The Pennsylvania State University. The Graduate School. College of Agricultural Sciences BILATERAL INFORMATION ASYMMETRY IN THE DESIGN OF AN AGRI-

Transcription

1 The Pennsylvania State University The Graduate School College of Agricultural Sciences BILATERAL INFORMATION ASYMMETRY IN THE DESIGN OF AN AGRI- ENVIRONMENTAL PROGRAM : AN APPLICATION TO PEATLAND RETIREMENT IN NORWAY A Dissertation in Agricultural, Environmental, and Regional Economics by Wonjoo Cho 2016 Wonjoo Cho Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy August 2016

2 The dissertation of Wonjoo Cho was reviewed and approved* by the following: David Blandford Professor of Agricultural and Environmental Economics Dissertation Advisor Chair of Committee James Shortle Distinguished Professor of Agricultural and Environmental Economics Katherine Zipp Assistant Professor of Environmental and Resource Economics Seth Blumsack Associate Professor of Energy and Mineral Engineering C. Daniel Azzara Alan R. Warehime Professor of Agribusiness Interim Head of the Department of Agricultural Economics, Sociology, and Education *Signatures are on file in the Graduate School

3 iii ABSTRACT Estimates indicate that agriculture is a significant source of global greenhouse gas (GHG) emissions. GHG mitigation through agri-environmental programs could be important in achieving emission reduction targets under the recently concluded U.N. global climate agreement. This study uses the principal-agent model to examine a peat land retirement program to reduce agricultural emissions in Norway. The focus is on the role of the government s private information in program design. Two cases are examined. First, optimal contracts are derived when farmers have private information about the costs of implementing peat land retirement, but the government reveals its private information on the resulting benefits through differentiated contracts. This corresponds to the standard targeting strategy with one-sided information asymmetry. In the second case, an informed principal model developed by Maskin and Tirole (1990) is employed to address bilateral information asymmetry. Using the informed principal model, the government offers the same menu of contracts to farmers in order to conceal its private information. Empirical results show that the government can achieve a higher payoff by using a pooling offer.

4 iv TABLE OF CONTENTS LIST OF FIGURES... viii LIST OF TABLES... x ACKNOWLEDGEMENTS... xii Chapter 1. Introduction Greenhouse Gas Emissions, Global Warming and Climate Change The Contribution of Agriculture to Global GHG Emissions Norwegian Agriculture and GHG emissions Environmental Policies for Agriculture and the Principal-Agent Model Research Objectives and Dissertation Outline Chapter 2. Design of Agri-Environmental Policy with One-Sided Information Asymmetry Contract Design using the Principal-Agent Model Agri-Environmental Policy with a Targeting Strategy Two Common Misunderstandings about the Principal-Agent Model Theoretical Model of Agri-Environmental Policy with Targeting Strategy... 28

5 2.4. Optimal Agri-Environmental Policy Design with Targeting Strategy v Ratio of Lagrange Multipliers Chapter 3. Design of Agri-Environmental Policy with Bilateral Information Asymmetry The Principal-Agent Model with Bilateral Information Asymmetry Literature Review on the Informed Principal Model Design of Agri-Environmental Policy with Bilateral Information Asymmetry Optimal Agri-Environmental Policy with Bilateral Information Asymmetry Pooling Offer Application to Peatland Retirement Program in Norway Managing Peatlands in Norway Theoretical Framework for a Peatland Retirement Program in Norway Chapter 4. Parameters Parameters Used for the Empirical Analysis Agricultural Sector Model for Norway (Jordmod) Carbon Price... 68

6 Time Horizon vi 4.2. Monetizing the Principal s Benefits Environmental Benefits of GHG mitigation Monetizing Agents Costs Income Foregone through Program Participation Approximation of Income Foregone Functions Chapter 5. Empirical Analysis of a Peatland Retirement Program Empirical Model of Peatland Retirement Characteristics of Optimum Allocations Optimal Contracts for Peatland Retirement with Unconstrained Farm Size Comparison of Optimum Allocations Comparison of Principal s Payoffs and Agent s Utility Comparison of Information Rent and Downward Distortion Optimal Contracts with a Farm Size Constraint Optimal Contracts for National Peatlands Retirement Chapter 6. Concluding Remarks

7 References vii Appendix A. Optimal Agri-Environmental Policy with Targeting Strategy Appendix B. Technical Description of Norwegian Agricultural Sector Model (Jordmod) Appendix C. Simulations of the Payoff Possibility Curve

8 viii LIST OF FIGURES Figure 1.1 Observed U.S. Trends in Heavy Precipitation... 2 Figure 1.2 Total Emissions of all GHG from the Different Source Categories Figure 1.3 GHG Emissions from Norwegian Agriculture (2011)... 7 Figure 2.1 Comparison of a Separating Contract and a Targeting Strategy Figure 3.1 Comparison of a Targeting Strategy and a Pooling Offer Figure 3.2 Drainage of Peatlands in Norway Figure 4.1 Depreciation (Subsidence) of Drained Peatlands Figure 4.2 Economic Lifespan of Peatlands Figure 4.3 Carbon Emissions from Drained Peatland Figure 4.4 Carbon Emissions from Rewetted Peatland Figure 4.5 Sorting Condition Figure 5.1 Proposed Contracts in a Targeting Strategy and a Pooling Offer Figure 5.2 Payoff Possibility Curve with the Pooling Offer Figure 5.3 National Optimum Allocations using Pooling Offer

9 Figure C.1 Simulation Result of Payoffs Possibility Curve ix

10 x LIST OF TABLES Table 1.1 Dissertation Outline Table 4.1 Summary of Parameters and Data Sources Table 4.2 GHG Emissions from Drained Peatland Table 4.3 GHG Emissions from Rewetted Peatland Table 4.4 GHG mitigation by Peatland Restoration Table 4.5 GHG Mitigation from Reduced Agricultural Production Table 4.6 Saving in Government Expenditure Table 4.7 Total Benefits of Peatland Retirement per Hectare Table 4.8 Simulated Representative Farm Incomes by Farm Size Table 4.9 Net Present Value of Income Foregone Table 5.1 Synthetic Parameter Values Used for Empirical Analysis Table 5.2 Comparison of Optimum Allocations without Farm Size Restriction Table 5.3 Comparison of Principal s Payoffs and Agent s Utility Table 5.4 Inefficiency of the Contract with a Targeting Strategy Table 5.5 Inefficiency of the Contract with a Pooling Offer (Case 1)

11 Table 5.6 Inefficiency of the Contract with a Pooling Offer (Case 2) xi Table 5.7 Optimum Allocations for Peatland Retirement with Farm Size Constraint Table B.1 Simulated Results Compared to Actual Figures (Base Year = 2004)

12 xii ACKNOWLEDGEMENTS I wish to acknowledge many individuals who have supported me during my graduate study at the Pennsylvania State University. In particular, I would like to express my deep gratitude to my advisor, Dr. David Blandford, for his encouragement and commitment. It was indeed my good fortune to work with him. This dissertation would not have been possible without his invaluable guidance and support. Also, I would like to thank Dr. James Shortle, Dr. Katherine Zipp, and Dr. Seth Blumsack for their insightful comments to improve my dissertation. In addition, I am indebted to Dr. Erling Vårdal and Dr. Ivar Gaasland at the University of Bergen who generously provided their data and helpful comments on my dissertation. And Dr. Arne Grønlund at the Bioforsk in Norway willingly shared his profound knowledge of a peat soil with me. As always, I am grateful to my family for their endless love and understanding throughout my life. Finally, I offer my heartfelt thanks to Mina. None of my achievements would have been possible without her.

13 1 Chapter 1. Introduction 1.1. Greenhouse Gas Emissions, Global Warming and Climate Change Anthropogenic greenhouse gases (GHG) emissions those that result from human activity have risen to the highest levels on record. The increased anthropogenic GHG emissions are mainly due to economic growth and an increasing population and are predominantly caused by fossil fuel combustion and changes in land use. Current atmospheric concentrations of GHGs (carbon dioxide, methane and nitrous oxide) have not been seen for the last 800,000 years. Atmospheric concentrations of GHGs are likely to have been the dominant source of observed global warming since the mid-20 th century. The globally-averaged combined land and ocean surface temperature has increased by 0.85 C over the period 1880 to 2012 (IPCC, 2014b). The primary way that most people experience climate change is through extreme weather events such as extremes in temperatures and precipitation and the effects of higher sea levels. Changes in extreme weather events have been observed since about 1950 (IPCC, 2014b). As an example, the heavy precipitation trend in the U.S. for is given in figure 1.1 (Melillo et al., 2014). The values in the figure 1.1 are measured by the once-in-five-year event index. The once-in-five-year event index counts the event when a two-day precipitation total is exceeded on average only once in a 5-year period. The results in figure 1.1 show that the observed frequency of such events has become much more common in recent decades.

14 2 Figure 1.1 Observed U.S. Trends in Heavy Precipitation Source: Melillo et al. (2014) Another example is the rise in global sea level. According to an IPCC report (IPCC, 2010), global sea level has risen about 0.2 meters (8 inches) since 1880 and is projected to rise another meters by One reason is that the oceans absorb over 90 percent of the increased atmospheric heat caused by anthropogenic GHG emissions. Increasing ocean temperature leads to water expansion and causes sea levels to rise. Another contribution to the rise in global sea level is melting glaciers and ice sheets caused by global warming. Global warming affects the timing of plant flowering, dates of breeding and migration patterns of insects and animals (IPCC, 2001). Changing precipitation trends and sea levels are altering hydrological systems and water resources in term of quantity and quality (IPCC, 2014b).

15 3 The first international attempt to limit GHG emissions was the Kyoto Protocol adopted in December The Kyoto Protocol extended the 1992 United Nations Framework on Climate Change (UNFCCC) to set binding emission reduction targets for Parties to the Convention (UN member countries). The goal of the Convention is to stabilize GHG concentrations in the atmosphere and thereby prevent dangerous human interference with the climate system. Under the Kyoto Protocol commitments, countries had to reduce GHG emissions on average by 5.2% over the period from 2008 to 2012 compared to 1990 levels. With the expiration of first commitment period under the Kyoto Protocol, an amendment to the Protocol was agreed at Doha on December Japan, New Zealand and Russia have not adopted new targets in the Doha amendment even though they had participated in the Kyoto Protocol. Canada withdrew from the Kyoto Protocol in 2012 and the US never ratified the Protocol. In the Doha amendment, 37 countries, including all European Union (EU) countries, Australia, Belarus, Croatia, Iceland, Kazakhstan, Norway, Switzerland and Ukraine, adopted legally binding targets. The reduction target is 18% in GHG emissions by 2020 compared to 1990 levels. In spite of these global efforts, tangible results in combating climate change have not been achieved. However, a milestone was reached through a new agreement at the UNFCC conferences in Paris in December The Paris Agreement is a proposed new agreement within the framework of the UNFCCC for handling GHG emissions mitigation, adaptation and finance starting in the year Through the Paris Agreement 195 nations reaffirm the seriousness of climate change. They agreed on actions and investment to achieve a low carbon, resilient and sustainable future. The Paris Agreement

16 4 is a major achievement in that all nations concur with a common cause of addressing climate change based on their historic, current and future responsibilities. A total of 195 countries submitted plans - their Intended Nationally Determined Contributions (INDCs) - to contribute to the objective of holding the increase in global average temperature to below 2 C compared to pre-industrial levels. Each country decides its own contribution in the context of national priorities, circumstances and capabilities. For example, Norway, the country that is the focus of empirical analysis in this study has set a target of at least a 40% reduction in its GHG emissions by 2030 compared to 1990 levels. Along with its INDC, Norway s long-term goal is to offset all GHG emissions by The Norwegian government plans that the country will be the first carbon-emissions-free country by 2050 (UNFCCC, 2015) The Contribution of Agriculture to Global GHG Emissions A recent assessment of agricultural GHG emissions shows that agriculture will need to play a pivotal role in reducing global GHG emissions. The IPCC has estimated that agricultural production causes approximately 14% of global GHG emissions, even though agriculture accounts for only 6% of world GDP (IPCC, 2007). Smith et al. (2008) estimate that agriculture accounts for 52% and 84% of global anthropogenic methane and nitrous oxide emissions, respectively. A recent study by the CGIAR (Consultative Group on International Agricultural Research) provides even more dramatic estimates. When the entire food system is taken into account, the study estimates that up to one-third of all human-made GHG emissions

17 5 stem from agriculture (Vermeulen et al., 2012). This research examines GHG emissions in all stages of the food system such as fertilizer manufacture, food storage and packaging and the release of N2O from agricultural soil by farming practices. According to the study, the global food system released 9,800 16,900 mega tons of CO2-equivalent in 2008, including indirect emissions from deforestation and land-use change. Estimates of GHG emissions by important sub-sectors of the food system are: agricultural production, up to 12,000 mega tons of CO2-equivalent per year; fertilizer manufacture up to 575 mega tons of CO2-equivalent per year; and refrigeration, 490 mega tons of CO2- equivalent per year Norwegian Agriculture and GHG emissions According to the country s Climate and Pollution Agency (2013), total GHG emissions in Norway in 2011 were 53.4 million tons of CO2-equivalent without taking into account emissions and removals from Land-Use, Land-Use Change and Forestry (LULUCF). From 1990 to 2011 total GHG emissions increased by 6 percent. CO2, CH4 and N2O emissions account for 84 percent, 8 percent and 6 percent of total GHG emissions, respectively. The remaining 2 percent of GHG emissions is explained by PFCs, HFCs and SF6 gases. When the impact of LULUCF is included, net GHG emissions had decreased by 25.8 million tons of CO2-equivalent in 2011 compared to The net GHG removal from LULUCF is 27.6 million tons of CO2-equivalent. Due to active forest management, the net GHG removal from LULUCF has increased by 26 percent since The major contributor to the total amount of sequestration by LULUCF is

18 6 remaining forest land. This accounted for 31.7 million tons of CO2 of sequestration in Land converted to forest land contributed 0.7 million tons of CO2-equivalent in 2011 (See Figure 1.2 for detail). Figure 1.2 Total Emissions of all GHG from the Different Source Categories Source: Climate and Pollution Agency (2013), Unit: Million tons CO2-equivalent The energy sector is a largest source of GHG emissions in Norway. It contributes 75 percent of total GHG emissions. 1 The largest single contributors are road traffic and offshore gas turbines. Other major sources are coastal navigation and energy commodities used for the production of raw materials. The second largest source of GHG 1 The energy sector includes the energy industries, transportation, energy use in manufacturing and construction, fugitive emissions from fuels, and energy combustion in other sectors.

19 7 emissions is industrial processes. This accounts for roughly 14 percent of total GHG emissions. Agriculture in Norway accounts for 8.4 percent of total GHG emissions. 2 Figure 1.3 GHG Emissions from Norwegian Agriculture (2011) Source: Blandford et al. (2015a) Note: The green columns are emissions under Chapter 4 Agriculture included in the Kyoto protocol. The black columns are emissions under Chapter 5 Land use, land use change and forestry. For agriculture these include emissions from cultivated soil. The column for fossil fuel combustion belongs to Chapter 1 Energy. According to Blandford et al. (2015a), GHG emissions from agriculture account for 13 percent of Norwegian GHG emissions when associated emissions from LULUCF and fossil fuel combustion are taken into account. Drained peatland accounts for 30 percent of the GHG emissions from agriculture. Drained peatland is the second largest 2 This figure does not include the impact of LULUCF.

20 8 source CO2 emissions from Norwegian agriculture with more than 1.5 million tons of CO2-equivalent per year (see Figure 1.3). The importance of reducing emissions from Norwegian agriculture is reflected in policy recommendations by the Norwegian Green Tax Commission. 3 This was set up to advise the Norwegian government on how to achieve its emission reduction commitments. The main policy recommendations relating to agriculture are follows. First, Norwegian agriculture should not be exempt from Norway s commitment to a 40 percent reduction in GHG emissions compared to In addition, agricultural emissions should be taxed at the same rate as that applied to other sectors beginning in The Commission proposes that carbon taxes should be imposed on the consumption of red meat and that government support for a red meat production should be reduced. Finally, the Commission proposes the imposition of carbon taxes on nitrogen in mineral fertilizers and on emissions resulting from major land-use changes, for example deforestation, draining, ditch digging and bog reduction, as well as peat removal (detailed discussion of the Green Tax Commission proposal is provided in Chapter 4). As results of the Norwegian Green Tax Commission s proposals and earlier economic analysis (Blandford et al., 2015a) it seems clear that efforts to remove peatlands from cultivation deserve serious consideration if GHG emissions from agriculture in Norway are to be reduced. 3 The Norwegian government established a Green Tax Commission in June 2015 to cope with the growing public concern about climate change. The primary objective of the Commissions was to assess whether and how tax reforms could be used to secure lower GHG emissions, improved resource utilization and ensure continued economic growth (Norwegian Green Tax Commission, 2015).

