Policy Research Working Paper 10201 Impacts of Transparent Online Auctions on Public Land Lease Revenue Evidence from Legal and Administrative Changes in Ukraine Klaus Deininger Daniel Ayalew Ali Roman Neyter Development Economics Development Research Group October 2022 Policy Research Working Paper 10201 Abstract Although millions of hectares of public land are transferred lease revenue. Had all public land Ukraine transferred since to private parties each year, often with unsatisfactory results, 2015 auctioned using post-reform mechanisms, local gov- evidence on mechanisms to achieve better outcomes is ernments would have received incremental lease revenue scant. This paper analyzes the impact of a 2021 reform in of US$500 million per year. In countries with large public Ukraine that -after earlier digitization efforts did not pro- land endowments, legal and regulatory reform to ensure duce desired results- mandated use of transparent online rights to public land are allocated competitively and in a auctions by local government to transfer rights to public decentralized way could possibly improve social, economic, agricultural land. The shift to a collusion-proof electronic and environmental outcomes. auction system led to a near-instantaneous doubling of This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at kdeininger@worldbank.org. or dali1@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Impacts of Transparent Online Auctions on Public Land Lease Revenue: Evidence from Legal and Administrative Changes in Ukraine Klaus Deininger1 Daniel Ayalew Ali1 Roman Neyter 2 Keywords: Ukraine, auctions, public land, governance, decentralization JEL codes: Q10, O13, H56, R14 1 The World Bank, 1818 H Street NW, Washington, DC 2 Kyiv School of Economics, Kyiv, Ukraine The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Funding support from the European Union (ENI/2017/387-093 and ENI/2020/418-654) is gratefully acknowledged. We thank Nataliia Kussul and Andrii Shelestov for support with data processing, Darina Marchak, Oleksandr Lihotop and Oleksii Sobolev from the ProZorro Sales team for explaining the auction process, and Denis Bashlyk, Markiyan Dmytrasevych, Baher el Hifnawi, Anthony Gaeta, Ben Hell, Thea Hilhorst, Roman Hrab, Sergyi Kubakh, Andrii Martin, Tymofiy Mylovanov, Oleg Nivievskyi, Taras Vysotskyi and Sergyi Zorya for helpful discussions and insightful comments and Daria Manzhura as well as Tania Khorzovskaya for outstanding coordination and administrative support. Impacts of Transparent Online Auctions on Public Land Lease Revenue: Evidence from Legal and Administrative Changes in Ukraine 1. Introduction With more than 10 million ha, Ukraine’s endowment of public agricultural land exceeds the total land area used for agriculture in most European countries. The effectiveness with which this asset -that is likely to appreciate further with climate change and growth in global commodity demand- will affect not only global food supplies but also local communities’ income and ability to provide public services. Yet, like in many developing countries with large endowments of public land or forests, finding ways to ensure the benefits from such land accrue to locals rather than bureaucrats has remained a challenge. Various initiatives trying to do so allegedly failed to achieve their objectives due to lack of capacity, elite capture, and corruption. This paper evaluates the impact of legal and regulatory reform that mandated use of an electronic auction platform for any public land leases and transferred full ownership of such land to local governments rather than having it administered (and auctioned) on their behalf by a central institution. The instantaneous and complete shift from offline to online auctions on November 1, 2021, and the fact lease decisions anticipating legal changes were implausible before laws had been adopted by Parliament and unlikely thereafter during the cropping season allows us to use a regression discontinuity design. Although in place for less than four months, use of a transparent e-auction platform by local communities doubled lease prices, a result robust to changes in bandwidth and a piecewise regression with kernel weights. Placebo tests using passage of the law as a pseudo-reform date do not show any statistically significant result throughout. While their joint adoption makes it difficult to disentangle ‘competition effects’ due to e-auction adoption from ‘decentralization effects’ resulting from elimination of an intermediary central government agency, descriptive analysis of earlier reform attempts provides clues as to the possibly underlying mechanisms. Adding data on (offline and online) auctions conducted in the 2015 to 2020 period serves not only as a robustness check but also allows us to assess the impact of a 2018/19 reform effort. This attempt, based on 2 a pilot resolution, temporarily allowed auctions to be conducted either in person or electronically on SETAM -a platform that continues to be used to liquidate assets seized during foreclosure proceedings- by State Service for Geodesy, Cartography & Cadaster (SGC) as exclusive administrator of electronic auctions. We find that use of SETAM is associated with lower lease prices by as much as 21% than offline auction, compounding the negative effect of having SGC as the auction organizer.1 In other words, a local government with the opportunity to do so would have obtained higher lease revenue by organizing an offline in person auction directly rather than via an online auction using SETAM organized by SGC. Regression results for the universe of auctions completed since 2015 provide a basis for a within-sample prediction to simulate, for each parcel that was transferred pre-reform, the predicted post-reform lease price had the auction been conducted on the Prozorro platform and organized by local government rather than SGC. Doing so suggests that per-hectare lease revenue would have tripled from US$98 to US$301. Auctioning all the 19,267 parcels comprising 312,813 ha transferred under the pre-reform regime using the modified arrangements would have implied an additional US$65.4 million in annual lease revenue for local governments. Our paper contributes to three strands of literature. First are the studies on the challenges of transferring use rights to public land in a transparent fashion. Empirical studies support the notion that, unless associated incentives and possible loopholes are carefully considered from the beginning, reforms aiming to stop corruption by making auctions mandatory may be ineffective (Cai et al. 2013). Given the size of the resources and networks at stake, failure to put in place auction mechanisms that are collusion-proof (Che & Kim 2009; Che et al. 2018) can trigger vast loss of resources and may corrode governance more broadly in ways that are difficult to rectify (Chen & Kung 2019). We build on these insights and show in addition that (i) it is possible to change auction formats rather quickly via a combination of legislative and regulatory action; (ii) such changes can have immediate, significant, and economically meaningful impact on auction revenue; (iii) use of an online auction format makes it easier to achieve these goals but is not a sufficient condition for greater transparency and higher auction revenue on its own. Second, the fact that far-reaching improvements in public land auction’s competitiveness and transparency were possible in a country that has been ranking in the bottom half of global rule of law and corruption rankings during the 2015-20 period,2 illustrates the potential of digitization to improve market functioning by (i) reducing barriers to entry, thus enhancing competition and transparency; (ii) eliminating the need for 1 While legal opinions on whether or not auctions on the SETAM platform could be organized by local governments, data reported in table 2 below show that 99% of auctions on SETAM were organized by SGC. While voluntary use of electronic auctions may raise selection issues, the coefficient on SETAM in a lease price regression with all auctions will be a lower bound estimate of this platform’s effect on realized l ease revenue if, as would be expected from rational agents, more valuable parcels were more likely to be auctioned electronically. 