Policy Research Working Paper 9399 Determinants of Property Tax Revenue Lessons from Empirical Analysis Rajul Awasthi Tuan Minh Le Chenli You Governance Global Practice & Macroeconomics, Trade and Investment Global Practice September 2020 Policy Research Working Paper 9399 Abstract Many developing countries have struggled with realizing Economic Co-operation and Development for 2006 to sufficient revenues from property tax. However, as devel- 2016, using a fixed effects model. The results show that oping countries experience economic growth, they are also increases in gross domestic product and population lead to seeing property values rising, providing a bigger tax base increases in property tax revenue and an increase in federal from which to realize revenues. Technology has made tax transfers decreases it. The outcomes of the empirical analysis administration easier and more effective and developing highlight the statistically significant impacts on property tax country governments have been improving their quality collection of a country’s state of development and its demo- of governance and considering introducing or enhanc- graphic, fiscal, and property tax–specific characteristics. A ing property tax revenue collection to diversify their tax critical question for further research is whether and how and fiscal revenues. This paper explores the determi- the empirical methodologies and specifications as applied nants of property tax revenue using data from the United to the set of developed economies would be replicated in States, Canada, Australia, Chile, and the Organisation for the context of developing countries. This paper is a product of the Governance Global Practice and the Macroeconomics, Trade and Investment Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank. org/prwp. The authors may be contacted at rawasthi@worldbank.org and tle@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Determinants of Property Tax Revenue: Lessons from Empirical Analysis Rajul Awasthi, Tuan Minh Le, Chenli You JEL Code: H71 Keywords: Local Government Revenue, Local Government Subsidies, Local Government Taxation, Local Tax, Property Tax, Sales Tax, State Government Subsidies, State Lottery, State Revenue, State Tax, Taxation Rajul Awasthi is a Senior Public Sector Specialist in Governance and Task Team Leader for the Strengthening Property Tax Systems in Developing Countries (ID: P169109) project. Tuan Minh Le is a Lead Economist in Macro, Trade, and Investment. Chenli You is an economist and a consultant. Determinants of Property Tax Revenue: Lessons from Empirical Analysis I. Introduction Property tax revenue plays an important role in tax collection in developed countries. It becomes revealing, theoretically and empirically, that property tax is the most efficient tax, encompassing a number of other virtues of a tax instrument: transparency, equity, and direct linkage to benefits. 1 The tax has become even more relevant in the evolving trends of urbanization and fiscal decentralization. 2 In some of the high-performing countries, the immovable property tax brings in as much as 3 percent of GDP in revenues. At the same time, many developing countries have struggled with realizing sufficient revenues from property tax. However, as developing countries experience economic growth they are also seeing property values rising, providing a bigger tax base from which to realize revenues. Technology has made tax administration easier and more effective and developing country governments have been improving their quality of governance and have been considering introducing or enhancing property tax revenue collection to diversify their tax and fiscal revenues. Taxation of immovable properties not only improves the tax revenue collections but also works as an instrument for income redistribution, given that it is progressive by nature. Exploring the determinants of property tax revenue using data from developed countries gives us important insights into understanding the extent to which these factors impact property tax collection. This research provides useful experiences for developing countries that are considering or planning to enforce the property tax collection. In this research, we review literature that has focused on the determinants of property tax revenue throughout the world. We then collect state/provincial-level data for the United States, Canada, Australia, and Chile, as well as country-level data for all the OECD countries. Our data allow us to build up a panel data set for sophisticated econometrics regressions and estimations. We use OLS, Fixed Effects Model, and Random Effects Models to estimate the impact of various determinants on the property tax revenue. Our econometric analysis allows us to measure the impact of key economic and demographic factors that influence property tax revenue. According to our research results (see Tables 5 and 6), we find that every 1 percent increase in GDP and population will increase the property tax revenue by around 0.13% and 0.76% on 1 OECD (2010), for example, ranks the order of efficiency (or least economic distortion, growth-impeding) tax types: property tax, VAT, personal income tax and corporate income tax. 2 Bahl (1999) reflects the criticality of property taxes in its direct linkage to efficiency in service delivery at the local government level. 2 average, respectively. 3 In the meantime, a 1 percent decrease in federal transfer and one person decrease in household size will increase tax revenue by 0.03% and 84%, respectively. Besides these determinants above, if the property tax rate increases by 1 percentage point, the tax revenue will increase by 8%. We attempted to measure the impact of governance on property tax revenue but found that the governance indicator does not have a significant impact on property tax revenue. 