21 9 Peatlands are waterlogged wetland which contains large amounts of carbon, accumulated over many centuries, stored in the soil. Due to the absence of oxygen under flooded conditions, carbon decomposition is restricted. Although peatlands cover only about 3% of the world s land area, they are a major source of carbon storage. Peatlands store approximately one-third of the world s total organic carbon, which is roughly equal to the total amount of carbon stored in the atmosphere or in all terrestrial biomass (Joosten and Clarke, 2002). Peatlands can be drained and used for agricultural production. Once the soils are drained, which aerates the soil, stored carbon begins to decompose. This results in high fluxes of CO2 and N2O. CH4 emissions are usually decreased after draining, but this effect is far outweighed by increasing N2O and CO2 emissions (Kasimir Klemedtsson et al., 1997). GHG emissions on drained peatlands can partly be controlled through soil management, and different drainage and remediation practices (Kløve et al., 2010). Nonetheless the definitive mitigation practice where GHG emissions are still high is to take peatland out of agricultural production and to re-wet it (Freibauer et al., 2004). In Norway, peat soils cover 6.5% of the land area and 85, ,000 ha of peat lands are used in agriculture (Maljanen et al., 2010). The major threat for the release of carbon from peat soils results from drainage for agriculture and forestry uses. Grønlund et al. (2008a) estimate that the carbon loss from cultivated peat soils in Norway is kg C m 2 year -1 and that million tons of CO2 year -1 are released due to peat degradation. This is equivalent to roughly 3-4% of total anthropogenic GHG emissions in

22 10 Norway. Despite this, cultivated peat soils have received little attention as a source of CO2 emissions. The retirement of peatland used in agriculture has the potential to make a significant contribution to Norway s emission reduction commitment Environmental Policies for Agriculture and the Principal-Agent Model Several policy instruments can be used to mitigate GHG emissions in agriculture: (1) environmental taxes; (2) command-and-control regulation; (3) integrated conservation and development projects (ICDPs); and (4) voluntary agri-environmental schemes (Engel et al., 2008). In particular, voluntary agri-environmental policy has attracted attention as a mechanism to translate the provision of environmental services into a financial reward for service providers. The major analytical approach used to examine the design of voluntary agrienvironmental policies is the principal-agent model. The model is one kind of Stackelberg game, i.e., the principal moves first and then the agent responds to the principal s behavior. The principal offers a contract to the agent to perform a certain task on behalf of the principal. In the context of voluntary agri-environmental policy, the government entrusts the provision of environmental services to the farmer. Given the contract proposed by the government, the farmer decides to accept or reject. The farmer has private information about some parameter of his/her utility function, e.g., the costs of supplying environmental services. This parameter determines the farmer s type. And the farmer s private information affects the government s payoff, at least indirectly. The principal-agent model addresses the asymmetric information problem so that the principal

23 11 secures its payoff against the self-interested agent. The main obstacle to the design of voluntary agri-environmental policy is how to deal with information asymmetry. Two types of problems can arise from asymmetric information: adverse selection (hidden information) and moral hazard (hidden action). Adverse selection occurs during the implementation of agri-environmental policy. Since farmers have better information than the government about the cost of supplying environmental services (private information) they have an incentive to disguise these costs. Farmers can procure higher payments than the costs of supplying services and this makes policy implementation more expensive. Adverse selection has been a focus in the literature (e.g., Wu and Bacock, 1995, 1996; Moxey et al., 1999; Smith and Tomasi, 1999; Gren, 2004; Ferraro, 2008; Peterson and Boisvert, 2001, 2004; Sheriff, 2009; Quillérou and Fraser, 2010; Arguedas and van Soest, 2011). Moral hazard can arise after the policy comes into effect. The government cannot monitor a farmer s compliance perfectly and farmers have an incentive to avoid the costs of complying with their contractual obligations. Moral hazard can limit the effectiveness of agri-environmental policy and increase the cost of monitoring compliance by an environmental service purchaser (e.g., Ozanne et al., 2001; Fraser, 2002; Hart and Latacz-Lohmann, 2005; Ozanne and White, 2008; Yano and Blandford, 2009, 2011; Bartolini et al., 2012; Elofsson, 2013). The asymmetric information problem can detract from the effectiveness of agri-environmental policy and make it expensive to implement (Ferraro, 2008).

24 12 The typical way in which the government can ameliorate inefficiency in contract design in agri-environmental policy is through a targeting strategy. A targeting strategy is implemented based on finer information privately gathered by the government. Laffont and Tirole (1993, pp.213) defined finer information as follows: An information structure is finer than another if it corresponds to a finer partition of the set of states of nature. This implies that if the government can disaggregate farmers by a finer information structure, the government offers more customized menus of contracts to the farmers depending on their types. Accordingly, the government can design more effective agri-environmental policy. In the context of targeting strategy, finer information can not only relate to the costs of supplying environmental services but also to the benefits from procuring these services. For instance, there may be spatial heterogeneities among agents in the provision of environmental benefits, such as reduction in water pollution or carbon sequestration, given the same level of abatement costs. Farmers who live in more environmentally valuable areas can provide higher levels of environmental services than farmers who live in less environmentally valuable areas. With such heterogeneity in environmental benefits, society can be better off if the government offers different contracts to the farmers in different regions. This problem has been examined in a number of studies (e.g., Ribaudo, 1989; Ribaudo et al., 1994; Babcock et al., 1996, 1997; Feather et al., 1999; Wu et al., 2001; Kirwan et al., 2005; Wünscher et al., 2008). Heterogeneity in the costs of supplying the same level of environmental services has also been a focus in the literature (e.g., Canton et al., 2009; Uthes et al., 2010; Doody et al., 2012; Coisnon et al.,

25 ). In addition, it has been shown that a targeting strategy using finer information can reduce the moral hazard problem by disaggregating farmers depending on their willingness to comply with a contract to supply environmental services (e.g., Fraser, 2004, 2012; Ferraro, 2004; Lankoski et al., 2010) Research Objectives and Dissertation Outline Through the targeting strategy the government can ameliorate the contractual inefficiency caused by asymmetric information. However, the use of differentiated contracts based on the targeting strategy reveals the government s private information to farmers. Given differentiated contracts, the farmers learn the government s private information that affects the benefits resulting from agri-environmental policy. Therefore, the information asymmetry is still one-sided. This means that the government s private information (i.e., principal s type) becomes common knowledge for farmers. Farmers still have an informational advantage in terms of private knowledge of the costs of supplying the environmental services when contracting. This raises two important questions. Should the government reveal its private information in order to improve the payoffs from a program? What is the optimal agrienvironmental policy when the government also has private information that can affect the payoffs? An important issue in the design of agri-environmental policy may be to mitigate GHG emissions. In cases where the interaction between an agent s activities and environmental outcomes is complex, the government may have private information about the impact of changes in these activities on emissions. The implications of an agent s

26 14 activities (production practices) for greenhouse gas emissions may not be known to the agent, but through the work of government scientists these implications may be known to the government agency that will offer an environmental contract to farmers. With this point of view bilateral information asymmetry may apply in the design and implementation of agri-environmental policy. Myerson (1983) and Maskin and Tirole (1990, 1992) pioneered the development of the principal-agent model under bilateral asymmetric information. They called it the informed principal model. Myerson (1983) studied the general properties of mechanism design for an informed principal and paid attention to the non-emptiness of the core (i.e., the existences of a solution to the problem) using cooperative game theory. Myerson shows that there is always equilibrium in the informed principal game. This result is called the Inscrutability Principle, which notes that without loss of generality we can restrict attention to the pooling offer in which the principal, whatever her/his type, offers the same contract to the agents. The Pooling Offer will be intensively dealt with in the rest of this study so a detailed explanation is not provided at this point. Maskin and Tirole (1990, 1992) analyzed bilateral asymmetric information under non-cooperation. In particular, they studied two forms of the informed principal model: private values and common values. The model of the informed principal with private values assumes that the principal s private information is not an argument in the agent s payoff. This implies that the agent s utility does not depend on the principal s private information. In contrast, the model of the informed principal with common values

27 15 assumes that the principal s private information is an argument of the agent s objective function. In the common values model, the agent s utility is directly affected by the principal s private information. Detailed explanations of private and common values are provided in chapters 2 and 3. In this study, contact issues in the design of a peatland retirement program in Norway are examined using two cases. First, a benchmark case is used in which the government reveals its private information through differentiated contracts, e.g., the government s knowledge of the impact of the retirement of peatland with a particular peat depth on GHG emissions. This implies that the government s type is common knowledge so information asymmetry is one-sided. Only the farmer has private information about the costs of implementing peatland retirement. This represents the standard design of agri-environmental policy with the targeting strategy. In the second case, it is assumed that the government has private information about the potential GHG mitigation of retirement of peatland with a particular peat depth, but this private information does not directly affect the farmer s payoff from land retirement. The farmer still has private information about her/his type (e.g., income foregone by program participation) that can cause adverse selection problems. This is an informed principal model with the private values assumption proposed by Maskin and Tirole (1990). The dissertation is structured as follows. Chapter 2 conducts a literature review on the design of agri-environmental policy and its use with targeting strategy. Comparisons between standard policy and targeting strategy help in understanding the overall structure

28 16 of the dissertation structure. As a benchmark case, the optimal design of agrienvironmental policy with targeting strategy is examined in Chapter 2. Using the ratio of Lagrange multipliers, it is explained why the privately informed government can be better off than the government with a targeting strategy. In Chapter 3, the theoretical model for agri-environmental policy with bilateral information asymmetry is established based on the informed principal model proposed by Maskin and Tirole (1990). A comparison between the policy design with targeting and that with bilateral information asymmetry is provided. Then the optimal design of agri-environmental policy with bilateral information asymmetry is derived. The implications for a peatland retirement program in Norway are discussed. Chapter 4 describes various parameters used for the empirical analysis in this study. In particular, a Norwegian agricultural sector model (Jordmod), that is used to generate several key parameters for the design of the peatland retirement program, is discussed. Also GHG emission factors from drained/rewetted peatlands are examined based on a recent IPCC report (IPCC, 2014a). Using the parameters derived in Chapter 4, empirical analysis of the optimal contract for Norwegian peatland retirement program is conducted in Chapter 5. This study concludes with a summary and policy implications in Chapter 6.

29 17 Table 1.1 Dissertation Outline Chapter Topics 1. Contract Design using Principal-Agent Model Chapter 2 2. Agri-Environmental Policy with Targeting Strategy (Benchmark Case: One-Sided Information Asymmetry) 1. Agri-Environmental Policy with Pooling Offer Chapter 3 (Bilateral Information Asymmetry) 2. Its Application to Peatland Retirement Program in Norway Chapter 4 Parameters Used for Empirical Analysis Chapter 5 Empirical Analysis of Peatland Retirement Program in Norway Chapter 6 Summary and Policy Implications

30 18 Chapter 2. Design of Agri-Environmental Policy with One-Sided Information Asymmetry 2.1. Contract Design using the Principal-Agent Model The major analytical approach used to examine the design of agri-environmental policies is the principal-agent model. This model posits that the principal (government) offers a contract to agents (farmers or land owners) to provide some type(s) of environmental service and farmers decide whether to accept or refuse the contract. If the agents accept, they carry out the required actions under the contract, e.g., implement production practices that provide the desired agri-environmental services. The main obstacle to the effective implementation of agri-environmental policy in the principalagent model is information asymmetry. Although agents cannot affect the structure of the contract, they have an informational advantage in terms of private knowledge of the costs of supplying the environmental services that are sought by the principal. In other words, government may not have access to the information relevant to the supply of services that is possessed by the agents and can only monitor their activities. Asymmetric information can detract from the effectiveness of agri-environmental policy and make it expensive to implement (Ferraro, 2008). As an example, assume that there are only two types of agents. Each type is defined by the costs of supplying the environmental services that are sought by the principal. An efficient (inefficient) type agent refers to the farmer who has low (high) costs in supplying the services. In the context of the principal-agent literature, type

31 19 relates to the nature of the player s private information. So an efficient or low type agent refers to a farmer with low supply costs for environmental services. And an inefficient or high type agent is a farmer with high supply costs of environmental services. Consider first the case in which there is no information asymmetry between the principal and the agent. In this perfect information case both the principal and the agent can perfectly observe the other player s information. Without asymmetric information, the agent does not have an informational advantage in terms of private knowledge of the costs of supplying environmental services. The principal can observe the agent s type and offer a contract that is tied directly to the costs of supplying environmental services. For instance, the principal offers a contract with a low monetary inducement to the efficient (low) type agent with low supply costs and a contract with a high monetary inducement to the inefficient (high) type agent with high supply costs. The optimal supply of agrienvironmental services is obtained by equating the marginal environmental benefits from the participation of each agent type and each agent s marginal costs of supplying environmental services. This is the first-best outcome which refers to the optimal procurement of environmental services under perfect information. To implement this contract, one additional condition is required. This is called the Individual Rationality (IR) constraint. The contract offered by the principal must generate a level of utility to each agent which is at least as great as the level of utility of each agent in the absence of the contract. Without this constraint, the agent does not have an incentive to conclude a contract with the principal.

32 20 By contrast, with asymmetric information, the agent has an informational advantage about the costs of supplying environmental services. Since the agent can conceal the cost of supplying environmental services, the adverse selection problem arises. The principal cannot offer a specific contract depending upon the type of agent, i.e., the costs of supplying environmental services. In other words, the principal cannot distinguish one type of agent from the other type. Because of this hidden information, the principal can only offer a single contract which covers all implementing costs, low and high, regardless of the agent s type. This leads to over-compensation of efficient (low) type farmers who have low supply costs. Therefore, optimal allocation under asymmetric information is inefficient. It differs from the first-best outcome under perfect information. This standard contract is called a pooling contract which means that all types of agents receive the same contract regardless of their actual supply costs. The principal-agent model handles this inefficiency through the Revelation Principle. The revelation principle implies that there exists an incentive compatible contract in which all types of agents truthfully reveal their private information to the principal (Laffont and Martimort, 2009). Under the incentive compatible contract, all types of agent can achieve the best outcome for themselves by revealing their true type. This implies that the contract based upon the revelation principle becomes a direct revelation game for which truth-telling is a non-cooperative Bayes-Nash equilibrium (Spulber, 1988). It gives rise to an additional condition in the design of contracts for agrienvironmental services which is the Incentive Compatibility (IC) constraint. Accordingly, the principal can offer a menu of contracts which satisfy individual

33 21 rationality (IR) and incentive compatibility (IC) constraints for the various types of agent. Each type of agent then selects only the contract that is designed for her/his type. This type of contract is called the separating contract. Although the separating contract provides a better payoff for the principal (supply of desired environmental services at lower cost) compared to the pooling contract, it is still inefficient compared to the first-best outcome. We can ascertain this inefficiency using two concepts: information rent and downward distortion. Well-known results of the contract design are: (1) efficient type agents are paid more than their costs of supplying environmental services - this additional payment is called information rent ; and (2) the level of environmental services provided by inefficient agent is distorted downwards relative to the first-best outcome - this difference is called downward distortion. Contractual inefficiency caused by asymmetric information can detract from the effectiveness of agri-environmental policy and make it expensive to implement (Ferraro, 2008). Reducing information rent and downward distortion plays a central role in improving the effectiveness of agri-environmental policy. With a budget constraint, the principal (government) can obtain more environmental benefits while minimizing information rent and downward distortion. These two concepts are used extensively in this dissertation to compare optimal contract design with one-sided information asymmetry to that with bilateral information asymmetry.

34 Agri-Environmental Policy with a Targeting Strategy The focus is on the specific case of agricultural land retirement to help in understanding the structure of contract design used in this study. Agricultural land retirement programs have been an important part of agri-environmental policies in the United States and in Europe, e.g., the Conservation Reserve Program (CRP) in the US; land set-aside policies in the European Union. These programs can provide environmental services such as habitat conservation, reduction in water pollution, and carbon sequestration through the agricultural land taken out of agricultural production (Hansen et al., 2015). CASE 1 CASE 2 Separating Contract Targeting Strategy Figure 2.1 Comparison of a Separating Contract and a Targeting Strategy

35 23 Figure 2.1 illustrates two different contract designs for a land retirement program. The government (principal) is a buyer of environmental services on behalf of the public and farmers (agents) are private sellers of these services. Agents (farmers) have an informational advantage in terms of private knowledge of the costs of supplying the environmental services on which society places a value. It is assumed that there are two types of farmers, differentiated by the costs of taking land out of production. One type of agent has a high income foregone, which is called the high type agent. The other type of agent has a low income foregone and is called the low type agent. Since the government cannot directly observe the farmer s type (whether the farmer is low or high cost), the asymmetric information problem can detract from the effectiveness of the land retirement policy. In addition, it is assumed that there is a spatial heterogeneity in environmental benefits from the land retirement program. There are two types of area, one of which is more environmentally valuable than the other, and these are simply labeled as more and less. In a separating contract, case 1 in figure 2.1, the government does not have information on whether particular areas of land that can potentially be retired will provide more environmental benefits or less environmental benefits. The uninformed government can only design a contract based on the average environmental benefits of land retirement. This leads to overcompensation for farmers who have less valuable land in terms of environmental benefits and the insufficient provision of more valuable land to the retirement program. Additional inefficiency arises from asymmetric information about farmers costs of land retirement. To minimize this inefficiency, the government

36 24 incorporates the Revelation Principle into the contract design so that government distinguishes the high cost type of agent from the low cost type of agent. With the revelation principle, the government offers a menu of contracts that satisfies the individual rationality (IR) constraint and incentive compatibility (IC) constraint. And each type of agent chooses the contract which is consistent with her/his type. Case 1 is a traditional separating contract using the revelation principle. The targeting Strategy is shown by case 2 in figure 2.1. Unlike the separating contract, the government has information on whether land that can be retired provides more environmental benefits or less environmental benefits. Due to the private information about the environmental benefits, the government also has its own type that depends upon the environmental benefits, e.g., more or less. Recall that type implies the character of the player s private information that affects the player s payoffs in the context of principal-agent literature, rather than different entities or individuals per se. So the more type principal denotes the contract designed for the farmers that have more environmentally valuable land and the less type principal stands for the contract designed for farmers that have less environmentally valuable land. However, the agents still have an informational advantage with respect to private knowledge of the opportunity costs of supplying environmental services. The principal s private information is about environmental benefits, not about the agent s cost of implementing a contract to provide these. Accordingly, this private information possessed by the

37 principal cannot fully eliminate the inefficiency associated with asymmetric information, but it can reduce this inefficiency compare to the separating contract in Case Since the principal in Case 2 can offer differentiated contracts to different farmers who provide different environmental benefits, the principal can minimize the inefficiency resulting from the differences in resulting environmental benefits. In other words, given the same levels of supply costs, the government can seek to enroll more land from farmers who possesses environmentally more valuable land than from farmers with less valuable land. For instance, the retired land size from the more/high contract is greater than the less/high contract. Contracts for the low type farmers are similar to high type farmers, e.g., more/low is larger than less/low. This is the main difference between the separating contract and a targeting strategy. Since the principal in Case 1 can only measure the environmental benefits from land retirement on average, retired land sizes are identical in more valuable land and less valuable land. By contrast, when the principal in Case 2 offers the contracts to the farmers, the principal can differentiate more valuable land from less valuable land. This raises the question: should the principal reveal its private information in order to improve its payoffs? The answer to this question is the main research objective of this study. This question will be examined more detail in the next chapter. But first, the targeting strategy is examined as a benchmark case in this chapter so that this can serve as a basis for a comparison with optimal policy design with bilateral information asymmetry. No further attention is given to case 1 in this study (the traditional separating

38 26 contract). Moxey et al. (1999) provide more details on case 1. In addition, a land retirement program continues to be used as an example for contract design Two Common Misunderstandings about the Principal-Agent Model To avoid confusion, two common misunderstandings of the principal-agent model, in particular when the principal has private information, are discussed before explaining the theoretical model of agri-environmental policy with targeting strategy. Type of Principle The notion of type is used when contract design is translated into the Principal- Agent model in order to represent the existence of private information in the contractual relationship. Type and the parties to a contract are independent concepts. So a different type implies neither different entities nor a single party. Simply stated, there always exists a certain type when private information exists in the context of the principalagent model. In this study, different principals do not correspond to different entities but to differentiated contracts based on the government s private information about the benefits of agri-environmental services denoted, for example, by more or less valuable in case 2 in figure 2.1. The government is assumed to be uninformed in the traditional principal-agent model, as in the separating contract. Because of this, the government could simply be equated with a single principal as in a traditional separating contract in case 1 in figure 2.1. Due to the absence of private information, the government is assumed to be identical

39 27 with the uninformed single principal in a traditional separating contract. When the government has private information that affects its payoffs, however, the government has more than one type depending on its private information. For the government in case 2 in figure 2.1, the two different types of principal, more and less, do not imply two different governments, but two differentiated contracts based on the government s private information. Information Asymmetry and Private Information The information asymmetry structure is not identical to structure of type, i.e., the existence of private information. It could be misunderstood that the opposite case to bilateral asymmetric information is the case of a privately informed agent contracting with an uninformed principal, as depicted by the separating contracts in case 1 in figure 2.1. This is incorrect. It should be viewed as a one-sided asymmetric information case where a privately informed agent makes a contract with an informed principal but the principal s private information is common knowledge, i.e., the principal fully reveals its private information through the proposed contract. A typical example of the latter case is a targeting strategy. Government has private information that affects its payoffs, e.g., spatial heterogeneity of environmental benefits, but the government reveals its private information on these benefits to the agent through the differentiated contracts. Accordingly, bilateral information asymmetry refers to the case in which a privately informed principal has a contractual relation with a privately informed agent.