2 Information as per the 2021 version of the worldwide governance indicators (Kaufmann et al. 2011) https://info.worldbank.org/governance/wgi/ 3 intermediaries whose incentives are not necessarily aligned with those of land owners and who in addition may suffer from conflict of interest; and (iii) providing all interested parties with access to the same set of information and allowing creation of an audit trail to unambiguously determine if any procedural rules were violated, thereby creating a level playing field and reducing opportunities for corruption. The limited impact of earlier reform efforts suggests digitization and use of sophisticated technology such as blockchain cannot substitute for laws and regulations to structure incentives and provide enforcement structures. Third, there is a general recognition that delaying macro-economic reforms often comes at a high price. For Ukraine, Havrylyshyn (2017) argues a 3-year delay of reform in 1991 imposed immense economic cost.3 We complement such qualitative statements that lack a clear counterfactual by providing a micro-economic argument and econometric evidence to show that (i) legal reform will have the desired effect only if relevant provisions are fully implemented, i.e., regulations to govern implementation are in place; (ii) delays in enacting reform laws or regulations can reduce the scope for decentralization and local democracy by squandering assets that should have provided a basis for local governments’ provision of public goods and services; (iii) reform delays are costly: had the 2021 reforms analyzed here been put in place in 2017 instead of the e-auction ‘pilot’ and applied to land transferred via free privatization, the incremental annual revenue of local councils would be about US$500 million or more than US$1 million per rayon. The rest of the paper is organized as follows: Section 2 provides context and motivation by describing the institutional setting for public land management and transfer in Ukraine and describing the legal reforms undertaken in 2021. Section 3 discusses the data, descriptive evidence, and analytical approach. Section 4 presents results regarding impact of legal and regulatory reform and uses these to explore the cost of delayed reform and discusses policy implications. Section 5 concludes by identifying implications for policy as well as areas for future research. 2. Motivation and approach In many developing countries, especially those with a socialist heritage or an open frontier, large areas of land are under public ownership. Non-transparent and centralized mechanisms often result in such land being transferred to private parties below market value and in ways that impede local development. Transparent auctions can prevent this although auctioneer corruption or bidder collusion pose challenges. We use this to discuss the challenges faced by auctions of public agricultural land in Ukraine and the legal and regulatory changes introduced to confront these. 3 Havrylyshyn (2017) argues that these costs accrued in three areas, namely (i) macro-economic deterioration and a decline in output (of 59%) without parallel in neighboring countries which the country recovered from only in the early 2000s; (ii) creation of an ‘oligarchate’ dependent on Russian hydrocarbons that would lose from and thus oppose any serious improvements in governance and rule of law; and (iii) inability to expeditiously embark on a path towards EU accession in concert with other accession countries. 4 2.1 Public land management and transfers in developing countries Increasing demand for land due to population growth, urbanization, and rising food and commodity prices increase demand on reserves of public or communal land and often implies that large tracts of such land are transferred to private parties. The modalities of such transfers will affect social and economic opportunities at local level and may generate external environmental effects. Auctions have been widely used successfully to transfer rights to valuable public assets including radio spectrum (Weber 1997), mineral exploration (Compiani et al. 2020), timberland (Athey et al. 2011), or debt obligations (Hortaçsu et al. 2018). Theory suggests that auctions will result in higher revenue than negotiated sales for sellers (Bulow & Klemperer 1996) in a wide range of situations (Bulow & Klemperer 2009). Auctions of public land in urban settings indeed resulted in prices above those of available alternatives in urban settings such as Beijing (Weidong & Xiaolong 2012), Singapore (Chow et al. 2015; Agarwal et al. 2018) or Hong Kong (Shen et al. 2019). They were also used to privatize or liquidate agricultural land (Hüttel et al. 2014; Hüttel et al. 2016; Croonenbroeck et al. 2020) in former East Germany. However, vast amounts of rural land have been transferred to outside investors in Africa (Anseeuw et al. 2012), often in centralized and non-transparent ways (Arezki et al. 2015) in ways that fail to increases local food security (Müller et al. 2021), deplete local resources (Rulli et al. 2013), and may cause environmental externalities. Potentially negative social effects of non-competitive centralized transfers of public land in rural areas to private individuals have been well-documented in Africa and Latin America. In Africa, such transfers were often motivated by political rather than economic objectives (Widengard 2019; Bélair 2021; Dieterle 2021), creating a danger of neglecting local rights (Maganga et al. 2016; Engström & Hajdu 2018) and safeguards (Cotula 2014; Nolte & Voget-Kleschin 2014), triggering conflict with existing land users (Lay et al. 2020; Sulle 2020) and possibly environmental damage (Shete et al. 2016). From an economic perspective, instead of activating markets to improve land use, negotiated transfers of public land may foster speculative land acquisition (Ali & Deininger 2022), fail to act as a catalyst for local development via access to technology (Ali et al. 2019), markets (Deininger & Xia 2016), or skill-intensive employment (Glover & Jones 2019; Nanhthavong et al. 2020; Anti 2021) as was the case elsewhere (Kraus et al. 2021). At the forest frontier, public forest may be cut for short-term political gain (Pailler 2018; Braganca & Dahis 2021; Cisneros et al. 2021) with little in ways that increase emissions, reduce revenue (Brito et al. 2019), clash with local rights or benefit elites (Faguet et al. 2020). Reliance on non-competitive centralized processes for transferring land is often justified with reference to elite capture or a lack of capacity to implement transparent auctions at local level. This is a valid concern; putting in place collusion-proof auction mechanisms (Che & Kim 2009; Che et al. 2018) to prevent 5 manipulation has been a challenge in many settings. For example, in urban China where income from public land leases accounts for large shares of local revenue, conduct of auctions for all leases was mandated to halt corruption in 2010. Yet, auction formats adopted were easy to manipulate (Cai et al. 2013) and collusive practices continued (Cai et al. 2017). In fact, non-transparent practices benefiting politically connected indivduals in return for promotion of local cadres led to vast revenue loss (Chen & Kung 2019). Use of electronic online auctions could, however, possibly alleviate such constraints and we will draw on the example of Ukraine to assess the extent to which this is possible in practice. 