4 II. Literature Review A series of research has focused on the importance and function of property and land tax revenues. After reviewing the history of the property tax in the United States, Wallis (2000) points out that the advantage of local governments is the ability to match taxpayers and beneficiaries. Since local governments are able to utilize the benefit tax features of the property tax, the property tax has been, and will continue to be, the main source of revenue for local governments in the United States. Bahl, FAO (2002) concludes that the demand for tax revenue will likely be growing for local governments as infrastructure needs and service demands grow. In the meantime, most major taxes – collected at the federal level in most countries -- are under pressure from international competition and the limits of public acceptance. So, there are limits to tax revenues and hence, fiscal transfers from federal to local governments. Consequently, property tax revenues become important as the potential yields can be increased in most countries. International Handbook of Land and Property Taxation (2004) reviews the taxation of land and property in 25 countries (five in each of five regions – OECD, Central and Eastern Europe, Asia, Africa, and Latin America). The handbook concludes that taxes on land and property are at best minor revenue sources in all countries studied. But property taxes are useful sources of subnational revenue in many countries, and more so in developing than in developed or transition countries. The main functions of the property taxes include financing local governments and affecting land use. A policy guide for land and property tax by UN-Habitat (2011) concludes that the potential of Land and Property Tax (LPTs) to contribute to the improvement of local communities is high. The realization of the potential requires policies and administrative procedures to be efficient to produce a fair and stable tax system that would yield between 1 and 2 percent of GDP on a 3 The results are calculated from the regression results in Tables 5 and 6, Column 4. The numbers are the average of the two coefficients from the two tables for each variable. 4 Our empirical results may underly a caveat when selecting the governance indicator for the sample of developed countries. The narrow variation in governance quality may not warrant the otherwise statistical impact on property tax collection. The selection of the sample thus indicates the trade-off between a reliable overall data set and the missing variation in some likely plausible variable. 3 sustainable basis. It requires that the policies and administrative procedures are adapted to the cultural views of property rights, to the ways in which property rights are acknowledged and defended in the communities, to the realities of land and property markets and the administrative capacities of the governments. Property taxation practices and the determinants of the tax across the world have also been explored by researchers. De Cesare (1998) concludes that public investment in urban areas often results in increased land value, benefiting only a small group of private owners. However, in a pioneering initiative, the city of Porto Alegre in Brazil uses the property tax as an instrument for capturing land value increments, which deters land speculation and promotes rational urban development. Franzsen (2003) briefly overviews the property tax systems in five of the member states of the Southern African Development Community (SADC): Botswana, Lesotho, Namibia, South Africa and Eswatini. The author finds that the property tax is not utilized optimally in any of the five countries. However, it is generally recognized that property tax could and should become a more important source of revenue for all five countries. It is crucial to improve capacity in the areas of professional, technical and management skills, training, computerization, collection and enforcement procedures. In a 2012 report, the Lincoln Institute of Land Policy proposed three categories of factors affecting revenue generation: socioeconomic (i.e. size of real estate stock, real estate prices, poverty levels, income distribution, urbanization, and land tenure), fiscal decentralization (i.e. intragovernmental transfers and subnational revenue), and institutional arrangements (e.g. sales tax). Bonet, Muñoz, and Pineda (2014) elaborate an approach to capture heterogeneity in property tax revenue across provinces within Argentina, using province GDP per capita, dependency on national transfers, export rights revenue, informality of house tenancy, property tax decentralization, and other taxes. Manaf, Hasseldine, and Hodges (2006) analyze the determinants of Malaysian land taxpayers’ compliance attitudes. They test a compliance model for land tax in Malaysia with four previously untested variables (race, positive incentives, land type and location) to identify characters that are associated with landowners’ compliance attitudes. The regression results show that age, race, level of education, level of income, occupation and ethics strongly impact land tax compliance attitudes. Martinez-Vazquez and Youngman (2008) draw four main conclusions from property tax practice experiences from developing and transitional countries. First, during a transition period, second- best approaches could be the right strategy. Market-assisted information could be an acceptable substitute for actual transaction data for valuation. Second, institutions matter greatly. It is essential for developing a workable property, investing records management, identifying the right collection machinery, and finding a way to gather reliable data on sales transactions. Third, accurate valuation of the tax base is the crucial to success with the property tax. Finally, and most importantly, the property tax can be a key point to strengthening local government finance, promoting rural development, and improving the fairness of the tax burdens distribution. 