40 28 Therefore, in both cases 1 and 2 in figure 2.1, information asymmetry is still onesided. Even if the government has private information about differences in environmental benefits, e.g., higher provision of environmental services from certain types of land enrolled in a program than other types of land, the government reveals its private information through the proposed contract. The government s type becomes common knowledge for farmers. This implies that the targeting strategy is basically a kind of separating contract because different type of farmers, whether they have high implementing cost or low implementing cost under the program, receive different contracts Theoretical Model of Agri-Environmental Policy with Targeting Strategy In this section, the theoretical background for designing agri-environmental policy with a targeting strategy is explained. Policy design is structured in terms of a three-stage game involving the principal (government) and the agent (farmer). In stage 1, the principal offers contracts to buy agri-environmental services from agents embodied in the removal of land from agricultural production. The contract consists of the amount of land to be retired and the monetary transfer to compensate the agent for the loss of income from removing that land from production. In the second stage, the agent updates prior belief about the principal s private information (type) based on the contract offered by the principal and decides whether to accept or reject that contract. Recall that a more type principal denotes the contract designed for the farmers with land in more environmentally valuable areas and a less type principal stands for the contract designed for farmers with

41 29 land in less environmentally valuable areas. If the agent rejects the contract, the game is over. If the agent accepts the contract, in the third stage the principal and the agent implement the proposed action and the monetary transfer is made. Agents retire areas of land specified in the contract and the principal pays compensation for the reduction in income that this creates. It is assumed that both the principal (government) and agent (farmer) have two independent types, i = 1, 2 and j = 1, 2, respectively. Superscript i stands for principal type i and subscript j denotes agent type j. The agent s common prior belief in type i principal is denoted by q i such that q 1 + q 2 = 1. Also, pj is the proportion of the type j agent such that p1 + p2 = 1. The risk-neutral principal (government) has a linear payoff function: V b y t i i i i j j (2.1) where y i j is the size of retired land by agent type j after accepting the contract offered by principal type i, t i j is a monetary transfer from principal type i to agent type j in compensation for income foregone through program participation. b i is principal i s marginal environmental benefit obtained from land retirement. Recall that the different principals reflect different environmental benefits on land retirement, b i, in order to represent the principal s private information. They do not correspond to different decision making entities, unlike the agent types. Each b i may reflect spatial heterogeneity of environmental services resulting from land retirement.

42 30 To avoid confusion, it is worthwhile to mention that the marginal environmental benefits from land retirement program, b i, determine the type of principal not the type of agent. Even if farmers have more environmentally valuable land it is assumed that this does not affect agricultural productivity or income foregone by program participation. This is due to the private values assumption in the informed principal model, which is that the agent s utility does not depend on the principal s private information. This is explained it more detail in chapter 3. With this assumption, the principal and the agent have two independent types. The private values assumption is not a strong assumption for this study. The government s private information that affects b i is the depth of peatland. And the farmer s income foregone by participation program is defined by taking peatland out of production. This will be discussed in chapter 3. In this case, the government s private information is unlikely to affect the farmer s income foregone. The only factor affecting the cost of implementing this program is the agricultural productivity of the land operated by the farmer. More productive farmers have to be paid higher compensation to retire land since their profits from agricultural production are larger than less productive farmers. Therefore, the types of principal and agent are assumed to be independent with the private values assumption. Finally, it is assumed that land retirement (y) is an observable and verifiable environmental service thereby ruling out the moral hazard problem in the model. For a land retirement program, the compliance is verifiable with negligible monitoring costs through satellite imagery. In addition, Norwegian social norms are likely to enforce the farmer s compliance by reporting from egalitarian society neighbors if a farmer breaks the rules.

43 31 The risk-averse agent (farmer) has a von Neumann-Morgenstern utility function Uj which is continuous, increasing and concave. 4 The agent s utility function has following form: i i U U( t ( y )) j j j j (2.2) where t i j is a monetary transfer received from principal i associated with income foregone through program participation and ψ j (y) is a retirement cost of y land by agent j. ψ j (y) is continuous, increasing and convex in y for every agent j. It is assumed that ψ 2 (y) > ψ 1 (y) so that type 1 agent has a lower income foregone and it is therefore more efficient to retire her/his land than that of the type 2 agent. The agent who has higher productivity is considered as an inefficient agent. Since the agent that has land with higher productivity earns more income than the less productive agent, income foregone by the higher productivity farmer is greater than that of the low productivity farmer. Because of this, a higher level of compensation will have to be paid to the more productive agents. Hereafter, the term an efficient agent implies an agent with less agriculturally productive land who would have less income foregone through land retirement and an inefficient agent is used to denote an agent with agricultural land of higher productivity who would have a larger income loss from program participation. For the sorting 4 The results of the principal-agent model with bilateral information asymmetry differ through the degree of risk aversion of principal and agent. Detailed examples of different results coming from the utility forms of principal and agent are reviewed in Chapter 3. This study follows Cella (2005) where the principal is riskneutral and the agent is risk-averse. In addition, the assumption of a risk-neutral principal and risk-averse farmer is common in theoretical models of the design of agri-environmental policy, e.g., Ozanne et al. (2001), Fraser (2002), Ligon (2004).

44 condition, ψ 2 (y) > ψ 1 (y) holds for any y. This ensures that the principal can distinguish between types of agent by offering an incentive compatible contract Optimal Agri-Environmental Policy Design with Targeting Strategy The benchmark case used to examine the optimal contract for a land retirement program is the targeting strategy. Since the principal reveals private information through the differentiated contracts, the principal s type is common knowledge in the targeting strategy. The government offers differentiated contracts based on its private information on the implications of land retirement. For instance, using case 2 in figure 2.1, a more type principal proposes more/high and more/low contracts, and a less type principal offers less/high and less/low contracts. Agents know the cost of implementing land retirement before contracting. It is posited that the retirement costs and the proportion of each type of agent are known but that it is not possible to observe directly the type of agent. Thus the agent s type is still private information. This corresponds to the standard design of agri-environmental policy with targeting strategy. The optimal contract for the land retirement program for each type i principal is found by solving the following problem (T i ): 2 i i i max ( ) i i p j b y j t j { yj, tj} i 1 i i i IR1 : U ( t1 1( y1 )) 0 i i i i ( T ) IR2 : U ( t2 2( y2)) 0 i i i i i IC1 : U ( t1 1( y1 )) U ( t2 1( y2)) i i i i i IC2 : U ( t2 2( y2)) U ( t1 2( y1 ))

45 33 where superscript i indicates principal type and subscript j denotes the agent type. IR (Individual Rationality) constraints imply that agents can guarantee at least the same utility level they would obtain before participation in the land retirement program. The RHS of the IR constraints is the agent s reservation utility and is normalized to zero. If IR constraints are not satisfied, no agent will participate in the program. IC (Incentive Compatibility) constraints mean that each type of agent prefers his/her designated contract to all other options. Due to the revelation principle, a menu of contracts is offered for all types of agent; agents truthfully reveal their type and choose the contract designated only for their own type. The optimal contracts obtained by solving the problem above are separating equilibriums where different types of agents receive different contracts. This is the separating contract. A well-known result of the adverse selection problem is that only IR2 and IC1 are binding constraints at the equilibrium (for details, see Moxey et al., 1999). The principal s maximization problem can be rewritten as follows: 2 i i i max ( ) i i p j b y j t j such that { yj, tj} i 1 i i i i i ( T ) IR2 : U ( t2 2( y2)) 0 ( ) i i i i i i IC1 : U ( t1 1( y1 )) U ( t2 1( y2)) ( ) (2.3) where ρ i and γ i are the Lagrange multipliers associated with the IR and IC constraints, respectively. Solutions of the problem above are the optimal contract for the land

46 34 retirement program when the contract is based on the targeting strategy. Contracts corresponding to each principal type i {(y 1 i, t 1 i ), (y 2 i, t 2 i )} will be offered as follows: 5 { y, t } { b ( y ), ( y ) ( y ) ( y )} i i i i i i i (2.4) p 2 { y i 2, t i 2} b i 2 ( y i 2) ( 1 ( y i 2) 2 ( y i 2)), 2( y i 2) p1 (2.5) The optimal contract shows that efficient farmers retire their land in line with marginal environmental benefits equal to marginal retirement costs. But the monetary transfer for efficient farmers (less agriculturally productive) is more than the retirement cost. This means that efficient farmers (less agriculturally productive) receive an information rent. On the other hand, the retirement level for inefficient farmers (higher agricultural productivity) is distorted downward from the environmentally efficient level. And inefficient farmers (higher agricultural productivity) only receive a monetary transfer equal to income foregone by implementing land retirement. The utility level of inefficient farmers is unchanged by participating in the land retirement program. This is a common result of the discrete-type adverse selection problem: (i) the efficient type agent has an efficient allocation and a positive surplus, information rent; (ii) the inefficient type agent has an inefficient allocation and zero surplus (Salanié, 2005). The optimal allocations in equations (2.4) and (2.5) can answer the following question: is there any possibility in a targeting strategy for the principal to conceal its 5 See Appendix A. for details.

47 35 private information, partially or fully, in order to obtain higher payoffs than those in a separating contract? This is not possible because the optimal allocations in a targeting strategy have an identical mathematical structure to a traditional separating contract. If the principal is uninformed, the only thing the principal can offer is a separating contract, the optimal allocations in a separating contract can be derived by replacing each b i in equation (2.4) and (2.5) with the average value of b i. This implies that in a targeting strategy each principal s best action, which can provide the maximum payoff from the land retirement program, is to reveal fully its private information through the differentiated contracts. This is equivalent to the case in which the principal does not have private information in a separating contract. 6 This results from the private values assumption, i.e., the principal s private information does not directly affect the agent s utility. Due to the private values assumption, the agent does not take into account the principal s payoffs obtained from the proposed contract. The agent only cares about his/her income foregone resulting from the required land retirement. If the proposed monetary transfers are enough to compensate for the loss of income from removing land from production, the agent accepts the proposed contract and carries out the required action. 6 Because of this, Maskin and Tirole (1990) call this the full information case even if the agent s type is still private information.

48 Ratio of Lagrange Multipliers It is worth noting that different types of principal imply a different ratio of Lagrange multipliers for the optimal contract with the targeting strategy. This is the reason why there can be higher payoffs with privately informed principals compared to the targeting strategy. This provides the theoretical background for designing optimal policy with bilateral information asymmetry. A more detailed theoretical description is provided in the next chapter. But here an explanation is provided of how this can be used to design agri-environmental policy with bilateral information asymmetry. From equation (A.3) and (A.4) in Appendix A, the ratio of Lagrange multipliers associated with the IR constraint (ρ i ) and the IC constraint (γ i ) is: i U ( t ( y )) p2 i U ( t ( y )) U ( t ( y )) U ( t ( y )) ( t ( y )) ( t ( y )) ( t ( y )) i i i i ( t2 1( y2)) p2 p1 1 i i i i i i i i i i i i and i p U( t ( )) ( t ( )) 1 i 1 i 1 y1 i i 1 1 y1 i i i i i i i i i U ( t1 1( y1 )) U ( 1( y1 ) 2( y2) 1( y2) 1( y1 )) U ( 2( y2) 1( y2)) i i i i i p1 U ( t2 2( y2)) p1 U ( t2 2( y2)) p1 U (0) (2.6) Since the Lagrange multipliers are the shadow prices for satisfying the IR and IC constraints, there are different relative costs in satisfying these constraints for different principal types. For instance, if ρ 1 /γ 1 < ρ 2 /γ 2, the IC constraint is relatively more costly for the type 1 principal and the IR constraint is relatively more expensive for the type 2 principal. Principal 1 can obtain a surplus which comes from relaxing the IC constraint

49 37 and enforcing the IR constraint. Principal 2, on the other hand, can achieve higher payoffs by relaxing the IR constraint and tightening the IC constraint.

50 38 Chapter 3. Design of Agri-Environmental Policy with Bilateral Information Asymmetry 3.1. The Principal-Agent Model with Bilateral Information Asymmetry Previous studies of the design of agri-environmental policy using the principalagent model assume that information asymmetry is one-sided. Only agents have private information that gives them an informational advantage when contracting. This hypothesis is too restrictive. In cases where the interaction between an agent s activities and environmental outcomes is complex, the government may have private information about the impact of changes in these activities. For example, the implications of an agent s land use activities for greenhouse gas emissions may not be known to the agent, but through the work of government scientists may be known to the government agency that will offer a land retirement contract to farmers. In the case study examined in this dissertation farmers are unlikely to have information about the potential effectiveness of land retirement for GHG mitigation, but the government is likely to have such information. In cases where there is a complex interaction between agents practices and environmental outcomes, bilateral information asymmetry may apply in the design and implementation of agri-environmental policy Literature Review on the Informed Principal Model Myerson (1983) and Maskin and Tirole (1990, 1992) pioneered the development of the principal-agent model under bilateral asymmetric information. They called it the

51 39 informed principal model. Myerson (1983) studied the general properties of mechanism design for an informed principal and paid attention to the non-emptiness of the core (i.e., the existences of a solution to the problem) using cooperative game theory. He showed that there is always an equilibrium in the informed principal game. This result is called the Inscrutability Principle. It implies that, without loss of generality, attention can be restricted to the pooling offer in which the same menu of contracts is offered to the agent regardless of the principal s private information. Maskin and Tirole (1990) explain that when the government has private information that affects its payoffs, the optimal strategy that guarantees at least the same payoffs that the government would obtain with no private information is achieved by offering the same menu of contracts regardless of private information. This is called the pooling offer and is done in order to obtain additional surplus. It implies that the contract offered to satisfy the type 1 principal also contains the allocation that is designed for the type 2 principal and vice versa. Since the same menu of contracts is proposed for both types of principal, agents cannot distinguish one type of principal from another. So when agents decide whether to accept or reject the contract in the second stage, the agent s belief about the principal s type remains the same as her/his prior belief. Because of this, agents only expect to satisfy their IR and IC constraints on average. The fact that principals are merely supposed to satisfy these constraints on average makes the informed principal relax the more costly constraint and tighten the less costly constraint. In so doing, the privately informed principal can be better off than the principal with targeting policy by using the pooling offer.

52 40 Maskin and Tirole (1990, 1992) analyzed bilateral asymmetric information under non-cooperation. In particular, they studied two forms of the informed principal model: private values and common values. The model of the informed principal with private values assumes that the principal s private information is not an argument in the agent s payoff. This implies that the agent s utility does not depend on the principal s private information. This is the reason for assuming that the principal s and the agent s type are independent in Chapter 2. In contrast, the model of the informed principal with common values assumes that the principal s private information is an argument in the agent s objective function. In the common values model, the agent s utility is directly affected by the principal s private information. Cella (2005) analyzed an informed principal model with the private values assumption when the principal is risk-neutral and the agent is risk-averse. The result showed that the risk-neutral informed principal can be better off than in the benchmark case where the principal s type is common knowledge. Fleckinger (2007) proved that even though both the principal and the agent have quasi-linear utility functions, the principal in the private value model can obtain additional economic surplus when the agent faces countervailing incentives. Cella (2008) examined the informed principal with private values model with both a risk-neutral principal and agent. She imposed the additional assumption that the principal s type is correlated with the agent s type. Results showed that even if both the principal and the agent are risk-neutral, the informed principal can achieve higher payoffs than the principal whose type is common knowledge since the contract can be used to signal the principal s type to the agent.

