MARCH 2019 PROMOTIN G TAX COMPL IAN C E IN KOSOVO WITH B EHAVIORAL IN SIGH T S Marco Hernandez - Jonathan Karver - Mario Negre - Julie Perng Kosovo: Promoting Tax Compliance with Behavioral Insights © 2019 International Development Agency / The World Bank 1818 H Street NW Washington, DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org This work is a product of the staff of The World Bank in cooperation with the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH and with external contributions. The findings, in- terpretations, and conclusions expressed in this work do not necessarily reflect the views of the GIZ, The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judg- ment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. Translations—If you create a translation of this work, please add the following disclaimer along with the attribution: This translation was not created by The World Bank and should not be considered an official World Bank translation. The World Bank shall not be liable for any content or error in this trans- lation. Adaptations—If you create an adaptation of this work, please add the following disclaimer along with the attribution: This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adaptation are the sole responsibility of the author or authors of the adaptation and are not endorsed by The World Bank. Third-party content—The World Bank does not necessarily own each component of the content con- tained within the work. The World Bank therefore does not warrant that the use of any third-party– owned individual component or part contained in the work will not infringe on the rights of those third parties. The risk of claims resulting from such infringement rests solely with you. If you wish to re-use a component of the work, it is your responsibility to determine whether permission is need- ed for that re-use and to obtain permission from the copyright owner. Examples of components can include, but are not limited to, tables, figures, or images. 02 Kosovo: Promoting Tax Compliance with Behavioral Insights Acknowledgments This report has been prepared at the request of the Government of Kosovo. The team is exceedingly grateful to the Ministry of Finance of Kosovo and the Tax Administration of Kosovo (TAK) for their fruitful cooperation in this initiative and for providing knowledge, data, and advice. This report was prepared jointly by staff at the World Bank and the Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) and benefited from contributions from staff at the Tax Administration of Kosovo. The initiative was financed by Deutsche Gesellschaft für Interna- tional Zusammenarbeit (GIZ) GmbH, on behalf of the German Federal Ministry for Econom- ic Cooperation and Development (BMZ) and was carried out in a partnership between the World Bank’s Mind, Behavior, and Development (eMBeD) unit and the GIZ Public Finance Project in Kosovo. The World Bank team was led by Jonathan Karver (Research Analyst), Marco Antonio Her- nandez Ore (Program Leader and Lead Economist), Mario Negre (Consultant), and Julie Perng (Research Analyst). The extended team included Abigail Dalton (Operations Officer), Monica Robayo (Economist), Asli Senkal (Economist), Agim Demukaj (Economist), and Lundrim Aliu (Communications Officer). The World Bank team worked under the overall guidance of Linda Van Gelder (Regional Director for the Western Balkans), Marco Mantovanelli (Country Man- ager for Kosovo and the former Yugoslav Republic of Macedonia), Renos Vakis (Co-Head of the Mind, Behavior, and Development eMBeD unit), Oscar Calvo-Gonzalez (Practice Manager and Supervisor of the Mind, Behavior, and Development eMBeD unit), Carlos Silva-Jauregui (Acting Practice Manager, Poverty and Equity), and Gallina Vincelette (Practice Manager, Macroeco- nomics, Trade, and Investment). The GIZ team was led by Kadri Jõgiste (Project Leader, Public Finance Project), and included Njomza Ilazi (Advisor, Public Finance Project), Ramadan Islami (Project Coor- dinator, Public Finance Project), Sylvia Van Ziegert (Senior Advisor, Sectoral Depart- ment) and Stefanie Rauscher (Senior Advisor, Sectoral Department). The GIZ team worked under the overall guidance of David Oberhuber (Country Director, Kosovo and the former Yugoslav Republic of Macedonia) and Henning Twesten (Principal Economist, Europe, Mediterranean Region, Central Asia Department). 03 Kosovo: Promoting Tax Compliance with Behavioral Insights We would especially like to thank the dedicated staff at the Tax Administration of Kosovo for their continuous support and contributions throughout the duration of the project. We are especially grateful to Ilir Murtezaj (Director General), Sakip Imeri (Former Director General), Fjolla Muja (Directorate for International Cooperation), Ardita Haxhnikaj Demi (Taxpayer Education and Service Department), Valdet Shala (Directorate for Debts and Tax Account Management), Njomëza Alaj (Information and Communication Technology De- partment), Kimete Bilalli (Information and Communication Technology Department), and Adelina Sutaj (Call Center). Excellent organizational and administrative assistance was provided by Ivana Bojic. Robert Zimmermann provided assistance in editing, and Lorena Guedes in designing the report. 04 Kosovo: Promoting Tax Compliance with Behavioral Insights Contents Acknowledgments.....................................................................................................................03 Abbreviations and Acronyms.................................................................................................. 05 Executive Summary................................................................................................................... 06 Introduction................................................................................................................................10 Tax Compliance in Kosovo....................................................................................................... 15 Behavioral Interventions in Tax Compliance: Three Experiments...................................21 Summary of Results................................................................................................................... 25 4.1 Letter Reminders to Taxpayers Late in Submitting their PIT Declarations............................................................. 26 4.2 E-mail Reminders to Firms Subject to the Value Added Tax...................28 4.3 SMS Reminders to Firms Subject to the Value Added Tax......................30 Process Makes Perfect: Lessons Learned from Implementation.....................................33 5.1 Implementation Successes..............................................................................34 5.2 Implementation Challenges............................................................................ 36 5.3 Opportunities Ahead........................................................................................38 Conclusion...................................................................................................................................42 References...................................................................................................................43 Notes............................................................................................................................ 45 Annexes........................................................................................................................47 Abbreviations and Acronyms EU European Union GDP Gross Domestic Product GIZ Gesellschaft für Internationale Zusammenarbeit PIT Personal Income Tax RCT Randomized Controlled Trial SMS Short Messaging Service TAK Tax Administration of Kosovo VAT Value Added Tax 05 Kosovo: Promoting Tax Compliance with Behavioral Insights Executive Summary As in many countries, tax collection is a develop- the World Bank and GIZ applied behavioral in- ment challenge in Kosovo. Kosovo is one of the sights to promote tax compliance among specif- poorest and youngest countries in Europe in terms ic groups of taxpayers. Three experiments were of gross domestic product (GDP) per capita and designed, implemented, and evaluated in 2018 both demographics and statehood. The lack of an that involved sending behaviorally informed re- independent monetary policy—given that Kosovo minders using letters, e-mails, and short mes- has adopted the euro as the national currency— saging service (SMS) messages to various groups means that ensuring the sustainability of fiscal of taxpayers to induce timely and honest decla- policy is critical. However, limited tax revenues rations and payments. The short-term objective hamper the government’s ability to address eco- of these trials was to increase the number and nomic cycles. Between 2011 and 2017, total gov- timeliness of tax declarations. ernment revenue amounted to about 14 percent of GDP, below the average of 19 percent among Simple, behaviorally designed messages were countries in Europe and Central Asia. Unlike other effective in inducing tax declaration in Kosovo. countries that collect significant nontax revenues Messages helped raise the tax declaration rate (for example, from natural resources), the Koso- by an average of around 3 percentage points var government relies on taxes for more than 85 during a period of between four and six weeks. percent of its revenues. Mobilizing tax revenues is Among personal income tax (PIT) declarations, therefore critical from both an efficiency perspec- this represents a 59 percent increase in compli- tive (to ensure that taxes are collected at the low- ance, equivalent to over 200 more annual tax est cost for the public administration) and an equi- declarations among participants. The likelihood ty perspective (to ensure that all taxpayers comply of payment rose in many instances, and no sig- by declaring and paying their due taxes). nificant difference was found in the amounts of taxes paid. The Tax Administration of Kosovo (TAK) re- quested assistance from the World Bank and Lessons from the tax experiments in Koso- the Deutsche Gesellschaft für Internationale vo highlight the benefits of rigorous impact Zusammenarbeit (GIZ) to address this challenge evaluation and the need to establish process- using an evidence-based approach. To this end, es that help integrate tax collection functions 06 Kosovo: Promoting Tax Compliance with Behavioral Insights and data systems. The effectiveness of tax col- site-based messaging or phone calls through a lection efforts is a function of a myriad of fac- call center. The populations of interest might tors, many of which are context specific. Estab- involve unregistered income earners and busi- lishing a process to test tax collection efforts, nesses. Future tax experiments might also learn learn from them, and adapt systems based on from the challenges outlined in this report by in- the lessons is at least as important as the impact vesting in the design and in the process equally of the interventions on revenue mobilization, and by relying closely on the results from each which represents a longer-term goal that itself trial to test, learn, adapt, and retest. is a function of changes in declaration behav- ior. The three experiments conducted in Kosovo generated lessons in two main areas: the quality of information systems and the existing com- munication infrastructure. Notwithstanding the resources available to the tax administration to reach out to taxpayers and to monitor tax com- pliance, an inability to locate and contact tax- payers can make or break a tax collection initia- tive. In other words, a communication campaign to remind taxpayers to declare on time and pay their due taxes will only be effective if taxpay- ers receive and read the reminders. Hence, the implementation of these tax interventions in Kosovo provides a critical learning experience and capacity-building exercise for policy makers. It is recommended that similar tax collection efforts in Kosovo in the future investigate ad- ditional communication channels and popula- tions of interest. The former might involve web- 07 Kosovo: Promoting Tax Compliance with Behavioral Insights Përmbledhje ekzekutive Si në shumë vende, mbledhja e tatimit është sellschaft fuer Internationale Zusammenarbeit sfidë zhvillimore në Kosovë. Kosova është një (GIZ) për ta adresuar këtë sfidë duke e përdorur nga vendet më të varfra dhe më të rejat (në as- një qasje të bazuar në dëshmi. Për këtë qëllim, pektin e demografisë dhe të shtetësisë) në Evropë. Banka Botërore dhe GIZ aplikuan njohuritë rreth Mungesa e një politike monetare të pavarur - duke sjelljes së njerëzve për ta promovuar pagesën e qenë se Kosova e ka aplikuar Euron si monedhën tatimeve në mesin e grupeve te veqanta të tatim- e saj - do të thotë se sigurimi i qëndrueshmërisë paguesve. Tri eksperimente u dizajnuan, u zbat- së politikës fiskale është kritik. Megjithatë, të uan dhe u vlerësuan në vitin 2018, që përfshinin hyrat e kufizuara nga tatimi pengojnë aftësinë e dërgimin e mesazheve përkujtuese bazuara ne qeverisë për t’i adresuar ciklet ekonomike. Midis informatat rreth sjelljes (përmes letrave, postet viteve 2011 dhe 2017 të hyrat e teresishme të elektronike dhe mesazheve te shkurtera SMS) qeverisë në Kosovë arritën në rreth 14 për qind grupeve të ndryshme të tatimpaguesve për ta të Bruto Prodhimit Vendor (BPV), nën mesataren nxitur deklarimin e saktë dhe pagesën me kohë. prej 19 për qind për vendet e Evropës dhe të Azisë Qellimi afatshkurter i ketyre tentativave ishte Qendrore. Ndryshe nga vendet tjera që mbledhin rritja e numrit dhe permiresimi i afatit kohor ne sasi të konsiderueshme nga të hyrat jo-tatimore deklarimet e tatimit. (p.sh., nga burimet natyrore), qeveria e Kosovës mbështetet në të hyrat tatimore për më shumë se Mesazhet e thjeshta, të disenjuara sipas sjelljeve 85 për qind të të hyrave të saj. Në këtë kontekst, ishin efikase në rritjen e deklarimeve tatimore mobilizimi i të