THE EFFECT OF RISK, TIME PREFERENCE, AND POVERTY ON THE IMPACTS OF FOREST TENURE REFORM IN CHINA

.................................................................................................................. ii ACKNOWLEDGEMENTS .......................................................................................... vi DEDICATION .............................................................................................................. ix PREFACE ...................................................................................................................... x TABLE OF CONTENTS .............................................................................................. xi LIST OF TABLES ...................................................................................................... xiv LIST OF FIGURES ................................................................................................... xvi

are often implemented in areas where the poverty rate is high. Those living in poverty are often assumed to have both high discount rates (i.e., impatient) and high levels of risk aversion, which make them less likely to make investments. Such characteristics may also hinder the intended effects of forest tenure reforms.
This study examines these issues in the context of rural China, where a large-scale reform of forest property rights is being implemented in areas where the poverty rate is still high. To examine these issues, this dissertation research has three objectives: 1) identify the impact of forest property rights reforms on forest management decisions and how individual risk and time preferences may augment those impacts; 2) examine the correlation between time and risk preferences and poverty; and 3) identify the effect of the forest tenure reform on household wealth. This dissertation research contributes to the literature on the effect of property rights reforms on natural resource management in developing countries in several ways. This study is the first to use a large-scale property rights reform to examine the heterogeneity of its impact on forest management due to risk and time preferences.
Moreover, it integrates experimental economics methods into natural resource management, which is an innovative approach to test the behavioral reactions to policy alternatives. This study is unique because it tests the theory by integrating field experiments to capture risk and time preferences and panel survey data to capture poverty and forest management decisions.
To achieve the research objectives, I first designed experiments to capture individual time and risk preferences and then ran them in the field with farmers in China. Then I integrated data the field experiment data with a household panel data in an econometric framework. In manuscript 1, I use this integrated data set to examine how preferences over time (present vs. future) and risk can affect households' forest management responses to strengthened forest property rights. I find that risk and time preferences impact households' forest management responses to forest plot certification. Specifically, in response to forest certification, more risk averse households used less labor for harvesting and more labor for applying inputs, while more loss averse households used more labor for harvesting. Households with higher discount rates (i.e., impatient) used less labor for applying inputs and spent less on forest inputs (chemical fertilizer, pesticides, and seeds) in response to receiving a forest certificate.
Manuscript 2 investigates the correlation between poverty and individual preferences for time and risk. The classic assumption is that the poor have both high levels of risk aversion and high discount rates. Contrary to this assumption, my research demonstrates that wealth does not have a significant effect on risk aversion or loss aversion (with the one exception that households with more forestland per capita are less loss averse). However, consistent with this assumption I find statistically weak evidence that households with lower wealth have higher discount rates (i.e., more impatient).
In manuscript 3, I examine the effect of forest tenure reform on household wealth. I find statistically weak evidence that the forest tenure reform has had a positive effect on household wealth, specifically, increased tenure security in the form of a forest certificate increased net worth per capita by 42% between 2000 and 2008. To further examine the source of increased wealth, I also examine the effect of the reform on household forest use. Results suggest that forest certification increased bamboo revenue, while obtaining a new plot (without a forest certificate) increased non-timber forest product revenue, although these results are statistically weak. Overall manuscript 3 provides weak evidence that forest tenure reform garners potential for improving poor rural households' livelihoods in China.
Overall this dissertation research demonstrates that time and risk preferences matter for forest management and responses to forest tenure reforms. This suggests that policymakers designing and implementing tenure reforms should consider the particular context of the reform and consider coupling the reform with appropriate programs and instruments to alleviate poverty and to help households' to deal with risks and make long-term investments to further stimulate the intended effects of the reform-increased investment in forest resources and improved livelihoods.  Table 2.9 Effect of alternative wealth proxy variables on loss aversion……….140 Table 2.10 Effect of alternative wealth proxy variables on time preference……141 Table 2.11 Robustness of time preference results……………………………….142

INTRODUCTION
Forests perform a multitude of ecosystem services and contribute directly to the livelihoods of more than one billion people living in extreme poverty (World Bank 2004). Despite the national and international efforts devoted to global problems of deforestation, forest degradation has steadily increased throughout much of the world .
