, in order to make a comparison with the HPI index constructed. The data used is an assessing district - level dataset obtaine d from the Michigan Department of Treasury. In its annual L - 4023 form, the Department of Treasury publishes the total assessed value of all properties and the number of total 34 Consistent with the criteria used in Chapter 3. 101 parcels of the state, each county, and assessing district. As mentioned in the pr evious section, the assessed value is 50 percent of its determined true cash value, which equals SEV in most municipalities . I calculate the SEV per parcel from 2008 to 2016 for each county and local government by dividing the total assessed value by the n umber of parcels. The SEV index measures fluctuations in the asses sed value for all parcels in an assessing district, not just for the single house family with a transaction from 2008 to 2016. Thus, the calculation involves all single house parcels in an assessing district, as opposed to the calculation of HPI, where only single - family residential properties sold between 2008 and 2016 are involved. 4. 4.3 Data used to calculate Median Sales Ratio I calculate the median sales ratio using individual parcel tax records and current sales data obtained from CoreLogic. The calculation of median sales ratio involves only parcels sold in the calculated year after trimming out non - arm's - length transactions, errors, and outliers. To ensure all sales ratios included in the calculation are correctly recorded arm's - length, open - market transfer, I eliminated all individual transactions with transaction types other than "resale" and "new construction ." Eliminated transaction types include non - arm's - length transactions, construction loan/financing, and seller carrybacks. Those transactions are not standard open - market transactions, and their sales prices are not proper reflections of their market values. I also elim inated all individual transactions with sales prices less than $1,000 or sales ratios greater than 20 to eliminate errors or non - arm's - length transactions. Since Michigan assesses property values at 50 percent of 102 market value, I calculate the sales ratios ; as two times the assessed value divided by market value. 4.5 Results and Discussion 4. 5.1 Time Dummies Hedonic Regressions The results of 124 - time dummies hedonic regressions are shown in Table 4. 1 in the Appendix, including 119 regressions for local gove rnments, and five regressions for the counties. The table shows the coefficients and standard errors of each year dummy, and each of the hedonic attributes of houses used in the regression, including the log of land area, the log of the garage area, and th e number of bathrooms. The coefficient of the sale year 2008 is not reported because 2008 is set to be the base year. The five - digit zip code dummies and sale month dummies are included in the regressions, but the coefficients are not shown for space reaso ns. Again, the parcels used in the hedonic regression are around 10 to 20 percent of total parcels in the majority of assessing units. However, some regressions only used a much smaller proportion, and thus the quality of the regression may be lower becau se: 1) the number of observations is smaller, which makes the regression less likely to generate reasonable - size effects with reasonable power ; 2) the sample may be unrepresentative because of selection bias. For example, in Monroe county, only Monroe town ship has satisfactory data, and thus the regression of Monroe county only reflects Monroe township. The average R - square for all regressions is 0.49, which offers a strong level of the explanatory power of the hedonic model. Note that the average R - square of the five - county regressions 0.64 is considerably higher than the average R - square. And if we 103 only count the three counties where the parcels used account for higher than 10 percent of total parcels (Macomb, Oakland, and Wayne), the average R - square becomes 0.72, which is even higher. One possible explanation for the results is that the sample in the county regressions are more varied, and a larger part of this variation can be explained by differences of attributes, including location differences. 4. 5.2 Comparison of Indices Table 4. 