hyrave tatimore në Kosovë është i në Kosovë. Mesatarisht, mesazhet ndihmuan në një rëndësie të veqantë: si nga perspektiva e efi- rritjen e normës së deklarimit të tatimit në rreth kasitetit (për të siguruar se tatimet mblidhen me 3 pikë përqindjeje në deklarimin e tatimeve në të koston më të ulët për administratën publike) ashtu ardhura përsonale (TAP), kjo paraqet 59 përqind edhe nga perspektiva e ekuitetit (për të siguruar se rrritje e përmbushjes, e barabartë me 200 të gjithë tatimpaguesit janë në pajtim me deklar- deklarime tatimore më shumë mes pjesmarrësve imin dhe pagesën e tatimeve). - gjatë periudhës prej rreth 4-6 javëve. Gjasat e pagesës ishin dukshëm më të larta në shumicën Administrata Tatimore e Kosovës (ATK) kërkoi e rasteve, dhe nuk gjetëm dallime statistikisht asistencë nga Banka Botërore dhe Deutsche Ge- të rëndësishme në lidhje me shumën e tatimeve 08 Kosovo: Promoting Tax Compliance with Behavioral Insights të paguara. i marrin dhe i lexojnë këto porosi përkujtuese. Prandaj, implementimi i ndërhyrjeve tatimore në Mësimet nga eksperimentet tatimore në Kosovë Kosovë siguron një përvojë kritike mësimore dhe nxjerrin në pah përfitimet nga vlerësimi rigoroz për ndërtim të kapaciteteve për politikëbërësit. i ndikimit dhe nevojën për të vendosur procese që ndihmojnë në integrimin më të mirë të funk- Rekomandohet që përpjekjet e ngjashme për sioneve të mbledhjes së tatimeve dhe të siste- mbledhjen e tatimeve në Kosovë në të ardhmen meve të të dhënave. Efektiviteti i përpjekjeve në të hulumtojnë për kanale shtesë të komunikimit grumbullimin e tatimeve është një funksion i disa si dhe grupeve të ndryshme të interesit. Të para faktorëve, shumë prej së cilëve janë specifikë në mund të marrin parasysh porositë e dërguara kontekst. Krijimi i një procesi për t’i testuar përp- përmes faqeve të internetit, ose telefonatat e jekjet e mbledhjes së tatimeve, për të mësuar prej bëra nga një qendër e telefonatave. Të fundit tyre, dhe për t’i përshtatur sistemet në bazë të mund të përfshijnë fituesit e të ardhurave dhe këtyre mësimeve është po aq e rëndësishme, sa bizneset e paragjistruara. Eksperimentet e ard- edhe ndikimi që këto ndërhyrje mund të kenë në hshme lidhur me tatimet mund të mësojnë nga mobilizimin e të hyrave, e që paraqet një qëllim sfidat e theksuara në këtë raport duke investuar më aftgjatë. Nga tre eksperimentet e zhvilluara në barabart si në proces edhe në dizajn dhe të ba- Kosovë, janë identifikuar një numër i mësimeve të zohen në rezultatet nga çdo provë për të testuar, nxjerra në dy fusha kryesore: cilësia e sistemeve mësuar, përshtatur dhe ri-testuar. të informacionit dhe infrastruktura ekzistuese e komunikimit. Pavarësisht burimeve në dispozi- cion të administratës tatimore pë të arritur tek tatimpaguesit dhe për të monitoruar përmbush- jen e tatimeve, pamundësia për t’i kontaktuar ose gjetur tatimpaguesit mund ta bëjë ose ta prishë nismën për mbledhjen e tatimeve. Thënë thjesht, një fushatë komunikimi ku autoriteti tatimor i përkujton tatimpaguesit që t’i paguajnë tatimet e tyre do të jetë efektiv vetëm nëse tatimpaguesit 09 Kosovo: Promoting Tax Compliance with Behavioral Insights 1. Introduction Kosovo is one of the poorest and youngest enues, accounting for about 88 percent of countries in Europe in terms of gross do- total revenues and grants during 2014 and mestic product (GDP) per capita and both 2018. They are thus critical in financing public demographics and statehood.26 The youth goods and services, such as education, health of the country represent a prospect for future care, and infrastructure. Moreover, given prosperity, but are hampered by structural is- that Kosovo uses the euro as a unit of mone- sues, such as informality. Coupled with the tary exchange, the government also has only population’s perceptions of institutional cor- limited autonomy in monetary policy. Fis- ruption, years of mistrust toward authorities cal policy—the use of government revenues preceding the formation of the republic, and and expenditures—is thus its main tool for low satisfaction with public service deliv- influencing the economy. The efficient mobi- ery, collecting taxes represents a substantial lization of tax revenues is therefore essential challenge (World Bank 2017).27 Mobilizing to ensuring economic sustainability. tax revenue efficiently and fairly is a priority of the government of Kosovo. Tax mobiliza- Tax collection in Kosovo has expanded in tion is crucial to ensuring sustainable devel- recent years, but a number of challeng- opment and fostering independence from es remain, and there is ample room for foreign aid over the long term (Mascagni improvement. Tax revenues as a share of 2015). Indeed, the 2015 Addis Ababa Action GDP rose from 21.1 percent in 2008, when Agenda, Financing for Development, calls on the country became independent, to an es- countries to step up their efforts to mobilize timated 23.7 percent in 2017. Despite this domestic resources, recognizing that much progress, there is ample room to improve tax of the public financing necessary to achieve mobilization in Kosovo, given that tax reve- the Sustainable Development Goals must be nues as a share of GDP are lower than in peer generated domestically.1 In Kosovo, taxes economies in Europe and in some economies are the most important source of public rev- at similar levels of income (figure 2, panel 10 Kosovo: Promoting Tax Compliance with Behavioral Insights b). For example, while the average tax revenue pay their taxes on time has involved a combina- as a share of GDP between 2014 and 2017 in tion of incentives (carrots) and sanctions (sticks). Kosovo stood at 22.4 percent, it stood at 24.4, The latter have traditionally been the focus of tax 26.1, and 35.4 percent in Albania, Montenegro, collection strategies, that is, tax enforcement pro- and Serbia, respectively. The challenge of low tax cedures, including fees and penalties. Traditional compliance is especially acute in the case of the means of enforcing compliance generally involve value added tax (VAT) and the personal income individual audits and residential or business visits, tax (PIT) because of a combination of tax eva- both of which require a level of human resourc- sion and weak enforcement capacity, that is, the es that is difficult to finance and may even involve inability to pursue noncompliant taxpayers. For changes in policy, which are commonly infeasible. instance, a World Bank (2014) study estimated In Kosovo, moreover, limited budgetary resources the size of the large tax gaps in the VAT, PIT, and and limited institutional capacity for reform in tax the corporate income tax, wherein the VAT gap administration make investment difficult in tradi- was estimated at about 34 percent of collections; tional enforcement mechanisms such as audits. the PIT gap was about three times the amount It is therefore important also to consider carrots, collected; and the corporate income tax gap was for example, reminders, in the design and imple- estimated at about 17 percent of collections. mentation of tax collection strategies because Low levels of tax compliance have negative con- these may be effective and less costly. How might sequences. For the private sector, for example, one improve tax collection in Kosovo at low cost tax evasion reduces collections and also creates and without changing legislation? Evidence-based an uneven playing field that concedes excessive strategies that respect the principles of testing, space to politically connected firms. This nega- learning, and adapting through the use of rigorous tively affects competition and productivity. Ul- analysis may be the answer. timately, perceptions that tax compliance is low, whether they are correct or not, can influence re- Behaviorally informed interventions represent ality by changing social norms around taxes. This an innovative, evidence-based approach to is because taxpayers who believe others are not improving tax compliance and the efficiency tax compliant may choose to avoid paying taxes of tax administration. They embody the social, themselves (Torgler 2007). psychological, and economic factors that affect how people think and act. They aim to identify Traditional measures to boost government efficiency gains given an existing bureaucratic revenues include changes in tax legislation and and technological environment. Behaviorally in- reforms in tax administration; however, such formed policy can provide creative solutions to measures can be costly, are often politically difficult challenges, often at low cost. This is true difficult to negotiate, and can take time to im- of tax collection, wherein compliance among in- plement. Encouraging individuals to declare and dividuals and firms is rarely only a function of the 11 Kosovo: Promoting Tax Compliance with Behavioral Insights expected probability of receiving an audit and a major part in taxpayer behavior demonstrate typically depends also on social norms and the that communication is a fundamental tool for im- framing of obligations, in addition to an evalua- proving compliance. By highlighting the benefits tion of risk. The World Bank has also been ex- of taxation; eliciting notions of fairness, respon- ploring the application of behavioral insights in sibility, and morality; emphasizing social norms; various policy spaces. World Development Report and framing information differently, taxpayer be- 2015: Mind, Society, and Behavior (World Bank havior can be nudged in a way that uncovers bot- 2015) highlights that the application of behavioral tlenecks unaddressed by traditional notions of science has the potential to amplify the effective- deterrence (see Hallsworth et al. 2017; Hernan- ness of public policy. People think automatically dez et al. 2017; Brockmeyer et al., forthcoming). and socially and often use mental models that are Evidence from a wide range of countries demon- unconscious. In other words, they apply heuris- strates the advantage of behavioral interventions tics and shortcuts that are particular to each con- to promote tax compliance because these gener- text. They also tend to think in terms of stories or ally involve modifications to existing systems and narratives rather than data points. These insights processes and thus tend to be cost-effective and can help policy makers more closely align their politically feasible. In recent years, the impact of communication strategies with the behavior of behavioral interventions has been measured us- citizens. Policy makers have historically struggled ing randomized controlled trials (RCTs), which are to approve and implement tax reforms because now widely applied in medicine, business, and in- of political opposition and bureaucratic inertia. ternational development. For example, the effec- Nonregulatory interventions rooted in the exper- tiveness of tax reminders has been tested using imental methodology of behavioral economics RCTs in a number of countries, including Argenti- can provide a valuable alternative to more con- na, Australia, Austria, Chile, Costa Rica, Denmark, ventional reform programs. Germany, Guatemala, Israel, Peru, Poland, Swit- zerland, the United Kingdom, the United States, Examples from international experience un- and Venezuela.3 Common to these RCTs is their derline the value of the behavioral approach to reliance on existing data collected and managed promote tax compliance, and the importance of by tax authorities, leading to rapid, low-cost im- establishing a process for testing, learning, and plementation. adapting.2 Traditional approaches to tax compli- ance, which revolve around deterrence messages Behavioral interventions are easily replicable like informing taxpayers about the probability of and scalable, stimulating a process of adaptive an audit, for example, represent a sizable portion learning. The success of behavioral messages of this literature (see, for example, Carrillo, Pomer- meant to incentivize greater tax compliance has anz, and Singhal 2017). However, recent stud- been clearly documented in recent research. Pi- ies emphasizing how psychological factors take oneering research by the Behavioural Insights 12 Kosovo: Promoting Tax Compliance with Behavioral Insights Team (BIT) and the tax authority in the United the country and tested the impact of reminder Kingdom highlights the efficacy of behaviorally messages on tax compliance outcomes. Fig- informed messages communicated by standard ure 1 displays a timeline of activities conduct- mail. Messages invoking social norms, such as ed during the collaboration in 2018. A behav- the average number of people who pay taxes ioral diagnostic was carried out in February on time, boosted the payment of declared tax 2018, and all three trials were designed and liabilities by up to 5.1 percentage points within implemented between March and September. 