Bringing property owners into the tax net: Avenues of fiscal capacity and local accountability

Within the state-building and taxation literature, the new fiscal sociology strand has emphasized the virtues of greater revenue mobilization at both the local and national level for the democratic future of developing countries. Taxation creates a bargaining arena in which taxpayers can work to foster accountable public institutions (Bräutigam et al, 2008; Martin et al, 2009). For local governance, increasing revenues through greater contributions from real estate is key to delivering public goods. As a salient tax levied for local government budgets, property tax is a prime candidate to achieve this objective. Due to rapid urbanization and substantial investments in owner-occupied and rental units, it offers huge potential (Bahl et al eds. 2008). However, as argued in Besley and Persson (2009), there is no basis for the assumption that developing countries have the necessary institutional capacity to raise taxes. This project, based on a collaboration with Senegal’s revenue authority (Direction générale des Impôts et des Domaines, DGID), aims at evaluating two different interventions which relate to these strands of the taxation and development literature. We have designed these interventions to help the DGID improve its capacity to administer property taxes.

Property taxes have many virtues. First, they ease the informational problem of taxation as the revenue authority computes the liability based on a visible and immobile base. Second, if well administered, they satisfy the two characteristics of a fair distribution of a tax burden, namely the ability-to-pay and the benefit principles. Linked to the first, Sennoga et al (2008) have shown the progressivity of property taxes in developing countries. Regarding the benefit principle, property taxes approximate a user fee as they fund local public goods (Bahl and Martinez-Vazquez, 2008). However, IMF and World Bank revenue statistics show that, between 2000 and 2012, property taxes represent on average of 0.1 to 0.2% of GDP in Sub-Saharan Africa (0.1% in Senegal). In OECD countries, the average range is between 2-3% of GDP. This poor performance suggests considerable scope for improvement. Raising statutory rates or enlarging its base are two options to increase revenue. Improved tax administration could achieve the same objective with lower efficiency losses, through a better system of tax assessment, collection, and enforcement (Keen and Slemrod, 2017).

Three main reasons explain the revenue gap in property taxes in developing countries such as Senegal. First, they are impossible to administer without a functioning fiscal cadaster. Second, if in place, revenue authorities manage fiscal cadaster data without an adequate information technology (IT) system. In particular, the manual management of records hinders the creation of an exhaustive cadaster capable of linking detailed parcel information to individual taxpayers in order to facilitate the production of a credible tax base, annual tax rolls and the surveillance of delinquent taxpayers (Norregaard, 2013). The manual management of records also makes it difficult to trace previous notifications. In addition, weak IT systems require annual field visits and render the surveillance of tax inspectors in charge of property taxes more difficult (Khan et al, 2016). The third reason relates to weak enforcement tools towards non-compliant taxpayers. Because of their salience compared to consumption taxes, property taxes are unpopular (Rosengard, 1998; 2012). Enforcement measures with lasting effects are thus necessary to make the property tax work.

Based on experiences in developing countries (Indonesia, Jamaica, The Philippines; see Rosengard, 1998; 2012), the data-led approach to property tax reform posits that technological upgrades in the collection and processing of property information could increase administrative efficiency. A second approach to property tax reform focuses on increasing voluntary compliance. In both approaches, a positive theory of property tax compliance would be different from the benchmark approach to tax evasion (Allingham and Sandmo, 1972). Unlike the income tax, which has a declaration component, the property tax is not self-assessed and is computed by the revenue authority. Therefore, according to this theory, the key parameter in compliance behavior is the perceived probability of credible enforcement. Our project will seek to elicit mechanisms, which can change this subjective probability.

We propose to assess the importance of these two approaches in an ideal context. DGID has developed an alpha version of a software for its fiscal cadaster censuses in order to phase-out the manual management of records, automate the storage and processing of data collected during field visits and the production of tax rolls. The present version needs to be developed into a mature fully reliable version.

---Our first intervention, with the help of professional applications developers, plans to achieve this objective with the introduction of this software in the property tax information system. We aim at measuring to what extent its use could increase property tax revenues both on the extensive and intensive margins, and to explore the underlying causal channels.

The first channel relates to the effort of tax inspectors. We ask the following question: how does the new informational structure created with this software affect the outputs of tax inspectors? With field information centralized in one database, the work of supervisors is facilitated as they can easily monitor the progress made in issuing notifications and conducting enforcement measures. We hypothesize that the centralization of property information for the production of tax rolls will limit: 1) tax inspectors’ ability to deploy low effort in issuing tax notifications; 2) the extent to which tax inspectors can engage in corruption by either omitting to issue or issuing incorrect notifications.