2 in the Appendix presents the housing price index, SEV index, and the median sales ratio for 108 local governments and the five counties. Indices for a few townships (including Lenox, Ray, New Haven, and eight other to wnships) are not shown because the hedonic regressions for these districts used less than 100 observations or 10 percent of total parcels, and thus less likely to generate a meaningful HPI. The left side of the following panel figure (Figure 4.1) shows the indices of three counties with high - quality data (Macomb, Oakland, and Wayne) . T he right side shows the city and township averages and two typical townships. In most districts, the housing price index went down dramatically in 2009, and remained nearly unchanged until 2011, and started to increase in 2012. Housing prices recovered to 2008 levels around 2013 or 2014 in most districts, and the rising trend continued through 2016. In 2008, SEV per parcel went down, but not as much as housing prices, which helped with tax revenue stabilization during the immediate crisis , t hough assessment delays also mean that property owners are paying too high a tax. However, SEV per parcel continued to fall until 2012 or 2013, to a level even lower than the lowest HPI le vel. I nterestingly, eight years past the financial crisis, assessed values are still far from their initial 2008 levels, though housing prices recovered to 2008 levels around 2013 or 2014. 104 Figure 4.1 Indices Comparison of Representative Counties and Cities This result is consistent with the literature that there is about a three - year lag from the housing price decline to tax assessment but challenges the literature that the elasticity of tax revenue to housing prices is far less than one. Although the property tax revenue is relatively s table in the short run, the problem is only delayed and not mitigated in the longer term. Assessed property values decreased more slowly than housing prices did in 2009, but fell to an even deeper level than housing prices when we look further out in 105 time. In fact, the problem becomes more severe because generally the lowest level of SEV per parcel is even lower than that of housing prices. 4. 5.3 Different behavior of over - estimating governments Anecdotally, the behavior of over - estimating governments is very different from other governments. In the 108 local governments reported in Table 4. 2 in Appendix, the average initial median sales ratio in 2008 was 1.83, which indicates a systematic overestimation in general. However, there are a few local governme nts that over - estimate to a much higher degree, including Detroit, Inkster, River Rouge, Ecorse, Highland Park, and Pontiac, with an initial median sales ratio of 6.9, 5.1, 7.8, 6.7, 6.3, and 6.1 , respectively. While the SEV index in most other local gover nments started to pick up around 2013 or 2014, these six cities never turned around, and where SEV continues to fall until 2016. The median sales ratio also continues to fall during the entire period in the six over - estimating governments . This finding is opposite to the other governments, In the following panel figure (Figure 4.2) , I compared the housing price and assessment behavior of two large over - estimating cities, Detroit and In kster, with two other large cities, Ann Arbor, Clinton. It is clear that the SEV index decreases from 2008 to 2016 in both Detroit and Inkster, and the decrease is quite sharp, without any signs of stopping. On the other hand, the SEV index reached its low est point in 2012 or 2013 in Ann Arbor and Clinton, which is similar to most other cities. This pattern demonstrates that assessments during and following the crisis served as a self - correcting scheme, as over - estimating governments took the housing price declines as a n opportunity to correct its over - assessment condition. 106 Figure 4.2 Comparison of Overestimating and Other Cities Th e crisis served to help correct the variation of assessment precision as well. From Table 4. 2 in the Appendix, we see that the range of median sales ratio is wide in 2008, from 0 35 to 7.8, but is considerably narrowed in 2016, from 0.2 to 3.9, with over 90 percent of local units falling between 0.8 and 1. Previously before the crisis, Michigan failed to assess properties a cross local governments equally : some places are way overestimated than others. As financially stressful governments likely to overestimate properties, this inequality can cause the state and local tax burden to be regressive, meaning that people in financ ially unhealthy cities such as Detroit (usually lower - income households) are paying a higher share of taxes. 36 After the crisis, this inequality is 35 York city in Washtenaw county. 36 Lower - m ore expensive properties are undervalued, while less expensive properties are overvalued , resulting that most state and local governments collect a larger share of the income of lower - income households than of upper - income households. - Income Americans Get Cheated on 107 significantly alleviated, meaning that people across different local governments and different financial backgrounds are now treated more equally than before . 