53 41 By adding the private information by the government to the principal-agent relationship, the informed principal model becomes a sequential game with incomplete information. The equilibrium concept of a sequential game with incomplete information will be a Perfect Bayesian Equilibrium (PBE). The perfect Bayesian equilibrium is characterized by the following properties: (i) the principal s offer is optimal given the agent s strategy and belief; (ii) the agent s strategy is optimal given her/his belief and the principal s strategy; (iii) both parties update their belief using Bayes rule, when applicable. These properties imply that a contract proposed by a privately informed government is a signaling device. The contract itself reveals the government s private information. Given this offer, the farmer updates her/his beliefs about the government s private information. The government updates her/his beliefs about the farmer s type (such as high income foregone or low income foregone) after observing the farmer s decision whether to accept or reject the contract. Using the PBE concept, Maskin and Tirole (1990) showed that informed principal with private values assumption can guarantee at least the same payoff he/she would obtain in the benchmark case where the principal s type is common knowledge. This benchmark case is the optimal policy design with targeting strategy in Chapter 2. They also showed that when both principal and agent have a quasi-linear utility function the principal neither gains nor loses if her/his private information is revealed before contracting takes place. In other words, the equilibrium payoff in the informed principal model with private values is equal to the payoff in the targeting strategy.

54 3.2. Design of Agri-Environmental Policy with Bilateral Information Asymmetry 42 To examine the implications of bilateral information asymmetry the starting point is an example of a targeting strategy and a pooling offer. The basic circumstances are the same as the example in Chapter 2. The only difference is whether the private information of the government is revealed through the proposed contracts. Again, the government is a buyer of an environmental service and farmers are sellers of this service. The service is generated by land retirement. Two types of agents have private information about the costs of implementing land retirement. Depending on the levels of supplying costs, the agent is called high type agent or low type agent, respectively. And there is a spatial heterogeneity in environmental benefits from the land retirement program. This is labelled more or less. The principal can only observe this information. The government only has private information which can differentiate more valuable land from less valuable land. Based on this private information, the principal is named as more type principal or less type principal. Since both the principal and the agent have their own private information, bilateral information asymmetry is a factor when designing the land retirement program

55 43 CASE 2 CASE 3 Targeting Strategy Pooling Offer Figure 3.1 Comparison of a Targeting Strategy and a Pooling Offer

56 44 As seen in Chapter 2, when a targeting strategy is employed, the government s private information about environmental benefits is revealed. The government can be better off than the government in a separating contract in case 1 in figure 2.1. However, Myerson (1983) and Maskin and Tirole (1990) provide a seminal approach on the use of the principal s private information in order to achieve a higher payoff than under a targeting strategy. Based on the Inscrutability Principle proposed by Myerson (1983), without loss of generality, attention can be restricted to the pooling offer in which the principal, regardless of type, offers the same menu of contracts to agents, regardless of type. This is described in case 3 in figure 3.1. If the government offers the same menu of contracts to all farmers, regardless of government s private information, the farmers cannot tell one type of the principal from the other type. This makes the government conceal its private information and bilateral information asymmetry applies when contracting. The farmer has private information about the costs of supplying environmental services, as in the separating contract and targeting strategy. In addition, the government has private information about the environmental benefits from the participation of agents in a contract. The Pooling offer reflects the selective disclosure of information by the government so that the farmer s IR and IC constraints can be relaxed. Because both types of principal offer the same menu of contracts, the farmers only expect their IR and IC constraints to hold on average when they accept or reject the contracts. If the farmers only expect to satisfy their IR and IC constraints on average, both types of principal can be better off by relaxing their more expensive constraint. As mentioned in Chapter 2, different types of principal have a

57 45 different ratio of Lagrange multipliers. This means that different types of principal have different relative costs in satisfying the agent s IR and IC constraints. Accordingly, each principal relaxes the more expensive constraint and enforces the less expensive constraint in order to obtain higher payoffs than those with a targeting strategy. Under these conditions, the pooling offer is a separating contract since the proposed contracts still satisfy the farmer s IR and IC constraints on average. Even though the government offers the same menus of contracts to the famers, each type of the farmer can determine which contract is designed for him/her. In case 3 in figure 3.1, for instance, high type farmers are aware of the contracts drawn with the bold line are appropriate for them. On the other hand, low type farmers perceive the contract with a dashed line as the contracts appropriate for them. This is because the pooling offer is made using the revelation principle so the farmer s IR and IC constraints are satisfied on average between the two contracts. Therefore, we can say that it is the pooling offer and a separating contract Optimal Agri-Environmental Policy with Bilateral Information Asymmetry In this section, the optimal contract is derived for the case in which the principal s type is private information, i.e., is not known by agents. Before deriving optimal contracts for the privately informed principal, it is shown that the equilibrium allocation in the pooling offer is Pareto superior to the allocation in the targeting strategy. The notation used follows that of Maskin and Tirole (1990). Defining i v as the principal i s i i i payoff in the targeting strategy, i.e., v p V ( ) and i { y i, t i } denotes the j j j j j j

58 46 equilibrium allocation for each type of principal and agent in the targeting strategy and i i and are Lagrange multipliers from the solution to the targeting strategy case. Consider the following perturbed version (one with slack variables) of the targeting strategy case (T i ): 2 i i i max ( ) i i p j b y j t j such that { yj, tj} i 1 i i i i i ( T* ) IR2 : U( t2 2( y2)) r 0 i i i i i i IC1 : U ( t1 1( y1 )) c U ( t2 1( y2)) where r i and c i are slack variables associated with the IR and IC constraints, respectively. These are zero in the targeting strategy case because IR2 and IC1 are binding. Since different types of principal offer different contracts in the targeting strategy, each type i of principal must satisfy IR2 and IC1. In contrast, if different types of principal offer the same menu of contracts, thereby concealing their private information, agents cannot observe the principal s type through the proposed contract and would expect to satisfy their IR and IC constraints on average. From the perspective of the type 2 agent, in the pooling offer, the IR constraint is satisfied in problem (T i ): IR : q ( U ( t ( y )) r ) q ( U ( t ( y )) r ) (3.1) And using same logic, the type 1 agent expects the IC constraint to hold as follows: IC : q ( U( t1 1( y1 )) U( t2 1( y2)) c ) q ( U( t1 1( y1 )) U ( t2 1( y2 )) c ) 0 (3.2)

59 47 From the IR and IC constraints, q r q r 0 and q c q c 0 which implies that slack variables need only be non-positive on average, and not for each type of principal. i Let v * be the maximized payoffs of principal i in the perturbed problem. By definition, the shadow prices i and i i, * v approximately equal: v 1 v 1 + ρ 1 γ 1 + γ 1 c 1 (3.3) v 2 v 2 + ρ 2 γ 2 + γ 2 c 2 (3.4) These can be rewritten as: v 1 v 1 ρ 1 γ 1 + γ 1 c 1 (3.5) v 2 v 2 ρ 2 γ 2 + γ 2 c 2 (3.6) If the LHS of equations (3.5) and (3.6) are positive, the informed principal can be better off than the principal in the benchmark case by using the pooling offer. To have positive values of the LHS of equations (3.5) and (3.6), two conditions must hold simultaneously: r c 0 (3.7) r c 0 (3.8) Rewriting the inequality condition (3.8) using q r q r 0 and q c q c 0, yields: q 1 2 q 1 q r c r c 0 q q q (3.9)

60 48 This is equivalent to: r c (3.10) Combining the inequality conditions (3.7) and (3.10) yields: c r (3.11) For inequality (3.11) to hold, the ratio of the Lagrange multipliers must be different across different principal types. It was demonstrated in Chapter 2 that different principal types have a different ratio of Lagrange multipliers when the principal is riskneutral and the agent is risk-averse. Therefore, it is shown that the informed principal can achieve a higher payoff than the principal in the targeting strategy case by permitting slack variables on the constraints. For a more detailed proof, see Proposition 1 in Maskin and Tirole (1990) Pooling Offer In the previous section, it was shown that the allocation i { y i, t i } in the j j j targeting strategy is dominated by the allocation i * { y i *, t i * }, which is the solution to the perturbed version of the benchmark case. This implies that the same menu of contracts will be proposed, regardless of principal type, which is called the pooling offer. Using this result, an optimal land retirement program can be defined under the assumption that the principal is risk-neutral and the agent is risk-averse.

61 49 Unlike the targeting strategy when principal s type is common knowledge, private information about environmental benefits from land retirement can only be observed by the government. This means there is now a sequential game of incomplete information in contracting. The equilibrium allocation resulting from the pooling offer should be a Perfect Bayesian Equilibrium (PBE). PBE requires that the equilibrium allocation is individually rational for the agent, and incentive compatible for the principal and the agent. To ensure this, two more constraints are specified such as ICP 1 and ICP 2, which are incentive compatibility constraints for two types of principal. The IR and IC constraints are modified such that they hold on average, as under the pooling offer. When the principal has private information about the environmental benefits from the land retirement program, each type i of principal will propose the same menu of contracts that is the equilibrium allocation of (P i ): 2 i i i max ( ) i i p j b y j t j such that { yj, tj} i 1 2 i i i IR2 : q U ( t2 2( y2)) 0 i 1 2 i i i i i i ( P ) IC1 : q [ U ( t1 1( y1 )) U ( t2 1( y2))] 0 i ICP : p j ( b y j t j ) p j ( b y j t j ) i 1 i ICP : p j ( b y j t j ) p j ( b yj tj) i 1 i 1 (3.12)

62 50 Since both principals propose the same menu of contracts, the agents cannot update their prior belief about the principal s type and accept the contract. This implies that there are five conditions for equilibrium allocation in the pooling offer: i b ( y ) i 1 1 (3.13) p U ( t ( y )) b y y y ( ( )) i i i i 1 i i ( 2) 2 ( 2) 1 ( 2) i i p2 U t1 1 y1 (3.14) 2 i 1 q U ( t ( y )) 0 i i i (3.15) 2 i 1 i i i i i q U ( t1 1( y1 )) U ( t2 1( y2)) 0 (3.16) U( t ( y )) U( t ( y )) p U ( t ( y )) U ( t ( y )) ( ( )) ( ( )) ( ( )) i i i i i i i i i i i i i i U t2 2 y2 p2 U t2 2 y2 U t2 2 y2 (3.17) A less constrained problem (P i ) can be used to prove that the optimal contract is a Perfect Bayesian Equilibrium (PBE). The less constrained problem is: 2 i i i max ( ) i i p j b y j t j such that { yj, tj} i 1 2 i i i i i ( P* ) IR2 : q U ( t2 2( y2)) 0 ( ) i 1 2 i i i i i i IC1 : q [ U ( t1 1( y1 )) U ( t2 1( y2))] 0 ( ) i 1

63 where i i and are Lagrange multipliers in the less constrained problem (P i ) associated with the IR and IC constraints, respectively. The only difference between the pooling offer (P i ) and the less constrained problem (P i ) is the absence of the incentive compatibility constraint for the two principal types. This implies that the equilibrium allocation from the less constrained problem naturally satisfies the IR and IC constraints for the agent. The logic of the proof is that even the principal in the less constrained problem does not have an incentive to offer the other principal type s allocation, the optimal contract in the pooling offer (P i ) is incentive compatible for both types of principal. By showing this, it can be proved that the equilibrium allocation from the less constrained problem (P i ) is indeed an equilibrium allocation in the pooling offer (P i ) and that the optimal contract by solving pooling offer (P i ) is a Perfect Bayesian Equilibrium (PBE). This proof follows from Cella (2005). 51 First, it is obvious that the optimal contract from the less constrained problem (P i ) also satisfies four conditions of the optimal contract in (P i ), which are equations (3.13)- (3.16), since these conditions come from IR and IC constraints for the agent that (P i ) and (P i ) have in common. Thus the optimal contract from the less constrained problem (P i ) is also incentive compatible for both types of principal. The ratio of Lagrange multipliers in the less constrained problem (P i ) can be obtained from its first-order conditions. The ratio of Lagrange multipliers is: p U ( t ( y )) U ( t ( y )) ( ( )) ( ( )) i i i i i i i i i i p2 U t2 2 y2 U t2 2 y2 (3.18)

64 The ratio of Lagrange multipliers above is equivalent to the RHS of equation (3.17). It is: 52 U( t ( y )) U( t ( y )) i i i i i i i i U( t2 2( y2)) (3.19) Note that agents expect their constraints to hold on average in the pooling offer. The numerator of the LHS in equation (3.19) is equal to the slack variable for the IC constraint, r i, and the denominator of the LHS in equation (3.19) is the same as the slack variable for the IR constraint, c i. Equations (3.20) and (3.21) can be obtained: r c (3.20) r c (3.21) And using the slack conditions that can be rewritten as follows: q r q r 0 and q c q c 0, equation (3.21) 1 q ( r c ) 0 2 q (3.22) Combining equation (3.20) and (3.22) yields: (3.23) Equation (3.23) implies that both types of principal have same relative cost of fulfilling the agent s IR and IC constraints at the optimal allocation in the less constrained

65 53 problem. Since the principals cannot obtain additional surplus from relaxing these constraints, both principals have no incentive to deviate from their own allocation. Furthermore, the optimal allocation from the less constrained problem (P i ) is Pareto optimal since is the slope of the value function at the equilibrium allocation. Each principal prefers its optimal allocation to the one for the other type. However, there is still the possibility that a principal offers an allocation which is neither an optimal allocation for its type nor for the other type. By using the FGP (Farrell (1985) and Grossman and Perry (1986)) refinement, Maskin and Tirole (1990) showed that no principal has an incentive to deviate from the solution to the pooling offer (P i ) thereby ruling out off-theequilibrium paths. Therefore, the optimal allocation from the pooling offer (P i ) is a Perfect Bayesian Equilibrium (PBE) and either principal type will offer the same menu of contracts Application to Peatland Retirement Program in Norway The previous section has developed the theoretical model of agri-environmental policy with bilateral information asymmetry. It is proved that the privately informed principal can be better off than the principal with a targeting strategy. Using this theoretical framework, the informed principal model is applied to the design of a peatland retirement program in Norway. In the first section an overview is presented of managing peatlands in Norway. The theoretical model for the peatlands retirement program is then defined. In addition, the principal s payoffs and agent s utility functions used for the empirical analysis are discussed.

66 Managing Peatlands in Norway Peatland is drained to be used for the agricultural production. Depending on the peat type, the drainage methods used are different. Bog-peat refers to peatland in which the peat layer is accumulated by atmospheric precipitation. In contrast, the accumulation of fen-peat is determined by the height by a high water table. Due to the different reasons for peat accumulation, in general, bog-peat is drained through the use of channels that prevent the accumulation of surface water, whereas fen-peat is drained by installing drainage tubes in the ground. Most of the peatland in Norway is bog-peat (Byrne et al., 2004) and is usually drained through the installation of channels. Figure 3.2 below shows the drainage procedure for peatlands in Norway. The drainage method is important when analyzing the management of peatlands. It not only affects drainage (and restoration) costs but also affects land-use decisions. When fen-peat is drained through installed drainage tubes, the lifespans of the tubes are affected by the type of farming, the distance between the top soil level and the drainage tubes and the depreciation rate of fen-peat. In general, a high level of drainage is required to cultivate vegetables, a medium level is needed for grain production and grass production is available with a low level of drainage. Depending on the level of drainage, the distance between the top soil level and the drainage tubes differs. The distance between the top soil level and the drainage tubes in high level of the drainage is shorter than medium and low level of drainage. And in medium level of drainage, the distance is shorter than in low level drainage. This leads to different depreciation rates for fen-peat. The shorter distance between the top soil level and the drainage tubes the more peat

67 55 subsides. This affects the lifespan of drainage tubes. 7 Because of this process, farmers decision making for fen-peat development varies. Under this process, Goetz and Zilberman (1995) examine the farmer s land-use decision making which focuses on the fen-peat development in Switzerland by using optimal control technique. Figure 3.2 Drainage of Peatlands in Norway Source: Grønlund and Weldon (2013) Once peatland is drained for agricultural production, which aerates the soil, stored carbon in the soil begins to decompose. This results in high fluxes of CO2 and N2O. CH4 emissions are usually decreased after draining, but this effect is far outweighed by increasing N2O and CO2 emissions (Kasimir Klemedtsson et al., 1997). GHG emissions on drained peatlands can partly be controlled with soil management, and different drainage and remediation practices (Kløve et al., 2010). Nonetheless the definitive mitigation practice is that peatlands are taken out of agricultural production and then rewetted where GHG emissions are still high (Freibauer et al., 2004). 7 Goetz and Zilberman (1995) explained that the maximum lifespans of installed drainage tubes are roughly 30 years.

68 56 In Norway, peat soils cover 6.5% of the land area and 85, ,000 ha of peat lands are used in agriculture (Maljanen et al., 2010). The major threat for the release of carbon from peat soils results from drainage for agriculture and forestry uses. Grønlund et al. (2008a) estimate that the carbon loss from cultivated peat soils in Norway is kg C m 2 year -1 and that million tons of CO2 year -1 are released due to peat degradation. This is equivalent to roughly 3-4% of total anthropogenic GHG emissions in Norway. According to Blandford et al. (2015a), GHG emissions from agriculture contribute 13 percent of Norwegian GHG emissions when GHG emissions from drained peatland are taken into account, and drained peatland accounts for 30 percent of GHG emissions from agriculture. CO2 emissions from drained peatland are the second largest source from Norwegian agriculture, accounting for more than 1.5 million tons of CO2- equivalent per year (see Figure 1.3). Along with this, ruminant animals (dairy cows, cattle and sheep) are a major source of GHG emissions from Norwegian agriculture. For instance, methane (CH4) emissions from enteric fermentation are a large source of GHG emissions from agriculture - almost 2.0 million tons of CO2-equivalent per year (see Figure 1.3). Methane (CH4) emissions from enteric fermentation and CO2 emissions from drained peatland are by far largest GHG emissions sources of Norwegian agriculture (see Figure 1.3). Therefore, reducing CH4 emissions from enteric fermentation and CO2 emissions from drained peatland will play a central role in the mitigation of GHG emissions from Norwegian agriculture.