23 days of delivery relative to a control group The diagnostic was meant to inform the design (Hallsworth et al. 2017). The World Bank and of the trials, specifically, the method, content, the BIT also worked closely with tax authori- and tone of the messages sent to taxpayers. ties in Guatemala to design better communica- This was accomplished through a combination tion strategies using these and other behavioral of semi-structured and unstructured interviews insights (Kettle et al. 2016). They found that a with taxpayers and various stakeholders, re- letter relying on a deliberate choice (omission spectively, as well as conversations with units versus commission) helped raise the average within the tax administration. This formative amount paid per taxpayer by $17.95 (269 per- research keyed on the success of the model that cent). Undoubtedly, the effectiveness of these views the taxpayer as client and that emphasiz- behavioral messages is not universal, and con- es voluntary tax compliance through a positive text is key. Merely because an intervention is tone. This ultimately informed the decision to successful in one setting does not guarantee promote positively framed messages, such as the success in another. Thus, while evidence on oth- benefits of paying taxes, rather than negatively er countries can provide insights to inform the framed messages, such as deterrence. design of behavioral interventions, there is still a need to experiment to determine whether be- This policy note summarizes the findings of havioral insights can promote tax compliance in these tax collection activities in Kosovo and fo- Kosovo. cuses on two key outcomes: the effectiveness of the trials in terms of the level of declarations The Tax Administration of Kosovo (TAK) has (the impacts) and the process for carrying out begun to apply behavioral insights to promote the trials (the implementation). Experimental tax compliance. In 2018, through a partnership trials have generally concentrated on measur- between the World Bank and the Gesellschaft able results, that is, how much additional rev- für Internationale Zusammenarbeit (GIZ), TAK enue can be attributed to some policy or how carried out the first three RCTs to improve tax the declaration rate may be increased by tweak- compliance, that is, to increase the number ing an existing framework. However, in Koso- and timeliness of tax declarations. These tar- vo, the focus was on a dimension that gener- geted taxpayers subject to PIT and VAT across ally receives less attention in the literature: the 13 Kosovo: Promoting Tax Compliance with Behavioral Insights effectiveness of implementation as an exercise taxpayer registry are associated with incorrect in assessing the operations of a tax adminis- addresses or no addresses at all. The limited in- tration, building capacity, and learning lessons formation on taxpayer addresses is especially rel- that may be extracted for future trials. Thus, evant because it clearly affects the analysis of the a key outcome of the trials has been the discov- reaction of the population of interest to the inter- ery that nearly two-thirds of the entries in the ventions and in changing the mindsets of policy Figure 1. TIMELINE OF ACTIVITIES DEC 2018 Results dissemination SEP 2018 I m p l e m e n t a t i o n o f VAT S M S t r i a l Te s t Le a r n JUL Ad a p t 2018 I m p l e m e n t a t i o n o f VAT e - m a i l t r i a l Re - t e s t ! M AY B ra i n s t o r m i n g f o r VAT t r i a l s , e a r l y 2018 results PIT trial MAR Implementation of PIT trial 2018 FEB 2018 Diagnostic of tax compliance 14 Kosovo: Promoting Tax Compliance with Behavioral Insights makers and implementers with regard to their tax clients. pared with other countries in Europe and Cen- 2. tral Asia. Moreover, a feature of the World Bank (2014) report is the analysis of the finding that the tax system is highly dependent on customs taxes collected at the border and that there is a need to shift from border taxes to domestic sources of revenue, including the VAT, PIT, the Tax Compliance in corporate income tax, and property taxes. Kosovo Tax compliance has generally improved over time, but the collection of VAT and PIT reve- nues remains problematic. Average tax revenues in Kosovo over the past seven years represent Kosovo has a simple tax system and relatively 14 percent of total revenues. Although the tax- low tax rates. A World Bank (2014) report ti- to-GDP ratio is similar in Kosovo and other coun- tled “Kosovo Public Revenues: Tax Policies, Tax tries in the Western Balkans, tax revenue is over- Evasion, and Tax Gaps” presents a detailed as- whelmingly sourced from indirect taxation (more sessment of the Kosovar tax system. The coun- than 80 percent), particularly VAT, excise, and in- try’s VAT rate is flat, at 18 percent, and the PIT ternational trade taxes. Meanwhile, PITs and tax- rate is progressive, at 0, 4, 8, and 10 percent es on profit account for 14 percent of tax revenue on income tax brackets of €0–€80, €81–€250, (less than 4 percent of GDP), about half the share €251–€450, and over €450 a month.4 Kosovar in other countries in the Western Balkans and in tax legislation is broadly aligned with European the EU accession countries. Union (EU) standards. Tax rates are low com- 15 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure 2. TAX REVENUE AS A SHARE OF GDP The heavy dependence on indirect taxation is a. Kosovo, by source consistent with the import intensity and transfer dependence characteristic of the Kosovar econo- 16 my. Indeed, remittances reach 14 percent of GDP, 14 higher than in all other Western Balkan countries. The share of the imports of goods and services 12 in GDP was around 52.5 percent in 2017, while Share of GDP 10 the trade deficit was 25.9 percent of GDP (World Bank 2017). The government budget deficit of 6.3 8 percent in 2017 and the mounting public debt, 6 at 16.6 percent of GDP in the same year, both 4 underscore the need to raise domestic revenues to engage along a financially sustainable path.5 2 The low revenue from income taxes derives from 0 2011 2012 2013 2014 2015 2016 2017 a combination of factors, mainly relatively low tax rates, the narrow tax base (because of low labor CIT PIT Domestic VAT Border VAT force participation and high unemployment), and the large informal sector (figure 2, panel a). The most recent tax gap estimates (2014) suggest b. Western Balkans, total that the productivity of revenue from income tax- es is low, particularly in the case of the PIT. Only 50 12 percent of the PIT tax base was actually taxed, and the corporate income tax rate was similarly 40 low (World Bank 2014). It has been suggested Tax revenues (% GDP) that PIT revenues are especially low in Kosovo 30 because of the PIT exemptions allowed on vari- ous sources of income, weak tax administration, and the size of the informal economy. In addition, 20 a low share of revenue from the PIT relative to the VAT suggests that the system exerts a regres- 10 sive distributional impact, at least on the taxation side, which is particularly the case considering the 0 Albania Bosnia & Kosovo FYR Serbia Montenegro maximum tax rate is only 10 percent, rather than Herzegovina Macedonia the more than 40 percent in many EU countries. 2008 2016 Despite improvements in revenue, important Source: World Bank calculations based on data from national authorities. 16 Kosovo: Promoting Tax Compliance with Behavioral Insights challenges persist. Tax revenue as a share of between 2011 and mid-2018. These records in- GDP increased from 13 percent to 15 percent clude contact information (addresses, e-mails, between 2014 and 2017; however, the vast and telephone numbers) that would be key to majority of this increase arose because of the any effort to establish a client-oriented tax ad- border VAT (figure 2, panel a). Tax revenues are ministration. Moreover, the registries contain projected to decline as a share of GDP over the information on corresponding tax centers (ad- medium term in the absence of measures to bol- ministrative districts), the types of taxpayers in ster domestic revenues to compensate for falling the PIT (natural person or individual business), border revenues. A gradual shift toward direct sectors and types of economic activity (if ap- taxation will be necessary as domestic produc- plicable), the gender of the taxpayer, and other tion expands and integration with the EU and details of the firm or individual business, such free trade agreements advance and thus reduce as the date the individual or firm went into busi- border revenues (World Bank 2017). Enhanced ness, the date the firm or individual registered tax administration is necessary to boost collec- with the tax authorities or the business registry, tions and raise compliance. and so on. Evaluating trends in these data, as well as the quality and consistency of the infor- Information in the tax registry provides a more mation, can provide a clearer picture of the fac- nuanced profile of compliance in Kosovo. The tors driving compliance. tax registry, which contains anonymized individ- ual records of taxpayers, including their declara- Tax registry records in Kosovo demonstrate that tion behavior in each tax system, allows for the individual compliance has improved in recent classification of taxpayers into different compli- years; yet, late declarations and late payments ance categories, particularly those who declare are frequent. Figure 3 plots key tax compliance or pay late and those who do not submit any tax indicators for the PIT and VAT systems, based payments. The tax registry also allows for the on tax registry records from 2010 through 2018. classification of firms or individual taxpayers into Two clear trends emerge: (1) late declarations groups defined by revenue and other measures and late payments are frequent in both systems, to distinguish between larger and smaller firms. but have decreased substantially; and (2) there Assessing the quality of data in the tax registry is substantial fluctuation in compliance behavior helps identify major shortcomings hindering the among VAT declarations month to month. In the tax collection effort and possible ways to ad- VAT, late payments are endemic among those dress them. The PIT registry contains records on taxpayers who actually make payments; an av- 17,696 taxpayers between 2010 and 2016 who erage of more than 90 percent of taxpayers sub- declared at least once during this period. The mit their payments after the monthly due date. VAT registry contains records on 27,929 firms submitting a monthly declaration at least once Strong disparities in tax compliance exist across 17 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure 3. TAX COMPLIANCE TRENDS, BY TAX SYSTEM, 2010–18 a. PIT declarations districts and show signs of persistence, particu- 60 larly in the VAT space. District differences seem to be responsible for most of the heterogeneity 50 in tax compliance, and this is especially true of late declarations—defined as declarations sub- taxpayers in period % of declaring PIT 40 mitted after the official due date—in the PIT 30 space (figure 4). The persistence of district dis- parities is true of the VAT much more than the 20 PIT, where reductions in the share of late dec- larations were largest among districts that, in 10 2010, were among the worst compliers, such as Ferizaj and Gjakova. In the VAT, compliance in 0 2010 2011 2012 2013 2014 2015 2016 terms of late declarations has either remained high (an average of over 40 percent) or increased Late declaration Late payment No payment in some districts, such as Gjakova, Pristina 2 and 3, and Prizren. b. VAT declarations Compliance is strongly correlated with the age 100 of the taxed businesses and the district of ope- ration across both the PIT and the VAT systems. 75 Across both the PIT and the VAT, late decla- taxpayers in period % of declaring VAT rations are a function of key individual or firm characteristics that persist, highlighting the chal- 50 lenges in tax collection that must be addressed. Among individuals shown in the annual PIT re- cords to have submitted declarations since 2010, 25 the share of late declarations is 2 percentage points higher among men than among women. This is also the case among less experienced tax- 0 2011m12 2012m12 2013m12 2014m12 2015m12 2016m12 2017m12 payers: each additional year an individual or a 2011m1 2011m6 2012m6 2013m6 2014m6 2015m6 2016m6 2017m6 2017m5 business has been registered is associated with Late declaration No payment Late payment a 1 percentage point decline in the probability of a late declaration. Taxpayers whose contact Source: World Bank calculations based on data from national information is not recorded in the registry are authorities. more likely to submit declarations late; those Note: “Late payment” is conditional on the payment having been made, whereas “no payment” refers to both without an e-mail declare late 47 percent of the those taxpayers who do not pay the tax burden and those taxpayers who report owing nothing (generally arising because of a credit in their favor from past periods). 