4.6 Conclusion M y findings are generally consistent with the literature that there is about a three - year lag between the housing price changes and changes in property assessments for tax purposes . Howev er, my research also suggests th at the earlier showing that the elasticity of property assessment (and thus property tax revenue) to housing prices is far less than one may not apply to Michigan. A ssessed property value decreased more slowly than housing p rices fell immediately after the crisis, but ultimately fell to an even deeper level than housing prices when we consider a longer time horizon. In other words, the decline in housing prices passes through completely to assessed values, and the elasticity might be greater than one , given enough time. Before the crisis, a number of Michigan communities tended to overestimate property values to support property tax revenue streams. I find that assessment adjustments to market declines in these overestimating places lagged in other fiscally healthier cities. On a positive note, my evaluation also shows that local authorities in these struggling communities used the financial crisis period as an opportunity to re align assessments more closely to actual market co nditions as per legal requirements. While the SEV index in most local governments began to once again grow in the wake of the crisis by 2013 or 2014, SEVs in struggling local governments such as Detroit continued to fall through 2016. https://www.nytimes.com/2021/04/03/opinion/sunday/property - taxes - housing - assessment - inequality.html 108 Given that SEV per parcel for most districts has not risen to their 2008 levels, it seems that the impact of the 2008 financial crisis on property assessment had not concluded until about 2016. Going forward, it would be interesting to extend this research with more recent data to examine whether the median sales ratios remained closely aligned with legal requirements. T o further explore the impact of housing price changes on property assessment, it would be interesting to calculate each assessing elasticity of assessed value regarding housing price changes in a relatively long term . The next step is to compare different districts , to see whether some districts have higher elasticity than other districts, and find out the reasons for pote ntial variations. Who are the districts responding mostly to housing price changes, and is there any geographic clustering of such districts? Why are they responding more drastically than others, is such response associated with the social and economic cha racteristics of the districts? Answering these questions would offer a more comprehensive view of Michigan property assessment and thus shed some light on how to improve property assessments in general. 109 APPENDIX 110 Table 4.1 Results of Time Dummies Hedonic Regressions 111 Table 4.1 ( Cont d ) 112 Table 4.1 113 Table 4.1 114 Table 4.1 115 Table 4.1 116 Table 4.1 117 Table 4.1 118 Table 4.2 Comparison between HPI and SEV index (base year = 2008) 119 Table 4.2 120 Table 4.2 121 Table 4.2 122 CHAPTER 5 : Conclusion In this dissertation , I examine and analyze several aspects of Michigan property assessment practice. In the first essay, I estimated a Michigan property assessment expenditure func t i on using data for most Michigan local government s. Assessment is more efficient when an assessing office is working with a larger number of propert i e s . For a local unit with 1700 parcels (the average size local unit in my dataset), each one percent increase in parcel number s will reduce assessment expenditure per parcel by 0.5 3 percent. Each one percent increase in average property assessed value is associated with a 0. 54 percent increase of assessment expenditure per parcel. My analysis suggests that hiring private assessors - assessors does not necessarily result in expenditure reduction . In the second essay, I evaluate the performance of three types of property - s, private assessors, and county assessor s , to test the hypothesis that neither the public n or the private sector is inherently superior to the other in the case of Michigan tax assessment . I also attempt to determine assessor type respective comparative advantages , identify a more optimal allocation of assessors, and compare a more optimal allocation to the existing arrangement to determine the degree of misallocation. Results show : (i) Neither private nor public assessors is alway s superior n or inferior to the other, and they both have relative strengths within different types of communities; (ii) they both tend to perform better in larger communities and communities with higher average property values; and (iii) there appears to b e a 123 misallocation of assessors in Michigan. Private contractors perform better in communities with more parcels but tend to be