4 Literature has also used panel data across the world to discuss sophisticated methodologies of verifying determinants of the property taxes and estimating their impacts on the property tax revenue. Davoodi and Grigorian (2007) attribute urbanization's importance to high demand for public services. Using data from Office of Real Property Tax Service and Office of State Comptroller’s Office of New York State in 1992, Tae Ho Eom (2008) examines a model for the determinants of property tax assessment quality. The author uses two stage least square (2SLS) regression to solve the endogeneity of the property tax administration variables and concludes that “the median tax share, the median property value as a share of median income, the ratio of state aid to total expenditures, and the share of adults with college or higher education have significant impacts on the quality of property tax assessment”. Sepulveda and Martinez−Vazquez (2012) estimate the determinants for property tax collection for nine Latin American countries during 1990-2007. They use fiscal decentralization, dependency of transfers, government size, income per capita, municipal cadaster, competition for public positions, and index of democracy as main independent variables. They find that fixed effect results yield statistically significant effects for fiscal decentralization, dependency on transfers, GDP per capita, competition for public positions and index of democracy. And for the random effects model, the rest of the variables are statistically significant (urbanization, municipal cadaster, government size). Hultquist and Petras (2012) employ a panel of 635 year-location observations covering the largest city in each U.S. state from 1997-2010 to measure the impact of the various political, fiscal, administrative, demographic, and economic factors on the use of fractional assessment of local property taxation. They use panel data techniques incorporating both year and location-specific fixed effects and conclude that fractional assessment is associated with political factors, such as higher levels of Republican legislative representation and divided government, suggesting that the fractional assessment of local property taxation may be the path of least resistance for enacting effective property tax relief. Norregaard (2013) discusses policy and administrative issues to be considered for successful reform. Based on panel data for 64 countries between 1990−2010, he uses GDP per capita, urbanization, openness of the economy, and a dummy if the country's legal heritage (is of Anglo−Saxon origin) to estimate recurrent immovable property revenue. He also estimates a Fixed Effects model to account for the influence the variations of income per capita have on property tax collection over time. He finds that a country's development over time has positive exponential effects over time. III. Data 5 We use the following data from different countries for our research. To run the econometrics model with panel data, we collect data through years 2006 to 2016. • United States: data within the United States are mainly from U.S. Census Bureau. According to the literature, the most important variables include property tax revenue, GDP, population, public finance (i.e. federal transfers 5), property tax rate, household size, and governance quality. 6 U.S. data have the most variables and the best quality among the data we have collected. • Canada: data are mainly collected from Statistics Canada, the statistics department of the government. Main variables include property tax revenue, GDP, population, public finance (i.e. federal transfers), and household size. Besides the United States, Canada is the only country that provides clear federal transfer data publicly online. • Australia: Australian Bureau of Statistics has the main variables we need, including property tax revenue, GDP, population, and household size. • Chile: variables for Chile are also property tax revenue, GDP, population, and household size, the same as Australia. And they are collected from the National Statistics Institute. • OECD: data on OECD countries are all from the official OECD website. Variables are the same as for Australia and Chile, including property tax revenue, GDP, population, and household size. IV. Methodology Most of the previous research simply use pooled Ordinary Least Squares (OLS), which is very likely to cause endogeneity (e.g. omitted variable bias), due to the different environments and situations in different countries (or states/provinces within a country). Since there exist some structural differences between countries, they are not consistently comparable using pooled OLS model. Therefore, we utilize Fixed Effects model 7 to estimate the impact of the important factors on property tax revenue, using only within-country (or within states/provinces) variations. This would provide a better estimator for the key determinants of property tax revenues even allowing extrapolation for developing countries. 5 Revenues of state and local governments that are received from federal government transfers. They are redistribution of income and wealth by means of the federal government making a payment, without goods or services being received in return. 6 The overall scores of the economic competitiveness for each state from The Beacon Hill Institute. The scores range from 1.9 to 8.1, with higher scores indicating better governance quality. 7 A fixed effects regression is an estimation technique employed in a panel data setting that allows researchers to control for time-invariant unobserved individual characteristics that can be correlated with the observed independent variables. It uses within-individual variations, so it is usually called “within estimator”. In our model, it controls for time-invariant unobserved individual characteristics within states/provinces in United States, Australia, Canada, and Chile, as well as within OECD countries. So, the results can be recognized as within- state/province/country estimators. 6 Due to the specialty of the data availability, we run our regressions on two data samples. Firstly, we only work on U.S. data, with 50 states across 11 years (2006-2016). As U.S. data have the most important variables and best quality, we are able to take advantage of these estimations to better understand the determinants of property tax revenue. We use both OLS and Fixed Effects models as shown in Equation (1) - (4). 