Easy come easy go? On the spillover effects of tax nudges among firms in Kampala

We conduct a large-scale tax message experiment with firms in Kampala, Uganda. For the experiment, we compiled a unique database by merging business tax information from the Uganda Revenue Authority and trade license information from the Kampala Capital City Authority. This novel database enables us to measure tax compliance in a multi-level tax system. 65,269 firms are allocated to a control group that does not receive a message, firms that receive a deterrence message either through the national (URA) or local tax authority (KCCA), firms that receive the message through both tax authorities, and firms that receive a message highlighting enforcement coordination between the tax authorities. We test whether (i) the effects of a prior message experiment persist four years post-intervention; (ii) the effects replicate after repeating the original intervention; (iii) the intervention shows different effects when applied to larger firms or previously non-compliant firms; (iv) the intervention shows similar effects when applied to local taxes paid by the same population of firms; and (v) whether treatment effects at one tax tier crowd out compliance at the other tier.

2023-11-30

2023-12-14

tax compliance and compliance spillovers in a multi-level tax system.

We will analyze the impacts of the message intervention on tax compliance measured at the extensive and intensive margin. For the former, we consider whether a firm filed a tax declaration (URA), renewed a trade license (KCCA), and whether it was renewed in time (KCCA). On the intensive margin, we consider the amount declared (URA) and the trade license amount paid (KCCA). To measure spillovers, we will consider how a message regarding KCCA trade license affects URA business tax compliance and vice versa.

For reference, the five research questions we are addressing are as follows:
1. Do tax message interventions remain effective in the long run?
2. Does their effectiveness change under repeated exposure?
3. Does the effectiveness of tax message interventions change when applied to a different tax tier (local trade licenses)?
4. How does the effectiveness of tax message intervention change when including larger firms under a corporate tax scheme and semi-informal firms that are currently not paying into any national business income tax scheme?
5. Do tax messages sent at one tax tier crowd out compliance within other tax tiers?

To analyze the first research question, we will revisit a previous tax message intervention three years post-implementation. Using the analysis database of tax payments up to 2022, we will examine if the treatment effects prevail. To analyze the second research question, we will replicate the intervention. Therefore, we send an (almost) identical deterrence message through URA’s bulk message platform to firms paying presumptive income taxes (PIT). In contrast to the original design that included firms in the entire country, our analysis is limited to firms in Kampala. To test for differential effects by firm size, research question 3, we send the same text messages to corporate firms (CIT) and firms that did not pay URA tax since 2020, but that had an active trade license during that period. To test if the message effects hold when applied to local taxes, we send a similar deterrence message through KCCA’s bulk message systems. Lastly, to examine spillover effects, we test how a message sent through one tax layer affects compliance at the other tax layer. In addition, we test whether a message that highlights coordination between tax authorities reduces possible crowing-out effects.
The wording of the message will be as close as possible to the original deterrence treatment. Like in the original research, we compare the effect of this treatment message to a control group that only receives regular URA communication. This design will allow us to test if the original results can be replicated (research question 1) and evaluate the treatment effects over time and under repeated exposure (research question 2).
To analyze if the results differ by firm size and compliance status, research question 3, we will send the messages also to CIT firms and firms that are active (i.e., that pay for a KCCA trade license) but have not paid into the URA since 2020. This will allow us to test if larger firms and partially informal firms respond differently to the intervention.

For reference, the five research questions we are addressing are as follows:
1. Do tax message interventions remain effective in the long run?
2. Does their effectiveness change under repeated exposure?
3. Does the effectiveness of tax message interventions change when applied to a different tax tier (local trade licenses)?
4. How does the effectiveness of tax message intervention change when including larger firms under a corporate tax scheme and semi-informal firms that are currently not paying into any national business income tax scheme?
5. Do tax messages sent at one tax tier crowd out compliance within other tax tiers?

To analyze the first research question, we will revisit a previous tax message intervention three years post-implementation. Using the analysis database of tax payments up to 2022, we will examine if the treatment effects prevail. To analyze the second research question, we will replicate the intervention. Therefore, we send an (almost) identical deterrence message through URA’s bulk message platform to firms paying presumptive income taxes (PIT). In contrast to the original design that included firms in the entire country, our analysis is limited to firms in Kampala. To test for differential effects by firm size, research question 3, we send the same text messages to corporate firms (CIT) and firms that did not pay URA tax since 2020, but that had an active trade license during that period. To test if the message effects hold when applied to local taxes, we send a similar deterrence message through KCCA’s bulk message systems. Lastly, to examine spillover effects, we test how a message sent through one tax layer affects compliance at the other tax layer. In addition, we test whether a message that highlights coordination between tax authorities reduces possible crowing-out effects.
The wording of the message will be as close as possible to the original deterrence treatment. Like in the original research, we compare the effect of this treatment message to a control group that only receives regular URA communication. This design will allow us to test if the original results can be replicated (research question 1) and evaluate the treatment effects over time and under repeated exposure (research question 2).
To analyze if the results differ by firm size and compliance status, research question 3, we will send the messages also to CIT firms and firms that are active (i.e., that pay for a KCCA trade license) but have not paid into the URA since 2020. This will allow us to test if larger firms and partially informal firms respond differently to the intervention.

Random assignment with a quasi random number generator.

firm level randomization

No

65.269 firms

65.269 firms

Pure Control:
12.406 firms
URA deterrence
12.406
KCCA deterrence
12.406
Both deterrence
12.406
Coordination
12.406