2.2 Challenges of public land management in Ukraine With about 10.4 million ha in 2017, Ukraine’s endowment of public agricultural land exceeds the total land area used for agriculture in most European countries.4 Public land cannot be sold but leased and the country’s decentralization policy views revenues from leasing public land as a key source to fund local service delivery. All public land, except what was located within settlement boundaries, was traditionally administered by a central agency, the SGC. Legislation passed in 2013 had required leases to public land to be auctioned to the highest bidder. The effectiveness of such auctions was, however, reduced by the fact that they had to be held in person and to be organized by SGC, an organization whose incentives were not necessarily aligned with those of local communities as ultimate beneficiaries. In fact, the multiplicity of roles played by SGC created potential conflicts of interest and opportunities for manipulation in three respects. First, as information on what land is surveyed and its ownership status is not public and controlled by SGC, it was difficult for local governments to identify public land that could potentially be auctioned or to assemble attractive lots that would fetch high prices in an auction. Second, most public land to be auctioned had still to be surveyed and registered. SGC could effectively regulate the supply of land via its de facto monopoly on land surveying and its ability to initiate procedures of free land privatization as an alternative to land auctions.5 Finally, as the auction organizer, SGC was able to influence the level of competition at any given auction, e.g., by disqualifying certain bidders or changing auction time or location at short notice. The challenges of implementing transparent auctions in this context are illustrated by the fact that the area of public land transferred via non-competitive means in 2017-20 exceeded that of land auctioned by at least 4 At the beginning of 2017, the total area of agricultural state land was about 10.4 million hectares, 3.2 of which was used by state-owned enterprises (including 0.47 by the National Academy of Sciences), 2.5 in reserve, and 4.7 under lease. This is larger than total agricultural land in smaller European countries (e.g., the Baltics, the Czech Republic, the Slovak Republic, and Hungary) and close to all agricultural land in larger ones (e.g., Romania, Poland, France, and Germany). 5 For land that had yet to be surveyed, an activity on which SGC had a de facto monopoly, interested individuals had to pay SGC to identify, survey and auction the land. 6 a factor of five (Nivievskyi 2020).6 Even results of completed auctions fell short of expectations and were marred by allegations of auctioneer corruption and bidder collusion not unlike those reported in Russia (Marshall & Marx 2009). Unclear procedures also affected ‘regular’ land markets where exercise of market power (Graubner et al. 2021) and manipulation by political elites (Zadorozhna 2020) were widely reported. A first reform attempt in 2018 involved providing an option to conduct auctions online on an electronic auction platform, SETAM, already used to liquidate bankrupt firms’ assets. However, although SETAM was linked to a private blockchain to bolster its credentials in terms of transparency, SGC’s monopoly on auction organization remained in place and concerns about potential for manipulation reduced trust.7 The SETAM for land auctions was thus discontinued at the end of 2019. Analysis of pre-2021 agricultural land auctions, SGC involvement is associated with 20-30% lower returns (Kvartiuk et al. 2022) and an 11 percentage point lower likelihood of successful completion (Neyter & Nivievskyi 2022) although a failure to control for locational attributes of land plots could affect such cross-sectional estimates. 2.3 Ukraine’s transition to mandatory transparent e-auctions To remedy the shortcomings described above, draft laws to mandate transparent electronic auctions and transfer all public land to local communities were prepared and cleared for voting by the Parliament’s agrarian committee in mid-2019 and eventually passed in 2021. Law 1423-IX (adopted on April 28, 2021) transferred ownership of all state land (except land owned by state enterprises) to local communities, allowing them to organize auctions directly without SGC involvement. Law 1444-IX (adopted on May 18, 2021) mandated exclusive use of an electronic trading system and transparent online auctions to lease out any public land. Building on this law, a cabinet resolution appointing the state enterprise Prozorro Sale under the Ministry of Economy as the sole electronic trading platform for land auctions and establishing detailed procedures was adopted on September 22, 2021. Software to conduct auctions was rolled out shortly thereafter and the first auctions using it were completed in November 2021. The Prozorro Sale system, which had been operational for other asset classes since 2017, was designed to adapt insights from auction theory to Ukrainian conditions (Mylovanov et al. 2017). For public agricultural land, a sealed bid English auction was chosen to be implemented in three steps.8 The auction modality 6 Under ‘free privatization’, a right anchored in the constitution and executed by SGC, every Ukrainian citizen has the right t o receive up to 6 parcels of land (of different land use types) for free. While this provision was rarely exercised in 2010-14, it was re-discovered in 2014, ostensibly to compensate war veterans. Anecdotal evidence -including by a former Minister of Agriculture who claimed that 5 million ha of state land having been privatized since 2017 based on an unpublished audit- suggests these provisions were widely abused. 7 See for example https://ti-ukraine.org/en/blogs/selling-seized-assets-what-needs-to-change-in-setam-system/. 8 The auction starts with the starting prices offered by each of the participants after which participants have 5 minutes to review (anonymous) competitors’ offers. In each round, each participant has 3 minutes to change their bid by increasing it by an increment of at least 1% of the starting price or leave it unchanged in sequence with the first to revise her offer is the bidder who submitted the lowest price in the previous round. That is, the participant who submitted the highest bid in each round has an advantage as she can see and outbid other participants’ offers in each of the rounds. See https://prozakupki.prom.ua/shho-obovyazkovo-treba-znati-dlya-efektivnoyi-kupivli-u-prozorro-prodazhi/ for a detailed description of auction rules and https://zakupki.prom.ua/sale_auction_sandbox to run a mock auction. 7 differs from those of alternatives in three ways: First, the system is accessed via Application Programming Interface (API) by trading platforms (referred to as electronic marketplaces) that provide services and assistance to users and earn commission on auctions they win, ensuring competition among platforms.9 The role of system administrator is strictly separate from that of individual auction platforms and a commission with representatives from Government, business, and civil society that is independent from Prozorro Sale hears and arbitrates any disputes. Second, in addition to trading platforms’ and users’ ability to monitor the progress of auction on the system in real time, detailed information on auction participants’ real identity and the documents they submitted is disclosed immediately after an auction ended. This allows interested parties to check results and, if collusion is suspected, challenge the auction by filing a case with the complaints commission or the court. Finally, to eliminate the danger of backdoors that could open opportunities for illegitimate access or manipulation by insiders, the software code for Prozorro Sales is open source and publicly available, so everyone can check and understand the working of the system.10 There are three reasons to believe that timing and content of these laws was unexpected and that landowners did not delay auctioning off land in anticipation of obtaining a better outcome once legal changes had become effective. First, almost 2 years elapsed between the Parliament’s approval of the two laws in first reading and these laws’ final adoption in second reading, so the timing of legal approval was difficult to anticipate and might as well have been delayed much longer. Following the victory of the ‘Servants of the People’ party in 2019 presidential (April) and parliamentary (July) elections, drafts for the two laws considered here, together with broader issues of opening the agricultural land sales market, were discussed intensively by candidates for Ministerial office and members of Parliament during the summer of 2019, formally submitted on October 1, 2019, once relevant parliamentary committees had been formed, and approved in first reading (as draft laws 2194 and 2195) on November 14, 2019. At the point, the expectation was that these would become law quickly to create the basis for transparent land governance needed for opening the agricultural land market, an action discussed under a separate bill that was also linked to the International Monetary Fund (IMF) program, to have the expected effect. Second, the laws’ content remained unpredictable as, after their approval in first reading, an unprecedented number of amendments to both laws, aimed to overturn virtually all the substantive provisions of the bills. More than 5,500 amendments were submitted for draft law 2194 (later passed as 1423-IX) and more than 2,000 for 2195 (1444-IX). Procedural rules required that every amendment be read, discussed, and put to a 9 The state enterprise Prozorro Sales is currently the system administrator and 47 electronic trading platforms are active on the system. 