8 Secondly, to utilize all the data we have, we combine data from all the countries, including Australia, Canada, Chile, USA, and OECD. State and province level data from Australia, Canada, and the United States are also pooled together to construct our panel data set. Since our data availability is limited, we can only use a few control variables in these specifications, such as GDP, population, and household size (we use government transfer for both the United States and Canada). Details are shown in Equations (5) - (8) 9 as below. Property Tax Revenue = constant + 1 . GDP + 2 . Population + 3 . Government Transfer + 4 . Property Tax Rate + 5 . Household Size + (1) Property Tax Revenue = constant + 1 . GDP + 2 . Population + 3 . Government Transfer + 4 . Property Tax Rate + 5 . Household Size + 6 . Governance + (2) Property Tax Revenue = constant + 1 . GDP + 2 . Population + 3 . Government Transfer + 4 . Property Tax Rate + 5 . Household Size + Year Dummy + State Dummy + (3) Property Tax Revenue = constant + 1 . GDP + 2 . Population + 3 . Government Transfer + 4 . Property Tax Rate + 5 . Household Size + 6 . Governance + Year Dummy + State Dummy + (4) Property Tax Revenue = constant + 1 . GDP + 2 . Population + 3 . Household Size + (5) Property Tax Revenue = constant + 1 . GDP + 2 . Population + 3 . Household Size + 4 . Government Transfer + (6) Property Tax Revenue = constant + β1 . GDP + β2 . Population + β3 . Household Size + Year Dummy + State/Province/Country Dummy + ϵ (7) Property Tax Revenue = constant + 1 . GDP + 2 . Population + 3 . Household Size + 4 . Government Transfer + Year Dummy + State/Province/Country Dummy + (8) Property Tax Revenue is total property tax revenues as million dollars; GDP is GDP as million dollars; Population stands for the population as thousands of people; Government Transfer is the government transfer as million dollars; 8 Equations 1 and 2 are OLS, while Equations 3 and 4 are fixed effects. 9 Equations 5 and 6 are OLS, while Equations 7 and 8 are fixed effects. 7 Property Tax Rate is the tax rate as percent of the value of the property 10; Household Size is the average number of persons living in each household; Governance Quality is the overall quality of governance. To better deal with the variables, we have conducted logarithm transformation on property tax, GDP, population, and government transfer. This serves three main purposes: (1) to solve the highly skewed distributions issue; (2) to directly estimate elasticities of property tax revenues by various determinants; and (3) to control for the wide variation in these variables. We also include both regional and time dummies. The regions are countries, states, and provinces. The time dummies are annual. According to economic intuition and literature, GDP and population are expected to have positive impacts on property tax revenue. Higher economic growth leads to an increase in demand for properties which then become more expensive and the size of the tax base becomes greater. An increase in population means an increase in demand for housing and increase in number of properties, again increasing the size of the tax base. Both factors lead to higher property tax revenue. As far as the property tax rate is concerned, clearly a higher rate is positively correlated with higher revenues. On the contrary, household size and government transfers are expected to have negative impacts on property tax revenue. As household size increases, more persons live in each property, reducing the demand for properties and consequently the tax base. Federal transfers tend to be negatively correlated with property tax revenues if they are raised by local governments; higher federal transfers reduce the importance of raising own-source revenues and are expected to have a negative impact on property tax revenues. V. Hausman Test There is a question whether the Random Effects model is more suitable for our research. However, under the circumstances, we think the Fixed Effects model is more appropriate for this research project. Generally, if omitted variables are uncorrelated with the explanatory variables that are in the model, then a random effects model is probably best. If there are omitted variables, and these variables are correlated with the variables in the model, then fixed effects models may provide a means for controlling for omitted variable bias. In our research we believe that some important explanatory variables (urbanization, e.g.,) might be missing and correlated with the existing variables, and thus a Fixed Effects model is found more suitable than a Random Effects model. To test the advantage of the Fixed Effects model, we conduct Hausman Tests for Equations (3), (4), (7), and (8). Table 4 displays the results of Hausman Tests. Since results show that Chi2 are generally big enough, so that the probabilities that are bigger than Chi2 are small, we reject the 10 In the regressions, it has been multiplied by 100 to be consistent with the logarithm transformation of the dependent variable. 8 null hypothesis that Random Effects models are appropriate. Therefore, this research has utilized the Fixed Effects Model. VI. Estimation Results Descriptive statistics for the 50 U.S. states are shown in Table 1. Among the 50 states, California has the highest property tax revenue, while New York and Texas rank second and third respectively. In terms of GDP and population, California also ranks first, while Texas takes the second place and New York takes the third. Table 2 displays the statistics through years, to demonstrate the data characteristics from another dimension. As GDP, population, and property tax rate are all increasing over time, property tax revenue has also been increasing over the years. Regression results of the 50 U.S. states are shown in Table 5. All the coefficients are basically in line with our predictions. Comparing between OLS and Fixed Effects Models would be helpful. In the OLS models (Columns 1 and 2), both property tax rate and governance quality have significant positive impacts on the property tax revenue. In the meantime, coefficients of federal transfer and household size are insignificant, while household size has positive impact on the dependent variable, which is a bit counter intuitive. The result might be because of running pooled OLS regressions on the sample, without dealing with differences between the states. When we run OLS regressions, we are pooling all the observations together for the estimation without controlling for time-invariant unobserved individual characteristics that can be correlated with the observed independent variables. 