The cause of this continued degradation is complex and multifaceted but there is a growing realization that a key cause, especially in developing countries, is insecure rights to ownership and use of forest resources FAO 2007;. Property rights to ownership and use of forest resources are often contested, overlapping or unenforced. This insecurity undermines sound forest management, for without secure rights forest holders have few incentives to invest in managing and protecting their forest resources. This realization coupled with a call for pro-poor forestry policy from international institutions, NGOs, and community organizations has stimulated the recent trend in forest policy toward strengthening property rights for forest resources by transferring property rights from the state to communities and individuals, giving them defined rights to manage and extract forest resources FAO 2003;FAO 2009). In the most forested developing countries, this trend has resulted in a doubling of the percent of forest owned or controlled by indigenous and rural communities between 1985 and 2000 . By 2050, 40% of the world's forest is expected to be managed or owned by communities and individuals (FAO 2003).
However, research to date presents conflicting conclusions regarding the impacts of tenure insecurity on forest management. For example, some empirical studies found that stronger land tenure facilitated investment in trees (e.g., Ghana  and Ethiopia (Holden, Deininger, and Ghebru 2009)). At the same time, other examples illustrate that these reforms have not led to their intended consequence of sustainable resource management (Ecuador , Indonesia , Russia (White and Martin 2001) and newly independent states of Eastern Europe (INDUFOR OY/ECO for the World Bank and World Wildlife Fund Alliance 2001) and others .
This dissertation research attempts to offer insight into why forest tenure reforms may not always lead to their intended effects by focusing on how an individual's preferences over time and risk affect individual responses to forest tenure reforms individuals' forest management decisions, as well as their responses to forest tenure reforms, will depend on: 1) how an individual perceives preference for income today versus in the future (time preference), and 2) an individual's attitude towards risk (risk preference). By taking individual preferences into account, we may find that forest tenure reform has the intended effects on individuals with some characteristics, while it does not have the intended effects on others. This heterogeneity in policy effect may be masked in studies that find negligible effect of the reforms.
In a developing country context, the potential correlation between individual preferences and the effect of forest tenure reforms is also associated with poverty. The intended effect of forest tenure reforms is to increase tenure security. Economic theory predicts that increased tenure security will give households greater incentives to invest without fear of expropriation. This increased investment should raise productivity and cash flows, which in turn should stimulate incomes as well as land values and general levels of economic activity, helping the rural poor escape from poverty (Demsetz 1997;World Bank 2003;Feder 1999;. This often-assumed premise is examined specifically in this dissertation. Moreover, households living in poverty are often assumed to be highly risk averse (Bardhan and Udry 1999; Stiglitz) and have high rates of discount rates (impatient), which keep them from making investments. This assertion is also tested in this dissertation. In sum, Economic theory predicts that increased tenure security from tenure reform will increase households' investment in forest resources. However, because investment and forest management decisions are dynamic, there are several possible paths by which households may reach a higher steady state investment level (I*) in response to the tenure reform (Figure 0.2). Households' investment paths may depend on factors such as their available capital, access to credit, other investment opportunities, forest product market conditions, forest plot characteristics, and individual preferences. For example, line (a) depicts a trajectory in which a household instantaneously increases investment to the optimal level in response to the tenure reform. A household with abundant capital or ability to borrow such that it can invest as soon as its land tenure is strengthened may follow an investment path like (a). Alternatively, a household without access to capital or credit or that has a high discount rate (i.e., impatient) may not increase investment even with strengthened tenure, as depicted by line (d). Lastly, in response to the tenure reform households may exhibit a period of transition during which the household increases the investment level towards a new steady-state level.
During the transition period, a household may increase investment gradually, as depicted by line (b) or in a series of steps (c), depending on their particular circumstances. A household's investment path may look like (c) if, for example, the household is risk averse and is unsure about the security of their new property rights, or if a household cannot access capital or credit quickly. Such a household may wait a period of time before increasing investment.