10 The SETAM platform has only one electronic trading platform that is run by the system administrator, creating a conflict of interest through the possibility of accessing auction data and using these to interfere in the auction process. SETAM reveals participants’ aliases but no t accompanying documents or personal identifiers, making it difficult to establish their identity. SETAM software is not open source, making it impossible to identify vulnerabilities, hidden backdoors, or the scope for manipulation and there is no third-party monitoring or recourse to independent arbitration. 8 separate parliamentary vote. Political leadership was thus put in a situation where these laws could only be passed by having Parliament spend weeks of precious session time debating these amendments. Legal opinions unequivocally advised that the alternative of holding a vote to change procedural rules retroactively was not desirable as it would have resulted in all these laws to be rendered ineffective by court challenges to their legality. In fact, although the two laws had been envisaged as necessary precursors to the opening of the market for agricultural land, the law (552-IX) to effect this change was adopted on March 31, 2020 without the legal basis for transparent land market functioning in place. COVID19-related restrictions created additional hurdles soon thereafter, adding further uncertainty to the laws’ timing and content. Eventually, the timing for passing the laws (in April/May of 2021), their entry into force (on May 24, 2021 for Law 1423-IX and July 6, 2021 for law 1444-IX), and passage of regulations (in September 2021) were dictated by the need to avoid missing deadlines that would have resulted in forfeiture of large amounts of multilateral financing. The substantive content of regulating e-auctions remained highly contentious; with public pronouncements from influential Ministries for the President to veto law 1444 or to block issuance of implementing regulations, thus rendering the law ineffective. Finally, both law 1423-IX and 1444-IX were passed and became effective in the middle of the growing season in April/May when entering into new leases for crop or pasture lands (more than 80% of the total) seems impractical. Once Russian attacked on February 24, 2022, martial law suspended new auctions of public land,11 providing us with a window of 3-4 months between November 1, 2021 and March 31, 2022, during which e-auctions were effective that we can use for our analysis. 3. Data, descriptive evidence, and discontinuity plots 3.1 Data sources Auction data for our analysis come from two sources. Information on 31,483 auctions of public land for agricultural use12 initiated before November 2021 were obtained from SGC’s website while information on 2,752 auctions of public land for the same land use classes conducted via the Prozorro Sales auction platform between November 2021 and March 2022 is available on the Prozorro Sales website.13 For each auction, we have information on the auction organizer (i.e., SGC or local government), if the auction was 11 While the Prozorro.Sale platform continues to operate during the war, a large number of public land auctions launched immediately before the war were unsuccessful due to lack of bidders. Legislation adopted under martial law thus suspended electronic auctions and instead provides local governments with far-reaching powers to issue one-year leases to unused public land for the 2022 cropping season on a non-competitive basis to ensure national food security. At the time of writing, the possibility of reinstating electronic auctions for the 2023 agricultural season is being debated in Parliament. We thus limit our analysis to auctions completed before March 31, 2022, all of which were initiated before the war had started and add a war dummy for those completed after February 24, 2022 to indicate the change in expectations associated with this event. 12 Data states the intended purpose of the land with included categories of commercial agricultural production ( для ведення товарного виробництва); individual farming (для ведення фермерського господарства); individual farmers (фермерське господарство); or individual rural farming (для ведення особистого селянського господарства) by private individuals. 13 All auctions completed by March 31, 2022, had been initiated before the start of the war. We add a war dummy for auctions completed after February 24, 2022, to control for the change in expectations regarding future revenue from such land associated with the war. 9 completed, i.e., resulted in a transaction, and the auctioned parcels’ normative monetary valuation as a proxy of its productive capacity.14 To obtain additional parcel attributes, we used auctioned parcels’ cadastral number to locate them on the public cadastral map and used this information to obtain the distance from the parcel centroid to Kyiv and the nearest main road, grain elevator, and city. A land use map, constructed based on remotely sensed data and machine learning (Kussul et al. 2017; Shelestov et al. 2020), was used to identify if, in 2021, a parcel was used for crops, pasture, forest, or other uses. For successful auctions, we have the price paid by the winning bidder in UAH which we, convert into US$ at the exchange rate prevailing in the auction month to ensure comparability of prices over time and contract length. Table 1 shows that regulations had become effective and mechanisms for implementation were available, the shift to the regime of e-auctions on the Prozorro Sales platform was complete and auctions completed after November 1 indeed relied exclusively on the Prozorro online platform. They were also conducted by local governments rather than SGC. This contrasts with the earlier online auction pilot where 6% and 51% of auctions in 2018 and 2019, respectively, were conducted electronically by SGC on the SETAM platform. We note that, while legal changes introduced by law 1423-IX implied that SGC-organized auctions ceased after November 1, 2021, their share had already been on the decline earlier. Table 2 provides descriptive statistics on auctions and characteristics of parcels transferred via successful auctions post January 1, 2021 (columns 1-3) and overall (columns 4-7). For the 2021/22 sample, the median annual rent paid per ha and year is about $87 for parcels auctioned offline as compared to $342 for those auctioned online through the Prozorro Sales platform, closely mirroring a similar jump (from $157 to $417) in the mean. This is mirrored by a modestly higher normative value of parcels auctioned offline and online (from $437 to $677), dividing the rental value by the normative value points towards an increase from 20% to 51% for the median and from 26% to 47% for the mean associated with the shift from offline to mandatory online auctions. Other parcel attributes, in particular land use and distances to infrastructure are not significantly different between parcels auctioned offline before and online after the legal change. Comparing the equivalent figures for the pre-2021 sample in col. 5 and 7 of table 2 suggests at, with a median of $81 and $94 and a mean of $133 and $137 for auctions conducted offline and online on the SETAM platform, a difference that may well be explained by a higher share of more valuable cropland being auctioned online (69%) as compared to offline (61%) rather than the auction modality. We also note that, with $88, the median rental price is above the rental price observed in the private market, a variable 14 A parcel’s ‘normative value’ is based on assessment of its productive potential and used to determine tax obligations. While such valuations are adjusted for inflation in the previous calendar year on January 1, many of the underlying valuations date back to Soviet times. 