18 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure 4. LATE DECLARATIONS, BY TAX CENTER DISTRICT time, compared with those with an e-mail, who a. PIT declarations are late only 26 percent of the time. Meanwhile, those with no address declare late 41 percent of 40 the time, compared with 27 percent of the time among those whose addresses are on file. Those assumed to be covered by an accountant because 30 period declairing late % of PIT taxpayers in of a shared e-mail address are nearly 4 percentage points more likely to declare late relative to those 20 with a unique e-mail address. Taxpayers with a tax burden are 15 percentage points less likely to declare late relative to those without a tax bur- 10 den (16 percent versus 33 percent). Among those with a burden, larger burdens (above the median) 0 Mitrovica Pristina 3 Pristina 2 Pristina 1 are less likely to declare late (15 percent versus Prizren 1 Prizren 2 Gjakova Ferizaj Gjilan Peć 18 percent). Sectoral differences represent an im- portant factor explaining differences in declaring 2010 2016 Avg. 2010 Avg. 2016 late: relative to those in the primary or second- b. VAT declarations ary sectors, taxpayers in the tertiary sector are less likely (4 percentage points) to submit their 40 declarations late. The predictive factors associ- ated with late declaration in the VAT are similar 30 to those in the PIT. Younger firms are also more period declairing late % of PIT taxpayers in likely to submit their declarations late; each ad- ditional year of appearance on the registry is as- 20 sociated with a 0.6 percentage point decline in the likelihood of late declaration. Firms in the 10 tertiary sector are 3 percentage points less likely to declare late relative to those in the primary and secondary sectors. Those with a tax burden 0 Mitrovica Pristina 3 Pristina 2 Pristina 1 Prizren 1 Prizren 2 Gjakova Ferizaj Gjilan Peć are 13 percentage points less likely to declare late relative to those without a burden, and larg- er firms (above the median of VAT output) are 2010 2016 Avg. 2010 Avg. 2016 11 percentage points less likely to declare late. The fluctuations in late declarations by month persist across years. The incidence of late dec- Source: World Bank calculations based on data from national authorities. larations is highest in December and lowest in Note: Late declarations are defined as declarations submitted after the due date, officially March 31st of the subsequent year for the PIT and the 20th of the subsequent month for the VAT. 19 Kosovo: Promoting Tax Compliance with Behavioral Insights January, April, and July, which highlights the im- to pay (nonpayers), and those taxpayers with portance of using reminders during key periods. outstanding debt (underpayers). The latter two are those who have submitted declarations, but TAK routinely communicates with taxpayers who have not completed their payments. In the- on noncompliance issues, ranging from missing ory, these groups are defined in each tax period declarations to unpaid or underpaid tax obliga- immediately after the due date, depending on tions among various types of taxpayers. These the system under consideration. At the moment communications have taken the form of auto- the trials were initiated, TAK did not have any matic e-mails generated through the electron- standardized communication strategy in opera- ic declaration system, but, more recently, have tion to remind taxpayers to declare in advance been dominated by personalized communica- of the due date. Taxpayers who were late on tions through the TAK call center. These calls declarations were generally eligible to receive are less about prevention than about treat- a phone call from the call center; however, ment: the vast majority of calls are made to the call itself was not based on a risk assessment. taxpayers with substantial outstanding debts Generally, the prioritization across taxpayers for or major inconsistencies in declaration behav- receiving calls was based on the tax burden and ior. Among those who do not declare, there are the number of declarations missed and was usu- many exceptions granted for special circum- ally intended to address a specific inconsistency stances, which are frequently rectified through in the records. This means that, effectively, TAK these communications. Taxpayers in the PIT was selective about who to contact, and one who earn below a specific threshold (described may assume that a baseline scenario involved no above) are not required to pay taxes and are communication. To guarantee this was the case, thus also not required to declare.6 during the implementation of the three exper- iments, no communications were undertaken TAK uses various communication tools to reach with taxpayers about current declarations. If out to taxpayers, but, in the past, these tools communications were made, these were related were not used regularly or evaluated system- to earlier declarations or a need for some other atically. This represents a unique opportunity clarification. to test the impact of various communication strategies in terms of both the method and the content. TAK classifies noncompliers as taxpay- ers, whether individuals or firms, who have not fulfilled a tax obligation for one or more tax pe- riods and one or more tax systems. Specifically, noncompliance may be categorized as failure to declare (nondeclarers or nondisclosers), failure 20 Kosovo: Promoting Tax Compliance with Behavioral Insights 3. Behavioral Interventions in Tax Compliance: Three Experiments Between February and October 2018, the of tax declarations submitted and their timeli- team supported the tax authority in the de- ness. The PIT space was assigned a priority be- sign, implementation, and evaluation of three cause of its annual cycle; PIT declarations must experimental trials to improve tax compli- be submitted by March 31st each year for the ance in Kosovo.7 The collaboration initially taxes corresponding to the previous calendar envisioned designing a single trial, but, thanks year. VAT declarations are required every month, to an effective use of resources and a fruitful permitting more flexibility around design and collaboration with TAK, the team was able to continuity. These types of trials, if adequately implement and evaluate three trials in two tax designed, constitute natural field experiments systems using three different communication in the sense that a treatment assignment was strategies. That expectations were exceeded randomly allocated to segments of the popula- in these trials highlights the adaptive learning tion without their knowledge. potential of randomized experiments in the tax space, that is, the ability to make marginal The scope of the work and the policy approach changes to the design at different stages of an revolved around the use of different communi- intervention, as well as the benefits of constant cation channels and content meant to address local engagement with the tax authority. the policy challenge, that is, to induce com- pliant taxpayer behavior at low cost without The objective of the three tax trials was to changes to legislation. Communication is only evaluate how innovative approaches to com- one dimension of many meant to address low munication with taxpayers using evidence and tax compliance in Kosovo. However, given the adaptive learning could improve timely and low cost and established communication pro- honest compliance behavior at low or zero cost. tocols, this dimension has the highest potential The VAT and the PIT were identified by TAK as efficiency gains, large expected impacts, and the priority areas in the effort to improve tax com- ability to shed more light on taxpayer behavior by pliance, specifically as this related to the number testing different instruments. To identify chang- 21 Kosovo: Promoting Tax Compliance with Behavioral Insights es in taxpayer behavior that could be directly, all reminders were behaviorally informed be- that is, causally attributed to the communica- cause they were designed with the clear intent tion channel and content, three experimental to make the framing of the messages more con- trials were developed over the course of eight ducive to the desired behavior.13 months. The team used an existing tax sched- ule, declaration infrastructure, and administra- To inform the design of the trials, the team car- tive data and randomly assigned taxpayers to ried out a behavioral diagnostic. A behavioral di- one of two situations in a given tax system and agnostic may be understood as an assessment of tax period: those who would be exposed to a factors that influence behavior in a given space, business-as-usual scenario (no specific remind- in this case, the factors that drive individuals er) and those who would receive one of two re- toward or away from tax compliance. The diag- minders with unique content.8 nostic was carried out in February 2018. In addi- tion to brainstorming with TAK technical teams, Each of the three trials involved a slightly dif- it involved informal discussions with accountant ferent taxpayer population.9 Each also involved associations, firm owners, and psychologists and a different design.10 However, all followed the semi-structured interviews with individual tax- same proposition: to apply behavioral insights to payers in Pristina and surrounding areas to gauge induce timely, honest declarations among a ran- perceptions of the tax administration and tax domly selected subset of the population of in- compliance in general. These interviews and di- terest.11 In each trial, the population of interest alogues highlighted the perception of corruption was divided into three segments: one segment and lack of transparency in the use of tax revenue, that received no reminder (business as usual); as well as the perception that larger firms evade one segment that received a simple, behavioral- taxes. Fieldwork revealed that taxpayers favored ly informed reminder; and one segment that re- the use of tax revenues to finance education and ceived the same behaviorally informed remind- health care. A parallel diagnostic in the jobs space er, complemented by additional language that highlights that poor public service delivery may addressed some behavioral bias or biases. The create incentives for informality (Cojocaru 2017). behaviorally informed reminder was written in According to Riinvest (2013), over 40 percent of simple, nontechnical language and contained a firm representatives believe tax evasion is justi- clear call to action, some personalization, and in- fied under certain circumstances, including in re- structions on carrying out the desired behavior. action to the (perceived) unsatisfactory supply of Figure 5 highlights the design aspects of each public services.14 Notwithstanding the low levels trial in more detail, and annex figures A.1a–A.4b of trust in public institutions, which is consistent provide snapshots of the letters and e-mails.12 with the findings of national surveys, the percep- Given that no standard letter, e-mail, or SMS re- tions in regard to TAK were positive, and the lev- minders were in place at the time of the trial, els of trust in TAK were relatively high.15 22 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure 5. DESIGN CHARACTERISTICS OF THE TAX TRIALS • 11,603 individuals • 23,622 firms with • 22,828 firms with late on 2017 PIT p e n d i n g d e c l a ra t i o n p e n d i n g d e c l a ra t i o n d e c l a ra t i o n o n J u n e 2 0 1 8 VAT o n A u g u s t 2 0 1 8 VAT • 7,747 letters sent • 10,340 e-mails sent • 11,096 messages sent • 2 letter types • 2 e-mail types • 2 SMS types Previously, we have Not paying your considered your failure taxes places an unfair Did you know 7 out of 10 firms to declare an oversight. burden on your that your VAT submit their However, if you do not fellow Kosovo citizens. contribution is declaration on time. declare now, we will Please don’t be an invested in Don’t wait,be part consider it to be your irresponsible citizen. your city? of the MAJORITY! active choice… Note: The signatures of two department directors were included at the bottom of both letters, and the director of one department at the bottom of the augmented behavioral e-mail. The first trial targeted letter reminders at addresses, which provided valuable information PIT taxpayers who were late in filing their on the efficiency of the tax registry. The eligible 2017 annual declarations. Taxpayers who, taxpayers (11,603 in total) were divided ran- by April 4, 2018, four days after the due date, domly into one of three groups of similar size: (1) had not yet submitted their PIT declarations those who would receive no letter; (2) those who for the 2017 fiscal year were identified in the would receive a simple, behaviorally informed let- tax registry and were eligible to receive a let- ter; and (3) those who would receive the same let- ter in the mail.16 These letters were posted by ter as group 2, but augmented with an additional private courier to record those who did not re- behaviorally informed message. The letters sent ceive the letter because of incorrect or outdated to group 3 included deliberate choice language, 23 Kosovo: Promoting Tax Compliance with Behavioral Insights which emphasized that a failure to act would be taxpayers fulfill their obligations (the social norms considered a choice of the taxpayer rather than an message). For the VAT e-mail trial, the TAK Informa- oversight. The private courier attempted to deliv- tion Technology Department sent out the messages er a total of 7,747 letters; multiple attempts were using their own software. All e-mails were intended made to locate proper addresses and deliver the to be sent within the same time frame, a single day letters. A recipient signature was requested, and for both the first and the second rounds of e-mails. the details of the delivery or the attempt to deliv- E-mails were sent in the second round regardless of er were recorded by the courier. The letters were whether the first e-mail was opened or the taxpay- intended to be delivered within three days of the er had already submitted a declaration. assignment of the recipients into the three groups. The third trial targeted all firms required to sub- The second trial targeted all firms required to mit August VAT declarations; it involved the use submit June VAT declarations; it involved the of SMS-based reminders a few days in advance use of e-mail–based reminders on two occa- of the due date. All firms in the VAT registry, sions, the first well in advance of the monthly excluding those that had already submitted due date. All firms in the VAT registry, excluding their declarations by September 17, 2018, or those that had already submitted declarations by for which there was no cell phone number on July 3, 2018, or for which no e-mail addresses were file, were considered eligible to receive an SMS on file, were considered eligible to receive an e-mail reminder, which was sent through a telecom- reminder from the TAK Information Technology munication firm in Kosovo. The eligible firms Department, first on July 4, then again on July 20, (22,828 in total) were divided randomly into the due date of the June declaration. The eligible one of three groups: (1) those that would re- firms (23,622 in total) were divided randomly into ceive no SMS message; (2)  those that would one of three groups of similar size: (1) those who receive a simple, behaviorally informed SMS would receive no e-mail; (2) those who would re- message; and (3) those that would receive the ceive a simple, behaviorally informed e-mail; and (3) same SMS message as group 2, but augmented those who would receive the same e-mail as group with a social norms message.18 2, but with reference to the unique fiscal numbers of the corresponding taxpayers and augmented with “Dear taxpayer, please declare and pay your positively framed, behaviorally informed