employed in communities with fewer parcels. Based on the results, I recommend merging small er assessing districts in Michigan to form larger assessing areas, and then allocate different types of assessors to move toward a more optimal arrangement. Scenario forecasts suggest that following my recommendation c ould improve assessment performance by 32 percent as compared to the current situation. In the last essay, I offer dynamic analysis of Michigan property assessments and connect assessment practices with housing price fluctuation s by construct ing and comparing a Hedonic Housing Price Index (HPI) and a State Equalized Value (SEV) index for 120 Michigan assessing units in the five - county region surrounding Detroit. My findings are generally consistent with the literature that there is about a three - year lag between housing price changes and changes in pr operty assessments for tax purposes. However, I also find that before the crisis struggling communities tended to over - estimate assessments to support property tax revenue streams. Assessment adjustments to market declines in these places also lagged in ot her fiscally healthier cities. On a positive note, the evaluation shows that local authorities in these struggling communities used the financial crisis period to align assessments more closely to actual market conditions as per legal requirements. While the SEV index in most local governments began to once again grow in the wake of the crisis in 2013 or 2014, SEVs in struggling local governments such as Detroit continued to fall through 2016. Several policy implications stem from this dissertation. First, my analysis suggests merg ing small assessing districts to form larger assessing areas to help reduce 12 4 expenditure s and improve productivity. I base thi s conclusion on the following: (i) as shown in the first essay , assessment expenditure per parcel is sma ller in bigger communities, and th e impact of size becomes more substantial the larger is the community ; and (ii) the second essay indicates, all types of assessors tend to perform better in larger communities. I also recommend that assessor types be reall ocated in a way th at improves productivity . Specifically, improvements can be achieved by using private contractors in communities containing more than 50,000 parcels and using in - house assessors in the rest of the communities. The reasons for this suggested allocation are: (i) as the first essay indicate s , hiring different types of assessors does not affect assessment expenditure ; and (ii) as the second essay indicates, a re allocation of assessors could potentially significantly enhance ass essment performance. However, the jurisdictions might have reasons for not taking optimal assessing regime s , as discussed in this paper. For example, they might be foregoing assessment performance for more control to the assessors, to deliberately over - or under - states properties . I t would be interesting to extend this research with focuses on government assessing objectives. Looking forward, I will continue to pursue research on Michigan property assessment. O ne possibility venue for further study is to explore the overestimation of property value in Michigan. With the calculated median sales ratio of each local government, it would be interesting to offer a dynamic overview of wh at types of local governments over - or under - assess propert y . In particular, I plan to map the geolocation of local governments with their median sales ratio, explore potential clustering of over - estimating governments, as well as identify trends in assessment over time. I also plan to 125 analyze what types of governments are more li kely to over - or under - state properties . This proposed additional analys i s will be an extension of this dissertation . Another possibility is to rank Michigan local governments by their technical efficiency in property assessment. An innovative meth odology Stochastic Frontier Model was used in this dissertation to measure the influences of various determinants o f assessment performance. It would be valuable to extend this analysis with the same methodology to calculat e technical e fficiencies for each government . This research will combine both the expenditure and performance components of assessment, and offer a more comprehensive view of Michigan property assessment. 126 REFERENCES 127 REFERENCES ing for Public Service. Urban Affairs Quarterly . 9 (3): 337 - 58. . 16 (1) : 1 - 1 6 . Allen, Joan W., Keon S. Chi, Kevin A. Devlin, Mark Fall, Harry P. Hatry, and Wayne Masterman . ( 1989 ) . The Private Sector in State Service Delivery: Examples of Innovative Practices. Washington, D.C.: The Urban Institute Press. Alm, J., Buschman, R. D., & Sjoquist, D. L. (2011). Rethinking local government reliance on the property tax. Regio nal Science and Urban Economics. 