Lorebgmipllsumdolyorsltzmhhabc Land Governance POLICY BRIEF WORLD BANK ✦ DEVELOPMENT RESEARCH GROUP ISSUE 5 ✦ AUGUST 2018 Using Satellite Imagery to Revolutionize the Creation of Tax Maps ✳✳ Daniel Ayalew Ali, Klaus Deininger, and Michael Wild Globally, cities rely on property taxes as a key source of revenues to finance the services that enhance its long-term competitiveness and counter the negative aspects of density. In developing countries, the technical com- plexity of ensuring that tax rolls are complete and valuations current is often perceived as a major barrier to bringing in more property tax revenues. This policy paper shows how high-resolution satellite imagery makes it possible to assess the completeness of existing tax maps by estimating built-up areas based on building heights and footprints. Together with information on sales prices from the land registry, targeted surveys, and routine statistical data, this makes it possible to use mass valuation procedures to generate tax maps. The example of Kigali illustrates the reliability of the method and the potentially Surface Model Derived Entirely from Satellite Imagery. | © Generated by GAF AG far-reaching revenue impacts. Estimates based on modelling show that heightened tax compliance and a move to a 1% ad valorem incentives for fraudulent under-declaration needs. Using spatial imagery for updat- tax would yield a tenfold increase in revenue of sales values. ing tax maps is also useful for places with from public land. rapidly expanding urban growth not yet The technical complexity and high fixed captured in official data, even although tax costs of ensuring that tax rolls are complete revenues could be quite buoyant. With a Municipalities need and valuations current are perceived as per km² cost of US$34 for imagery and $112 property taxes to finance major barriers to bringing in more property for processing of footprints and building tax revenues. Often, less than 50% of tax- heights the total cost for our study area of service delivery able properties are on the tax roll, because 340 km² was about US$40,000, with ample The ability for cities to raise revenues in a maps to identify these are incomplete, out scope for further reductions. non-distortionary way for effective urban of date, or access to them is fragmented service delivery and infrastructure is essen- between government units. Using the tial to realizing the potential of urbanization. example of Kigali, in our latest working Using land registration As most benefits from these investments paper, we demonstrate how high-resolu- to enhance property tax will be capitalized in surging land values, tion satellite imagery makes it possible to recurrent taxes on land and other real assess the completeness of existing tax collection in Rwanda property can be an incentive-compatible maps, enhances their distributional fair- Rwanda stands out among African coun- financing method. In developing countries, ness and facilitates the use of spatial mass tries for having established a complete taxes on land and property are still far below valuation models or CAMA to impute the and fully digital registry of rights in 2012. those of developed countries, even in rela- potential yield from tax policy changes. This nationwide first-registration program tive terms. Instead, cities often rely heavily recorded spatial and textual data for each on land transaction taxes, but these impose Municipalities can use this approach to of the 11.6 million parcels, assigned a frictions on land market operations, push improve tax maps and build scenarios for Unique Parcel Identifier (UPI) to each, with transactions into informality, and create identifying policies that best meet their all data captured in the Land Administration Information System (LAIS). This UPI that is costs of tax compliance (e.g., automated (see Figure 1). We overlaid the result, with also used for the mortgage register and also billing, reminders, mobile payment, apply- boundaries of cadastral parcels obtained the Rwanda Revenue Authority (RRA) uses ing peer pressure) may thus have high from LAIS to estimate building footprint the same UPI and can access the LAIS. The returns. and compute building volume. Together land registry is, somewhat surprisingly, not with information on sales prices for 85,000 used for fiscal purposes and setting lease properties from the land registry, targeted fees—the equivalent of property taxes— Modeling changes in tax surveys, and routine statistical data, this to identify properties, automatically bill fee structures makes it possible to develop spatial hedonic owners for land lease fees owed or to send regression analysis for property prices for reminders. Moreover, the lease fees charged To assess the effects of changing the fee use in mass valuation models to generate are still based on self-declaration and fail to structure from self-declaration to market tax maps and impute the potential yield consider changes in land values, albeit, reli- values, we estimate hedonic property from changes in tax fees and exemptions, able price data are available in the LAIS for values by combining remote sensing thus laying the foundation for better-in- some 85,000 recently transacted properties. and administrative data. As the property formed policy. registry in Rwanda does not contain data on building characteristics needed for a Using the current tax system, potential lease Tax collection efficiency mass valuation exercise, we drew