The second channel relates to taxpayers’ beliefs about the efficiency of the revenue authority in issuing notifications and taking enforcement steps (Powers, 2008). This leads to the following research questions: does the improvement fiscal cadaster survey technology change taxpayers’ perception of the administrative effectiveness in property taxation? Does awareness of the new technology affect their compliance decisions through an increase in the perceived probability of credible enforcement? We will answer these questions through variations in the deployment of the software and through variations in the salience of new technology. We achieve the latter with the delivery of official messages about its capabilities and subsequent administrative effectiveness in tracking delinquent taxpayers. We hypothesize that if taxpayers internalize the improved enforcement capacity of the revenue authority, these communications should increase voluntary compliance.

---Our second intervention relates to tax compliance and more specifically to deterrence and non-deterrence messages in the context of property taxes. A wide range of theoretical and empirical research has studied taxpayer behavior when the perceived probability of enforcement is raised (deterrence) or in response to a range of non-deterrence related communications often defined as tax morale (Slemrod et al, 2001; Dwenger et al, 2016; Luttmer and Singhal, 2014). We plan to use a deterrence message with extracts and provisions of the tax codes on fines, penalties, and escalated enforcement. For non-deterrence messages, we will prioritize the reciprocity (or fiscal exchange) channel. With the growing demand for provision of local public goods, a reciprocity letter aiming to change prior beliefs about local government expenditures could induce greater compliance. In this process, we will answer three questions. First, how do deterrence and reciprocity messages affect compliance in our context? Second, given that compliance related communications have an important dynamic component: do the two approaches lead to sustained changes in the beliefs of taxpayers about the institutional environment beyond the year of the first nudge? Finally, we seek to highlight how these two types of communications interact with the technological improvement of our first intervention to improve compliance.

2018-01-01

2019-12-31

Our first contribution relates to the evaluation of technological modernizations of tax administration with randomized controlled trials (Pomeranz, 2015; Carrillo et al, 2017; Eissa and Zeitlin, 2014). There is a dearth of tax natural field experiments in developing countries, particularly in Sub-Saharan Africa (Mascagni, 2017). There are also very few studies evaluating the returns to investments in the administrative capacity to raise local taxes. An exception to that is Gadenne (2017) who studies the effects of a local taxation capacity-building program in Brazilian municipalities, including both technological improvements and enforcement efforts. She finds substantial positive effects on revenue collections. However, the reform includes multiple components and it is impossible to disentangle their relative contributions. Our design will allow a more precise identification of the channels at play.
Key variables: returns to investment in new technology on property tax collection.

Our second contribution relates to the literature in public finance which focuses on mechanisms to increase taxpayers’ voluntary compliance (Hallsworth, 2014; Luttmer and Singhal, 2014; Mascagni, 2017). As far as we are aware, there is no study on compliance messages on property taxation in Sub-Saharan Africa, but at the same time there is strong empirical support in more developed economies that deterrence is effective (Slemrod et al, 2001; Kleven et al, 2011). Reciprocity is the most relevant non-deterrence channel in our setting. We believe it could be effective with local taxes because they raise revenues for expenditures on local goods and services (Del Carpio 2014; Castro and Scartascini 2015). Yet, the empirical evidence on the reciprocity channel remains mixed (Luttmer and Singhal, 2014; Slemrod et al, 2001; Torgler, 2007). In the case of property taxes, Castro and Scartascini (2015) evaluate the relative effects of a deterrence message, information on peers’ compliance, and a reciprocity message. They only find significant average effects for the deterrence channel. However, peer and reciprocity messages affect compliance depending on the underlying beliefs of taxpayers. Importantly, the literature highlights that the effect of these channels depend on prior beliefs about taxation and government (Luttmer and Singhal, 2014; Del Carpio, 2014; Castro and Scartascini, 2015; Jibao and Prichard, 2016; Cummings et al., 2009). We aim to elicit these heterogeneous effects. Finally, our contribution is unique in the sense that it lies at the intersection of the two strands of literature, evaluating both the separate and the interacted effects of the technological improvement and the compliance messages.
Key variable: impact of various compliance messages on property tax collection.

---For intervention 1, we divide the three cities of Dakar, Mbour, and Thies into cadaster sections (clusters). The ten tax centers in these cities can conduct cadaster surveys in 200 sections per year. We will randomly assign one-half of the sections to a treatment group (T) for which we use the new cadaster application software. Our stratification will ensure that each city has treatment and control groups. Field agents in charge of the treated sections will use tablets with the new application installed. Subsequently tax inspectors will use the data collected with the software to issue tax notifications to taxpayers in those sections. The other half of the sections will be the control group (C) where the tablet (and the software) is not used.