69 57 In Norway, drained peatland is mainly used to produce grass for livestock production. Among livestock farming types (such as dairy cattle, beef, and sheep) dairy farming is the major agricultural activity in Norway. Dairy farms account for 50 percent of total number of livestock farms (Statistics Norway, 2016). This means that peatland retirement in dairy farms is an effective way of helping to achieve Norway s GHG reduction target of 40%. Because of this, in this study the focus is on the retirement of peatland used for dairy farming. Through peatland retirement in dairy farming, GHG emissions from drained peatland are reduced in two ways: (1) Mitigation of GHG emissions by peatland restoration; (2) Mitigation of GHG emissions from reduced production of ruminants. Mitigation of GHG emissions by peatland retirement According to an IPCC report (2014a), drained peatland in the boreal zone used to produce grass emits 5.7 tons CO2-C ha -1 yr -1, tons N 2O-N ha -1 yr -1 and ton CH4 ha -1 yr -1. Using molecular weight and GWP (global warming potential) from the Fifth Assessment Report (IPCC, 2013) these emissions can be converted into tons of CO2-equivalent per hectare per year. Total GHG emissions from drained peatland used for grass production are 25.4 tons CO2-equivalent per hectare per year. After peatland is rewetted, GHG fluxes are changed. Rewetted grass peatland absorbs tons of CO2-C per hectare per year from the atmosphere and emits tons of CH 4-C per hectare per year in the air. The emissions of N2O from rewetted peat soil are negligible. Total GHG emissions from rewetted peatland for grass production are -1.3 tons CO2-equivalent per hectare per year. Negative emissions mean that rewetted peatland acts as a carbon sink.

70 58 Therefore, annual net reductions (emissions avoided plus carbon sink after rewetting) of GHG emissions from rewetted grass peatlands are 26.7 tons of CO2-equivalent per hectare. In addition, peatland retirement induces GHG mitigation from reduced agricultural production from ruminants. Milk cows directly emit GHGs through enteric fermentation and manure. For instance, CH4 and N20 are released directly from manure and the use of manure as fertilizer is assumed to be the source of N20 emissions. Estimated GHG emissions (CO2-Equivalent) per 1 kg of milk production are 0.66kg (Blandford et al., 2014). Based on simulations by using Norwegian agricultural sector model, tons CO2-equivalent are reduced from reduced dairy production per hectare per year. Detail information is provided in Chapter 4. In this study, the government is assumed to have private information which affects the payoff from peatland retirement. The nature of the private information determines the type of the principal. The private information which government alone relates to the depth of peatlands. Depending on the depth of peatlands, two factors that affect government benefits are determined: (1) the amount of carbon stored in the soil 8 ; (2) the economic lifespan of peatlands. It is straight forward to think that since thicker peatlands can sequester more carbon in the soil than thinner ones, thicker peatlands are environmentally more valuable than thinner peatlands. 8 The depth of peat is assumed to be the only factor that determines the carbon stored in the peat soil. All other factors affecting carbon stored in the peat soils are excluded such as temperature, moisture, bulky density, and mineral contents (Grønlund et al., 2008).

71 59 In addition, the depth of peatland determines its economic lifespan in agricultural production. The thicker the peatland, the longer it will be used for agricultural production. Once peatland is drained, the peat layer starts to degrade. Farmers use peatland only while the peat layer remains. When the peat is completely depleted, the land loses its economic value in agriculture and is abandoned. This lifespan of peatland is closely related to the second and third components of government benefits above. The mitigation of GHG emissions from reduced dairy production on peatland results in budget savings from production-related support. The annual-based valuation of mitigation of GHG emissions from reduced dairy production depends on what farmers produce by using peatland, not on its depth. Since thicker peatland can be used in agriculture for a longer time period, however, higher total GHG mitigation is to be expected from retiring thicker peatland than thinner peatland. In this sense, thicker peatland is more environmentally valuable than thinner peatland in terms of GHG mitigation from reduced agricultural production. By the same logic, more government expenditure can be saved by retiring thicker peatland than thinner peatland due to its longer economic lifespan. It is assumed that thicker peatlands have a depth of 200 cm (78.7 inches) and thinner ones have a depth of 100 cm (39.3 inches). This is because the average peat depth in Norway is 150 cm (59 inches). 9 These factors define two types of principal in the land retirement model. Type 1 principal represents thick peatland (200 cm depth) and type 2 9 This information is obtained by personal communication with Dr. Arne Grønlund from Bioforsk (The Norwegian Institute for Agricultural and Environmental Research).

72 60 principal stands for thin peatland (100 cm). It is assumed that the average depth of peatland in Norway is common knowledge, i.e., farmers have information about the average depth of peatland in Norway but they do not know whether the peatland that they farm is thick or thin. Recall that much of the peatland in Norway is used for permanent pasture rather than being tilled, so peatland depth is not easily observable. Savings of government expenditure on output subsidies through peatland retirement When peatland is taken out of the production, there is an additional benefit resulting from government budget savings. Due to the national policy aim of food security, Norwegian agriculture is maintained by huge government subsidies. Without subsidies, farming in Norway is barely sustainable because of small farm size, high production costs and a limited home market. According to the Blandford et al. (2015a), the Norwegian government can reduce its budgetary support for agriculture by 7% by removing peatland from agricultural production. When grassed peatland in dairy farming is retired, the government does not pay subsidies on agricultural output since that output is no longer generated by the peatland. By using simulation results from a Norwegian agricultural sector model, it is estimated that the government can save 13,600 Norwegian Krone per hectare per year (roughly $1,660 at recent exchange rates) when grass peatland used in dairy farm is retired. Detailed information is provided in Chapter 4.

73 Theoretical Framework for a Peatland Retirement Program in Norway Principal s Payoffs Function On the basis of the above discussion, it is assumed that the benefits obtained from a peatland retirement program, as perceived by the government, consist of three components: (1) Mitigation of GHG emissions from peatland by peatland restoration; (2) Mitigation of GHG emissions from reduced agricultural production on peatland; (3) Savings of government expenditure on output subsidies through reduced agricultural production on peatland. We can probably expect additional environmental benefits through peatland restoration, such as biodiversity conservation, the provision of habitat, flood control and water quality control. In this study, however, attention is restricted to the value of GHG mitigation and saving of output-related expenditure when the principal s payoffs are defined. So we can conclude that the method used adopts a conservative valuation of the benefits from a peatland retirement program. It is assumed that the government s private information that affects program benefits is the depth of peatland. Depending on the depth of peatland, the government is defined in terms of two types of principal and the payoff functions for the two types of principals reflect the benefits from a peatland retirement program. Given the importance of dairy farming in Norway, the empirical application will focus on dairy farming, so the two types of agents will relate to two types of dairy farms (farms in the North and farms in the South of Norway). This is discussed in Chapters 4 and 5.

74 62 Recall that risk-neutral principal s payoff function from Chapter 2 is: V b y t i i i i j j where y i j is hectares of peatland retired by agent type j after accepting the contract offered by principal type i, t i j is a monetary transfer from principal type i to agent type j to compensate for income foregone by taking peatland out of production. The definition of the two principal types reflects the fact that different areas of peatland (defined by peatland depth) imply different marginal valuations in a retirement program. This is reflected in b i. Type 1 principal (thick peat) has larger b i than type 2 principal (thin peat), i.e., b 1 > b 2. The difference between b 1 and b 2 results in different amounts of retired peatland, y i, and different levels of monetary transfer, t i, from the government to farmers. Hence y 1 2 j > y j and t 1 2 j > t j for each type farmer j = 1 and 2. Since the type 1 principal can obtain higher benefits than type 2 principal given the amount of peatland retired, a higher amount of peatland will be required for retirement by agents to satisfy the type 1 principal than the amount required by the type 2 principal. Agent s Utility Function The risk-averse dairy farmer has a von Neumann-Morgenstern utility function Uj which is continuous, increasing and concave. The dairy farmer s utility function has following form: i i U U( t ( y )) j j j j

75 63 where t i j is a monetary transfer received from principal i associated with income foregone by retiring peatland from dairy production. ψ j (y) is an income foregone function for y hectares of peatland retired by agent j. ψ j (y) is continuous, increasing and convex in y for every agent j. It is assumed that ψ 2 (y) > ψ 1 (y) so that the type 1 dairy farmer has a lower income foregone since the type 1 dairy farmer has lower productivity than the type 2 dairy farmer. Hereafter type 1 dairy farmer is called the efficient type agent and the type 2 dairy farmer is called the inefficient type agent. The specific income foregone is a farmer s private information and gives rise to asymmetric information in agrienvironmental policy. The government cannot directly observe the farmer s type. The government can only realize the farmer s type after the farmer chooses the contract from a menu of contracts. This is because the contract offered by government must be an incentive compatible contract.

76 64 Chapter 4. Parameters 4.1. Parameters Used for the Empirical Analysis In this chapter, parameters used for the empirical analysis of the design of a peatland retirement program for Norway are explained. The focus is on how to monetize those parameters from available data sources. Recall that the government s payoffs consist of three components: (1) GHG mitigation through peatland restoration; (2) GHG mitigation from reduced dairy production; (3) Savings of government expenditure on output subsidies through by peatland retirement. And the government s payoffs vary across the principal s type, i.e., private information about benefits of peatland retirement program. Specifically, the government expects more benefits from the retirement of thick peat than thin peat. To calculate GHG mitigation through peatland restoration, GHG emissions factors are used from the IPCC 2013 Supplement (IPCC, 2014a). Since the 2013 Supplement was prepared to fill gaps in the existing literature on emissions, the 2013 Supplement suggests the wetlands GHG emissions factors by land-use category, nutrient-level and climate zone. Using the GHG emissions factors from the 2013 supplement provide more accurate data for the design of land retirement contracts for Norway. The second and third components relate directly to the current agricultural situation in Norway. Parameters to quantify those components are derived from

77 65 simulation results from a Norwegian agricultural sector model (Jordmod) 10 developed and maintained by economists at the University of Bergen in collaboration with economists at the Agricultural Economics Institute in Oslo (NILF) and the University of Bonn (Germany). The model is described in detail below. This sectoral model is also used in estimating agents cost functions. Representative models for dairy farms that are incorporated into Jordmod are used to approximate income foregone functions for two types of dairy farms through program participation. In addition, an estimate of the carbon price needed to monetize the net present value of the program (net present value of benefits and costs) is derived from the model. A summary of required parameters and data sources is given in Table 4.1. This chapter is organized as follows. First, the Norwegian agricultural sector model (Jordmod) is discussed and estimates of GHG mitigation by reducing dairy production, government savings and farmers income forgone are derived. Then the remaining data requirements and issues, e.g., carbon price, and the time-horizon for the valuation of benefits and costs are explained. 10 Jordmod was developed in the mid-1980s and has since been used in a number of analyzes relating to changes in the policy framework for Norwegian agriculture. This project has particularly focused on farmers' adaptation to changed policy instruments and on the food industry which is in a special relationship to primary agriculture. The project has been funded through research funds from the Agricultural Agreement and the Research Council in the framework of the program JORDMAT (Mittenzwei and Gaasland, 2008). For this study, Professor Erling Vårdal and Ivar Gaasland at the University of Bergen have collaborated with the author and Professor David Blandford to estimate GHG mitigation from reduced dairy production, savings in government expenditure and farmer s income foregone by program participation by using Jordmod.

78 66 Table 4.1 Summary of Parameters and Data Sources Player Parameters Data Sources Both Carbon Price Time Horizon Norwegian Green Tax Commission (2015) and Blandford et al. (2014) Depth of Peatland Principal (Government) GHG Mitigation by Peatland Restoration GHG Mitigation by Reduced Dairy Production 2013 Supplement (IPCC, 2014a) Simulation results from Jordmod Saving Government Expenditure Simulation results from Jordmod Agent (Farmer) Income Foregone Simulation results from Jordmod Agricultural Sector Model for Norway (Jordmod) The Norwegian agricultural sector model, Jordmod, is a price-endogenous partial equilibrium model of the type proposed by McCarl and Spreen (1980) and McCarl and Schneider (2001). This model is constructed on the basis of representative farm models to represent Norwegian agriculture (Brunstad et al., 1999 and 2005). Jordmod has been widely utilized to address agricultural policy questions. Brunstad et al. (1999, 2005) used the model to explain the provision of public goods in Norwegian agriculture. Blandford et al. (2010) employed the model to analyze the effects of trade liberalization. For climate change issues, Blandford et al. (2014, 2015a) explored the trade-off between food

79 67 production and GHG mitigation in Norwegian agriculture. In addition, Blandford et al. (2015b) examined the interconnection between trade liberalization and reductions in GHG emissions. Jordmod incorporates various policy instruments in order to analyze food policy issues and the linkage between the food and agricultural industries and the rest of the economy. Trade barriers and subsidies are included to examine domestic and foreign market conditions. Tax related variables are also included, e.g., value added taxes, excise taxes, import levies, payroll taxes and wage taxes. The treatment of agriculture is detailed but the rest of the economy is represented in an aggregated form. In the model, Norwegian agriculture is represented by 11 farm types, such as extensive beef, sheep, combined milk and beef, grain farms, etc. The representative model farms are distributed over 32 production regions. Yields and soil types vary across each production region. There are 25 primary input factors included such as land, capital, labor, seeds, pesticides, etc. and a total of 34 products, comprised of 22 final products and 12 intermediate products. The outputs of the model farms are used as inputs by processing plants and their final products are then offered on the market. Domestic demand for final products is split into five separate demand regions and each region has its own demand function. Domestic supply is the sum of domestic production and imports. Domestic production occurs on the representative farm models with fixed input and output coefficients. Input/output coefficients are constructed based upon extensive farm

80 68 surveys collected by a government agency. 11 Domestic and imported commodities are assumed to be perfect substitutes. Heterogeneity in different production regions allows for regional variation in climatic and topographic conditions and differences in productivity. A more detailed description is provided in Appendix B Carbon Price Two methods are mainly used for estimating the unit value of carbon dioxide emissions: the Social Cost of Carbon (SCC) and the Shadow Price of Carbon (SPC). The social cost of carbon (SCC) is defined as the marginal cost of the damage caused by carbon dioxide emissions to the atmosphere. The SCC pays attention to the economic damage caused by climate change. By contrast, the shadow price of carbon (SPC) is the marginal cost of reducing emissions associated with a certain reduction goal. The SPC focuses on the opportunity cost of achieving the targeted reduction goal. In this study, two estimates of the shadow price of carbon are used as the basis for the figure adopted for the empirical application of the land retirement model. The Norwegian SPC has been estimated by Blandford et al. (2014) and the Norwegian Green Tax Commission (2015). Blandford et al. (2014) estimate the SPC in Norwegian agriculture by using the sectoral model, Jordmod. Blandford et al. (2014) adopted the GHG mitigation scenario proposed by Norway to the UN climate change conference in Copenhagen in November That scenario involved a reduction in 11 Data used for the Jordmod stem from the Economic Accounts for Agriculture and the Farm Accounts (both administered by NILF, Norsk Institutt for Landbruksøkonomisk Forskning), the Subsidy Database maintained by the Norwegian Agricultural Authority and Agricultural statistics prepared by Statistics Norway (Mittenzwei and Gaasland, 2008).

81 69 GHG emissions of 30% compared to the emissions levels in The Norwegian Green Tax Commission (2015) evaluated the SPC associated with Norway s commitment to achieve at least a 40% reduction of emissions by 2030 compared to 1990 levels as part of the UN climate change agreement reached in Paris in December Blandford et al. (2014) estimated that the shadow price of carbon to achieve a 30% reduction in agricultural GHG emissions is 300 Norwegian Krone (NOK) per metric ton of CO2 (roughly $36 at recent exchange rates). With that carbon tax, agricultural production would be decreased by 23%. This is less than the targeted emissions reduction. Ruminant meat production (beef and sheep meat), which is a major source of emissions, suffers a larger reduction in output than white meat (pork and poultry) and milk production which involve less emissions per unit of output. Despite the imposition of the carbon tax of 300 NOK per metric ton of CO2, on emissions from agricultural production, economic welfare (the sum of producers and consumer s surplus and taxpayer expenditure) increases by 19% compared to the base period level. This is mainly due to savings of government expenditure on agricultural support. Norwegian agriculture is supported by large subsidies. As a result, the government can reduce its expenditure by 26% after imposing carbon tax on the GHG emissions due to reductions in subsidies with reductions in production. This implies that the social marginal abatement cost for Norwegian agriculture is negative, i.e., Norway can achieve a 30% reduction in agricultural GHG emissions with a positive gain in social welfare. Norway has a long experience with environmental taxation. Taxes have been introduced for the purpose of reducing negative environmental externalities such as air

82 70 and water pollution, waste treatment, and health damaging chemical products (Norwegian Government Official Website, 2007). With the growing public concern about climate change, the Norwegian government established a Green Tax Commission in June The primary objective of the Commissions was to assess whether and how tax reforms could be used to secure lower GHG emissions, improved resource utilization and ensure continued economic growth (Norwegian Green Tax Commission, 2015). To achieve these policy goals, the Commission has proposed an expansion of the use of carbon taxation in Norway. Implicit carbon taxes are imposed on some sectors through the imposition of emissions caps in certain industries and the trading of emissions permits through the European Union s (EU) Emissions Trading Scheme. Under the Commission s proposals a carbon tax would be imposed on emissions from sectors that are not included in the ETS. The proposed tax rate is 420 NOK (roughly $50) per metric ton of CO2-Equivalent. It is noteworthy that the Commission s proposed rate of tax of 420 NOK per metric ton of CO2 is very close to the computed carbon tax rate in Blandford et al. (2014) if the carbon tax rate is linearly increased along with the reduction target, i.e., the reduction target for agricultural GHG emissions is increased from 30% to 40%. For this study, therefore, it is assumed that the appropriate carbon price is 420 NOK per metric ton of CO2-equivalent in line with the Commission s proposed tax rate.