messag- August VAT correctly and on time!,” read the es placed strategically throughout the body of the simple reminder, and the social norms language e-mail.17 These short behavioral messages includ- added “7 of 10 firms submit their declarations on ed (1) language to instill feelings of guilt for non- time. Don’t wait, be part of the MAJORITY!”19 A compliance, (2) language highlighting the benefits telecommunication company was hired to send all that taxes provide (the public goods message), and the SMS messages during a one-hour window on (3) language highlighting how the majority of other September 18, two days before the due date. 24 Kosovo: Promoting Tax Compliance with Behavioral Insights 4. Summary of Results All three behaviorally informed communication tract (the shared principles that serve to regulate, trials succeeded in producing statistically sig- define, and redistribute national resources). Over- nificant impacts on the declaration rate or the all, the effectiveness of the trials can be sum- payment rate in the short term, within four to marized by answering the following questions six weeks. The behavioral messages managed on each trial: not only to induce more people to submit their declarations on time, but also to encourage • Did the various reminders directed at tax- more people to submit declarations.20 These payers encourage them to declare and pay in impacts hold under various specifications, includ- a timely fashion? ing limiting samples to specific subgroups among • Did the message content lead to a separate which impacts may be expected to be low.21 This change in behavior among taxpayers that highlights the value of testing, learning, and adapt- was associated with each communication ing in tax interventions. The result is all the more method? encouraging considering the low take-up rate in two of the three trials. In those two trials, many • Did taxpayers respond differently to the letters did not reach the intended recipients, and reminders and the content of the messages? e-mails were often not read. Moreover, the total The results of each of the three trials are pre- amount of taxes declared or paid did not rise in sented below. Many of the people who should this first, limited exercise, partly because many have received reminders did not receive them be- of the intervention recipients were not obligat- cause of the incorrect or outdated postal address- ed to submit declarations. The short-term objec- es or because they ignored the e-mails. The asso- tive of these trials was to increase the number ciated intervention effects are therefore examined of individuals and firms that submitted timely on the basis of those who should have received declarations. Declaration is the first step toward the reminders (intent-to-treat) and those who increasing tax revenues because it builds a so- actually did receive the reminders (average treat- cial norm around compliance and the social con- 25 Kosovo: Promoting Tax Compliance with Behavioral Insights ment-on-the-treated). (Most SMS reminders were the declaration rate among late declarers, which received and viewed by the intended recipients.) increased by between 91 and 160 percent (fig- Had more addresses been correct or had more ure 6). The difference in the magnitude of effects people opened e-mails, the general results would on both groups—those assigned to receive a let- have been better because they would have cor- ter and those who actually did receive one—is responded to the greater incidence of effects ob- important. This is because, in the first case, the served on those who did receive the reminders. overall effect is analyzed in the trial, whereas, in the second case, the effect on the treated group is analyzed, and this indicates what might have 4.1 been the overall outcome if the take-up problem Letter Reminders had been dealt with by correcting and updating postal addresses, for example. to Taxpayers Late in Submitting their The trial had different effects across the groups. The heterogeneity in the effects on PIT Declarations the declaration rate was appreciable; im- pacts were substantially larger among some groups. This was particularly marked across the Letter reminders to PIT taxpayers late in sub- districts in which the taxpayers resided and ac- mitting their annual declarations resulted in cording to the past declaration behavior of the small, but statistically significant impacts on the taxpayers. For example, in relative terms, the im- declaration rate. The reminders were successful pact was considerably larger among natural per- in inducing more people to submit declarations sons than among individual businesses. While the closer to the due date. Among those who were share of natural persons who had submitted dec- randomly assigned one of the two letters (the larations on time rose three- to fourfold (between intent-to-treat group), there was an increase of 270 and 431 percent increase), the correspond- 2 to 4 percentage points in the declaration rate ing share of individual businesses was more mod- within one month of the transmission of the let- est, but still substantial, between 48 percent and ters.22 Only half the people assigned to receive a 94 percent increase. However, the latter group letter did receive one. This shortfall was caused reported a much higher rate of declaration in the by the large number of incorrect addresses in absence of treatment than the former (11 percent the records. If only the group is examined that versus 1 percent). actually received the letters (the treatment-on- the-treated estimate), the increase in the decla- Despite the significant impacts on the dec- ration rate is almost double, at 4 to 7 percent- laration rate and the payment rate, no ef- age points.23 This represents a large increase in 26 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure 6. SHORT-TERM IMPACT OF LETTERS ON PIT DECLARATIONS AMONG TAXPAYERS FAILING TO DECLARE BY THE DUE DATE, 2017 a. Letter assigned b. Letter received 20 20 20 those failing to declare by due date those failing to declare by due date those failing to declare by due date PIT declaration rate (%) among PIT declaration rate (%) among PIT declaration rate (%) among 16 16 16 160% 1 *** *** 12 12 12 91% 91% 73% 73% *** *** 44% *** 44% *** 8 *** 8 8 *** 4 4 4 0 0 0 No reminder No Simple reminder reminder + Simple Simple reminder + reminder Simple reminder No reminder No Simple reminder reminder+ Simple Simple reminder+ reminder Simple re deliberate choice deliberate choice deliberate choice deliberate choice c. Declaration rate, by group, March–July 2018 10 those failing to declare by due date PIT declaration rate (%) among 8 6 4 2 0 03-May 07-May 11-May 15-May 19-May 23-May 27-May 31-May 28-Mar 01-Apr 05-Apr 09-Apr 13-Apr 17-Apr 21-Apr 25-Apr 29-Apr 04-Jun 08-Jun 12-Jun 16-Jun 20-Jun 24-Jun 28-Jun 02-Jul 06-Jul 10-Jul No reminder Simple reminder Simple reminder + deliberate choice Source: World Bank calculations based on data from national authorities. Significance level: *** = 1 percent. ** = 5 percent Bars with arrows represent confidence intervals. 27 Kosovo: Promoting Tax Compliance with Behavioral Insights fects were identified on payment amounts. outcome may derive from the way the deliberate While the rate at which taxpayers declared and choice message was framed in the augmented made a payment was substantially higher among letter. The phrasing might have been interpreted those receiving a letter (1 to 2 percentage points by some recipients as a veiled threat. Most like- higher), the amounts paid were not different ly, however, the unexpectedly low impact of the among those who received a letter. This is not deliberate choice message may be the result of surprising, however, because the population of operational difficulties during implementation. interest in this trial plausibly consisted of indi- viduals who did not actually have anything to declare. Indeed, among those who were late 4.2 in submitting declarations, but who did end up submitting in the absence of the reminder, only E-mail Reminders 18 percent claimed they owed something in tax- to Firms Subject to the es. Conditional on owing something, those who received a letter did, however, submit a nonzero Value Added Tax payment at a higher frequency relative to those who did not receive a letter. In other words, The e-mail reminders sent to firms in July did not among those who owed taxes and submitted lead to a substantial increase in the submission declarations, the group that received reminders of declarations for June taxes, in payment rates, was larger than the group that did not receive or in the amount of taxes paid.24 The e-mail re- reminders. minders led to an average 0.5 to 1.2 percentage point rise in the on-time declaration rate and an The simple behavioral letter was as effective as average 1.0 to 1.6 percentage point rise in the the behavioral letter augmented by a deliberate declaration rate within two weeks after the due choice message. This outcome is not predicted date, though neither of these changes was statis- by theory. Relevant experience in other coun- tically significant. A small share, 25 percent, of the tries suggests that the inclusion of the deliberate e-mails received by firms were actually opened. choice message, along with the simple reminder, One might expect that the firms at which e-mails tends to be significantly more effective than the were opened would show a larger effect. Yet, this simple reminder alone. The reason is that the ad- was not entirely the case (figure 7). Firms at which ditional message highlights that it is an individu- e-mails were received and opened did submit dec- al’s choice to fail to submit a declaration and that larations on time more often than the other firms, failure to submit is not merely an unintentional but this could not be captured in a statistically sig- error. However, in the trial, the magnitude of nificant manner because the number of firms at the impact was larger in the case of the simple which e-mails were opened was so small. Similar letter compared with the augmented letter. This results were obtained in the case of the positively 28 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure 7. DECLARATION RATE, BY E-MAIL ASSIGNMENT AND STATUS OF RECEPTION framed e-mails. The declaration rates were higher a. By asignment to rceive the e-mail over time among firms at which positively framed 80 e-mails were received and opened, but the share VAT declaration rate (%), June of these firms was so small that the results were not statistically significant, and causality cannot be 60 claimed with high confidence. The e-mails including the additional positive framing were opened by only 40 7 percent of the firms receiving them, whereas the simple e-mails were opened by half the recipient firms. Among the 7 percent of firms at which the 20 e-mails with the augmented positive framing were opened, 74 percent submitted declarations within two weeks following the due date, compared with 0 01-Aug 03-Aug 05-Aug 07-Aug 09-Aug 11-Aug 13-Aug 15-Aug 17-Aug 19-Aug 04-Jul 06-Jul 08-Jul 10-Jul 12-Jul 14-Jul 16-Jul 18-Jul 20-Jul 22-Jul 24-Jul 26-Jul 28-Jul 30-Jul 67 percent among firms at which the e-mails were not opened. This difference was substantially small- No reminder Simple reminder Simple reminder + positive framing er—67 percent among firms at which the e-mails were not opened and 70 percent among firms at which the e-mails were opened—among firms re- b. Among those for which e-mails were opened ceiving the simpler e-mail reminder. 80 The augmented behavioral message did, howev- VAT declaration rate (%), June er, have a positive impact on the share of taxpay- 60 ing firms submitting a declaration and making a payment within a month and a half after the due date. Conditional on submitting a declaration by 40 early September, those firms that received the e-mail with a positive framing of the benefits of tax- es were 2.2 percentage points more likely to make 20 a payment (figure 8). This is an important finding because payment is generally made substantially 0 later than the declaration and rarely by the due 01-Aug 03-Aug 05-Aug 07-Aug 09-Aug 11-Aug 13-Aug 15-Aug 17-Aug 19-Aug 04-Jul 06-Jul 08-Jul 10-Jul 12-Jul 14-Jul 16-Jul 18-Jul 20-Jul 22-Jul 24-Jul 26-Jul 28-Jul 30-Jul date. Conditional on opening the e-mail reminder with augmented behavioral language, this impact No reminder Simple reminder Simple reminder + positive framing increases to a remarkable 32 percentage points. Notwithstanding the rise in the payment rate, no Source: World Bank calculations based on data from national increases were observed in the payment amounts authorities. 29 Kosovo: Promoting Tax Compliance with Behavioral Insights as a result of either of the two e-mails. differential impacts. E-mail reminders had different effects across taxpayer subgroups; more analysis is needed 4.3 to understand this heterogeneity. Reminders had substantial positive impacts on the declara- SMS Reminders to tion rate among firms below the median of VAT Firms Subject to the output, that is, the amount of VAT charged to consumers, and among firms that had recent- Value Added Tax ly failed to submit declarations, for example, in the previous few months. However, the ef- SMS reminders sent before the due date led to fect of the e-mail assignment was negative in a 2.1 to 3.7 percentage point increase in the on- the case of some groups. More analysis of the time declaration rate and a 2.2 to 3.9 percent- tax registry data is needed to understand these age point increase in the declaration rate (figure Figure 8. PAYMENT RATE, BY E-MAIL 9). Within one month of the due date, both SMS ASSIGNMENT AMONG FIRMS THAT variants led to a statistically significant rise in the SUBMITTED DECLARATIONS declaration rate. Given the high level of treat- ment compliance (80 percent of the messages 20 were received by taxpayers), it is not surprising that these results are practically the same if they are conditioned on actually receiving the SMS VAT payment rate (%), June 16 messages. Payment rates (the likelihood that a firm submitted any payment) were not high- 12 er among firms that received an SMS message; so, while the reminders induced firms to declare 8 sooner, they did not induce them to submit pay- ments sooner. 