10 the value of which increased significantly over the period from about $27 in 2015 to about $82 in 2020,15 a difference that is likely to reflect a size premium for public land which, with average parcel sizes of 16.03 ha in the pre-2021 period and 10.9 ha in 2021 and 2022, is well above the 3 ha for privately owned land parcels, lowering renters’ cost of land assembly. 3.2 Analytical approach and robustness checks Given the mandatory nature and clear cut-off date for the shift from offline to online auctions on November 1, 2021, a regression discontinuity design framework (Lee & Lemieux 2010) for auctions completed between January 1, 2021 and March 31, 2022 is exploited to assess if this reform resulted in changes in rental prices. Following Calonico et al. (2015), we produce discontinuity plots with weekly mean rental values centered on the reform date of November 1. To assess the association between e-auctions and lease prices parametrically, we index completed auctions by a, rayons by i, and years by t to estimate a rayon fixed effects regression = + 1 + 2 + + + (1) where is the winning bid received for auction a in rayon i in year t; Pat is a dummy indicating whether the auction was conducted electronically on the Prozorro platform, SGCa is an indicator on whether the (offline) auction had been organized by SGC or not; Xit a vector of parcel characteristics that includes the parcel’s normative value, its land use (crops, pasture, forest, or other), altitude, slope, and distances to the nearest city, elevator, primary road and to Kyiv, and a war dummy to takes the value of one for auctions initiated before but completed after February 24, 2022 (and is zero otherwise); is rayon fixed effects; is year fixed effects; and is a random error term. Standard errors in all the specifications are adjusted for clustering at rayon level throughout. The coefficient of interest is 1 , the estimated effect of regulations mandating use of the Prozorro electronic trading system. To check the robustness of the results obtained, this regression is also run for a subsample of auctions completed in the 10-weeks before and after the shift to mandatory online auctions. To assess the robustness of our results, we implement a placebo test that hypothetically lets the start date of Prozorro Sale auctions coincide with the passage of law 1444-IX on May 18, 2021 that mandates these and use only data for offline auctions -either for the entire January-October 2021 period or only for the 20- week window centered around the start data of the hypothetical reform date. A second robustness check involves the use of a segmented regression, drawing from the regression discontinuities in time literature (Hausman & Rapson 2018), that includes a weekly time trend and its interaction term with Pat -both for 15 See https://kse.ua/wp-content/uploads/2022/06/eng_jan2022_land_monitoring.pdf; http://www.kse.org.ua/en/research-policy/land/governance- monitoring/yearbook-2016-2017/ (accessed August 22, 2022). 11 the 2021/22 sample and narrow window restricting to days within 10 weeks of the start of the online Prozorro auction. The placebo test is repeated for this specification as well. A triangular kernel weight is used in all the segmented regressions. While our focus is on obtaining an estimate of the reform effect that can be given a causal interpretation if the conditions for regression discontinuity analysis apply, three additional aspects can help provide additional descriptive evidence to put findings into context: First, to the extent that the prohibition of local governments organizing auctions for land outside their settlement boundaries was enforced and other relevant parcel attributes are controlled for, 2 , the coefficient on (offline) auction organization by SGC can provide a lower bound estimate of the extent to which agency costs affected auction outcomes. Second, we can re-estimate (1) using data for the 2018/19 period when the pilot allowing online auctions on the SETAM platform was active to check the association between auction outcomes and use of SETAM-based online auctions on outcomes (in addition to getting an independent estimate of SGC effects using a different sample). Finally, we can estimate (1) using the entire 2015-22 sample and use the results to make a counterfactual prediction of the price that would have been obtained if parcels auctioned offline (via local government or SGC) would instead have been auctioned on Prozorro Sales using the post-reform process. 4. Results Analysis suggests that, in contrast to the legal reform on its own, regulatory reforms that mandated use of a transparent e-auction platform resulted in a jump in auction starting prices that almost doubled the lease price. Regression analysis supports this result which is robust to various robustness checks and a placebo test of a hypothetical reform date coinciding with the law’s passage rather than its implementation. Equivalent regression on pre-2021 data suggests e-auctions on the SETAM platform and auction organization by SGC are associated with lower revenue. 4.1 Discontinuity plots, regression results, and robustness checks Following Calonico et al. (2017), figure 1 presents discontinuity plots for weekly means of auction prices in the January 1, 2021 to March 31, 2022, i.e., the 44 weeks before the reform came into force and 20 weeks thereafter together with fitted polynomials of order 1 to 4. All plots point towards a significant and discontinuous upward shift of auction prices (from 4.5 to 5.5 in log terms), roughly equivalent to a doubling of lease prices. Note that, as the war started in week 15, the last five periods (week 15-19 after the actual implementation of Prozorro Sale auction) with lower prices refer to auctions completed under war conditions. In the absence of anticipatory action, characteristics of auctioned parcels not affected by reforms should not display any discontinuity around the temporal threshold (Hausman & Rapson 2018). Discontinuity plots 12 for parcel size in appendix figure 1 and normative value in figure 2 point to little if any discontinuity coinciding with the reform date -a small increase of the normative value is most likely triggered by the automatic annual adjustment of normative values for inflation. To allay fears about differences in parcels’ productive capacity as captured by the normative value driving our findings, we divide the auction price by the parcel’s normative value. The corresponding discontinuity plot in figure 3 points towards a discrete shift of this variable that coincides with the initiation of e-auctions on the Prozorro Sales platform. To complement evidence from discontinuity plots, we use parametric rayon fixed effects regressions. Table 3 displays results from the basic regression for the entire time span of 59 weeks comprising the 10 months before and the 5 months after the implementation of Prozorro Sale e-auction platform (col. 1). To avoid confounding effects from other policy or institutional changes, we report results from a narrower band of 20-weeks centered around the reform date (col. 2). In both cases, the highly significant point estimate for the post-reform dummy (0.946 in col. 1 and 0.999 in col. 2) suggests that shifting from an offline to an online auction on the Prozorro Sales platform organized by local government almost doubles the rental price received. To assess the reliability of our results, we conduct several robustness checks. First, we use a placebo where instead of the actual reform auctions, we set a hypothetical start date of Prozorro Sale auctions at the date of the passage of law 1444-IX (mandating exclusive use of transparent online auctions to lease out any public land) and use only data for offline auctions -either for the entire January-October 2021 period or only for the 20-week window centered around the start data of the hypothetical reform date. The coefficient on the hypothetical e-auction reform is insignificantly different from zero, consistent with the notion that legal reform on its own had no impact on auction lease prices. Signs and levels of significance for the rest of the coefficients remain unchanged. A second robustness check involves the use of a segmented regression that includes a weekly time trend and its interaction term with Pat. While the point estimate of 1 , remains virtually unchanged, neither the trend nor its interaction are significant and repeating the same placebo test as reported earlier does not materially affect the results. Beyond 1 , significant coefficients on the parcel’s normative value (0.59), its area (0.176), and its use for crops (0.259) or forest (-0.231) as compared to pasture as the omitted category, and a war dummy (-0.41) that suggests a steep decline in leases after the war started are consistent with expectations. At a descriptive level, the coefficient (-0.43 for the entire period and -0.57 for the 20-week window) on offline auctions organized by SGC suggests that even after controlling for land quality, SGC involvement is associated with much lower lease prices, possibly due to the agency costs and opportunities for corruption described above. 