83 Time Horizon Environmental benefits from agri-environmental policy can be expected to continue over a long period, so the time horizon is central to evaluate the net present value of the benefits and costs associated with policy. For instance, an infinite time horizon is usually assumed for evaluating the environmental benefits of soil carbon sequestration. In this study, however, the calculation of net present value is limited to the economic lifespan of the use of peatland in agriculture rather than an infinite time horizon. As explained in section 3.4, as peatland is used in agriculture it decays. When the peat is fully depleted the land becomes unusable for agriculture and is abandoned. Hence the relevant time horizon for computing the benefit of land retirement is the depletion horizon for the peatland. The retirement of land avoids direct and indirect emissions of GHGs resulting from the use of the land in agriculture and determines the total benefits and costs associated with that removal. With this assumption, the environmental benefit of GHG mitigation from peatland retirement and restoration can be defined as the value of net savings (emissions avoided plus carbon sink after rewetting peatlands) over the economic lifespan of the peatland rather than the value of soil carbon sequestration over an infinite time horizon. This assumption is consistent with the theoretical framework of this study. Unlike other principal-agent literature, it is assumed that the principal (government) has private information, in this case, about the depth of peatland. By contrast, the agent (the farmer) only knows the average depth of peatland. This asymmetric information about peat depth leads to a discrepancy in time horizon between the principal and the agent. The principal

84 72 can evaluate the benefit of peatland retirement depending on the specific peat depth, so the time horizon for the net present value varies with the economic lifespan of peatland. On the other hand, the agent evaluates income foregone based on the average economic lifespan of peat soil because the agent does not have a specific information about the depth of peatland. The economic lifespan of peatland is determined by two factors: depth and depreciation (subsidence) rate. The average annual depreciation rate in Norway is estimated to be 1.63cm per year. 12 If linear depreciation is assumed, as in IPCC (2014a), the economic lifespan of 200 cm depth peatland, representing thick peat in this study, is 120 years (see Figure 4.1 below). But the actual depreciation curve is most certainly geometric. Grønlund et al. (2008a) show the geometric depreciation of peatland at Smøla Island 13 in Norway. This study assumes a linear depreciation curve as in IPCC 2013 Supplement (2014a). With linear depreciation, the economic lifespan of thin peat (100 cm depth) is 60 years. The average economic lifespan of peatland, which is known to the agent, is assumed to be 90 years (see Figure 4.2. below). Therefore, the time horizon for calculating the net present value of type 1 principal (thick peat) is 120 years and of type 2 principal (thin peat) is 60 years. The agent s net present value of income foregone is monetized over a 90-year time horizon. 12 This information was obtained from personal communication with Dr. Arne Grønlund, Bioforsk, Norway. 13 Smøla island is located in Smøla Municiplity in Møre og Romsdal county. This is to the west of the northern city of Trondheim, the third largest city in Norway and located at over 63 degrees North. Almost all cultivated soils in Smøla island are peat soil (Kløve et al., 2010).

85 73 Figure 4.1 Depreciation (Subsidence) of Drained Peatlands Figure 4.2 Economic Lifespan of Peatlands

86 Monetizing the Principal s Benefits Environmental Benefits of GHG mitigation GHG mitigation through peatland restoration The 2013 Supplement (IPCC, 2014a) provides GHG emissions factors for peatland by land-use category, climate zone and nutrient-level 14. The emission factors are obtained by using two methods: (1) the Soil CO2 Flux measurement and (2) the Depreciation (Subsidence) method. Norway s GHG emissions from peatland are measured by the second method, the monitoring depreciation rate (Grønlund et al., 2008a). Peatland is drained to be used for agricultural production. Once the soil is drained, which aerates it, stored carbon in the soil decomposes. This leads to the depreciation (subsidence) of peatlands. The depreciation rate is related to bulk density and soil carbon content. Dairy farmers in Norway use peatland for the production of grass. New emissions factors for drained grass peatlands in the boreal zone 15 (the zone in which Norway is located) from the 2013 Supplement (IPCC, 2014a) are 25.4 CO2-equivalent per hectare per year. Using the IPCC linear depreciation method, annual GHG emissions from drained grass peatlands are constant until the peatland is completely depleted (See Figure 4.3). 14 Nutrient-level that is used in the 2013 Supplement is defined in the GPG-LULUCF and the 2006 IPCC Guidelines (IPCC, 2006 and 2014a). In general, bog-peat (ombrogenic peat) is assumed to be nutrient-poor and fen-peat (minerogenic peat) is characterized as nutrient-rich. 15 According to Wieder et al. (2006) and Walter and Breckle (2002), the definition of the boreal zone is the true boreal zone commences at the point where the climate becomes too unfavorable for the hardwood deciduous species, i.e, when summers become too short and winters too long.

87 75 Figure 4.3 Carbon Emissions from Drained Peatland According to the 2013 Supplement, drained grassland in the boreal climate zone emits 5.7 tons of CO2-C per hectare per year, tons of N2O-N per hectare per year and tons of CH4 per hectare per year. Using molecular weight and GWP (global warming potential) from the Fifth Assessment Report (IPCC, 2013) these emissions can be converted into tons of CO2-equivalent per hectare per year. The GWP is assumed to be: CH4 = 34, N2O = 298 and CO2 = 1. A summary of the results is given in Table 4.2.

88 76 Table 4.2 GHG Emissions from Drained Peatland Land-Use Category CO 2 N 2O CH 4 Total Emissions Drained Grassland (ton CO 2-C/ha) (ton N 2O-N/ha) (ton CH 4/ha) (ton CO 2-Equiv/ha) GHG Emissions 5.7 a b c 25.4 Source: IPCC (2014a); a: Table 2.1; b: 2.3; c: 2.5 If a farmer joins a peatland retirement program, the farmer blocks the drainage channel for the land. The water level rises over time until the land finally is returned to pristine peatland. We have little knowledge about annual GHG emissions from the soil during this period. Estimates are site specific and the average effects in terms of GHG emissions during the rewetting processes to the return to pristine peatland are unknown. Due to the limited information, the simplifying assumption is made that GHG emissions from rewetted peatlands instantly drop to equilibrium levels after the drainage channels are blocked (See Figure 4.4). The equilibrium levels of GHG emissions from rewetted peatland are -1.3 ton CO2-equivalent per hectare per year. Negative emissions mean that there are carbon sinks in rewetted peatlands. Rewetted peatlands absorb tons of CO2-C per hectare per year from the atmosphere and emit tons of CH 4-C per hectare per year in the air. The 2013 Supplement assumes that emissions of N2O from rewetted peat soil are negligible (See Table 4.3).

89 77 Figure 4.4 Carbon Emissions from Rewetted Peatland Table 4.3 GHG Emissions from Rewetted Peatland Rewetted Grassland CO 2 N 2O CH 4 Total Emissions (tons CO 2-C/ha) (tons N 2O- N/ha) (tons CH 4-C/ha) (tons CO 2-Equiv/ha) GHG Emissions a b -1.3 Source: IPCC (2014a); a: Table 3.1; b: Table. 3.3 Annual net reductions (emissions avoided plus carbon sink after rewetting) of GHG emissions from rewetted grass peatland are 26.7 tons of CO2-equivalent per hectare. The emissions factors are not affected by the depth of the peat. Regardless of the depth, annual net savings are constant. To calculate the net present value for the net savings of GHG emissions, the social and private discount rates are assumed to be 2.0 % and 2.5%,

90 78 respectively. 16 Using this rate, the net present value of GHG mitigation from peatland restoration over the two economic lifespans can be calculated (Table 4.4). The type 1 principal (thick peatland) can expect a 61.3 thousandus dollar value (per hectare) of GHG mitigation over 120 years and the type 2 principal (thin peatland) can expect a 47.3 thousand US dollar value (per hectare) of GHG mitigation over 90 years. Table 4.4 GHG mitigation by Peatland Restoration Net Mitigation NPV Type of Principal (ton CO 2-Equiv/ha) (1000 NOK/ha) NPV (1000 US$/ha) Thick Peat (Type 1 Principal / 120 years) Thin Peat (Type 2 Principal / 60 years) GHG Mitigation from Reduced Agricultural Production Retiring peatland results in GHG mitigation from reduced dairy production. Milk cows emit GHGs through enteric fermentation and from their manure. A reduction in farm size through land retirement leads to reduced dairy production and the mitigation of GHG emissions. For instance, CH4 and N20 released directly from manure and the use of manure as fertilizer. Estimated GHG emissions (CO2-equivalent) per 1 kg of milk production are 0.66kg (Blandford et al., 2014). Using Jordmod estimates of net 16 From personal communication with Professor Erling Vårdal at the University of Bergen, recommended social and private discount rate in Norway were obtained. Professor Erling Vårdal suggests that 2% is proper level of the social discount rate in Norway and the private discount rate is 0.5% higher than the social discount rate, i.e., 2.5%. There is supporting literature in Norwegian for these estimates (see also, Sandmo and Dreze, 1971; Goulder and Williams III, 2012).

91 79 mitigation of GHG emissions from reduced dairy production can be calculated (Table 4.5). Government can mitigate tons CO2-equivalent by retiring one hectare of peatland. GHG mitigation from reduced dairy production has slightly different values by the type of agent. Type 1 farmer (efficient type farmer) and type 2 farmer (inefficient type farmer) reduce emissions by 7.03 and 9.36 tons CO2-equivalent by retiring one hectare of peatland, respectively. In this study, the valuation of GHG mitigation is focused on the government s private information about the depth of peatland rather than by differences in GHG mitigation by the type of agent. Thus the average value of GHG mitigation from reduced agricultural production is used. Net present values of GHG mitigation are between 10.6 and 12.4 thousand US dollars per hectare over the economic lifespan of peatland. Table 4.5 GHG Mitigation from Reduced Agricultural Production Net Mitigation NPV NPV Type of Principal (ton CO 2-Equiv/ha) (1000 NOK/ha) (1000 $/ha) Thick Peat (Type 1 Principal / 120 years) Thin Peat (Type 2 Principal / 60 years) Reductions in Government Support Expenditure Peatland retirement leads not only to GHG mitigation but also results in lower government expenditure on agricultural support on an annual basis. Norwegian agriculture is sustained with large subsidies. According to Blandford et al. (2014), government expenditure on agricultural support is reduced by 26% by imposing a carbon

92 80 tax of 300 NOK per metric ton of CO2 on the GHG emissions from agricultural production. When peatland is taken out of dairy production, government support is reduced. The reduction differs by the type of agent. The annual reduction in government support expenditure from type 1 farmer (efficient type farmer) and type 2 farmer (inefficient type farmer) are and (1000 NOK/ha), respectively. As for GHG mitigation from reduced agricultural production, average values are used in calculating the net present value of savings in annual government support expenditure. The reduction in government support expenditure is assumed to be same regardless of type of principal. The reason is that since the dairy farmers use a peatland for grass production, there is no productivity difference (amount of milk produced) by depth of peatland. Using simulation results from Jordmod it is estimated that the government can save thousand US dollars per hectare over 120 years as a result of the elimination of dairy production from thick peatlands and thousand US dollars per hectare over 60 years from thin peatlands (Table 4.6). The difference is due to the difference in economic lifespan by depth of peatland. Table 4.6 Saving in Government Expenditure Annual reduction in NPV Type of Principal Expenditure (1000 NOK/ha) (1000 NOK/ha) Thick Peat (Type 1 Principal / 120 years) Thin Peat (Type 2 Principal / 60 years) NPV (1000 US$/ha)

93 81 Total Benefits Obtained from Peatland Retirement The government can expect two environmental benefits from the retirement of peatland: net savings from elimination of GHG emissions from peatland decay and GHG mitigation as a result of reduced dairy production, plus one economic benefit through reduced expenditure on agricultural support. The environmental benefits account for roughly 52% of the total benefit per hectare with the economic saving contributing the remaining 48%. The net present value of both benefits is calculated over the economic lifespans of peat soil (Table 4.7). Thick peatland retirement (type 1 principal) yields thousand US dollars of benefits per hectare and thin peatland retirement (type 2 principal) yields thousand US dollars of benefits per hectare. Since the assumption of risk neutrality with linear payoff functions is used, marginal benefits per hectare are constant over all ranges of retired peatland (See equation 2.1). Type of Principal Thick Peat (Type 1 Principal) Thin Peat (Type 2 Principal) Table 4.7 Total Benefits of Peatland Retirement per Hectare Environmental Benefits Peat Restoration Reducing Ag. Production Economic Benefits Saving Expenditure Total Benefits (1000 US$/ha)

94 Monetizing Agents Costs Income Foregone through Program Participation The design of the peatland retirement program is based on a farmer s voluntary participation (chapter 1). Unlike command-and-control policy, voluntary agrienvironmental schemes cannot compel the farmer to provide agri-environmental services that are in accord with public purposes. Without the active involvement of the farmer, the Norwegian government cannot achieve its targeted mitigation of GHG emissions. Although many factors will contribute to the success of a land retirement policy, appropriate compensation plays a central role in farmers participation in the program. A proper payment system is important for designing an incentive compatible land retirement contract. Given the asymmetric information structure, the fact that the farmer can conceal her/his private information may lead to inefficiency, such as information rent and downward distortion in the peatlands retirement program. Accordingly, the income foregone that results from retiring peatland must be measured by the farmer s type. To estimate income foregone functions according to productivity, the representative farm models in Jordmod (Norwegian agricultural sector model) are used. Type 1 farmer represents a lower income foregone farmer (efficient agent) that operates a dairy farm in Bodø in Nordland county, which is in the north of Norway. The type 2 farmer with higher income foregone (inefficient agent) corresponds to the representative farm model in Stavanger on Rogaland county where is the intensive dairy farm area in the south of Norway. The difference between high and low income foregone for the two

95 83 types of dairy farms is mainly attributable to differences in climate. Stavanger has more favorable weather conditions for dairy farming than Bodø (particularly, milder temperatures). It is assumed that the farmers that live in the same region (north or south) will have same income foregone function so that heterogeneity of income foregone within each region is ruled out. For each type of farmer, income foregone is generated using representative farm models in Jordmod. To generate income foregone by program participation, the size of the representative farm is varied for the two types of dairy farms (north and south) under three hypothetical scenarios. As a base solution, model farm income is derived without intervention (no reduction in the area of land farmed). This gives reference income levels of each type of dairy farmer. In a second scenario, the model farm size is reduced to75% of the base solution. The income differences between the base solution and the second scenario are assumed to reflect the retirement of 25% of the amount of land farmed (peatland) compared to current farm size. Finally, in scenario 3 farm size is reduced by 50% compared to the base solution. As before, it is assumed that the 50% reduction in farmed area is due to the retirement of peatland. The simulated results are given in Table 4.8. When type 1 farmer removes 8.8 hectares of peatland from dairy production (25% of the farmed area) the income loss is 20.8 thousand US dollars. And when 17.5 hectares of peatland are retired (50% of the farmed area), the type 1 farmer loses 42.4 thousand US dollars. The corresponding figures for a type 2 farmer are 5.8 hectares and 13.9 thousand US dollars; and 11.6 hectares and 28.6 thousand US dollars, respectively.

96 Table 4.8 Simulated Representative Farm Incomes by Farm Size Type 1 Farmer Type 2 Farmer Farm size (percent of current) 100% 75% 50% 100% 75% 50% Total land use (ha) Tillable Peatland (ha) Income from Peatland (1000 NOK) Income from Peatland (1000 US$) Based on the simulation results in Table 4.8, the net present value of income foregone from peatland retirement is calculated. Recall that the assumption is made that a farmer only knows the average depth of peat soil and that a 90-year economic lifespan is expected regardless of actual depth. For the type 1 farmer, the net present value of income foregone by retiring 8.7 hectares of peatland is 748 thousand US dollars. When the type 1 farmer retires 50% of the land farmed, 1,523 thousand US dollars of income foregone is expected over 90 years. When a type 2 farmer takes 25% of the farmland out of production, income decreases by 501 thousand US dollars over 90 years, and when 11.6 hectares of land are retired 1,028 thousand US dollars of income foregone is expected (Table 4.9).

97 Table 4.9 Net Present Value of Income Foregone Type 1 Farmer Type 2 Farmer Farm size (percent of current) 100% 75% 50% 100% 75% 50% Retired Land (ha) Income Foregone (1000 NOK) Income Foregone per Retired Land (1000 NOK/ha) Income Foregone per Retired Land (1000 US$/ha) NPV for Income Foregone 0 6, , , ,566.8 (1000 NOK, 90 years) NPV for Income Foregone (1000 US$, 90 years) , , Approximation of Income Foregone Functions The theoretical model assumes several properties for the farmer s income foregone function (Chapters 2 and 3). First, ψ j (y) is continuous, increasing and convex in y for every farmer j where y is the size of retired peatland by type j farmer. In addition, for the sorting condition, ψ 2 (y) > ψ 1 (y) holds for any y. This ensures that the principal can distinguish between types of agent by offering an incentive compatible contract. It is assumed that ψ 2 (y) > ψ 1 (y) so that type 1 dairy farmer has less income foregone per unit of land retired. This implies that the land retired by the type 1 dairy farmer has lower productivity than the land of the type 2 farmer. Finally, each type of farmer loses nothing from non-participation in the program, so ψ j (0) = 0 for every farmer j. If a farmer

98 86 chooses not to retire peatland from agricultural production, he/she can maintain her/his original income level. To satisfy these properties, it is assumed that the farmer s income foregone function has a quadratic form (Spulber, 1988). The quadratic income foregone function is: ψ(y) = ay 2 + by + c (4.1) where a, b, and c are the quadratic coefficient, the linear coefficient and the constant term, respectively. Also c is equal to zero because of ψ j (0) = 0. With the quadratic form assumption, the two unknown coefficients, a and b, are determined using the income foregone data in Table 4.9. The quadratic income foregone functions are approximated based on the data in Table 4.9. Approximate income foregone functions are: ψ 1 (y) = 0.17 y y ψ 2 (y) = 0.54 y y where y is hectares of retired peatland and ψ j (y) is the net income foregone by reducing farm size (1000 US$). In addition to income foregone, the land restoration costs (costs of blocking the drainage) are included. These are 120 US$/ha. 17 Since this cost is incurred when the farmer initially blocks the drainage channels to rewet the peatland, it is taken into account only once. Therefore, the approximate total income foregone functions are: 17 This information is obtained from personal communication with Dr. Arne Grønlund, Bioforsk, Norway.