11 This could cause potential omitted variable bias due to unobserved heterogeneity when this heterogeneity is constant over time. Therefore, we use Fixed Effects Model to remove heterogeneity from the data through differencing (i.e. removing any time invariant components of the model). When we move to Columns 3 and 4, regression results from the Fixed Effects Model, GDP and population are still significantly positive, while property tax rate and governance now turn to insignificant. The reason that coefficients of property tax rate and governance change to insignificant mainly comes from controlling for time-invariant unobserved state characteristics. When we run OLS regressions, property tax rate and governance have important impacts on the tax revenues because we consider the variations across states. But when we use the Fixed Effects Model, we estimate the impacts within states, so that impacts from the tax rate and governance diminish significantly, since tax rates and governance within each state maintain very similar levels across time. The most important changes are in the observed coefficients of federal transfer and household size. Now, both of them show negative impacts on property tax revenue as expected and are statistically significant. Since Fixed Effects Models are more sophisticated to deal with the 11 Some plausible indicators can be cited—especially those related to the indexation applied in perception surveys to reveal the intrinsic pricing of local benefits provided by property taxes. 9 potential endogeneity issues coming from the structural differences between different countries (or provinces or states), results from Columns 3 and 4 are more reliable. To interpret the results in terms of the determinants of the property tax revenue in the United States, we look at Column 4 of Table 5. From the coefficients in the table, we can see that every 1% increase in GDP and population will increase the property tax revenue by around 0.18% and 0.63%, respectively. In the meantime, 1% increase in federal transfer will decrease the tax revenue by about 0.04%. The next three control variables are not transformed by natural logarithm and its correlation with property tax revenues (expressed in natural log form) indicates some interesting results. If the property tax rate is increased by one percentage point, the tax revenue would be expected to respond with an increase by 8%. And as explained above, the governance index exhibits negligent variation across states and that leads to the minimal impact (though positive) of this variable on tax revenues. On the other hand, a one-unit decrease in household size tends to exert substantial impact on tax intake. The governance indicator does not have much of an impact on tax revenue. 12 Descriptive statistics for the full sample are shown in Table 3. In this table, all the OECD countries, Australian and Canadian provinces, and U.S. states have been included. Provinces in Chile are also included in our regressions, but not in the descriptive statistics table. The reason is that data from Chile use their own currency, so that they are not comparable to other countries. But this will not cause any problem for the regressions as our primary regression models are fixed effects models that use “within variation” to obtain the “within estimator." Regression results of the full sample are shown in Table 6. As we discussed before, all the coefficients are in line with our predictions, for both OLS and Fixed Effects Models. The only issue is that both GDP and federal transfer return insignificant coefficients in Column 4 from the Fixed Effects Model. But since this model is using less observations and within estimator, we believe it is not a big issue, as the coefficients do display the expected sign. To interpret the results in detail for the full sample, we look at Column 4 of Table 6. From the coefficients in the table, we can see that every 1% increase in GDP and population will increase the property tax revenue by around 0.09% and 0.90%, respectively. In the meantime, 1% increase in federal transfer will decrease the tax revenue by about 0.02%, while one person decrease in the household size on average will increase the tax revenue substantially. 13 We have also included results of Random Effects Model for Equations (3), (4), (7), and (8) in Tables 5 and 6. Columns 3 and 4 of these two tables show coefficients of Random Effects regressions. There are some differences between Fixed Effects model and Random Effects model, 12 More specifically a decrease by one unit (one person) in household size would result in the increase in property tax revenue by almost 57 percent [or, Exp(0.449)-1]. 13 Similar to the previous footnote, a one unit decrease in household size would relate to the increase of property tax revenue by almost 111 percent [or, Exp(0.747)-1]. 