13 4.2 Descriptive evidence on earlier reforms Having seen that mandatory e-auctions significantly increased land prices, we estimate (1) using data from the entire 2015-22 period as an additional robustness check and using data from 2018/19 when e-auctions using the SETAM platform were an option. For the entire period (table 5, col. 2), the coefficient on Prozorro Sales platform use is virtually indistinguishable from that obtained earlier while the coefficient on SGC organization (-0.381) is slightly smaller whereas the coefficient on SETAM (-0.207) is negative and significant. This would imply that pilot use of the SETAM e-auction platform reduced competition compared to offline auctions. The fact that information technology, even if linked to a blockchain, can reduce transparency shows that careful attention to implementation modalities is warranted. Restricting the sample to the months in 2018/19 when SETAM was in operation (see table 5 col. 1 for results) reduces the precision of the estimates but produces results that are otherwise equivalent. Leasing out public land at a rate that is less than what could have been obtained via a transparent process constrains local governments’ fiscal space and their ability to provide local public goods or social services for the duration of the contract. Using the coefficients estimated above allows us to quantify the amount of incremental income that could have been obtained had the two policy reforms discussed above -i.e., putting local governments rather than SGC in charge of auction organization and using the Prozorro electronic auction platform- been implemented earlier. We do so by predicting, for each parcel auctioned in the pre- reform period, the lease price under if, instead of conducting them offline or on SETAM, auctions would instead have been conducted on the Prozorro platform and organized by local government rather than SGC. Results presented in table 6, show that the mean annual lease revenue gain from shifting to the post-reform modality of leasing would have been almost US$203/ha. Summing this up over the 19,267 parcels (covering 312,813 ha) auctioned before November 2021 suggests local governments could have obtained additional annual lease revenue of US$63 million in each of the 9 years for which leases were on average concluded. Beyond the 0.3 million ha of land that was leased via auction, at least 1.5 million ha of public land was transferred to individuals through ‘free privatization’ in 2017 – 2020 (Nivievskyi 2020). Assuming average quality of these parcels is equal to parcels auctioned, lost annual lease revenue from free privatization would amount to US$4.5 million. Thus, had the reforms eventually enacted in 2021 been put in place in 2017 and extended to cover land subject to free privatization, annual lease revenue generated would be about US$500 million, providing every rayon with about US$1 million per year, in addition to allowing to pay a perpetual income of US$100 per ha to those who received such land in line with legal requirements. 14 5. Conclusion and policy implications As secular trends increase land values, use rights to millions of hectares of public land are transferred every year to rural investors or urban developers, often using centralized and non-competitive mechanisms that may be justified with reference to local capacity gaps or elite capture. Limited success of past efforts to make public land management and disposal in Ukraine more transparent shows these challenges are real and may be exacerbated by ill-considered digitization, as happened with adoption of the SETAM e-auction platform that was associated with even worse outcomes than offline auctions. It is thus encouraging to find that even in such a weak governance setting, legal reform and provision of a transparent open-source e-auction platform facilitated a doubling of lease revenue, likely by improving market functioning in terms of (i) reducing barriers to entry and increasing competition and transparency; (ii) a leveling of the playing field to provide all interested parties with the same set of information and create audit trails to further reduce corruption risk; and (iii) eliminating intermediaries whose incentives may not be aligned with those of land owners and who may suffer from conflict of interest. While this experience holds important lessons for countries beyond Ukraine, the principles that have guided mandatory introduction of the Prozorro Sales e-auction platform are likely to be applicable to improve land market functioning (Acampora et al. 2022), spatial planning, and revenue collection by local governments, all issues that will acquire even higher importance in a context of reconstruction. 15 Table 1: Summary of data and their distribution between online and offline auctions No. of auctions Transferred Share of auctions.. Total Successful area (ha) ..online ..org. by SGC 2015 1,935 1,530 28,345 0 0.54 2016 2,817 1,916 42,754 0 0.87 2017 3,388 2,139 43,358 0 0.84 2018 5,108 3,519 65,859 0.06 0.85 2019 8,710 5,327 83,269 0.51 0.69 2020 2,431 847 13,408 0 0.50 2021 7,805 5,232 58,442 0.09 0.38 Pre-Prozorro 7,094 4,774 54,425 0 0.41 Prozorro 711 458 4,017 1 0 2022 2,041 1,043 9,964 1 0 Total 34,235 21,553 345,399 0.22 0.61 Source: Own computation from SGC and Prozorro data as described in the text. 16 Table 2: Descriptive statistics by auction type Sample I (2021/22) Sample II (2015-22) Total Offline Online Total Offline Online All online SETAM Panel A: All auctions Total # of parcels 9,846 7,094 2,752 34,235 26,773 7,462 4,710 of which successful 0.64 0.67 0.55 0.63 0.65 0.55 0.55 Organized by SGC 0.3 0.41 0 0.61 0.61 0.62 0.99 N. value (US$/ha) med. 462.78 423.75 579.29 522.57 505.48 559.83 548.08 N. value (US$/ha) avg. 756.88 680.93 952.51 810.68 818.50 795.77 704.38 Parcel area (ha) 13.22 11.82 16.81 16.47 16.34 16.91 16.97 Land use Crops 0.41 0.41 0.42 0.56 0.57 0.51 0.57 Pasture 0.39 0.4 0.37 0.35 0.33 0.42 0.45 Forest 0.16 0.16 0.16 0.12 0.12 0.13 0.1 Distance in km to Main road 8.99 8.85 9.37 9.25 9.24 9.29 9.24 Nearest city 15.67 15.78 15.36 16.28 16.27 16.32 16.89 Grain elevator 13.69 13.83 13.31 13.92 13.81 14.34 14.97 Kyiv 311.94 316.06 300.97 319.37 318.57 322.3 335.13 Panel B: Successful auctions Total # of parcels 6,275 4,774 1,501 21,553 17,471 4,082 2,581 Organized by SGC 0.27 0.36 0.00 0.58 0.57 0.62 0.99 Parcel area (ha) 10.90 11.40 9.31 16.03 16.41 14.38 17.32 Rent (US$/ha) med. 116.02 87.21 341.69 87.59 81.21 139.77 93.81 Rent (US$/ha) avg. 219.36 157.28 416.79 153.17 132.96 239.65 136.64 NV (US$/ha) med. 485.50 437.34 677.06 582.25 563.82 626.33 596.59 NV (US$/ha) avg. 681.47 616.34 888.47 760.05 763.78 751.98 672.76 Rent/NV med. 0.239 0.199 0.505 0.150 0.144 0.223 0.157 Rent/NV avg. 0.322 0.255 0.469 0.202 0.174 0.319 0.203 Contract length (years) 8.61 8.33 9.51 8.84 9.04 7.98 7.08 Land use Crops 0.48 0.48 0.47 0.65 0.66 0.61 0.69 Pasture 0.35 0.34 0.36 0.29 0.27 0.37 0.38 Forest 0.15 0.15 0.13 0.09 0.09 0.08 0.06 Distance in km to Main road 8.65 8.48 9.22 9.31 9.30 9.32 9.38 Nearest city 15.64 15.66 15.58 16.58 16.59 16.57 17.16 Grain elevator 13.39 13.55 12.84 13.56 13.50 13.84 14.43 Kyiv 297.22 295.17 304.04 312.33 309.85 323.26 334.55 Source: Own computation from SGC and Prozorro data as described in the text. 17 Table 3: Regressions and placebo test for effect of Prozorro Sale online auction on lease prices Actual reform (reform on Nov1, 2021) Placebo (reform on May 18, 2021) all observations 20-week window all observations 20-week window Online auction on Prozorro Sales 0.946*** 