4 Similar to the letter and e-mail trials, there was 0 no impact on payment amounts in the SMS trial. 01-Aug 03-Aug 05-Aug 07-Aug 09-Aug 11-Aug 13-Aug 15-Aug 17-Aug 19-Aug Payment amounts were no higher among firms 04-Jul 06-Jul 08-Jul 10-Jul 12-Jul 14-Jul 16-Jul 18-Jul 20-Jul 22-Jul 24-Jul 26-Jul 28-Jul 30-Jul that received the SMS reminders than among No reminder Simple reminder Simple reminder + positive framing those that received no reminder. This outcome may have arisen because of the tendency to make payments a substantial time after submitting the Source: World Bank calculations based on data from national authorities. 30 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure 9. IMPACT OF SMS MESSAGES ON VAT DECLARATIONS AFTER ONE MONTH a. Letter assigned b. Letter received 75 75 75 5% 5% 6% VAT declaration rate (%), August *** *** VAT declaration rate (%), August 3% *** 4% VAT declaration rate (%), August 70 70 70 ** ** 65 65 65 60 60 60 55 55 55 50 50 50 er + Simple reminder No reminder Simple reminder + Simple reminder No reminder Simple reminder + Simple reminder No rem social norm social norm c. Declaration rate, by treatment group 80 VAT declaration rate (%), August 60 40 20 0 02-Oct 04-Oct 06-Oct 08-Oct 10-Oct 12-Oct 14-Oct 16-Oct 18-Oct 20-Oct 16-Sep 18-Sep 20-Sep 22-Sep 24-Sep 26-Sep 28-Sep 30-Sep No reminder Simple reminder Simple reminder + social norm Source: World Bank calculations based on data from national authorities. Significance level: ** = 5 percent, *** = 1 percent. Bars with arrows represent confidence intervals. 31 Kosovo: Promoting Tax Compliance with Behavioral Insights declaration, or it may simply reflect the fact that the messages served more as nudges to do some thing sooner, but not necessarily better. A simple follow-up with the SMS message recip- ients would reveal if there is an effect on pay- ments or a longer-term effect on declaration rates overall and on on-time payments. Heterogeneity in the impacts of the SMS re- minders was found among select subgroups of taxpayers, though more analysis is needed to understand this heterogeneity. The effects of the SMS reminders on the declaration rate were particularly marked among firms in the pri- mary and secondary sectors. Thus, there was a 5 percentage point rise in the declaration rate associated with the simple SMS message. More analysis of the tax registry data is needed to un- derstand these differential impacts. 32 Kosovo Promoting Tax Compliance with Behavioral Insights 5. Process Makes Perfect: Lessons Learned from Implementation A crucial component of effective experimental more frequency than larger, more well estab- trials to improve tax compliance is an under- lished firms, and this was related to the belief standing of the implementation and capaci- that many firms underreport or altogether evade ty-building process. The effectiveness of an in- the taxes that they owe. This is also highlighted tervention is dependent, after all, on successful in the taxpayer survey conducted in Kosovo in implementation. An analysis of this process is 2017 (UBO Consulting 2018). perhaps even more important than the evalu- ation of impacts because it uncovers the risks Brainstorming with TAK helped support this and challenges faced in tax compliance in gen- narrative and direct the message content to- eral, which are undoubtedly associated with ward positively framed statements. Insights behavioral factors that may not be revealed and communicated by the TAK teams supported addressed through the exploration of impacts the importance of positively framed messages, alone. The approach to designing the behav- specifically, messages that placed less empha- ioral trials followed the process illustrated in sis on enforcement and more on positive rein- figure 10. forcement by highlighting the benefits that tax- es provide and targeting the moral inclinations The pretrial diagnostic revealed that a taxpayer of individuals. The capabilities and efforts of as client model is an effective way to improve TAK greatly facilitated the trial process, espe- perceptions of the tax administration. Semi- cially with regard to existing data systems and structured interviews with taxpayers and in- processes that allowed for a baseline to be es- formal dialogue with key stakeholders revealed tablished and the process of treatment assign- that attitudes toward TAK are generally positive, ment in the PIT and VAT trials to be simplified. highlighting the voluntary nature of tax collec- tion that the institution values.25 Some firm owners and accountants indicated that small- er, less accomplished firms are pursued with 33 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure 10. APPROACH TO A BEHAVIORALLY INFORMED TRIAL 01. 03. Definition Implementation & Diagnosis & Evaluation 05. Re-define & Re-diagnose 02. 04. Design Adapt 5.1 demonstrate the benefit of using behavioral science approaches to public policy. The World Bank was able to provide expertise in solutions Implementation Successes to implementation and evaluation problems based on evidence associated with behavioral insights, while advancing its current engage- The partnership with the GIZ was motivat- ment with the Ministry of Finance to improve ed by the recognition that this type of pol- tax collection in general. The GIZ was strate- icy challenge could most effectively be ad- gically positioned to buttress implementation dressed through a collaboration embracing and effectively manage the coordination ef- the strengths of each institution in the field. fort with the tax administration, given the Many implementation successes were achieved GIZ-supported Public Finance Project team’s through the three trials, both in the trials them- long-standing relationship and continued en- selves and in the overall coordination among the gagement in capacity building with the tax World Bank, the GIZ, and jointly with TAK. Con- administration. This was the first partnership sidering the positive impacts, these successes between the World Bank team and the GIZ in 34 Kosovo: Promoting Tax Compliance with Behavioral Insights Kosovo, and the initiative provided important nalities by nudging the policy makers. The lessons for future engagement between the team was able to coordinate the content of two institutions in Kosovo and beyond. messages and communication channels with the TAK director general and his team, which Coordination between the GIZ and the World (1) guaranteed that the trials were informed by Bank proved that a partnership embracing the the knowledge and preferences of TAK and (2) strengths of each team can capitalize on these allowed TAK to witness the design of a behav- strengths and exploit synergies to achieve iorally informed intervention from the ground successful results. The GIZ acted as a liaison up and how adjustments could be made to re- with TAK on day-to-day aspects of the trials, flect the different realities of taxpayers. By par- maintaining an open line of communication ticipating in the design and implementation of with leadership and technical teams and tak- the trials through brainstorming sessions with ing advantage of an existing dialogue with the the TAK Director General’s Office and the TAK administration. Moreover, the GIZ played a piv- Information Technology Department, TAK staff otal role in supporting a behavioral diagnostic were able to come to understand how behav- to inform the design of the trials, including car- ioral science tools can change the mindsets of rying out semistructured interviews alongside taxpayers and tax administrators by highlight- members of the extended World Bank team and ing nuances in the communication with taxpay- following up with multiple stakeholders. While ers. The coordination with TAK also underlined initial differences in the understanding of the how TAK data systems could be pushed to the division of labor among the two teams needed limit to identify new ways of doing business to be addressed, including the communication and interacting with taxpayers. of objectives and the ways to reach them, the insistence by each team on investing in the bi- Considering the unique implementation chal- lateral dialogue proved invaluable in the suc- lenges that arose, the engagement was a cess of the project. The availability of a team remarkable success in building capacity, improv- on the ground that has a close relationship with ing future policy, and establishing important the client, while incorporating frequent feed- collaborations. The GIZ was able to coordinate back from multiple stakeholders, is key to successfully with two firms to deliver letters reaching intended outcomes. across the country and to send SMS messages even when information was either erroneous or Close coordination with TAK allowed for incomplete. The GIZ’s frequent communication quick action and feedback and demonstrat- with both firms on the delivery of messages ed that a behaviorally informed nudge might and recording observations during the process encourage a population of interest toward a made the team’s evaluation of information on common goal, but also create positive exter- the success of the trials substantially easier. 35 Kosovo: Promoting Tax Compliance with Behavioral Insights 5.2 ceived a letter actually signed for it, suggesting that the intended recipient did not necessarily Implementation receive or even read the letter (figure 11). Challenges Moreover, because of operational difficulties, the timing of letter delivery was different for Despite the successes of the three trials, im- each of the treatment arms, which is not ideal. plementation challenges created bottlenecks The delivery of the standard letter started one in the design and, ultimately, the evaluation full day before the delivery of the deliberate of the trials; these challenges should be ad- choice letter. The deliberate choice letter arrived dressed in future trials in Kosovo and else- an average of more than a half day later and clos- where. These challenges were observed in er to the due date. Nearly half the standard let- each of the trials rather than in the coordina- ters were delivered by the fifth day, whereas only tion across the World Bank, the GIZ, and the a third of the deliberate choice letters were deliv- TAK teams. The challenges faced during each ered by the same date. This is relevant because trial are briefly described below. the letter urged action by a specific date, and the later the letter was received, the more likely a tax- The implementation challenge of the PIT letter payer would presumably give up on the process, trial centered on the lack of reliable contact believing that there was no time to act. information for the delivery of messages to taxpayers. Among the population of interest, The implementation of the VAT e-mail trial also that is, taxpayers who, by the due date, had faced a number of technical challenges. While a not submitted tax declarations, the postal ad- larger share of VAT taxpayers had an e-mail ad- dresses of 27 percent were not on file or were dress on file relative to the postal addresses in the outdated or erroneous and could not be used PIT trial, around 20 percent of the e-mails origi- in the trial. Among the remaining 11,600 tax- nally sent out by the TAK Information Technolo- payers with an address on file, only 48 percent gy Department bounced back because the e-mail of the 7,747 taxpayers randomly selected to address was not valid. This may have occurred receive a letter could be found despite multi- because the e-mail addresses were no longer ac- ple attempts by the courier service, including tive or were incorrectly entered into the registry phone calls to the taxpayers to validate contact by the taxpayers. This problem of treatment non- information. This revealed an unfortunate issue compliance was exacerbated by the fact that few- with the tax registry: the quality of informa- er than 25 percent of the e-mails that arrived at tion on addresses is poor, including addresses valid e-mail addresses were opened, despite two that were ultimately located by the courier. In attempts at different dates, that is, at the begin- addition, less than two-thirds of those who re- ning of the month and around the due date, 15–20 36 Kosovo: Promoting Tax Compliance with Behavioral Insights days later. The rate at which e-mails were opened The implementation of the VAT SMS trial, was substantially higher in the case of the stan- while relatively more successful in terms of dard e-mail (more than 50 percent) relative to the enhancing compliance than the letters or the positively framed e-mail (only 7 percent). Many of e-mail reminders, was nonetheless hindered the e-mails may not have been opened because by issues related to data quality. The SMS trial of issues in the delivery. The first e-mail remind- displayed the highest exposure of all three trials: er included inconsistent subject lines across the 80 percent of all SMS messages were success- two variants, as well as unbalanced timing: some fully delivered to the phone numbers on file, e-mails were sent early in the morning; others though a small share of the phone numbers were late in the evening or overnight. That they were invalid. While the telecommunication company opened with much less frequency is not surpris- sending the SMS messages had no way of know- ing. The positively framed e-mail was also sent an ing if the messages were read, it is fair to assume average of more than a half day later. The second that the reminder was available on the phones e-mail reminder was also poorly timed. Moreover, because most phones allow the