10 both in magnitude and significance, especially for “log GDP” and “property tax rate”. But the signs are the same for all the coefficients of both models. As discussed above, omitted variables are likely to be correlated with existing variables and based on the results from Hausman Tests, we are confident that the results from the Fixed Effects model provide more appropriate estimates. VII. Conclusions To study the determinants of property tax revenue, we utilize OLS, Fixed Effects Models, and Random Effects Model to research on data from the United States, as well as Australia, Canada, Chile, and all the OECD countries. As expected, the Fixed Effects model fits the data better due to potential endogeneity issues, although they all return estimates as predicted. The outcomes of our empirical analysis highlight the statistically significant impact on property tax collection of such determinants presenting a country’s state of development and its demographic, fiscal and property tax-specific characteristics. Specifically, the income level, the population density, and property tax rate correlate significantly and positively with property tax revenue, whereas federal transfers and household size have significantly negative impact. In our study, we are able to measure the impact of each of these factors and our research provides coefficients which can be used to make estimates of potential property tax revenues. Our empirical analysis draws in some relevant ‘food for thought’ pertinent to property taxation. First, design of fiscal transfer matters. As noted, property tax reforms typically go in conjunction with deepening fiscal decentralization. Any ‘soft budget constraints’ embedded in fiscal transfer schemes (equalization transfers as a case in point) can influence the fiscal policies of recipient states. Within this frame, lax behavior by states in mobilizing own revenue, most credibly through property tax, tends to be unintendedly compensated by the increased magnitude of equalization or other types of redistributive transfers. 14 The broader policy relevancy that we would venture to highlight here is: The efficiency in property taxation is directly related to or dependent upon the structure of the fiscal decentralization. Second, economic and population growth exerts both push and pull effects on tax policy makers. As countries climb up the development ladder, the overall restructuring and rebalancing of the tax mix would inevitably pull in the vision for better designing and administering a property tax, an efficient tax. Higher population, combined with higher state income, would extend the base for property tax on the one hand, and push for revenue-enhancing measures to finance rising demand for public services, most visibly at local levels. A third policy issue relevant to property taxation is emerging from demographic trends observed around the world; the one most pertinent to this research is the decline in household sizes around the globe. 15 Estimated trends indicate that in most countries the average household size has 14 See, for example, Vigneault (2007) and Smart (2007) for more discussion of the disincentives from soft budget constraints and distributive transfers. 15 Patterns and trends in household size and composition: Evidence from a United Nations dataset, United Nations, 2019. 11 declined over recent decades, according to the UN report. In some developing countries – for example, Botswana, Brazil, and Peru – the decline has been sharp, reducing from over 5 per household to just above 3, but the trend appears to be universal in the developing world. This is significant from the point of view of exploiting the potential of property taxation. As our model shows, reduction in the size of households has a significant, positive impact on property tax revenues. A critical question for further research is whether and how our empirical methodologies and specifications as applied to the set of developed economies would be replicated in the context of developing countries. The sample of the latter, undoubtedly, would pose substantial challenge in terms of data availability and quality (particularly, the issues of data consistency and measurement errors). Time-series data for a specific developing country would present a possibility to explore the extent of the country’s own property tax capacity and be helpful for policy design. Further research is needed to fill the analytical gap in comparative empirical analysis for a wide set of developing countries. 12 References 1. Norregaard, John. 2013. "Taxing Immovable Property Revenue Potential and Implementation Challenges." IMF Working Papers 13/129, International Monetary Fund. 2. Sepulveda, Cristian and Jorge Martinez-Vazquez. 2012. “Explaining property tax collections in developing countries: the case of Latin America.” in G. Brosio and J.P. Jimenez (eds.) “Decentralization and Reform in Latin America: Improving Intergovernmental relations.” Chetelham: Edward Elgar. 3. De Cesare, Claudia. 2012. “Improving the Performance of the Property Tax in Latin America.” Lincoln Institute of Land Policy. 4. Bonet, Jaime, Andrés Muñoz, and Carlos Pineda. 2014 “The Hidden Potential: Property Tax determinants and opportunities in Latin America.” Inter-American Development Bank. 5. Grigorian, David, and Hamid Davoodi. 2007. "Tax Potential vs. Tax Effort; A Cross-Country Analysis of Armenia's Stubbornly Low Tax Collection," IMF Working Papers 07/106, International Monetary Fund. 6. Andy Hultquist, and Tricia L. Petras. 2012. “Determinants of Fractional Assessment Practice in Local Property Taxation: An Empirical Examination.” Proceedings. Annual Conference on Taxation and Minutes of the Annual Meeting of the National Tax Association, vol. 105, 2012, pp. 146–152. 7. Bahl, Roy. 1999. “Implementation Rules for Fiscal Decentralization.” International Studies Program Working Paper Series 99-01, Andrew Young School of Policy Studies, Georgia State University. 8. Bonet, Jaime, Andrés Muñoz, and Carlos Pineda. 2014. “The Hidden Potential: Property Tax determinants and opportunities in Latin America.” Inter-American Development Bank. 9. De Cesare, Claudia, 1998, “Using the property tax for value capture: A case study from Brazil”, Land Lines, 10(1):5 ff. 10. De Cesare, Claudia. 2012. “Improving the Performance of the Property Tax in Latin America.” Lincoln Institute of Land Policy. 11. Eom, Tae Ho, 2008. “A Comprehensive Model of Determinants of Property Tax Assessment Quality: Evidence in New York State. Public Budgeting & Finance”, Vol. 28, Issue 1, pp. 58- 81. 13 12. Food and Agriculture Organization, 2002, “Rural property tax systems in Central and Eastern Europe”, Rome: Food and Agriculture Organization of the United Nations. 