Increased investment should raise productivity and cash flows, which in turn should stimulate forest revenue. As with changes in investment, changes in forest revenue may take different paths towards a higher steady state level of forest revenue, R* (Figure 0.3). Households' forest revenue depends on both their investment and harvesting decisions, which in turn depend on a variety of factors such as their available capital, access to credit, other investment opportunities, forest plot characteristics, product market conditions, and personal preferences. For example, a household that is able to invest in its forest plot immediately following the reform may see a gradual increase in forest revenue over a transition period, as depicted by line (a), before it reaches the long-term steady state level of forest revenue, R*.
Alternatively, a household that is unable to invest following the reform may not experience any increase in forest revenue, as depicted by line (b). Furthermore, a household may experience a temporary dip in forest revenue during the transition period, as depicted by line (c). This temporary dip may occur if a household with strengthened land tenure now waits until the new, longer optimal harvesting age is reached. The next subsection will explain how the three manuscripts in this dissertation investigate these hypotheses.

Outline of the three manuscripts
In Manuscript 1, I examine the relationships in H1 and H2. Specifically, I examine how forest tenure reform affects harvesting behavior and investment in forest plots.
Increased tenure security gives households confidence that if they invest in their plot (planting, maintenance, etc.) then they will be able to obtain the benefits from those efforts in the future . As such, I hypothesize that households that were subject to forest tenure reform will have an incentive to invest in their forests resources and delay harvest until the optimal harvesting threshold is reached (H1).
Furthermore, in manuscript 1, I investigate the heterogeneity in the effects of the forest tenure reform by focusing on risk and time preferences as the source of heterogeneity (H2). Households making forest management decisions face many uncertainties. Furthermore, decisions about forest management often involve a long time horizon ). As such, households' risk and time preferences take an important role in their forest management decisions , and may affect their responses to forest tenure reforms. Certain types of households' risk and time preferences may lead forest tenure reform to affect forest management in ways that are consistent with what policymakers intended. For example, assuming that with forest tenure reform a household believes that there has been a reduction in the risk of expropriation of its forest plot, then a household that is more risk averse may make more investments on the forest plots than a risk neutral or risk seeking household. Alternatively, households' risk and time preferences may cause them to respond to receiving a forest tenure reform in a way that is contrary to what policymakers intended. For example, in response to forest tenure reform, a household that has a high discount rate (impatient) may make less investment than a household with a low discount rate. Relatedly, a household that is more loss averse (i.e., has a tendency to strongly prefer avoiding losses to acquiring gains and to dramatically overweight losses relative to gains) may make less investment after a forest tenure reform than a household that is less loss averse. Therefore the intended effect (increased investment) that policymakers expected in response to a forest tenure reform may be weak or may not be exhibited by households with high discount rates or a high degree of loss aversion.
In manuscript 2, I examine the association between poverty and individual preferences (H3). Households living in poverty are often assumed to be highly risk averse and have high discount rates (impatient), characteristics that keep them from making investments Lipton 1968;Lumley 1997;Bardhan and Udry 1999;Fafchamps 2003 (2010)). Manuscript 2 offer new empirical evidence.
Finally, manuscript 3 examines whether or not household wealth has increased as a result of the forest tenure reform (H4). In addition to increasing investment in forest resources, an additional goal of the forest tenure reform in many developing countries is to improve households' livelihoods. Economic theory predicts that with more secure property rights, households will have a greater incentive to invest in their forest resources without fear of expropriation, which will stimulate income Feder 1999;Coning and Deb 2007). There is growing evidence that forest tenure reforms cause changes in local livelihoods but those changes have been both positive and negative (Shackleton and Campbell 2001;Edmunds and Wollenberg 2003;Jagger, Pender and Gebremedhin 2005;Sikor and Nguyen 2007). Manuscript 3 offers new empirical evidence based on a large-scale forest property rights reform.
Specifically, I examine how changes in forest land tenure affect households' net worth per capita, and further examine to see if the source of this effect is from changes in revenue from bamboo and non-timber forest products.