first portion of a a glitch in the batch e-mail delivery system re- message to be displayed even if the full message sulted in a delay, causing half the e-mails to be is not opened. (The full message was limited to sent six days after the due date. The batch e-mail 160 characters.) Timing was largely fixed across procedure required a member of the information all treatment arms. All SMS messages were sent technology team to send each batch individually, between 10:00 a.m. and 11:00 a.m. on the same which meant that some language and treatment day. This was not the case with e-mails; so, all group combinations were sent at more opportune taxpayers who received an SMS message did so times than others (figure 12). Independent of im- at a moment when they might be more likely to plementation issues, factors that drive an individ- read it, that is, earlier during the workday. De- ual’s decision to open messages were identified spite these positive highlights, multiple phone in both rounds of e-mail delivery. The most im- numbers exist for each taxpayer, and simple portant determinant in whether or not an e-mail rules had to be established to identify a pri- was opened was the treatment assignment, which mary cell phone number, which did not always was highly correlated with the time the e-mail result in a correct phone number and possibly was sent. Those receiving the extended behav- meant these messages were sent to the wrong ioral e-mail were around a third as likely to open person. Moreover, the quality of the numbers the message as those receiving the standard be- meant that some could not be used at all in the havioral message, after one controls for the hour trial (some had the incorrect number of digits, during which the e-mail was sent. Another import- for example); for others, assumptions had to be ant factor was the number of years a firm has been made on leading 0s, country codes, and so on to registered. Accountants representing longer-ten- identify a valid phone number. ured firms were less likely to open the e-mail. 37 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure 11. DELIVERY OF LETTERS 5.3 Opportunities Ahead a. By district 70 4000 60 Determining the most effective communication Number of PIT letters prepared 3200 channel is context specific and may not align 50 % of addresses found with the priorities of a tax administration. For 2400 40 example, e-mail was identified as the most effec- tive method of communication with taxpayers, 30 1600 but the majority of the e-mails were not read by 20 the recipients. SMS messages were more suc- 800 cessful in delivery, opening, and reading rates. 10 Given cultural considerations, SMS messages 0 0 Prizren Mitrovica Ferizaj Pristina Gjakova Gjilan Peć might be the most effective in future trials simply because of the reliability of the contact informa- tion and despite the low impact of the messages Address found Number of letters on declaration behavior. The prominence of tax b. By delivery date accountants in helping individual taxpayers and firms submit their declarations is also an import- ant factor to consider because the intended audi- 100 ence might not be receiving the message, which was likely the case in many e-mail and some SMS % of all PIT letters delivered 75 communications. Existing communication strategies and the 50 method of communication employed might not be optimal. Consider, for example, the 25 use of e-mails to increase VAT tax compliance. Only one e-mail in five was actually opened. In the future, test e-mails might be sent to help 0 06-Apr 07-Apr 08-Apr 09-Apr 10-Apr 11-Apr 12-Apr 13-Apr assess whether the e-mail addresses are val- id and whether the e-mails are opened and to Simple reminder Simple reminder + deliberate choice ask taxpayers to complete a short question- naire on their favored method of communica- Source: World Bank calculations based on data from national tion. Enhancing the quality of the data on postal authorities. 38 Kosovo: Promoting Tax Compliance with Behavioral Insights addresses, e-mail addresses, and phone num- to elicit an undesirable response from taxpayers bers is fundamental to improving communica- in Kosovo, evidenced by complaints communi- tion with taxpayers. cated to call centers, even though the intention was to keep messages more positively framed, Continuing the dialogue in Kosovo has potential as was the case with the deliberate choice lan- in establishing a testing ground for the realization guage. Moreover, the behaviorally augmented of advances in the development and implementa- messages (beyond using simplified language and tion of communication strategies. Two methods a clear call to action) did not show substantial- of communication that were not tested could be ly greater impact than simple reminders, sug- considered in future trials. Throughout the project, gesting that messages need to be tested before it became clear that the more personal or individu- implementation at scale. Future interventions alized the communication, the more likely a nudge might involve focus groups to test various mes- would work. Individualized phone calls from the sages and inform message content. TAK call center might provide a useful, albeit re- source-dependent means to test behavioral mes- The support of local partners is indispensable sages among taxpayers. TAK already reaches out for successful implementation and long-term to select taxpayers by phone to inquire about late capacity building. Supporting the implementa- declarations and late payments; so, a trial could tion of tax trials requires a hands-on approach take advantage of the existing infrastructure and by the project team, which is feasible only with test different scripts. Website-based messaging local support. Because decisions are generally through pop-ups or the use of CAPTCHAs, for ex- made in a tight time frame and data are shared ample, could also be effective. This would entail on the fly, daily communication with the tax au- building a communication infrastructure within thority during implementation is crucial for suc- the declarations platform and testing how vari- cess in these types of interventions. By closely ous message formulations change declaration and collaborating during the trials, the World Bank payment behavior among taxpayers who log in to and the GIZ were able to foresee the challeng- the website (see Kettle et al. 2017). Such a com- es posed by the project and make quick adjust- munication strategy would be low cost; most of ments if needed. Ongoing cooperation between the cost would be incurred in establishing, rather these two institutions is essential. Partners than maintaining the infrastructure, which can be can learn from this implementation success, tweaked at low marginal cost. especially if the collaboration with a govern- ment counterpart is communication intensive. Close collaboration between the teams creates Message content and wording are also con- mutual learning and effective adaptation, in- text specific. Negatively toned messages, which stilling a behavioral mindset among the GIZ and were successful in a similar trial in Poland, seem 39 40 Share of e−mails delivered 0.0 0.2 0.4 0.6 0.8 1.0 04-Jul 12:00 04-Jul 13:00 04-Jul 14:00 04-Jul 15:00 04-Jul 16:00 04-Jul 17:00 04-Jul 18:00 04-Jul 19:00 04-Jul 20:00 Simple reminder 04-Jul 21:00 04-Jul 22:00 04-Jul 23:00 05-Jul 00:00 # Emails 05-Jul 01:00 05-Jul 02:00 05-Jul 03:00 05-Jul 04:00 0 100 200 300 400 500 600 700 800 900 05-Jul 05:00 0:00 05-Jul 06:00 05-Jul 07:00 1:00 05-Jul 08:00 2:00 05-Jul 09:00 Sent a. First round of e-mails (early July) 05-Jul 10:00 3:00 05-Jul 11:00 4:00 05-Jul 12:00 05-Jul 13:00 5:00 05-Jul 14:00 Kosovo: Promoting Tax Compliance with Behavioral Insights 6:00 05-Jul 15:00 05-Jul 16:00 Simple reminder + positive framing 7:00 05-Jul 17:00 8:00 05-Jul 18:00 Opened Figure 12. TIMING OF E-MAIL DELIVERY 05-Jul 19:00 9:00 10:00 11:00 12:00 13:00 Share of e−mails delivered 14:00 Source: World Bank calculations based on data from national authorities. 0.0 0.2 0.4 0.6 0.8 1.0 15:00 16:00 20 Jul 08:00 17:00 20 Jul 18:00 18:00 21 Jul 04:00 19:00 20:00 21 Jul 14:00 21:00 22 Jul 00:00 Simple reminder 22:00 c. Sent vs. opened e-mails, first round of e-mails 22 Jul 10:00 23:00 22 Jul 20:00 23 Jul 06:00 23 Jul 16:00 24 Jul 02:00 24 Jul 12:00 24 Jul 22:00 25 Jul 08:00 b. Second round of e-mails (late July) 25 Jul 18:00 26 Jul 04:00 Simple reminder + positive framing 26 Jul 14:00 27 Jul 00:00 Kosovo: Promoting Tax Compliance with Behavioral Insights TAK partners and ensuring that future policies and programs incorporate behavioral design at the outset. The bigger challenge—the elephant in the room—in tax administration is the need to address the informal economy. Using the evi- dence-based insights outlined in this brief, fu- ture efforts should also be directed at expand- ing the size of the tax registry by integrating informal income earners and businesses in the system. Simple messages can be effective at en- couraging individuals and firms to do their part by registering with the tax and business regis- tries. By using the existing data infrastructure through the census or public service providers, similar strategies could expand the size of the tax base and address the more serious fiscal challenges in Kosovo. 41 Kosovo: Promoting Tax Compliance with Behavioral Insights 6. Conclusion Institutionalizing a behavioral approach to tax Technological advances are reducing the already collection could enable Kosovo to develop in- low cost and limited administrative demands of terventions that are effective locally at low cost the application of behavioral insights to tax col- and with only modest demands on administra- lection and public policy more generally. Tax au- tive and human resources. The potential positive thorities could leverage existing technologies impact of behavioral interventions more than jus- such as websites or smartphones to promote tifies the cost of experimentation. Establishing a compliance with modest budgetary resources. team that conducts rigorous evaluations would These types of experiments also produce gran- not only help identify what works, but also enable ular data on taxpayers, which can inform pol- policy makers to understand taxpayer behavior. icies. Thus, building institutional capacity for Determining whether interventions create per- experimentation has the potential to generate sistent and lasting effects among taxpayers is large returns at comparatively low cost, espe- important. The effectiveness of behavioral inter- cially in lower-middle-income countries such as ventions may diminish over time; thus, remind- Kosovo. er messages may become routine, and the per- ceived threat of enforcement may fade. Adapting communication strategies by changing the con- tent of messages and contacting taxpayers in dif- ferent ways may signal the continued attention of the authorities, and fostering this perception may promote taxpayer compliance independent- ly of the messages. 42 Kosovo: Promoting Tax Compliance with Behavioral Insights References A D I Arcos Holzinger, Lilia, and Nicholas Biddle. 2016. “Behavioural Insights of Tax Compli- Del Carpio, Lucia. 2013. “Are the Neighbors IMF (International Monetary Fund). 2013. ance: An Overview of Recent Conceptual and Cheating? Evidence from a Social Norm Ex- Republic of Kosovo: 2013 Article IV Consulta- Empirical Approaches.” TTPI Working Paper periment on Property Taxes in Peru.” Work- tion. IMF Country Report 223. Washington, 8/2016 (October), Tax and Transfer Policy ing paper, Princeton University, Princeton, DC: IMF. Institute, Crawford School of Public Policy, NJ. College of Asia and the Pacific, Australian K National University, Canberra. Dwenger, Nadja, Henrik Kleven, Imran Ra- sul, and Johannes Rincke. 2014. “Extrinsic vs Kettle, Stewart, Marco Hernandez, Simon Ariel, Barak. 2012. “Deterrence and Moral Intrinsic Motivations for Tax Compliance: Ev- Ruda, and Michael Sanders. 2016. “Behavior- Persuasion Effects on Corporate Tax Com- idence from a Randomized Field Experiment al Interventions in Tax Compliance: Evidence pliance: Findings From A Randomized Con- in Germany.” Paper presented at the German from Guatemala.” Policy Research Working trolled Trial.” Criminology 50 (1): 27–69. Economic Association’s Annual Conference Paper 7690, World Bank, Washington, DC. 2014, “Evidence-Based Economic Policy,” Hamburg, September 7–10. Kettle, Stewart, Marco Hernandez, Michael B Sanders, Oliver Hauser, and Simon Ruda. E 2017. “Failure to CAPTCHA Attention: Null Brockmeyer, Anne, Marco Hernandez, Stewart Kettle, and Spencer Smith. Forth- Results from an Honesty Priming Experi- coming. “Casting a Wider Tax Net: Experi- EBRD (European Bank for Reconstruction ment in Guatemala.” Behavioral Sciences 7 mental Evidence from Costa Rica.” American and Development). 2016 “Life in Transition: (2): 28.Kleven, H Henrik Jacobsen, Martin Economic Journal: Economic Policy. https:// A Decade of Measuring Transition.” EBRD, B. Knudsen, Claus Thustrup Kreiner, Søren www.aeaweb.org/articles?id=10.1257/ London. https:/ /www.ebrd.com/publications/ Pedersen, and Emmanuel Saez. 2011. “Un- pol.20160589&&from=f. life-in-transition-iii. willing or Unable to Cheat? Evidence from a Tax Audit Experiment in Denmark.” Econo- H metrica 79 (3): 651–92. C Carrillo, Paul, Dina Pomeranz, and Monica Hallsworth, Michael, John A. List, Robert D. Singhal. 2017. “Dodging the Taxman: Firm Metcalfe, and Ivo Vlaev. 2017. “The Behav- Misreporting and Limits to Tax Enforcement.” ioralist as Tax Collector: Using Natural Field American Economic Journal: Applied Econom- Experiments to Enhance Tax Compliance.” ics 9 (2): 144–64.Castro, Lucio, and Carlos Journal of Public Economics 148 (April): 14– Scartascini. 2013. “Tax Compliance and En- 31. forcement in the Pampas: Evidence from a Field Experiment.” IDB Working Paper IDB- Hernandez, Marco, Julian Jamison, Ewa WP-472 (December), Inter-American Devel- Korczyc, Nina Mazar, and Roberto Sormani. opment Bank, Washington, DC. 2017. Applying Behavioral Insights to Improve Tax Collection: Experimental Evidence from Po- Cojocaru, Alexandru. 