13. Franzsen, Riel C.D. 2003, “Property Taxation within the Southern African Development Community (SADC): Current Status and Future Prospects of Land Value Taxation.” Lincoln Institute of Land Policy. 14. Grigorian, David, and Hamid Davoodi. 2007. “Tax Potential vs. Tax Effort; A Cross-Country Analysis of Armenia's Stubbornly Low Tax Collection,” IMF Working Papers 07/106, International Monetary Fund. 15. Manaf, Nor, Hasseldine, John, and Hodges, Ron. 2006. “The Determinants of Malaysian Land Taxpayers' Compliance Attitudes.” Public Economics: Taxation, Subsidies, & Revenue eJournal. 16. Norregaard, John. 2013. “Taxing Immovable Property Revenue Potential and Implementation Challenges.” IMF Working Papers 13/129, International Monetary Fund. 17. OECD. 2010. Tax Policy Reform and Economic Growth. OECD Tax Policy Studies. 2010. No. 20. 18. Richard M. Bird and Enid Slack (ed.), 2004. “International Handbook of Land and Property Taxation,” Books, Edward Elgar Publishing, number 3304. 19. Roy Bahl, Jorge Martinez-Vazquez and Joan Youngman, eds., 2008, “Making the property tax work: Experiences in developing and transitional countries”, Cambridge, MA: Lincoln Institute of Land Policy 20. Sepulveda, Cristian and Jorge Martinez-Vazquez. 2012. “Explaining property tax collections in developing countries: the case of Latin America.” in G. Brosio and J.P. Jimenez (eds.) “Decentralization and Reform in Latin America: Improving Intergovernmental relations.” Chetelham: Edward Elgar. 21. UN-Habitat, 2011, “Land and Property Tax - A Policy Guide”, United Nations Human Settlement Programme, Nairobi (Principal author: Lawrence Walters). 22. United Nations, 2019, “Patterns and trends in household size and composition: Evidence from a United Nations dataset” 23. Wallis, John Joseph, 2000, “A History of the Property Tax in America”, State Tax Notes. 14 Table 1. Descriptive Statistics for the States in U.S. (Averaged by years) Property GDP Population Federal Transfer Property Household States Tax Revenue Governance (Million $) (Thousands) (Million $) Tax Rate Size (Million $) Alabama 2,448 182,397 4,771 9,170 0.0039 2.52 3.62 Alaska 1,306 52,434 713 2,892 0.0102 2.77 4.98 Arizona 6,791 268,778 6,476 11,368 0.0065 2.70 5.01 Arkansas 1,756 106,734 2,919 6,045 0.0059 2.51 4 California 53,002 2,162,854 37,556 73,316 0.0074 2.93 4.87 Colorado 6,712 274,227 5,109 7,134 0.0060 2.53 6.42 Connecticut 9,224 243,089 3,567 3,144 0.0146 2.55 4.69 Delaware 693 61,082 905 1,599 0.0049 2.60 5.2 Florida 26,028 792,266 19,179 23,337 0.0100 2.56 5 Georgia 10,479 447,658 9,755 12,357 0.0089 2.70 4.75 Hawaii 1,331 72,254 1,371 2,059 0.0029 2.93 4.36 Idaho 1,365 58,532 1,576 2,382 0.0071 2.65 5.81 Illinois 23,990 703,272 12,805 13,200 0.0175 2.63 4.25 Indiana 6,726 293,299 6,497 9,378 0.0095 2.53 4.25 Iowa 4,285 152,068 3,059 5,527 0.0132 2.40 5.98 Kansas 3,893 135,409 2,851 3,842 0.0129 2.49 5.68 Kentucky 3,007 171,493 4,346 9,043 0.0076 2.49 4.2 Louisiana 3,430 222,905 4,529 10,288 0.0045 2.62 3.81 Maine 2,379 52,794 1,328 2,593 0.0110 2.34 4.73 Maryland 8,031 326,434 5,819 9,035 0.0090 2.64 4.88 Massachusetts 13,287 431,900 6,608 8,278 0.0104 2.52 7.43 Michigan 13,798 418,479 9,943 17,024 0.0149 2.53 4.6 Minnesota 7,449 288,727 5,339 8,165 0.0103 2.46 6.36 Mississippi 2,530 97,831 2,964 7,914 0.0060 2.63 3.19 Missouri 5,614 265,009 5,984 7,151 0.0096 2.47 4.61 Montana 1,354 40,172 996 2,014 0.0074 2.42 4.97 Nebraska 2,928 97,913 1,837 2,788 0.0170 2.46 6.01 Nevada 2,967 131,667 2,720 2,682 0.0075 2.68 4.73 New Hampshire 3,422 67,009 1,323 1,732 0.0179 2.49 6.07 New Jersey 24,947 515,628 8,788 11,759 0.0187 2.71 3.7 New Mexico 1,335 85,888 2,050 5,565 0.0060 2.64 3.96 New York 45,596 1,274,318 19,422 40,021 0.0135 2.62 4.62 North Carolina 8,635 439,453 9,583 13,595 0.0079 2.51 5.18 North Dakota 789 42,238 695 1,549 0.0129 2.28 6.93 Ohio 13,562 535,710 11,551 12,018 0.0147 2.47 4.17 Oklahoma 2,285 165,534 3,770 7,849 0.0082 2.53 4 Oregon 4,904 176,240 3,863 7,283 0.0089 2.49 5.53 Pennsylvania 16,640 626,841 12,692 23,191 0.0138 2.47 4.35 Rhode Island 2,250 51,298 1,056 2,470 0.0132 2.49 5.01 South Carolina 4,845 175,678 4,658 7,672 0.0055 2.53 4.2 South Dakota 977 40,815 822 1,410 0.0122 2.44 5.96 Tennessee 5,074 277,086 6,378 11,560 0.0074 2.51 4.24 Texas 40,071 1,349,514 25,564 33,099 0.0176 2.82 5.59 Utah 2,558 127,667 2,790 3,186 0.0061 3.11 6.36 Vermont 1,371 27,804 625 1,664 0.0157 2.36 5.56 Virginia 11,332 434,326 8,060 8,395 0.0079 2.58 5.83 Washington 8,790 394,926 6,799 8,597 0.0090 2.53 5.94 West Virginia 1,408 66,055 1,844 3,908 0.0050 2.41 3.65 Wisconsin 9,242 268,659 5,689 9,967 0.0168 2.42 5.19 Wyoming 1,287 37,703 562 1,246 0.0055 2.47 5.58 15 Table 2. Descriptive Statistics for the Years (Averaged by 50 states) Property Federal GDP Population Property Household Year Tax Revenue Transfer Governance (Million $) (Thousands) Tax Rate Size (Million $) (Million $) 2006 7,267 272,869 5,956 7,373 0.0095 2.56 5 2007 7,748 285,437 6,013 7,585 0.0095 2.55 5 2008 8,156 290,440 6,013 7,764 0.0095 2.56 5 2009 8,661 284,998 6,070 9,260 0.0095 2.55 5 2010 8,842 295,640 6,174 11,245 0.0095 2.54 5 2011 8,880 306,426 6,219 11,354 0.0099 2.55 5 2012 8,884 319,421 6,265 10,239 0.0103 2.56 5 2013 9,025 331,133 6,308 9,975 0.0106 2.57 5 2014 9,265 345,758 6,354 10,315 0.0108 2.58 5 2015 9,642 359,644 6,401 11,346 0.0106 2.58 5 2016 10,017 369,290 6,448 11,665 0.0106 2.58 5 16 Table 3. Descriptive Statistics for the Full Sample (Averaged by years) 16 Property Federal GDP Population Household State/Country Tax Revenue Transfer (Million $) (Thousands) Size (Million $) (Million $) Alabama 2,448 182,397 4,771 9,170 2.52 Alaska 1,306 52,434 713 2,892 2.77 Alberta 8,181 311,243 4,006 4,593 2.60 Arizona 6,791 268,778 600 11,368 3.48 Arkansas 1,756 106,734 2,919 6,045 2.51 Australia 2,497,703 969,510 302 2.83 Australian Capital Terrotory 466 44,754 387 2.55 Austria 203,638 361,340 8,452 2.28 Belgium 1,380,628 435,520 106 2.67 British Columbia 8,385 216,345 2,079 5,867 4.05 California 53,002 2,162,854 37,556 73,316 2.93 Canada 5,213,817 1,406,995 34,222 2.54 Chile 320,555 336,077 17,471 3.45 Colorado 6,712 274,227 5,109 