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

We partnered with the tax administration (DGID) and with a private local software company to define a modernized property taxation protocol, and to develop an application with both Android and Web components which encompass all stages of the fiscal chain. The application can be utilized in all areas of Senegal in the future, but the RCT focuses on the region of Dakar (cities of Dakar, Pikine, Guediawaye, Rufisque).

The intervention corresponds to a comprehensive property tax census carried out in treated areas by the tax administration, using the new application. The intervention seeks both to detect unregistered properties and to update the valuation roll. The census is accompanied by sensitization activities carried out in coordination with municipalities.

Within treated areas where the property tax census will occur, there are two possibilities for the way the tax base -- the value of the property -- is assessed, leading to two treatment branches. In the first treatment branch, tax agents in the field assess the value of the property using their discretionary judgement. This involves their knowledge of the area and potential interactions with residents. In the second treatment branch, the application allows for the semi automatized estimation of the property value based on observable characteristics that the agent fills in (rule based, or formula based valuation method). This valuation method is inspired from CAMA methods and Points Based Valuation methods, adapted to the Senegalese context.

2019-07-15

2023-08-31

For the comparison T and C: tax payment; tax revenues.

For the comparison T1 and T2: tax assessment, tax payment; tax revenues.

Tax notification distribution
Compliance Rate
Accuracy of property valuation
Adjustment of tax assessment to taxpayer's capacity to pay
Corruption/Collusion
Taxpayer satisfaction
Taxpayer attitudes toward taxation and public institutions
Resident knowledge of local taxation
Resident demand for local government accountability
Implication of neighborhood delegate in local governance dynamics
Pass-through of the tax to rents
Demand for, and Regularization of property ownership

The region of Dakar is divided into around 700 cadastral sections. Among these, 194 sections have been identified as eligible for the program -- eligibility being defined as: having up to date cadastral information (plot identifiers); having a tax potential; not being a traditional village, an informal settlement nor an industrial area. These 194 sections have been randomly divided into treatment and control group. The property tax census using the new application is the treatment that will be implemented in the 97 treatment sections. Within these treated sections, another randomization was done to assign 49 sections for the discretionary (or current) valuation method, and 48 sections for the rule based valuation method. The software is configured accordingly for each section.

Not available

Randomization done in office by a computer

Our unit of randomization is the cadastral section. The region of Dakar is divided into around 700 sections. The average number of plots in a section is 397.

Yes

The planned number of clusters (cadastral sections) is 194.

Administrative data: 194 clusters, 77,000 plots. Survey data: We plan to survey 20 property owners in each section. Total number of observations: 3,880.

Administrative data:
Control -- 97 clusters, 38,000 plots.
Treatment 1 Discretion -- 49 clusters, 20,500 plots
Treatment 2 Rule -- 48 clusters, 18,500 plots

Survey data:
Control -- 97 clusters, 1,940 respondents.
Treatment 1 Discretion -- 49 clusters, 980 respondents.
Treatment 2 Rule -- 48 clusters, 960 respondents.

Effect of fiscal census program on extensive margin of tax compliance: our baseline survey showed that 13% of eligible property owners paid the property tax in 2018. With 97 sections in T, and 97 in C, 397 properties per section on average (total number of plots in T and C sections: 77,208), and based on an intra cluster correlation of 0.08 (as observed in baseline data) and a power of 0.80, we can detect an effect of 4 percentage points (from to 13 to 17%) of the program on tax compliance . This is realistic considering the existing literature on property tax. This minimum detectable effect is also sufficient from a policy point of view. Indeed, given the significant cost of modernizing the tax system, an effect below 4 percentage points would not be cost effective. We use the following stata command: “clustersampsi,binomial detectabledifference p1(0.13) m(397) k(97) rho(0.08)” Effect of rules vs discretion on accuracy of tax assessment: In the baseline data, for 22% of cases property value implicitly assessed by the tax administration based on the amount of tax paid (as declared by the owner) is within 30% of the “objective” annual rental property value assessed by a real estate expert. With 48 sections in the “Rule” treatment arm and 49 sections in the “Discretion” treatment arm, with a survey sample size of 20 properties in each section, assuming a power of 0.80 and based on an intra-cluster correlation of 0.01 (observed in the baseline data), we can detect a difference of 6 percentage points (from 22 to 28%) in the proportion of “correctly assessed” properties between the two assessment regimes. Effect of fiscal census program on local governance outcomes: This will be measured using taxpayer survey data, clustered at the section level. Our baseline survey showed that 46% of property owners were in touch with a representative of local government (municipality, neighborhood delegate) in the past 6 months. Assuming an intra-class correlation of 0.04 (from baseline data), and a power of 0.80, with 97 sections in T and 97 in C, with a survey sample size of 20 respondents in each section, we can detect an effect of 6 percentage points (increase from 46 to 52%) on taxpayer interactions with local administrations. This corresponds to a 13% increase, which we believe is realistic considering the literature. We use the following stata command: “clustersampsi,binomial detectabledifference p1(0.46) m(20) k(97) rho(0.04)”