0.999*** -0.0224 -0.0504 (0.0987) (0.154) (0.0547) (0.0656) Offline auction organized by SGC -0.434*** -0.576*** -0.386*** -0.288*** (0.0725) (0.199) (0.0801) (0.0937) Normative value (US$/ha) 0.590*** 0.554*** 0.659*** 0.660*** (0.0323) (0.0641) (0.0342) (0.0420) Parcel area (ha) 0.176*** 0.139*** 0.169*** 0.188*** (0.0251) (0.0484) (0.0247) (0.0268) Dist. to primary road (km) -0.0361* -0.0114 -0.0304 -0.0257 (0.0208) (0.0310) (0.0243) (0.0278) Crop land 0.259*** 0.164*** 0.276*** 0.248*** (0.0354) (0.0470) (0.0376) (0.0534) Forest -0.231*** -0.273*** -0.192*** -0.209*** (0.0437) (0.0622) (0.0501) (0.0721) Other land class -0.126 -0.154 -0.0898 -0.195 (0.0953) (0.160) (0.112) (0.130) Post-war dummy -0.410*** (0.103) Distance to nearest city (km) 0.0192 0.00487 0.0245 0.0202 (0.0209) (0.0329) (0.0248) (0.0295) Distance to Kyiv (km) -0.373 0.198 -0.322 -0.346 (0.375) (0.738) (0.439) (0.482) Distance to elevators (km) 0.0400 0.0513 0.0237 0.0116 (0.0394) (0.0562) (0.0398) (0.0430) Slope -0.00925 -0.00435 -0.00649 -0.00385 (0.00607) (0.00792) (0.00673) (0.00870) Altitude (m) 0.0817** 0.131 0.0591** 0.0375 (0.0322) (0.0863) (0.0267) (0.0249) 2022 year dummy -0.0436 0.0432 (0.106) (0.135) Constant 2.207 -0.959 1.612 1.831 (2.097) (4.092) (2.461) (2.654) N 6134 2124 4716 2800 R2 (within) 0.371 0.349 0.323 0.294 Note: Dependent variable is the lease price (US$/ha) in logs. Regressions include rayon and year fixed effects. Covariates that are not indicator variables (i.e., normative value, distance to road) are in natural logs. The observations in the narrow window regressions are limited to days within 10 weeks of the start of online Prozorro auction. For the placebo test (col. 3 and 4), data is restricted to offline auctions in 2021 comparing prices on either side of the cutoff date May 18, 2021 (i.e., the date when the law that mandates electronic auction was passed). Robust standard errors adjusted for clustering at the rayon level in parentheses. * p<0.10, ** p<0.05, *** p<0.010. 18 Table 4: Regressions and placebo test with segmented regression and weekly trend Actual reform (reform on Nov1, 2021) Placebo (reform on May 18, 2021) all observations 20-week window all observations 20-week window Online auction on Prozorro Sales 1.035*** 0.781*** -0.0374 0.0171 (0.0774) (0.177) (0.0636) (0.0822) Weekly trend -0.00303 0.0239 -0.00491 -0.0216** (0.00219) (0.0200) (0.00511) (0.0109) Prozorro # Weekly time trend -0.0119 -0.0265 0.00432 0.0180 (0.00758) (0.0242) (0.00730) (0.0153) Normative value (US$/ha) 0.599*** 0.597*** 0.684*** 0.687*** (0.0210) (0.0379) (0.0256) (0.0366) Offline auction organized by SGC -0.522*** -0.584*** -0.430*** -0.384*** (0.0503) (0.197) (0.0526) (0.0712) Parcel area (ha) 0.167*** 0.132*** 0.172*** 0.187*** (0.0137) (0.0255) (0.0157) (0.0225) Dist. to primary road (km) -0.0344** -0.0128 -0.0367** -0.0602** (0.0143) (0.0245) (0.0185) (0.0265) Crop land 0.199*** 0.185*** 0.267*** 0.280*** (0.0301) (0.0490) (0.0370) (0.0535) Forest -0.259*** -0.263*** -0.166*** -0.141** (0.0389) (0.0633) (0.0478) (0.0700) Other land class -0.138 -0.174 -0.146 -0.192* (0.0894) (0.162) (0.0939) (0.116) Post-war dummy -0.333*** (0.0897) Distance to nearest city (km) 0.0315** -0.00572 0.0330* 0.0551* (0.0160) (0.0286) (0.0198) (0.0296) Distance to Kyiv (km) -0.286 -0.144 -0.498* -0.949** (0.228) (0.405) (0.282) (0.443) Distance to elevators (km) 0.0563** 0.0304 0.0293 -0.0124 (0.0238) (0.0430) (0.0292) (0.0432) Slope -0.00383 -0.00129 -0.00145 0.00792 (0.00523) (0.00865) (0.00678) (0.00957) Altitude (m) 0.0972*** 0.148** 0.0443** 0.0332* (0.0240) (0.0680) (0.0173) (0.0200) Constant 1.519 0.789 2.503 5.033** (1.268) (2.302) (1.578) (2.478) N 5,905 2,038 4,567 2,731 R2 (within) 0.621 0.738 0.560 0.588 Note: Dependent variable is the lease price (US$/ha) in logs. Regressions include rayon dummies but not reported. Covariates that are not indicator variables (i.e., normative value, distance to road) are in natural logs. The observations in the narrow window regressions are limited to days within 10 weeks of the start of online Prozorro auction. For the placebo test (col. 3 and 4), data is restricted to offline auctions in 2021 comparing prices on either side of the cutoff date May 18, 2021 (i.e., the date when the law that mandates electronic auction was passed). Triangular kernel weights are used in all the specifications and robust standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.010. 19 Table 5: Comparing effects of different e-auction platforms Auctions conducted in 2018/19 2015-22 e-auction on Prozorro Sales 0.931*** (0.111) e-auction on SETAM -0.147** -0.207*** (0.0614) (0.0416) Offline auction organized by SGC -0.279*** -0.381*** (0.108) (0.0352) Normative value (US$/ha) 0.655*** 0.594*** (0.0321) (0.0213) Parcel area (ha) 0.121*** 0.113*** (0.0157) (0.0164) Dist. to primary road (km) -0.0323 -0.0367*** (0.0209) (0.0121) Crop land 0.222*** 0.302*** (0.0492) (0.0284) Forest -0.149** -0.105*** (0.0626) (0.0336) Other land class 0.285 0.106 (0.291) (0.0977) Post-war dummy -0.348*** (0.119) Distance to nearest city (km) 0.00986 0.0115 (0.0252) (0.0150) Distance to Kyiv (km) -0.714 -0.337* (0.522) (0.188) Distance to elevators (km) 0.0144 0.00574 (0.0375) (0.0231) Slope -0.00985 -0.0138*** (0.00759) (0.00366) Altitude (m) -0.00652 0.100** (0.0530) (0.0421) 2022 year dummy 4.083 1.523 (3.026) (1.128) No. of obs. 4,480 20,688 R2 (within) 0.286 0.296 Note: Dependent variable is the lease price (US$/ha) in logs with rayon and year fixed effects. Covariates that are not indicator variables (i.e., normative value, distance to road) are in natural logs. The observations for the SETAM sample are restricted to months when the SETAM platform was operational (2018/19). Robust standard errors adjusted for clustering at the rayon level in parentheses. * p<0.10, ** p<0.05, *** p<0.010. 20 Table 6: Predicted revenue gain from transferring parcels transacted pre-reform under reform modality All parcels Auction modality Offline SETAM Predicted lease price ($/ha per year) without reform 98.09 97.89 99.42 Predicted lease price ($/ha per year) with reform 301.04 299.94 310.10 Total area auctioned January 2015-Novmber 2021 (ha) 312,813 271,654 41,079 Number of parcels 19,267 16,852 2,415 Potential revenue gain from reform (US$ millions) 65.42 56.41 9.01 Source: Own computation based on regression results column 4 in table 5. 21 Figure 1: Discontinuity plots of mean weekly winning bids with polynomials of degree 1-4 22 Figure 2: Discontinuity plots of mean weekly normative value with polynomials of degree 1-4 23 Figure 3: Discontinuity plots of mean weekly winning bid/normative value with polynomials of degree 1-4 24 Appendix figure 1: Discontinuity of mean weekly parcel size with polynomials of degree 1-4 25 References: Acampora, M., Casaburi, L., Willis, J., 2022. Land rental markets: Experimental evidence from Kenya. University of Zurich, Zurich Agarwal, S., Li, J., Teo, E., Cheong, A., 2018. Strategic Sequential Bidding for Government Land Auction Sales--Evidence from Singapore. Journal of Real Estate Finance and Economics 57, 535-565 Ali, D., Deininger, K., Harris, A., 2019. Does Large Farm Establishment Create Benefits for Neighboring Smallholders? Evidence from Ethiopia. Land Economics 95, 71-90 Ali, D.A., Deininger, K., 2022. Institutional determinants of large land-based investments’ performance in Zambia: Does title enhance productivity and structural transformation? World Development 157, 105932 Anseeuw, W., Bache, W., Bru, T., Giger, M., Lay, J., Messerli, P., Nolte, K., 2012. Transnational land deals for agriculture in the global South: Analytical report based on the land matrix database. Bern, Montpellier, Hamburg Anti, S., 2021. Land Grabs and Labor in Cambodia. Journal of Development Economics 149 Arezki, R., Deininger, K., Selod, H., 2015. What Drives the Global “Land Rush”? The World Bank Economic Review 29, 207 - 233 Athey, S., Levin, J., Seira, E., 2011. Comparing open and Sealed Bid Auctions: Evidence from Timber Auctions*. Quarterly Journal of Economics 126, 207-257 Bélair, J., 2021. Farmland investments in Tanzania: The impact of protected domestic markets and patronage relations. World Development 139, 105298 Braganca, A., Dahis, R., 2021. Cutting Special Interests by the Roots: Evidence from the Brazilian Amazon. PUC, Rio de Janeiro Brito, B., Barreto, P., Brandão, A., Baima, S., Gomes, P.H., 2019. Stimulus for land grabbing and deforestation in the Brazilian Amazon. Environmental Research Letters 14, 064018 Bulow, J., Klemperer, P., 1996. Auctions versus negotiations. American Economic Review 86, 180-194 Bulow, J., Klemperer, P., 2009. Why Do Sellers (Usually) Prefer Auctions? American Economic Review 99, 1544-1575 Cai, H., Henderson, J.V., Zhang, Q., 2013. China's Land Market Auctions: Evidence of Corruption? RAND Journal of Economics 44, 488-521 Cai, H., Wang, Z., Zhang, Q., 2017. To build above the limit? Implementation of land use regulations in urban China. Journal of Urban Economics 98, 223-233 Calonico, S., Cattaneo, M.D., Farrell, M.H., Titiunik, R., 2017. rdrobust: Software for Regression Discontinuity Designs. Stata Journal 17, 372-404 Calonico, S., Cattaneo, M.D., Titiunik, R., 2015. Optimal Data-Driven Regression Discontinuity Plots. Journal of the American Statistical Association 110, 1753-1769 Chassang, S., Ortner, J., 2019. Collusion in Auctions with Constrained Bids: Theory and Evidence from Public Procurement. Journal of Political Economy 127, 2269-2300 Che, Y.