7,134 2.53 Connecticut 9,224 243,089 3,567 3,144 2.55 Czech Republic 125,634 288,359 741 2.77 Delaware 693 61,082 905 1,599 2.60 Denmark 453,754 245,804 5,599 2.14 Estonia 9,299 31,777 1,334 2.30 Finland 252,398 209,248 5,374 2.10 Florida 26,028 792,266 19,179 23,337 2.56 France 9,057,050 2,378,147 65,379 2.40 Georgia 10,479 447,658 9,755 12,357 2.70 Germany 2,988,512 3,303,394 80,935 2.01 Greece 631,452 298,465 10,994 2.66 Hawaii 1,331 72,254 1,371 2,059 2.93 Hungary 239,479 224,986 9,954 2.53 Iceland 30,056 13,045 315 2.57 Idaho 1,365 58,532 1,576 2,382 2.65 Illinois 23,990 703,272 12,805 13,200 2.63 Indiana 6,726 293,299 6,497 9,378 2.53 Iowa 4,285 152,068 3,059 5,527 2.40 Ireland 402,556 216,183 4,491 2.73 Israel 757,883 236,276 7,908 3.47 Italy 4,925,787 2,073,733 59,623 2.38 Japan 11,480,156 4,518,232 127,600 2.53 Kansas 3,893 135,409 2,851 3,842 2.49 Kentucky 3,007 171,493 4,346 9,043 2.49 Korea 4,552,579 1,552,496 49,888 2.88 Latvia 37,906 41,013 2,106 2.53 Lithuania 21,400 67,073 3,103 2.39 Louisiana 3,430 222,905 825 10,288 2.84 Luxembourg 130,066 43,647 510 2.54 Maine 2,379 52,794 162 2,593 2.71 Manitoba 2,060 57,389 1,274 3,428 2.50 Maryland 8,031 326,434 5,819 9,035 2.64 Massachusetts 13,287 431,900 6,608 8,278 2.52 Mexico 537,764 1,827,143 1,024 2.80 Michigan 13,798 418,479 9,943 17,024 2.53 Minnesota 7,449 288,727 5,339 8,165 2.46 Mississippi 2,530 97,831 2,964 7,914 2.63 Missouri 5,614 265,009 5,984 7,151 2.47 Montana 1,354 40,172 996 2,014 2.42 Nebraska 2,928 97,913 1,837 2,788 2.46 16 Data for Chile are included in the regression, but not in the descriptive table, since its currency unit is different from other countries. This is not a problem in the regression, as we only use within variations in fixed effects models. 17 Table 3 (continued). Descriptive Statistics for the Full Sample (Averaged by years) Property Population Federal GDP Household State/Country Tax Revenue (Thousand Transfer (Million $) Size (Million $) s) (Million $) Netherlands 1,090,903 751,763 16,644 2.27 Nevada 2,967 131,667 2,720 2,682 2.68 New Brunswick 1,156 29,410 759 2,603 2.30 New Hampshire 3,422 67,009 1,323 1,732 2.49 New Jersey 24,947 515,628 8,788 11,759 2.71 New Mexico 1,335 85,888 2,050 5,565 2.64 New South Wales 6,461 517,175 7,484 2.60 New York 45,596 1,274,318 19,422 40,021 2.62 New Zealand 276,381 140,136 4,387 2.67 Newfoundland and Labrador 423 31,026 527 820 2.35 North Carolina 8,635 439,453 9,583 13,595 2.51 North Dakota 789 42,238 695 1,549 2.28 Northern Terrotory State 103 27,095 239 2.90 Northwest Territories 71 4,542 44 1,177 2.75 Norway 350,416 294,052 4,997 2.15 Nova Scotia 1,306 34,969 942 2,919 2.30 Nunavut 16 2,316 36 1,392 3.65 Ohio 13,562 535,710 898 12,018 2.88 Oklahoma 2,285 165,534 3,770 7,849 2.53 Ontario 28,634 657,340 13,571 19,234 2.60 Oregon 4,904 176,240 3,863 7,283 2.49 Pennsylvania 16,640 626,841 12,692 23,191 2.47 Poland 1,115,706 808,202 38,371 2.90 Portugal 326,250 280,899 10,490 2.65 Prince Edward Island 166 5,210 145 537 2.35 Quebec 13,166 335,959 8,119 19,224 2.30 Queensland 4,604 327,366 4,671 2.60 Rhode Island 2,250 51,298 1,056 2,470 2.49 Saskatchewan 1,519 76,810 7,101 1,384 3.34 Slovak Republic 57,007 132,907 5,408 2.89 Slovenia 34,009 56,344 2,038 2.70 South Australia State 2,036 105,744 1,676 2.40 South Carolina 4,845 175,678 4,658 7,672 2.53 South Dakota 977 40,815 822 1,410 2.44 Spain 3,497,143 1,476,801 46,356 2.59 Sweden 434,973 405,278 9,542 2.14 Switzerland 837,033 420,128 7,889 2.32 Tasmania 462 30,962 317 3.12 Tennessee 5,074 277,086 6,378 11,560 2.51 Texas 40,071 1,349,514 25,564 33,099 2.82 Turkey 1,452,402 1,391,945 73,519 3.67 United Kingdom 9,319,319 2,348,265 62,989 2.38 United States 48,492,621 15,714,450 313,700 2.58 Utah 2,558 127,667 2,790 3,186 3.11 Vermont 1,371 27,804 1,781 1,664 2.54 Victoria 6,337 395,677 5,867 2.60 Virginia 11,332 434,326 8,060 8,395 2.58 Washington 8,790 394,926 6,799 8,597 2.53 West Virginia 1,408 66,055 1,844 3,908 2.41 Western Australia State 2,819 210,705 2,458 2.60 Wisconsin 9,242 268,659 5,689 9,967 2.42 Wyoming 1,287 37,703 562 1,246 2.47 Yukon 36 2,430 37 856 2.35 18 Table 4. Hausman Test Chi2 Prob>Chi2 Equation 3 34.56 0.0028 Equation 4 28.37 0.0286 Equation 7 242.00 0.0000 Equation 8 22.99 0.0843 19 Table 5. Regression Results for the U.S. Sample Dependent Variable: Property Tax Revenue log GDP 0.836*** 0.781*** 0.179** 0.176** 0.291*** 0.280*** (0.076) (0.081) (0.060) (0.061) (0.058) (0.060) log Population 0.164* 0.204* 0.617*** 0.627*** 0.750*** 0.762*** (0.082) (0.084) (0.183) (0.187) (0.070) (0.072) log Federal Transfer -0.044 -0.021 -0.036* -0.036* -0.0473** -0.0465** (0.027) (0.030) (0.017) (0.017) (0.017) (0.017) Property Tax Rate 0.580*** 0.575*** 0.079 0.081 0.252*** 0.253*** (0.031) (0.031) (0.057) (0.057) (0.045) (0.045) Household Size 0.046 0.066 -0.447*** -0.449*** -0.331** -0.336** (0.067) (0.071) (0.114) (0.114) (0.107) (0.107) Governance 0.030* 0.002 0.005 (0.012) (0.007) (0.007) Constant -3.042*** -3.152*** 4.075 3.984 1.344* 1.272* (0.301) (0.304) (2.761) (2.793) (0.606) (0.618) Fixed Effects NO NO YES YES NO NO Random Effects NO NO NO NO YES YES N. of Obs 550 550 550 550 550 550 Adjusted R2 0.946 0.946 0.742 0.742 0.916 0.917 Standard errors in parentheses * p<0.05, ** p<0.01, *** p<0.001 20 Table 6. Regression Results for the Full Sample Dependent Variable: Property Tax Revenue log GDP 0.859*** 1.062*** 0.275*** 0.091 0.812*** 0.151** (0.006) (0.060) (0.038) (0.055) (0.013) (0.053) log Population 0.740*** 0.068 1.074*** 0.896*** 0.704*** 0.919*** (0.028) (0.056) (0.124) (0.147) (0.044) (0.060) Household Size -0.538*** -0.716*** -0.628*** -0.747*** -0.406*** -0.725*** (0.148) (0.051) (0.045) (0.090) (0.052) (0.083) log Federal Transfer -0.0786** -0.024 -0.025 (0.026) (0.016) (0.016) Constant -10.56*** -2.916*** -7.130*** -4.276* -9.713*** -5.365*** (0.559) (0.278) (1.745) (2.070) (0.712) (0.511) Fixed Effects NO NO YES YES NO NO Random Effects NO NO NO NO YES YES N. of Obs 1275 654 1275 654 1275 654 Adjusted R2 0.892 0.947 0.684 0.75 0.892 0.932 Standard errors in parentheses * p<0.05, ** p<0.01, *** p<0.001 21