2017. “Kosovo Jobs land. Washington, DC: World Bank. Diagnostic.” Job Series 5, World Bank, Wash- ington, DC. 43 Kosovo: Promoting Tax Compliance with Behavioral Insights M P W Mascagni, Giulia. 2015. “Tax Experiments Pomeranz, Dina. 2015. “No Taxation without Wenzel, Michael. 2005. “Misperceptions of in Developing Countries: A Critical Review Information: Deterrence and Self-Enforce- Social Norms about Tax Compliance: From and Reflections on Feasibility.” CDI Practice ment in the Value Added Tax.” American Eco- Theory to Intervention.” Journal of Econom- Paper 11 (March), Center for Development nomic Review 105 (8): 2539–69. ic Psychology 26 (6): 862–83. Impact, Institute of Development Studies, Brighton, U.K. Pomeranz, Dina, and José Vila-Belda. 2019. World Bank. 2014. “Republic of Kosovo, “Taking State-Capacity Research to the Public Sector Revenues: Tax Policies, Tax Mascagni, Giulia, Christopher Nell, and Nara Field: Insights from Collaborations with Tax Evasion, and Tax Gaps.” Report 89195–XK Monkam. 2017. “One Size Does Not Fit All: A Authorities.” Annual Review of Economics 11 (June), Poverty Reduction and Econom- Field Experiment on the Drivers of Tax Com- (August). ic Management Unit, Europe and Central pliance and Delivery Methods in Rwanda.” Asia Region, World Bank, Washington, DC. ICTD Working Paper 58 (January), Interna- R https://openknowledge.worldbank.org/ tional Center for Tax and Development, Insti- handle/10986/22269. tute of Development Studies, Brighton, U.K. Riinvest. 2013. To Pay or Not to Pay: A Busi- ness Perspective of Informality in Kosovo. Pris- — Ministry of Finance. 2018. “Annual Bulletin tina, Kosovo: Riinvest Institute. 2017: On Public Debt.” April, Debt Manage- World Bank. 2015. World Development ment Division, Ministry of Finance, Pristina, T Report 2015: Mind, Society, and Behavior. Kosovo. https://mf.rks-gov.net/desk/inc/me- Washington, DC: World Bank. dia/233CE729-B312-4F03-9081-E6193AE- Torgler, Benno. 2007. Tax Compliance and Tax 2A2E9.pdf. World Bank. 2017. Republic of Kosovo: Morale: A Theoretical and Empirical Analysis. Cheltenham, UK: Edward Elgar Publishing. Systematic Country Diagnostic. Report O 114618–XK (January). Washington, DC: U Europe and Central Asia Region, World Ortega, Daniel, and Pablo Sanguinetti. 2013. Bank. http://documents.worldbank.org/ “Deterrence and Reciprocity Effects on Tax curated/en/282091494340650708/pdf/ UBO Consulting. 2018. “Taxpayer Survey Compliance: Experimental Evidence from Kosovo-SCD-FINAL-May-5-C-05052017. 2017 Kosovo.” March, Gesellschaft für Inter- Venezuela.” CAF Working Paper 2013/08 pdf. nationale Zusammenarbeit, Pristina, Kosovo. (December), CAF Development Bank of Latin America, Caracas. 44 Kosovo: Promoting Tax Compliance with Behavioral Insights Notes 1 Changes in the level 4 A reduced VAT rate of 8 9 The exclusion of taxpayers 11 In the reminders, the and composition of taxes percent is applied to the supply from a given trial was generally translation of “honestly” was and government spending and import of basic foods and nonrandom and represented not deemed appropriate for the can affect macroeconomic utilities, among other items. a small share of the total, messages given the limitations variables, including economic typically revolving around those of the word. “Correctly” was 5 See Ministry of Finance activity, consumption, savings lacking contact information. used in its place. (2018); WDI (World and investment, and income Exclusion criteria was unique Development Indicators) 12 Letters were sent in distribution. On the importance to each trial, but, in all cases, (database), World Bank, English only. E-mails and SMS of tax collection in economic taxpayers without the relevant Washington, DC, http://data. messages were sent in either development, see “Helping contact information (address, worldbank.org/products/wdi. Albanian or Serbian, depending Governments and Stakeholders e-mail address, phone number) on the language preference Make the SDGs a Reality,” 6 The tax administration were excluded from the trials. of the firm recorded in the Sustainable Development Goals is considering obliging 10 Randomization among registry. If this was not defined, Knowledge Platform, United all taxpayers to submit the population of interest the messages were sent in Nations, New York, https:// declarations even if these was repeated multiple times Albanian. sustainabledevelopment.un.org. exemptions are available (rerandomization) to ensure a because exemption status 13 The small sample size 2 For a review of the balance across treatment arms typically varies from one year meant, in most cases, that theoretical and empirical on observed characteristics to the next. the teams were limited in literature on tax compliance, during randomization. The their ability to test many see Arcos Holzinger and Biddle 7 Experimental trials refers level of randomization and different messages in each (2016); Pomeranz and Vila- to interventions in which use of stratification depended trial, including, for example, Belda (2019). a comparison in outcomes on the method and the a reminder using standard across two or more groups heterogeneous quality of 3 See Ariel (2012); language that had been applied in the context of a randomly communication: clustering Brockmeyer et al. in past communications. assigned condition (in this case, was necessary for e-mails (forthcoming); Castro and receiving a reminder) could and SMS messages because 14 When asked about the Scartascini (2013); Del Carpio guarantee that any differences contact information for each fairness of the tax system given (2013); Dwenger et al. (2014); in the outcome are caused by method was not unique to the level of public services, 60 Hallsworth et al. (2017); the assigned condition rather single taxpayers. The most percent believe they should Hernandez et al. (2017); Kettle than other factors. common explanation for this pay less in taxes; 35 percent et al. (2016); Kleven et al. was that an e-mail or phone believe they should pay the (2011); Ortega and Sanguinetti 8 The business-as- number presumably belonged same amount of taxes; and only (2013); Pomeranz (2013); usual scenario in practice to an accountant rather than 5 percent think they should pay Torgler (2007); Wenzel (2005). involves some nonrandom a firm owner or individual more taxes. communication with outlier taxpayer. Because the quality taxpayers (those with large of addresses varies across the debts, multiple missing country, stratification during declarations, and so on). the PIT trial was performed However, during the trials, TAK at the district level to assure suspended all communication within-district balance. The with taxpayers within the optimal number of treatment population of interest during arms was identified through the design and evaluation of power calculations using a the trials. combination of historical data and results from similar trials, though some qualitative data informed these decisions as well. 45 Kosovo: Promoting Tax Compliance with Behavioral Insights 15 The Life in Transition Survey 19 In Albanian, the simple 23 Because treatment highlights the low institutional reminder with the social norms noncompliance was one-sided trust and the lack of satisfaction message read as follows: (some taxpayers assigned to with social services (EBRD “Mesazh nga ATK: deklaroni, treatment did not receive the 2016). Notwithstanding this paguani TVSH Gusht 2018 me letter, but no taxpayers assigned mistrust, a large share of the korrektësi, me kohë! 7 nga 10 to the control group received population, over 80 percent, firma i dorëzojnë deklaratat e a letter), the local average are willing to give more money tyre me kohë. Mos prisni, jeni treatment effect is reduced to to improve social services and pjesë e SHUMICËS!” the average treatment on the help the needy, and attitudes treated. Thus, because many of 20 Statistically significant toward government spending those who should have received here and throughout the text priorities mostly revolve around a letter did not receive one, the indicates that the causality link problems in social services and total effect of the treatment is at least 95 percent likely to the assistance to the poor. is the effect on those who did be true. Statistically insignificant receive the letter. 16 The population of interest means that causality cannot also included taxpayers who be confirmed with this degree 24 VAT declarations for a given were still in the system despite of certainty, though it does month are filed the following their inactivity since 2010. not imply that the there is no month; so, the declaration for These individuals were identified causality. June was to be filed in July. through their unique fiscal 21 Robustness checks were 25 Moreover, the declaration numbers. carried out that included adding process (which has entirely 17 The random assignment various controls to the estimate migrated to e-filing) was to one of the three groups of treatment impacts and highlighted as being generally was carried out using the excluding certain groups from easy to manage for the average e-mail addresses rather than the regressions. taxpayer. Forms are simple and the taxpayer names because use relatively clear and concise 22 The impact of the letters each message that would make language. As a result, the on the declaration rate by April reference to all corresponding experiments did not consider 13th, the due date highlighted unique fiscal numbers would be form-simplification to reduce the in the letters, was 2 percentage sent to only one e-mail address. cognitive burden of submitting points for the simple + deliberate tax declarations. 18 The random assignment choice letter and 3 percentage to one of the three groups was points for the simple letter (both carried out at the level of cell statistically higher than the group phone number rather than the that received no reminder). taxpayer name because each SMS message could only be communicated to a cell phone. 46 Kosovo: Promoting Tax Compliance with Behavioral Insights Annex Figure A.1. MACROECONOMIC INDICATORS FOR KOSOVO, 2012–18 Income and economic growth 2012  2013  2014  2015  2016e 2017e 2018f GDP growth (annual %) 2.90 3.43 1.20 4.09 4.06 4.23 4.46 Composition  Consumption 6.78 3.84 5.45 2.70 4.70 1.49 1.18   Private consumption 6.41 3.30 5.86 3.76 5.67 1.57 0.85   Public consumption 0.37 0.54 −0.41 −1.06 −0.97 −0.08 0.33  Investments −4.37 −0.10 −1.37 2.90 2.07 1.52 3.52  Net exports 0.48 −0.32 −2.88 −1.51 −2.71 1.21 −1.85   Exports −3.92 −0.38 1.39 0.43 0.52 3.97 3.11   Imports 4.40 −6.52 −4.27 −1.94 −3.23 −2.76 −4.96 GDP per capita growth 2.03 2.36 1.33 5.25 3.23 3.40 3.63 (annual %) GDP per capita 3,600.87 3,877.96 4,054.64 3,745.75 3,698.47 3,896.96 4,046.76 (current US$ millions) Private consumption growth 7.75 3.95 6.98 4.34 6.63 1.83 2.90 (annual %) Gross investment 26.03 24.82 23.24 25.81 25.54 27.23 29.64 (% of GDP) Gross investment - public 10.88 9.93 7.39 6.95 7.31 7.41 9.13 (% of GDP) Gross investment - private 15.33 14.78 15.69 17.98 17.33 19.83 20.51 (% of GDP) Note: e = estimate. f= forecast. 1) Fiscal balance includes the regular fiscal deficit within the 2 percent fiscal rule (lower than 2 percent projections here) and the deficit due to investment clause Sources: Ministry of Finance, Central Bank of Republic of Kosovo, Agency of Statistics of Kosovo, World Bank calculations. 47 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure A.2. FISCAL INDICATORS FOR KOSOVO, 2012–18   2012 2013 2014 2015 2016e 2017e 2018f Fiscal indicators Revenue (% of GDP) 26.13 24.64 23.95 25.08 26.30 26.83 26.98 Expenditure (% of GDP) 28.49 27.94 26.60 27.08 27.75 28.08 30.16 Current (% of GDP) 17.37 17.75 18.95 19.85 20.13 20.34 20.91 Capital (% of GDP) 11.09 9.82 7.39 6.95 7.19 7.44 8.99 Overall fiscal balance −2.37 −3.30 −2.65 −1.99 −1.45 −1.24 −3.18 (% of GDP) /1 Primary fiscal balance −2.12 −3.04 −2.38 −1.71 −1.13 −0.95 −2.84 (% of GDP) /1 Overall fiscal balance per −2.37 −3.30 −2.65 −1.99 −1.45 −1.21 −1.76 regular budget (% of GDP) /1 Total public and publicly 8.10 9.01 10.56 12.83 14.41 16.51 17.66 guaranteed debt (% of GDP) Total public debt 8.10 9.01 10.39 12.66 14.08 15.81 16.91 (% of GDP) Domestic public debt 1.45 2.86 4.61 6.50 7.89 9.09 10.36 (% of GDP) External public debt 6.65 6.15 5.78 6.16 6.19 6.72 6.54 (% of GDP) Note: e = estimate. f= forecast. 1) Fiscal balance includes the regular fiscal deficit within the 2 percent fiscal rule (lower than 2 percent projections here) and the deficit due to investment clause. Sources: Ministry of Finance, Central Bank of Republic of Kosovo, Agency of Statistics of Kosovo, World Bank calculations. 48 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure A.3a. SIMPLE REMINDER LETTER (ALBANIAN) 49 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure A.3b. SIMPLE REMINDER LETTER (ENGLISH TRANSLATION) 50 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure A.4a. SIMPLE REMINDER WITH DELIBERATE CHOICE MESSAGE LETTER (ALBANIAN) 51 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure A.4b. SIMPLE REMINDER WITH DELIBERATE CHOICE MESSAGE LETTER (ENGLISH TRANSLATION) 52 Kosovo: Promoting Tax Compliance with Behavioral Insights FIGURE A.6A. SIMPLE REMINDER WITH POSITIVE FRAMING E-MAIL (ALBANIAN) 53 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure A.6b. SIMPLE REMINDER WITH POSITIVE FRAMING E-MAIL (ENGLISH TRANSLATION) 54 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure A.5a. SIMPLE REMINDER E-MAIL (ALBANIAN) 55 Kosovo: Promoting Tax Compliance with Behavioral Insights Figure A.5b. SIMPLE REMINDER E-MAIL (ENGLISH TRANSLATION) 56 World Bank Publications, The World Bank Group. P RO M OT I N G TAX CO MP LI ANCE IN KO SOVO W I T H BEHAVI O RAL I NSIGHTS Washington, DC. March 2019.