41(4) : 320 - 331. Anderson, N. (2006). Property tax limitations: An interpretive view. National Tax Journal, 59(3) : 685 - 694. Aigner, D. J., Lovell, C. A. K., and Schmidt, P. (1977). Stochastic Fro ntier Production Function M 6(1): 21 - 37. Bagi , F. S. and Huang, C. J. (1983). Estimating Production Technical Efficiency for Individual Farms in Tennessee. Canadian Journal of Agricultural Economics/Revue canadienne d'agroe conomie . 31 (2) : 249 - 256. Barr, Stephen and Bill McAllister. (1997) . Downsizing Cuts Federal Union Representation; More Postal Employees Covered. The Washington Post . Battese, G. E., and Coelli, T. J. (1998). - Level Technical Efficienc y with a Generalized Frontier Production Function and Panel Da Econometrics. 38 (3): 387 - 99. Battese, G.E. and Corra, G.S. (1977) . Estimation of a Production Frontier Model: With Application to the Pastoral Zone off Eastern Australia. Australian Journal of Agricultural Economics . 21 (3) : 169 - 179. Bell, E. (1984). Administrative Inequity and Property Assessment: The Case for the Traditional Approach. Property Tax Journal . 3(2): 123 - 131. Bendick, Marc. ( 1989 ) . Privatizing the Deli very of Social Welfare Service. Privatization and the Welfare State edited by Sheila B. Kamerman and Alfred J. Kahn. Princeton: Princeton University Press. 128 Bernstein, Nina. (1996) . Giant Companies Entering Race to Run State Welfare Programs. The New York Ti mes. Sunday September 15, 1996, section 1, page 1. Bowman, John H., and John L. Mikesell. (1978) Returns from Institutional Remedies: 31 ( 2 ) : 137 - 52 . Brunori, D. (2007). Local tax policy: A federalist perspective (2 nd ed.). Washington, DC: Urban Institute Press. Butler, Stuart M., ed. ( 1985 ) . The Privatization Option: A Strategy to Shrink the Size of Government. Washington, DC: The Heritage Foundation Cameron , Trivedi . (2005) . Microeconometrics: Methods and Applications . Cambridge: Cambridge University Press. Case, K. E., & Shiller, R. J. (1989). The efficiency of the market for single oÀ family homes. American Economic Review, 79(1) : 125 - 137. Cheng, P. (1974). Property Taxation, Assessment Performance, and Its Measurement. Public Finance . 29(3): 268 - 284. Chernick, H., Langley, A., & Reschovsky, A. (2011). The impact of the great recession and the housing crisis on the financing of America's largest cities. Regional Sc ience and Urban Economics . 41(4) : 372 381. Clapp, J. (1990). A New Test for Equitable Real Estate Tax Assessment. Journal of Real Estate Finance and Economics . 3(3): 233 - 249. Cornia, G. C., & Walters, L. C. (2006). Full disclosure: Controlling property tax increases during periods of increasing housing values. National Tax Journal . 59(3) : 735 - 749. David, I. ( 1987 ) . Privatization in America. Washington, D.C.: Touche Ross. es . 20 (1) : 223 - 26. . 14 (1) : 149 - 65. De Haan, J. (2004). Hedonic Regression: The Time Dummy Index As a Special Case of the Imputation Törnqvist Index. Paper Presented at the 8th Ottawa Group Meeting, Helsinki. 2004b. Public Choice . 34(3 - 4): 381 - 97. 129 and Scope in Rural Low - . 70 ( 3 ) : 459 - 65. Doerner, W. M., & Ihlanfeldt, K. R. (2011). House pric es and city revenues. Regional Science and Urban Economics . 41(4) : 332 - 342. Donahue, John D. ( 1989 ) . The Privatization Decision: Public Ends, Private Means. New York: Basic Books, Inc. Dorsey, R. E., Hu, H., Mayer, W. J., & Wang, H. oÀ C. (2010). Hedonic versus repeat oÀ sales housing price indexes for measuring the recent boom oÀ bust cycle. Journal of Housing Economics . 19(2) : 75 - 93. Public Production, with an Application to Fire Economics . 52 ( 1 ): 49 - 72. Goodman, C. (2018). House prices and property tax revenues during the boom and bust: Evidence from small oÀ area estimates. Growth and Change . 49(4) : 1 - 21 . Greene . (2012) . E conometric A nalysis . (7 th ed), Upper Sa ddle River, N.J: Prentice Hall. Gurin, Arnold. ( 1989 ). Governmental Responsibility and Privatization: Exa mples for Four Social Services. Privatization and the Welfare State, edited by Sheila B. Kamerman and Alfred J. Kahn. Princeton: Princeton University Press. Hatry, Harry P. ( 1983 ) . A Review of Private Approaches for Delivery of Public Services. Washington, D.C.: Urban Institute Press. Hatry, Harry P. and Eugene Durman. ( 1985 ) . Issues in Competitive Contracting for Social Services. Falls Chur ch, VA: National Institute of Governmental Purchasing, Inc. Hilke, John. (1993). Cost Savings from Privatization: A Compilation of Study Findings. Los Angeles, CA : Reason Foundation . efuse . 47 (1) : 87 - 93. Hodge, Tim and McMillen, Daniel and Sands, Gary and Skidmore, Mark L. (2017). Estate Economics . 