-K., Condorelli, D., Kim, J., 2018. Weak cartels and collusion-proof auctions. Journal of Economic Theory 178, 398-435 Che, Y.-K., Kim, J., 2009. Optimal collusion-proof auctions. Journal of Economic Theory 144, 565-603 Chen, T., Kung, J.K.-s., 2019. Busting the 'Princelings': The Campaign against Corruption in China's Primary Land Market. Quarterly Journal of Economics 134, 185-226 Chow, Y.L., Hafalir, I.E., Yavas, A., 2015. Auction versus Negotiated Sale: Evidence from Real Estate Sales. Real Estate Economics 43, 432-470 Cisneros, E., Kis-Katos, K., Nuryartono, N., 2021. Palm Oil and the Politics of Deforestation in Indonesia. Journal of Environmental Economics and Management 108 Compiani, G., Haile, P., Sant'Anna, M., 2020. Common Values, Unobserved Heterogeneity, and Endogenous Entry in US Offshore Oil Lease Auctions. Journal of Political Economy 128, 3872-3912 Cotula, L., 2014. Testing Claims about Large Land Deals in Africa: Findings from a Multi-country Study. Journal of Development Studies 50, 903-925 Croonenbroeck, C., Odening, M., Hüttel, S., 2020. Farmland values and bidder behaviour in first-price land auctions. European Review of Agricultural Economics 47, 558-590 Deininger, K., Jin, S., 2006. Tenure Security and Land-Related Investment: Evidence from Ethiopia. European Economic Review 50, 1245-1277 Deininger, K., Xia, F., 2016. Quantifying Spillover Effects from Large Land-based Investment: The Case of Mozambique. World Development 87, 227-241 Dieterle, C., 2021. Global Governance Meets Local Land Tenure: International Codes of Conduct for Responsible Land Investments in Uganda. The Journal of Development Studies, 1-17 26 Engström, L., Hajdu, F., 2018. Conjuring ‘Win -World’ – Resilient Development Narratives in a Large-Scale Agro-Investment in Tanzania. The Journal of Development Studies 55, 1-20 Faguet, J.-P., Sánchez, F., Villaveces, M.-J., 2020. The perversion of public land distribution by landed elites: Power, inequality and development in Colombia. World Development 136, 105036 Glover, S., Jones, S., 2019. Can Commercial Farming Promote Rural Dynamism in Sub-Saharan Africa? Evidence from Mozambique. World Development 114, 110-121 Graubner, M., Ostapchuk, I., Gagalyuk, T., 2021. Agroholdings and Land Rental Markets: A Spatial Competition Perspective. European Review of Agricultural Economics 48, 158-206 Hausman, C., Rapson, D.S., 2018. Regression Discontinuity in Time: Considerations for Empirical Applications. Annual Review of Resource Economics 10, 533-552 Havrylyshyn, O., 2017. The Political Economy of Independent Ukraine: Slow Starts, False Starts, and a Last Chance? Palgrave MacMillan, London. Hortaçsu, A., Kastl, J., Zhang, A., 2018. Bid Shading and Bidder Surplus in the US Treasury Auction System. American Economic Review 108, 147-169 Hüttel, S., Jetzinger, S., Odening, M., 2014. Forced Sales and Farmland Prices. Land Economics 90, 395-410 Hüttel, S., Wildermann, L., Croonenbroeck, C., 2016. How do institutional market players matter in farmland pricing? Land Use Policy 59, 154-167 Kaufmann, D., Kraay, A., Mastruzzi, M., 2011. The Worldwide Governance Indicators: Methodology and Analytical Issues. Hague Journal on the Rule of Law 3, 220-246 Kraus, S., Heilmayr, R., Koch, N., 2021. Spillovers to manufacturing plants from multi-million dollar plantations: evidence from the Indonesian palm oil boom. Mercator Research Institute on Global Commons and Climate Change, Berlin Kussul, N., Lavreniuk, M., Skakun, S., Shelestov, A., 2017. Deep Learning Classification of Land Cover and Crop Types Using Remote Sensing Data. IEEE Geoscience and Remote Sensing Letters 14, 778-782 Kvartiuk, V., Herzfeld, T., Bukin, E., 2022. Decentralized public farmland conveyance: Rental rights auctioning in Ukraine. Land Use Policy 114, 105983 Lay, J., Nolte, K., Sipangule, K., 2020. Large-scale farms in Zambia: Locational patterns and spillovers to smallholder agriculture. World Development, 105277 Lee, D.S., Lemieux, T., 2010. Regression Discontinuity Designs in Economics. Journal of Economic Literature 48, 281-355 Maganga, F.P., Askew, K., Odgaard, R., Stein, H., 2016. Dispossession through Formalization: Tanzania and the G8 Land Agenda in Africa. Asian Journal of African Studies 40, 3-49 Marshall, R.C., Marx, L.M., 2009. The vulnerability of auctions to bidder collusion. Quarterly Journal of Economics 124, 883- 910 Müller, M.F., Penny, G., Niles, M.T., Ricciardi, V., Chiarelli, D.D., Davis, K.F., Dell’Angelo, J., D’Odorico, P., Rosa, L., Rulli, M.C., Mueller, N.D., 2021. Impact of transnational land acquisitions on local food security and dietary diversity. Proceedings of the National Academy of Sciences 118, e2020535118 Mylovanov, T., Shapoval, N., Zapchelnyuk, a., Huffman, D., Vohra, R., Sobolev, O., 2017. Selling toxic assets in Ukraine: Overview of the current situation and suggestions for improvement of the auction design. Kyiv School of Economics and Prozorro.Sale, Kyiv Nanhthavong, V., Epprecht, M., Hett, C., Zaehringer, J.G., Messerli, P., 2020. Poverty trends in villages affected by land-based investments in rural Laos. Applied Geography 124, 102298 Neyter, R., Nivievskyi, O., 2022. Local versus central public land governance: Evidence from spatial analysis of land auctions in Ukraine. Kyiv School of Economics, Center for Food and Land Use Research, Kyiv Nolte, K., Voget-Kleschin, L., 2014. Consultation in Large-Scale Land Acquisitions: An Evaluation of Three Cases in Mali. World Development 64, 654-668 Pailler, S., 2018. Re-election Incentives and Deforestation Cycles in the Brazilian Amazon. Journal of Environmental Economics and Management 88, 345-365 Rulli, M.C., Saviori, A., D'Odorico, P., 2013. Global land and water grabbing. Proceedings of the National Academy of Sciences of the United States of America 110, 892-897 Shelestov, A., Lavreniuk, M., Vasiliev, V., Shumilo, L., Kolotii, A., Yailymov, B., Kussul, N., Yailymova, H., 2020. Cloud Approach to Automated Crop Classification Using Sentinel-1 Imagery. IEEE Trans. Big Data 6, 572-582 Shen, J., Pretorius, F., Li, X., 2019. Does Joint Bidding Reduce Competition? Evidence from Hong Kong Land Auctions. Journal of Real Estate Finance & Economics 58, 111-132 Shete, M., Rutten, M., Schoneveld, G.C., Zewude, E., 2016. Land-Use Changes by Large-Scale Plantations and Their Effects on Soil Organic Carbon, Micronutrients and Bulk Density: Empirical Evidence from Ethiopia. Agriculture and Human Values 33, 689-704 27 Sulle, E., 2020. Bureaucrats, investors and smallholders: contesting land rights and agro-commercialisation in the Southern agricultural growth corridor of Tanzania. Journal of Eastern African Studies 14, 332-353 Weber, R.J., 1997. Making more from less: strategic demand reduction in the FCC spectrum auctions. Journal of Economics & Management Strategy 6, 529-548 Weidong, Q., Xiaolong, L., 2012. Assessing the Performance of Chinese Land Lease Auctions: Evidence from Beijing. Journal of Real Estate Research 34, 291-310 Widengard, M., 2019. Land Deals, and How Not All States React the Same: Zambia and the Chinese Request. Review of African Political Economy 46, 615-631 Zadorozhna, O., 2020. Clientelism and Land Market Outcomes in Ukraine. Eastern European Economics 58, 478-496 28