on sat- fees based on land value from all residential ellite imagery to generate information on properties in urban Kigali sold in 2013–16, Overlaying tax and registry data made it built-up area. A spatial hedonic regression and for which the prices are available, would possible to assess the tax collection gap. To model was subsequently developed to esti- be US$ 552,923. A 1% flat rate of land value compare potential to actual revenue from mate property values for urban Kigali. Our could raise some US$2.6 million—more land lease fees, and estimate the potential analysis of property values displayed that than four times the current property tax for a more effective tax collection, we draw neighborhood characteristics and building yield (see Figure 1). Extrapolating this to all on 2015 parcel-level data for land tax rev- volume affect residential property prices urban residential properties in Kigali sug- enue from the RRA, while using the UPI to that, in fact, are not accounted for in the gests that a move from the current lease link RRA tax collection and the LAIS data. computation of current lease fees which is fee based on self-declaration at the time of Our analysis shows that collection rates are highly regressive and hence fails to maxi- registration, to a 1% updated value-based below a potential of $6.74 million, as only mize property tax revenue. tax could increase revenue to between US$ 30% of taxable residential parcels in urban 16 and US$ 19.3 million—almost 10 times Kigali paid lease fees in 2015. Thus, in Kigali Data on building heights and the volume what is currently collected, and would also alone, closing the tax collection gap could of built-up area were generated from pro- spread the tax burden more equally. more than triple revenue. Efforts to reduce cessing high-resolution satellite imagery Figure 1. Estimated building height a. Central Kigali b. Higher-Density Neighborhood Building Height High: 7.4 meter Low: 2 meter Costs of tax exemptions Policy implications Exemptions significantly reduce tax yield. The case of Kigali illustrates that high resolu- ancillary benefits in terms of ownership For example, in Kigali city 60% (67,000) of tion remotely sensed imagery can be used documentation and planning. Using such all residential parcels and 99% of agricul- to prepare a property inventory or “tax map”, imagery, together with information on land tural parcels are exempted from paying and reliably check the completeness of val- prices, could reduce the cost of preparing lease fees. A proposal being discussed uation rolls and generate data on property tax maps and ground the debate on tax in government circles is to levy separate values that are needed to run automated reform by providing more reliable data on fees on land and on buildings and adjust mass valuation models at a fraction of the which to base scenarios. exemptions. The hedonic regression model time and resources required by more tra- developed allows to explore what would ditional technologies. The imputed values happen if taxes were levied only on land or on land and property, together with admin- if certain exemptions were adopted. Figure istrative records on taxes collected, allows 2 projects the tax revenues for the various us to provide information on (i) potential exemption options that are being consid- revenue gains from the full collection of ered by government. The model shows that current property taxes fees; (ii) likely yields while there may be scope for exempting from alternative rates, in the case of Rwanda those at the very bottom, those currently a uniformly applied 1% valuation tax; and being discussed for ‘low-cost housing’ (i.e., (iii) the implicit cost of exemptions. This less than RWF 30 million) are too generous: approach can enable cities in developing they would leave only 5% of properties with countries to not only augment the financial any building tax obligation. resources at their disposal but also realize Figure 2. Predictions for estimated tax revenue in Millions of Rwanda Franc (RWF) from a 1% property tax with various exemptions on building tax levels and control areas before and during the project Current lease fee No exemption, 1% p. tax < 30 mn RWF < 5 mn RWF < 9 mn RWF < 13 mn RWF 0 5 10 15 Millions of Rwandan Franc (RWF) Land lease fee Building tax Total Note: Results are based on the spatial error model and apply to our study area. 1. Current lease fee rates imply that 0.45% pay RWF 5/m2, 0.16% pay 10/m2, 14% pay RWF 30/m2, and 86% pay RWF 70/m2. 2. Building tax is 1% of building value exempting all structures with values less than the exemption threshold. 3. Threshold values: RWF 30 mn ((US$ 38,120), RWF 5 mn (US$ 4,927) is first quartile, RWF 9 mn (US$ 8,277) is median value, and RWF 13 mn (US$ 11,055) is mean value. 4. Land tax is RWF 70/m2 for area < 300 m2 plus RWF 105/m2 for area > 300m2. This policy brief is based on Ali, Daniel Ayalew; Deininger, Klaus W.; Wild, Michael. 2018. Using Satellite Imagery to Revolutionize Creation of Tax Maps and Local Revenue Collection. Policy Research Working Paper; No. 8437. World Bank, Washington, DC.  https://bit.ly/2GfXsXH. Daniel Ali is a senior economist, Klaus Deininger a lead economist, Daniel Monchuk a consultant, all at the World Bank, Washington DC. ✦  CONTACT: kdeininger@worldbank.org. The views presented are those of the authors and do not necessarily represent those of the World Bank, its Executive Directors or the member countries they represent.