45(2): 237 - 258. Hughes, Polly Ross. (1997) "Ads Seek to Block Welfare Privatization ," Houston Chronicle, April 17 , page A1. 130 Ihlanfeldt, K. R. (2011). How do cities and counties respond to changes in their property tax base? The Review of Regional Studies . 41( 1) : 27 - 48. International Association of Assessing Officers (IAAO). (1978). " Improving Real Property Assessment: A Reference Manual. Chicago: IAAO. Jeter, Jon. (1997). "A Winning Combination in Indianapolis." Th e Washington Post, September 21 . Jorge Martinez - Administration . 33 ( 12 - 13 ): 601 - 19 . K. P. Kalirajan & J . C. Flinn. ( 1983 ) . The Measurement Of Farm - Specific Technical Efficiency. Pakistan Journal of Applied Economics, Applied Economics Research Centre . 2(2 ): 167 - 180. Kenneth L. Wertz. (1979). - Administering Agenc National Tax Journal . 32(2) : 143 - 56 Kinsey, M arcia. (1997). Privatization. Austin, TX: Cente r for Public Policy Priorities. . 42(2) : 213 - 233. Is small beautiful? Village level taxation of natural resources in . 27(4) : 307 - 18. Lutz, B. F. (2008). The connection between house price appreciation and property tax revenues. National Tax Journal . 61(3) : 555 - 572. Lutz, B. F., Molloy, R., & Shan, H. (2011). The housing crisis and state and local government tax revenue: Five channels. Regional Science and Urban Economics . 41(4) : 306 - 319. Mark Sunderman, John Birch, Roger Cannaday, and Thomas Hamilt on . (1990) . Testing for Vertical Inequity in Property Tax Systems. Journal of Real Estate Research . 5( 3 ): 319 - 334. Meeusen, W., and van den Broeck, J. Efficiency Estimation from Cobb - Douglas Production Functions with Composed Error. Internation al Economic Review . 18 (2) : 435 - 44. Mehta, Shekhar; Giertz, J Fred . (1996). Measuring the Performance of the Property Tax Assessment Process. National Tax Journa l . 49 ( 1 ): 73 - 85. Michigan Assessed at August 2017 : 131 https://www.michigan.gov/documents/treasury/State_Tax_Commission_General_Ru les_276567_7.pdf Michigan State Tax Commission. August 2017 : https://www.michigan.gov/documents/treasury/STC_How_to_Become_an_Assessor _in_Michigan_555268_7.pdf n: Quality Effects of . 11(1) : 101 - 116. Mikesell, J. L., & Liu, C. (2013). Property tax stability: A tax system model of base and revenue dynamics through the great recession and beyond. Public Finance & Management . 13(4): 310 - 334. New York Times Editorial Board - Income Americans Get Cheated The New York Times. Sunday April 3 , 2021. Accessed at: https://www.nytimes.com/2021/04/03/opinion/sunday/property - taxes - housing - assessment - inequality.html Osborne, David and Ted Gaebler. ( 1992 ) . Reinventing Government: How the Entrepreneurial Spirit is Transforming the Public Sector. Reading, Mas s.: Addison - Wesley Publishing Co., Inc. Paglin, M. and M. Fogarty. (1972). Equity and the Property Tax: A New Conceptual Focus. National Tax Journal . 25(4): 557 - 565. Ross, J., & Yan, W. (2013). Fiscal illusion from property reassessment? An empirical test of the residual view. National Tax Journal . 66(1) : 7 - 32. Schmidt, P., and Sickles, R. C. (1984). Production Frontiers and Panel Data. Journal of Business and Economic Statistics . 2 (4): 367 - 74. Schlesinger, M., R. Dorwart, and R. Pulice. (1986). Policy Analysis and Management . 5(2):245 - 63. Silver, M. (2016). How to better measure hedonic residential property price indexes. IMF Wor king Paper. 16 ( 213 ): 1 - 89 . Sjoquist, David L., and Mary Beth Walker. (1999). "Economies of Scale in Property Tax Assessment." National Tax Journal . 52 ( 2 ) : 207 - 20. Skidmore, M., Reese, L., & Kang, S. H. (2012). Regional analysis of property taxation, educ ation finance re Regional Science and Urban Economics . 42( 1 2 ): 351 - 363. 132 Stevens, B. (1984). Delivering Municipal Services Efficiently: A Comparison of Municipal Services Efficiently: A Comparison of Municipal and Private Service Delivery. Washington, D.C.; U.S. Department of Housing and Urban Development. Stocker, F. (1973). Revenue System. Chicago: International Association of Assessing Officers. Cited in R. W. Poole, Jr., Cutting Ba ck City Hall. New York: Universe Books. Strumpf, K. S. (1999). Infrequent assessments distort property taxes: Theory and evidence. Journal of Urban Economics . 46(2) : 169 - 199. Sunderman, M., J. Birch, R. Canady and T. Hamilton. (1990). Testing for Vertical Inequity in Property Tax Systems. Journal of Real Estate Research . 5(3): 319 - 334. Vlaicu, R., & Whalley, A. (2011). Do housing bubbles generate fiscal bubbles? Public Choice . 149(1 2) : 89 - 108. icipal Police Services: The . 54( 4 ) : 431 - 8. Wooldridge, Jeffrey M. (2 012). Introductory econometrics : a modern approach. Mason, Ohio : South - Western Cengage Learning.