99 ψ 1 (y) = 0.17 y y (4.2) 87 ψ 2 (y) = 0.54 y y (4.3) Sorting Condition The sorting condition is a key element of contract design. Because of the revelation principle, the proposed contract is incentive compatible only when the sorting condition holds. In reality, it is an empirical question whether the sorting condition actually holds (Peterson and Boisvert, 2001). Before empirically analyzing the optimal peatland retirement program in Chapter 5, therefore, it is necessary to check whether the approximate income foregone functions satisfy the sorting condition. Recall that the sorting condition is ψ 2 (y) > ψ 1 (y) and holds for any y. Figure 4.5 below shows the result of the sorting condition. The horizontal axis is y, hectares of retired peatland, and the vertical axis is the first derivative of the income foregone function, ψ j (y). Given the quadratic income foregone function, the first derivative of the income foregone function is constant. Figure 4.5 shows that the sorting condition does not hold when y is below This implies that if y is less than 1.74 ha in the proposed contract, the type 2 farmer may act like a type 1 farmer. In other words, the type 2 farmer chooses the contract that is designed for the type 1 farmer when y is less than 1.74 ha. This means that the government cannot distinguish the type 1 farmer from the type 2 farmer within this range. This defines a lower bound for y so that the proposed contract is incentive compatible. In the empirical

100 88 analysis, therefore, the minimum size of retired peatland in the contract is restricted to 1.74 hectares. ψ'(y) Type 1 Farmer Type 2 Farmer Y Figure 4.5 Sorting Condition

101 89 Chapter 5. Empirical Analysis of a Peatland Retirement Program 5.1. Empirical Model of Peatland Retirement This section provides empirical analysis to illustrate the key results of the policy of land retirement with a targeting strategy and with bilateral information asymmetry. The application is based as much as possible on specific parameters estimated for Norway as described in Chapter 4. Some of the parameters required for the analysis do not exist for Norway. Some values are based on earlier studies and in other cases they are assumed. Consequently, the empirical results may be viewed as a first approximation to the design of an actual land retirement contract for Norwegian peatland. Using the parameters derived in Chapter 4, each risk-neutral principal in the model has a linear payoff function defined as: Type 1 Principal: V 1 = y j 1 t j 1 (5.1) Type 2 Principal: V 2 = y j 2 t j 2 (5.2) where y i j is the amount of retired peatland in hectares by agent type j after accepting the contract offered by principal type i, t i j is a monetary transfer from principal type i to agent type j in compensation for income foregone through program participation. The different types of principal reflect the different marginal benefits of peatland retirement, i.e., b 1 = (in 1000 $) and b 2 = (in 1000 $). The different marginal benefits represent different valuations of peatland retirement based on the depth of the peat

102 90 involved. The type 1 principal (thick peat) has larger marginal benefits than the type 2 principal (thin peat). The difference between b 1 and b 2 leads to different amounts of retired peatland of the two types, y i, and different levels of monetary transfer, t i, from the government to the farmer. The risk-averse agent (farmer) has a von Neumann-Morgenstern utility function Uj which is continuous, increasing and concave. It is assumed that the agents (farmers) have an exponential utility function so that constant absolute risk aversion (CARA) represents their attitude to risk: 2 U( ) 1 exp( ), U ( ) exp( ), U ( ) exp( ) (5.3) t ( y) (5.4) j j where π is a farmer s income and λ stands for the degree of risk aversion. The farmer s income π is a function of the monetary transfer, t, under the peatlands retirement program and income foregone function, ψ(y). The agent s reservation utility is normalized to zero. There is no information on risk aversion among Norwegian farmers, so the degree of risk aversion, λ, is assumed to be (Babcock et al., 1993). The Arrow-Pratt coefficient of constant absolute risk aversion is: 2 U ( ) exp( ), 0 U ( ) exp( ) (5.5) Two types of farmer can be distinguished from a policy perspective: efficient and inefficient farmers depending on the income foregone through peatland retirement. The

103 91 type 1 farmer represents an efficient agent (lower income foregone and therefore requiring lower payments by the government to compensate for retiring land) and type 2 farmer is an inefficient agent (higher income foregone requiring higher compensation), i.e., ψ 2 (y) > ψ 1 (y). As mentioned in chapter 4, the size of retired peatland on an individual farm, y, is restricted to be greater than 1.74 hectares due to the sorting condition. Quadratic income foregone functions are assumed as follows: Type 1 Agent: ψ 1 (y) = 0.17 (y 1 ) (y 1 ) (5.6) Type 2 Agent: ψ 2 (y) = 0.54 (y 2 ) (y 2 ) (5.7) Although the government cannot directly observe the agent s type (efficient or inefficient), it is assumed that the government knows the probabilities of the agent s type j in order to derive the optimal policy for peatlands retirement program. This is a common assumption in design of agri-environmental policy (Moxey et al., 1999). The probabilities of the agent s type j are denoted by pj. The p1 and p2 are the probabilities of efficient type farmer and inefficient type farmer, respectively. These values are assumed to be p1 = p2 = 0.5. In a pooling offer, since the government has private information about the depth of peatland, the principal-agent relationship becomes a sequential game with incomplete information. The optimal allocations in a pooling offer are Perfect Bayesian Equilibrium (PBE). This implies that the agent updates its belief in the type of principal based on the proposed contracts using Bayes rule. Because of this, unlike in a targeting strategy, the agent s prior belief in the type of principal is required. So q i stands for the agent s common prior belief in type i principal. The q 1 and the q 2 are the farmer s prior

104 92 belief in thick and thin peat, respectively. These values are assumed to be q 1 =q 2 =0.5. Since the farmer does not know about the depth of peatland, which is private information possessed by the government, these assumption of v equal prior belief in thick and thin peatland is a reasonable assumption. Synthetic parameter values are summarized in Table 5.1. Table 5.1 Synthetic Parameter Values Used for Empirical Analysis p1 p2 q 1 q 2 λ Characteristics of Optimum Allocations Since the marginal benefits of land retirement for type 1 principal are greater than those of type 2 principal, the ratio of Lagrange multipliers (as derived in Chapter 2) is ρ 1 /γ 1 < ρ 2 /γ 2. This implies that the IC constraint is relatively more expensive for the type 1 principal and the IR constraint is relatively more expensive for the type 2 principal. In the pooling offer, principal 1 will relax the IC constraint and enforce the IR constraint. Principal 2, on the other hand, will relax the IR constraint and tighten the IC constraint. Since the IR and IC constraints hold on average, there are slack variables on the IR and IC constraints for both types of principal:

105 IR : q 1 U( t 1 ( y 1 )) q 2 U( t 2 ( y 2 )) 0 (5.8) IC : q U ( t1 1( y1 )) U( t2 1( y2)) q U ( t1 1( y1 )) U( t2 1( y2 )) 0 (5.9) 0 0 From equation (5.8): (1) The type 2 agent who chooses the contract offered by the type 1 principal increases her/his utility level compared to the targeting strategy (Contract #2 in figure 5.1); (2) The type 2 agent who selects the contract offered by type 2 principal decreases her/his utility level. This type 2 agent has a negative utility level, i.e., below the reservation utility (Contract #4 in figure 5.1). From equation (5.9): (3) The type 1 agent who accepts the contract proposed by the type 1 principal decreases her/his utility level compared to the targeting strategy but that utility level is still greater than the reservation utility. That implies that the principal can reduce the information rent paid to an efficient type agent (Contract #1 in figure 5.1); (4) The type 1 agent who accepts the contract proposed by the type 2 principal increases her/his utility level compared to the targeting strategy and this type 1 agent receives some information rent from type 2 principal (Contract #3 in figure 5.1).

106 94 Targeting Strategy Pooling Offer Figure 5.1 Proposed Contracts in a Targeting Strategy and a Pooling Offer

107 95 Before examining the empirical results for optimal allocations in the peatland retirement program, the possibilities are illustrated using figure 5.1 and table 5.2. This helps in interpreting the results. Each contract is numbered in the bottom of figure 5.1. The type 1 principal is labelled thick (referring to the valuation placed on thick peatland) and the type 2 principal is thin (referring to the valuation placed on thin peatland). The type 1 agent is denoted low (referring to the low income foregone by peatlands retirement) and type 2 agent is high (referring to the high income foregone by peatland retirement). The Contracts with a Targeting Strategy Using a targeting strategy (as depicted on the left-hand side of figure 5.1), differentiated contracts corresponding to each type of principal are offered based on the government s private information about the depth of peatlands. The amount of retired peatland determined by the thick type principal is greater than the thin type principal. In other words, the size of retired peatlands, y, from contract #1 (thick/low) is larger than y from contract #3 (thin/low) given same type of the agent. Along with this, payments in contract #1 (thick/low) are also greater than payments in contract #3 (thin/low). Using the same logic, y from contract #2 (thick/high) is greater than y from contract #4 (thin/high) and compensation in contract #2 (thick/high) is also larger than compensation in contract #4 (thin/high). Through the differentiated contracts proposed, the agents learn the principal s private information about the benefits of this program.

108 96 The Contracts with a Pooling Offer In the pooling offer, the same menu of contracts is offered for both types of principal. The menu of contracts contains the contract #1, #2, #3 and #4 simultaneously in order to conceal the principals private information. In reality, contract #1 and #2 are offered for the type 1 principal (thick) and contract #3 and #4 are proposed for the type 2 principal (thin). Although the government cannot directly observe the type of agent, each type of agent realizes which contracts are designed for his/her type. This is because the pooling offer satisfies the revelation principle as in a targeting strategy. For instance, given a pooling offer, type 1 agents know that contract #1 and #3 are designed for them. But they do not care which contracts correspond to which principal because contract #1 and #3 already satisfy their IC constraint on average. Thus a type 1 agent can choose either of these. After they carry out the required action, i.e., retire a certain amount of peatland specified by the contract, however, they realize their true utility. Due to equation (5.9), the type 1 agent who chooses the contract corresponding to the type 1 principal decreases her/his utility level compared to the targeting strategy but that utility level is still greater than the reservation utility. On the other hand, the type 1 agent who accepts the contract corresponding to the type 2 principal increases her/his utility level compared to the targeting strategy. This implies that when the principals private information is concealed by using a pooling offer, it is only necessary to satisfy the agent s IR or IC constraint until the agent

109 97 chooses the contract. Maskin and Tirole (1990) called this an interim stage when the agent selects a contract. In the tautological sense, from equation (5.8), the type 2 agent who chooses the contract corresponding to the type 1 principal increases her/his utility level compared to the targeting strategy and the type 2 agent who selects the contract corresponding to the type 2 principal decreases her/his utility level. The requirement that the agent s IR and IC constraints only need to be satisfied at the interim stage yields a higher payoff to the government than by using a targeting strategy Optimal Contracts for Peatland Retirement with Unconstrained Farm Size In an initial application of the model for contract design no constraint is imposed on the amount of land that could be retired by an individual farmer. Examining the optimum allocations without the farm size restriction enables us to check the features of optimal contracts in a targeting strategy and a pooling offer more easily. Thus the focus is on the difference of optimum allocations between two approaches rather than the specific amount of peatlands and corresponding compensation. This is done to facilitate a comparison of results obtained from targeting and pooling. In a subsequent application of the model in section 5.4 the amount of land per farm that can be retired is constrained to reflect the actual situation in Norwegian agriculture. In section 5.4, the amount of retired peatland and corresponded compensation from the government to the agent are discussed more detail. Based on the optimum allocations with the farm size constraint, the

110 principal s payoffs possibility curve is derived. Using this, optimal contracts for national peatland retirement are also obtained in section Given the principal s payoffs function in equation (5.1) and (5.2) and the agent s income foregone function in equation (5.6) and (5.7), optimum contracts with a targeting and with a pooling offer are obtained from solving equation (2.3) and (3.12), respectively. The GAMS/CONOPT solver was used in solving these problems Comparison of Optimum Allocations According to data from Jordmod, representative type 1 and type 2 farmers have 34.9 ha and 23.1 ha of peatland, respectively (see Table 4.8). Empirical results in a targeting strategy and a pooling offer with no constraint on land retirement per farm are showed first so that we can test the feature of optimal contracts with a targeting strategy and with a pooling offer. Under unconstrained farm size, the focus is on the differences in the optimum allocations between the two approaches in contract design. Empirical results with unconstrained farm size are given in Table 5.2. Empirical results with constrained farm size are provided in the next section. An infinite number of equilibrium allocations that are feasible for the pooling offer exist. The set of equilibrium allocations is also a Pareto set so that at least the same payoff can be obtained under the pooling offer for each principal as under the targeting strategy. Two extreme cases are shown in Table 5.2. In the first, all the additional payoff 18 The problems were also solved using GAMS/BARON so that the optimum allocations from this study are guaranteed to be a global optimum. The differences of optimum allocations between two solvers are negligible, e.g., identical or differences below three decimal places.

111 99 from the pooling offer accrues to principal 1 (case 1) and in the second all the surplus accrues to principal 2 (case 2). Contracts are identified by the numbers given in the first column for principal/agent combinations of amounts of land retired and payments. Table 5.2 Comparison of Optimum Allocations without Farm Size Restriction Proposed Contracts Targeting Strategy Pooling Offer Contract Number Allocations Principal 1 Principal 2 Principal 1 takes all surplus (Case 1) Principal 2 takes all surplus (Case 2) y 1 1 (ha) t 1 1 (1000 US$) 25, y 2 1 (ha) t 2 1 (1000 US$) 4, y 1 2 (ha) t 1 2 (1000 US$) 10, y 2 2 (ha) t 2 2 (1000 US$) 1, When the targeting strategy is employed, two contracts such as ( ha, US$25,113,861) and (39.78 ha, US$4,170,213) are offered to satisfy the type 1 principal and the type 2 principal ( ha, US$10,462,287) and (20.33 ha, US$1,917,658), respectively. Contracts #1 and #3 are designed for the type1 agent. Due to the different marginal benefits for the principal, the required peatland retirement and compensation in contract #1 is greater than for contract #3. By the same logic, contract #2 has larger retirement and higher payments compared to contract #4.

112 100 When the same menu of contracts is offered, type 2 agent s IR constraint is relaxed to satisfy the type 1 principal by using contract #2. Compared to contract #2 with targeting, the payment with a pooling offer in contract #2 is increased to satisfy type 2 agent s IR constraint. Type 1 agent s IC constraint is enforced to satisfy the type 1 principal using contract #1. In contract #1, even when the same amount of peat retirement is required to satisfy the type 1 principal in a targeting strategy and a pooling offer, the amount of compensation through the pooling offer decreases. Since the agent s IC constraint is relatively more expensive for the type 1 principal, the agent s IC constraint is enforced in contract #1. In contrast, the agent s IC constraint is relaxed to satisfy the type 2 principal by using contract #3. Although the type 1 agent retires the same amount of peatland in a targeting strategy and a pooling offer, the payment for income foregone in a pooling offer is greater than that in a targeting strategy, regardless of the distribution of the surplus. Also the agent s IR constraint is strengthened using contract #4 to satisfy the type 2 principal. Compared to the contract #4 in a targeting strategy, the payment in a pooling offer diminishes more than proportionally. The different ratio of Lagrange multipliers allows the more expensive constraint for each principal to be relaxed and the less costly constraint to be tightened.

113 Comparison of Principal s Payoffs and Agent s Utility The summary of the feature of optimal contracts in a targeting strategy and a pooling offer helps in understanding the discussion below. The optimal contract in a targeting strategy has the following features: (1) Type 1 agent (efficient type farmer) is bound by the IC constraint and receives information rent; (2) Type 2 agent (inefficient type farmer) is bound by the IR constraint so the utility level is equal to the reservation utility. The optimal amount of retired peatland is distorted downward from first-best outcomes. The optimal contract in a pooling offer has the following features: (1) Type 1 agent (efficient type farmer) and type 2 (inefficient type farmer) are bound by the IC and IR constraints on average, respectively. Due to different ratio of Lagrange multiplier: (2) The agent s IC constraint is enforced to satisfy the type 1 principal since the IC constraint is relatively more expensive for the type 1 principal. And the agent s IR constraint is relaxed by the type 1 principal because the IR constraint is relatively cheaper; (3) The agent s IC constraint is loosened to satisfy the type 2 principal since the IC constraint is relatively cheaper for the type 2 principal. And the agent s IR

114 102 constraint is tightened by the type 2 principal because the IR constraint is relatively more costly to the type 2 principal. Principal s Payoffs in a Targeting Strategy and a Pooling Offer The characteristics of the pooling offer can be explored further by using the principal s payoffs and the agent s utility. Comparisons of principal s payoffs and agent s utility for the various contracts are given in Table 5.3. In the targeting strategy, type 1 and 2 principal s payoffs are US$ 4,312,676 and US$ 1,063,233, respectively. When a pooling offer is used, however, at least the same payoffs can be guaranteed for each principal as under the targeting strategy where the principal s type is common knowledge. If type 1 principal takes the entire additional payoffs resulting from the pooling offer (case 1), the type 1 principal s payoff increases from US$ 4,312,676 to US$ 4,596,563 compare to a targeting strategy. If the type 2 principal takes the entire additional payoffs from the pooling offer (case 2), the type 2 principal s payoff increases from US$ 1,063,233 to US$ 1,203,657 compare to a targeting strategy. Agent s Utility in a Targeting Strategy Type 1 agent utility with contract #1 is 0.847, shown in row (1), and with contract #3 it is 0.36, shown in row (6). As shown in rows (3) and (8), the agent s IC constraint is binding for the type 1 agent. In addition, the type 2 agent utility is zero because the IR constraint is binding for that agent. This is shown in rows (4) & (5) and (9) & (10).