---For intervention 1bis, during the census: a randomization program included in the cadaster application will prompt the field agent to deliver to selected property owners (group T-t) the letters. The other group receiving no message is group T-c. Once the cadaster censuses are completed, we have the following three groups of taxpayers: T-t, T-c and C.

---For intervention 2, we will allocate each of these groups into two treatment groups (intervention 2a and 2b) and a control group (intervention 2c). Since field agents deliver tax notifications to all taxpayers, we will take advantage of this practice to deliver our different compliance messages to the treated taxpayers.

Not available

Randomization done in office by a computer

Our unit of randomization for intervention 1 are the cadaster sections. The ten tax centers in the 3 cities selected in Senegal can conduct cadaster surveys in 200 sections. We will randomly assign one-half of the sections to a treatment group (T) (100 sections) for which we use the new cadaster application software (intervention 1). Another 100 sections will be used as control. Within each cadaster section we will randomly select 45 taxpayers for our intervention 1bis and intervention 2. That makes a total sample of 9000 taxpayers (45 individuals*200 sections).

Yes

We divide the three cities of Dakar, Mbour, and Thies into cadaster sections (clusters). The ten tax centers in these cities can conduct cadaster surveys in 200 sections. We will randomly assign one-half of the sections to a treatment group (T) (100 sections) for which we use the new cadaster application software. Another 100 sections will be used as control. Our stratification will ensure that each city has treatment and control groups.

Within each cadaster section we will randomly select 45 taxpayers. That makes a total sample of 9000 taxpayers (45*200).

A group T of taxpayers will be treated for intervention 1 (introduce a software for fiscal cadaster surveys) and a group C will not. A random sample will receive an official letter during the census presenting the improved system of collecting cadaster data. The letter will explain that from then on property information is stored systematically and electronically for all taxpayers. This will allow us to evaluate whether taxpayers’ knowledge about improved technology is a channel through which the new technology’s effects on compliance can be increased. We refer to the group of taxpayers receiving this letter T-t while the one not receiving this letter is group T-c.
After finalizing the tax roll, we will send two types of messages to randomly selected taxpayers in each of the three groups T-t, T-c and C. The messages will be included in the tax notifications delivered by DGID agents:
Treatment arm 2a: Deterrence message
Treatment arm 2b: Reciprocity message
Treatment arm 2c: No message.

We thus form the following nine groups: T-t-2a, T-t-2b, T-t-2c, T-c-2a, T-c-2b, T-c-2c, C-2a, C-2b, C-2c.
According to our power calculations presented below we need 885 taxpayers per arm. Taking into account potential attrition we aim at taking 1000 taxpayers per arm adding up to a total of 9000. Our stratification will ensure that each cluster (cadaster section) will have similar proportions of individuals exposed to each t, c, 2a, 2b and 2c arm.

Effect of messages on compliance: According to the DGID, the proportion of compliant taxpayers in areas targeted by a census with the current method is 40%. For a minimum detectable effect on the proportion of compliant taxpayer of 5 percentage points (so for a compliance of 45%; 5pp is comparable to results in Castro and Scartascini, 2015 and Del Carpio, 2014), for both types of compliance message, assuming a power of 0.80 and with one baseline and two post intervention measures (baseline data will allow us to explain 30 percent of our outcome variance, a conservative and often made assumption in our previous RCT work): we need 885 taxpayers per arm (figures computed using the Stata command ‘sampsi’ using the ANCOVA method). Taking into account potential attrition we aim at taking 1000 taxpayers per arms adding up to a total of 9000. Effect of new technology on compliance: For intervention 1, we randomize at the level of the cadaster section, which means that we will have to adjust our standard errors for the clustering. The DGID has the capacity to conduct a cadaster census in 200 sections in a given fiscal year. With 100 sections in T and 100 in C, with 45 taxpayers per section (for a total sample of 9000), assuming a power of 0.80, we can identify a minimum effect of 7 percentage points (from 40% to 47%) of the technological improvement on compliance which is realistic given the results in the literature (Gadenne, 2017). For this, we assume an intra-cluster correlation (rho) of 0.1 for our main outcome. We think this is a conservative assumption given that the clusters defined by cadaster sections are relatively large geographical units (around 2400m2). If we assume a rho of 0.2 then we have detect a minimum effect 0.09 with the same power of 0.8). In addition, we will conduct a baseline survey and assume that the baseline data will allow us to explain 30 percent of our outcome variance (which is common practice). We used the Stata command ‘clustersampsi’ (to compute MDE).