115 103 Agent s IC and IR constraints in a Targeting Strategy Using the targeting strategy as a reference, the IC constraint is binding for the type 1 (low type) agent. This can be seen using rows (1) through (3) and (6) through (8). The type 1 agent utility with the targeting strategy is as in row (1) and row (2) is its IC constraint. We can check that the IC is binding in row (3). And the type 1 agent on contract #3 is also bound by the IC constraint in row (7) and (8). In addition, the IR constraint is binding for the type 2 (high type) agent. From rows (4) and (5) and (8) and (9), it is clear that the IR constraint is binding for the inefficient agent in the targeting strategy. Agent s Utility in a Pooling Offer (1) Type 1 Agent: Related Contracts are #1 (thick/low) and #3 (thin/low) The type 1 agent in contract #1 experiences a decrease in utility from to or to in row (1) of each pooling offer. The type 1 agent in contract #3 has an increase in utility from 0.36 to or to in row (6). The reason is that the IC constraint is strengthened by the type 1 principal using contract #1 in row (1) and the IC constraint is loosened by the type 2 principal using contract #3 in row (6) in both pooling offer cases. Because different types of principal have a different ratio of Lagrange multiplier, the costs of fulling the agent s IR and IC constraints differ by the type of principal.

116 104 (2) Type 2 Agent: Related Contracts are #2 (thick/high) and #4 (thin/high) By the same logic, the type 2 agent has an increase in utility from zero to or to with contract #2 and the type 2 agent has a reduction in utility from zero to or to with contract #4 in row (8) with both pooling offers. This is because the IR constraint is relaxed by the type 1 principal using contract #2 and the IR constraint is tightened by the type 2 principal using contract #4. Agent s IC constraint in a Pooling Offer Since both types of principal offer the same menu of contracts, the IC constraint is binding on average for the type 1 (low type) agent in the pooling offer. When the additional payoff from a pooling offer accrues to the type 1 principal, the IC constraint is non-binding for the type 1 agent in contract #1(more/low). The utility of the type 1 agent who selects the contract offered by the type 1 principal is in row (1). However, if the type 1 agent chooses the contract designed for the type 2 agent by the type 1 principal (contract #2 in figure 5.1), which is the IC constraint for the type 1 agent, the type 1 agent s utility is Thus the IC constraint is enforced to satisfy the type 1 principal using contract #1 (thick/low). This is shown in row (1) through (3). If the type 1 agent accepts the contract offered by the type 2 principal, the type 1 agent s utility is in row (6). But if the type 1 agent chooses the contract designed for the type 2 agent by the type 2 principal (contract #4 in figure 5.1), the utility of the type 1 agent is Because of this, the IC constraint is relaxed to satisfy the type 2 principal using contract #3 (thin/low). This is described in rows (6) to (8). But the total effects by enforcing and

117 105 relaxing the agent s IC constraint is zero (0= ). Thus the type 1 agent s IC constraint only holds on average. By the same logic, the case in which the type 2 principal gets all the additional payoff from the pooling offer is identical to the case of the type 1 principal. Agent s IR constraint in a Pooling Offer The type 2 agent is similar to type 1 agent but the IR is the binding constraint. The type 2 agent choosing the contract offered by the type 1 principal has an increase in utility level compared to the reservation utility, which is normalized to zero in section 2.4. It is which is shown in rows (4) and (5). From row (9) and (10), on the other hand, the type 2 agent receives negative utility in selecting the contract offered by the type 2 principal. However, this is neutralized on average (0 = ). In other words, the IR constraint for the type 2 agent is only satisfied on average. By the same logic, the case in which the type 2 principal gets all the additional payoff from the pooling offer is identical to the case of the type 1 principal.

118 106 Principal s Payoffs Table 5.3 Comparison of Principal s Payoffs and Agent s Utility Benchmark Principal 1 Principal 2 Principal 1 takes all surplus (Case 1) Pooling Offer Principal 2 takes all surplus (Case 2) 1 V V Agent type 1 Utility Contract 1: Agent type 1 chooses the contract offered by principal type 1 (1) 1 1 U ( t1 1( y1 )) (2) IC: U( t2 1( y2)) (3) (1) - (2) Agent type 2 Utility Contract 2: Agent type 2 chooses the contract offered by principal type 1 (4) 1 U (5) (4) Reservation Utility Agent type 1 Utility Contract 3: Agent type 1 chooses the contract offered by principal type 2 (6) 2 2 U ( t1 1( y1 )) (7) IC: U ( t2 1( y2 )) (8) (6) (7) Agent type 2 Utility Contract 4: Agent type 2 chooses the contract offered by principal type 2 (9) 2 U (10) (9) Reservation Utility

119 Comparison of Information Rent and Downward Distortion Since the agent s constraints are relaxed or enforced in order to achieve higher payoffs in pooling offer, we can observe an interesting feature of this type of contract that cannot be found in the targeting strategy. Recall from section 2.1 that in a targeting strategy only the efficient type of agent (type 1) receives an information rent 19 and the optimal amount of retired peatland by the inefficient agent (type 2) are distorted downwards from the first-best outcomes (see Table 5.4). This is a well-known result in contract design (Salanié, 2005). However, with the pooling offer, the efficient agent (type 1) is paid less information rent by the type 1 principal and more information rent by the type 2 principal. Inefficient (type 2) agent obtains a positive information rent from the type 1 principal and a negative information rent from the type 2 principal. And downward distortion by the inefficient agent (type 2) is decreased with the type 1 principal (See Table 5.5 and 5.6). Downward distortion 20 by the inefficient (type 2) agent is increased with the type 2 principal (see Table 5.5 and 5.6). The reason for this is as identified earlier. Due to the different ratio of Lagrange multipliers, the agent s IR constraint is relatively more expensive for the type 2 principal and the IC constraint is relatively more costly for the type 1 principal in a targeting strategy. 19 Information rent is the difference between received monetary transfer and actual income foregone. 20 Downward distortion means that the amount of retired peatland is distorted downwards relative to the first-best outcome.

120 108 Table 5.4 Inefficiency of the Contract with a Targeting Strategy Agent Type Inefficiency Principal 1 Principal 2 Agent 1 Information Rent (in 1000 US$) Downward Distortion (hectares) Agent 2 Information Rent (in 1000 US$) Downward Distortion (hectares) Table 5.5 Inefficiency of the Contract with a Pooling Offer (Case 1) (Type 1 Principal takes all the surplus) Agent Type Inefficiency Principal 1 Principal 2 Agent 1 Information Rent (in 1000 US$) Downward Distortion (hectares) Agent 2 Information Rent (in 1000 US$) Downward Distortion (hectares) Table 5.6 Inefficiency of the Contract with a Pooling Offer (Case 2) (Type 2 Principal takes all the surplus) Agent Type Inefficiency Principal 1 Principal 2 Agent 1 Information Rent (in 1000 US$) Downward Distortion (hectare) Agent 2 Information Rent (in 1000 US$) Downward Distortion (hectare)

121 Optimal Contracts with a Farm Size Constraint As seen from the empirical results in Table 5.2, a peatland retirement program appears to be a promising approach to achieving a GHG mitigation plan. Through a pooling offer, the type 1 principal offers contracts with 205 ha and 65 ha of peatlands to be taken out of production to efficient (lower income foregone) and inefficient (higher income foregone) dairy farmers, respectively. And type 2 principal proposes 101 ha and 15 or 19 ha of retired peatlands to type 1 and type 2 dairy farmers, respectively. And the type 1 principal compensates US$ 24,720 and US$ 24,998 (in 1000 $) for the type 1 dairy farmer corresponding to the peat retirement. US$ 7,999 and US$ 8,290 (in 1000 $) is proposed by the type 1 principal as a compensation for peatland retirement by the type 2 dairy farmer. The type 2 principal offers compensation of US$ 10,484 and US$ 10,329 (in 1000 $) for the type 1 agent and US$ 1,243 and US$ 1,584 (in 1000 $) for the type 2 agent. The type 2 dairy farmer retires less peatland that the type 1 dairy farmer due to the higher income foregone, but at least 15 ha of peatland per farm is taken out of dairy production. These contracts are likely to be unrealistic for many farms in Norway. Agriculture in Norway is characterized by small-scale family farms. According to the farm size data in the sector model (Jordmod), the representative dairy farms used for this study have 23 ~ 35 ha of peatlands among 35 ~ 48 ha of total farm size (see Table 4.8). Thus the proposed retirement of peatland per farm is so large that the optimum allocations presented in the previous section are considered to be infeasible. This natural restriction is an additional constraint on the amount of peatland that could be retired at the farm

122 110 level. Based on the sector model, the type 1 dairy farmer s y is limited to 34.9 ha and the type 2 dairy farmer s y is limited to 23.1 ha. With the sorting condition, this defines the lower and upper bound for peatland retirement per farm: 1.74 ha y 1 1, y ha for type 1 farms 1.74 ha y 2 1, y ha for type 2 farms Using these constraints, the optimum allocations for peatland retirement program at the farm-level are recomputed. Also one additional constraint is imposed with farm size restriction. That is an equal payment constraint. When both types of principal offer the same amount of retired peatland, the payment levels offered by both types of principal must be equal. With these additional constraints, the empirical results are given in Table 5.7.

123 Table 5.7 Optimum Allocations for Peatland Retirement with Farm Size Constraint 111 Proposed Contracts Targeting Strategy Pooling Offer Contract Number Allocations Principal 1 Principal 2 V 1 = V 2 = Principal 1 takes all surplus (case 1) V 1 = V 2 = Principal 2 takes all surplus (case 2) V 1 = V 2 = y 1 1 (ha) t 1 1 (1000 US$) y 2 1 (ha) t 2 1 (1000 US$) y 1 2 (ha) t 1 2 (1000 US$) y 2 2 (ha) t 2 2 (1000 US$) Due to the farm size restriction, optimal contracts in a targeting strategy and a pooling offer are similar to each other. Results show that an offer to retire the entire amount of peatland used in dairy production is made to both types of agent in targeting strategy and pooling offer to satisfy the type 1 principal. To satisfy the type 2 principal all the peatland used by the type 1 agent is retired. Only the type 2 agent who chooses the contract to satisfy the type 2 principal can keep a small portion of his/her peatland in dairy production. With the exception of this case, the government will seek to retire all the peatland used by dairy farmers. The payoffs to the principals are measured in both cases.

124 112 Farm size restrictions create some interesting features of optimal contracts with a pooling offer. First, consider the contract proposed by the type 1 principal to the type 1 agent (contract #1). In both cases of a pooling offer, the type 1 principal offers the same amount of peatland retirement to the type 1 agent, 34.9 ha. That is the maximum amount of peatland possessed by the type 1 agent. However, the compensation for peatland taken out of production varies across case 1 and 2. These are US$ 3, and US$ 3, (in 1000$), respectively. The reason is that in case 1 the type 1 principal takes all the additional payoff from a pooling offer so the type 1 principal maximizes its payoff. The type 1 principal tightens agent 1 s IC constraint as much as possible in case 1. However, due to the equal payment constraint, type 1 principal cannot fully utilize its additional payoffs. Therefore, the compensation in case 1, which is US$ 3, (in 1000$), is slightly greater than that in case 2, which is US$ 3, (in 1000$). And the amount of compensation in the pooling offer are smaller than in the targeting strategy, which is US$3, (in 1000 $). By the same logic, the type 2 principal enforces the agent 2 s IR constraint as much as possible in case 2 (contract #4). However, type 2 principal cannot fully utilize its additional payoffs because of an equal payment constraint. Thus the compensation in case 2, which is US$ 1, (in 1000$), is a bit greater than in case 1, which is US$ 1, (in 1000$) even though almost same amount of peatland is retired, which is 18.9 ha. Along with tightening the agent 2 s IR constraint in a pooling offer by the type 2 principal, the amount of retired peatland in a pooling offer is smaller than that in a targeting strategy, which is ha.

125 113 From contract #2, the type 1 principal relaxes the agent 2 s IR constraint in a pooling offer. Because of this, the compensation with a pooling offer in both cases is greater than with a targeting strategy, which is US$ 2,213 (in 1000 $), even though the same amount of peatland is retired, which is 23.1 ha. This is the maximum amount of peatland possessed by the type 2 agent. In case 2 in a pooling offer, the type 2 principal takes all the additional payoff from a pooling offer. As a counter effect, the type 1 principal loosens the agent 2 s IR constraint as much as possible. Thus the compensation in case 2, which is US$ 2,248 (in 1000$), is greater than in case 1, which is US$ 2,243 (in 1000$). From contract #3, since type 2 principal loosens the agent 1 s IC constraint in a pooling offer, the compensation with a pooling offer in both case 1 and 2 is higher than with a targeting strategy, which is US$ 3,287 (in 1000 $), even though the same amount of peatland is retired, which is 34.9 ha. In case 1 in the pooling offer, the type 1 principal takes all the additional payoff. As a response, the type 2 principal relaxes the agent 1 s IC constraint as much as possible. Therefore, the compensation in case 1, which is US$ 3, (in 1000$), is little bit greater than in case 2, which is US$ 3, (in 1000$) given same amount of retired peatland, 34.9 ha. In the targeting strategy, the type 1 principal (thick peat) generates a payoff of US$ 1,709,652 and the type 2 principal expects a payoff of US$ 686,544. These define the lower bounds for payoffs in the pooling offer. The pooling offer can guarantee at least the same payoff for each principle as under targeting strategy. In the pooling offer, when all the surplus accrues to principal 1, the payoff increases from US$1,709,652 to

126 114 US$ 1,711,834 while the payoff to principal 2 remains the same as with the targeting strategy. This means that US$ 1,711,834 is an upper bound for the type 1 principal s payoff under the pooling offer. On the other hand, US$ 688,702 accrues to the type 2 principal compared to US$ 686,544 with targeting if all the surplus goes to the type 2 principal. This reflects the upper limit of the payoff to the type 2 principal in the pooling offer Optimal Contracts for National Peatlands Retirement Based on these upper and lower bound for payoffs to the principals, a Payoff Possibility Curve (PPC) can be defined. As indicated earlier, an infinite number of equilibrium allocations (combinations of payoffs to the principals) exist for the pooling offer, all of which are Pareto superior to the allocations in the targeting strategy. This implies that if one principal s payoff in the pooling offer is determined, the other principal s maximum payoff in the pooling offer is also determined through the equilibrium allocation. To derive the principals payoffs possibility curve, principal 2 s payoffs are maximized subject to principal 1 s payoffs. The upper and lower bounds of each principal s payoff have already been derived (Table 5.7). Given type 1 principal s payoffs, type 2 principal s maximum payoffs are derived. And type 1 principal s payoffs are gradually increased by its upper bound. Whenever it increases, the optimum payoff for the type 2 principal is recomputed. The Type 1 principal s payoffs start at the payoff level with the targeting strategy, i.e., the lower bound of US$ 1,709,652. Type 1 principal s payoffs are increased until they reach the upper bound of payoffs

127 115 US$ 1,711,834. This is the same as the payoff when all surplus in the pooling offer accrues to the type 1 principal. Details on the simulation results are given in Appendix C. The horizontal axis in figure 5.2 is the type 1 principal s payoff and the vertical axis is the type 2 principal s payoff. Point a in figure 5.2 denotes the lower bound of each principal s payoff. These are the same as the payoffs in the targeting strategy. Point b is type 1 principal s payoff when the entire surplus from the pooling offer accrues to the type 1 principal and the type 2 principal s payoffs are equal to the payoffs with targeting. Point c is the opposite case. This is an upper bound of principal 2 s payoff because it only occurs when all the surplus from the pooling offer accrues to the type 2 principal. The shaded area in figure 5.2 denotes the feasible equilibrium allocations of payoffs in the pooling offer. It implies that: (1) infinitely many feasible equilibrium allocations exist in the pooling offer; (2) the equilibrium allocations in a targeting strategy are dominated by the equilibrium allocations in the pooling offer. Any point in the shaded area guarantees at least the same payoff to principals as under the targeting strategy. The line between point b and c is the combination of maximum payoffs in both principal. This frontier is the Payoff Possibility Curve.

128 116 Figure 5.2 Payoff Possibility Curve with the Pooling Offer Using the payoff possibility curve (PPC), optimal allocations for national peatlands retirement can be defined. To do this, a national benefits function must be defined. The national benefits function represents the total benefits that government can achieve through a peatland retirement program. And national optimum allocations provide maximum benefits from this program. The national benefits function assumed is: National Benefits Function = N 1 V 1 + N 2 V 2 where N1 is the total number of dairy farmers with thick peatland and N2 is the total number of dairy farms with thin peatland. V 1 and V 2 are principal s payoffs with the farm size restriction associated with the retirement of thick and thin peatland, respectively. The

129 117 percentage of thick and thin peat in Norway is assumed to be 40 % and 60% (Table 4.3 in Grønlund et al., 2008b). 21 To conduct a precise analysis of national optimum allocations, it is necessary to know the correlation between the principal s and the agent s types. For instance, N1 is the number of type 1 farmers with thick peatland plus the number of type 2 farmer with thick peatland. N2 is the number of type 1 farmers with thin peatland plus the number of type 2 farmers with thin peatland. Unfortunately, specific information about the correlation between principal s and agent s type is not available. To derive a rough estimation, it is assumed that the percentage of thick and thin peatland used by dairy farms is the same as the proportion of thick and thin peatland nationally. Thus the percentages from Grønlund et al. (2008b) can be used directly. With the assumption of correlation between the principal and the agent, the slope of the function is the proportion of thick peat and thin peat. Thus the national benefits function has a slope of -0.67(= 0.4/0.6), which is obtained from the proportion of thick and thin peat, in principals payoffs space. This means that the slope of the national benefits function is flatter than the slope of the payoffs possibility curve. Because of this, the optimal contracts for national peatlands retirement are defined at point c. Point c means that the type 2 principal (thin peat) takes all the additional payoff from a pooling offer and the type 1 principal s (thick peat) payoff remains the same as the targeting strategy. Therefore, the optimal contracts for national peatlands retirement with a pooling offer are the last column in Table The report only provides details on the proportion of the less than 100 cm depth peat soil. That is 60% of total peat soil in Norway. Thus it is assumed that the more than 100 cm depth peat soils are all thick peat which has 200 cm depth. So the percentages of thick and thin peat soil are 40 % and 60 %, respectively.

130 Figure 5.3 National Optimum Allocations using Pooling Offer 118