I examine these issues empirically in the context of China's collective forests, where a large-scale reform of forest property rights began in 2003 in rural areas where the poverty rate is still high. The reform was aimed at delegating responsibility of forest management from the collective (by townships and villages) to households and strengthening property rights with the distribution of forest certificates that establish the use of a specific forest plot for a period of 30-70 years and expand rights to include those of land transfer, inheritance, and mortgaging ). An advantage of this study is that we utilize this actual change in forest property rights whereas previous studies have used proxy variables (e.g., number of conflict with abutters and duration of residence in a village) that are either subjective or indirect measures and may not accurately measure tenure security (Godoy et al 1998;Godoy et al. 2001;Hagos and Holden 2006). China's collectively owned forests total approximately 100 million hectares and are home to more than 400 million people, which arguably makes these reforms the largest one undertaken in modern times both in terms of forest area and people affected ). In China, many people living in or near forests are poor , and while there has been a dramatic reduction in the poverty rate in China over the last decades, poverty is still a serious problem, particularly in rural areas (Chen and Ravallion 2008;). The recent rapid and dramatic changes in forest tenure in poor regions in China makes it an ideal context to study how individual preferences affect forest management decisions and the implications for the effectiveness of strengthening property rights to stimulate investment in forest resources and improve households' livelihoods.
To test these hypotheses, I combine original field experiment data on risk and time in previous related studies (Godoy et al. 1998;Godoy et al. 2001;Hagos et al. 2006).
Furthermore, our risk preference experiment design follows a recently developed methodology that expands the classic lottery experiment of Holt and Laury (2002) to allow for estimation of a more flexible and richer description of a person's risk preference as described under prospect theory-the degree of risk aversion, the degree of loss aversion, and a nonlinear probability weighting measure ).
In examining these hypotheses there are econometric challenges that must be addressed. For example, in manuscript 1, I aim to identify how heterogeneity in households' time and risk preferences may impact the average effect of forest plot certification on household forest management (H1). To identify this effect, the ideal would be to compare forest management outcomes under the counterfactual of no forest certification. But plots cannot both receive a forest certificate and not receive a forest certificate, and so actual counterfactuals cannot be observed. Instead we need to estimate the value of this unobserved counterfactual's outcomes by obtaining a comparison group of plots that did not receive a forest certificate. The identification problem is that it is difficult to identify a reliable comparison group for those receiving a forest certificate because of non-random placement of forest plot certification and/or self-selection of households into forest plot certification. Without a carefully selected comparison group, we risk incorrectly attributing differences in measured forest management outcomes between those plots for which households received a forest certificate and plots for which households did not receive a forest certificate to forest plot certification when in fact they may be due to initial differences in observed (e.g., education of the head of household) and unobserved characteristics (e.g., entrepreneurial ability) between the two groups . To address these sources of biases, we use a variety of econometric techniques throughout the three manuscripts, including: preprocessed matched data in a difference-in-differences framework, fixed effects, and instrumental variable approach.
The outcome of this research has implications for policymakers in China and elsewhere by informing when they can expect property right reforms to stimulate investment in the resource and when they may not as a result of heterogeneity in households' risk and time preferences. The results may indicate that instruments to deal with risk, time preferences, and poverty need to be coupled with such reforms.
Although this research is conducted in the context of forests, the general finding may also apply to other natural resources where lack of property rights have been recognized as a key barrier to sustainable management of natural resources. Notes: T* indicates the optimal rotation age under well-defined property rights. This figure assumes that the optimal rotation time with well-defined property rights is longer than with weaker property rights and that prior to the reform the rotation is shorter than optimal. I* is the optimal steady state investment level. (a) indicates a trajectory in which a household instantaneously increases investment to the optimal level after the tenure reform. Note: T* indicates the optimal rotation age under well-defined property rights. This figure assumes that the optimal rotation time with well-defined property rights is longer than with weaker property rights and that prior to the reform the rotation is shorter than optimal. R* indicates the optimal steady state forest revenue level. (a) indicates a transition period after the tenure reform during which the forest revenue increases towards the new steady state level. (b) indicates a trajectory where the tenure reform does not affect forest revenue. (c) is a trajectory where forest revenue falls temporarily during the transition period.  Results show that risk and time preferences impact households' forest management responses to forest plot certification. Specifically, in response to forest certification, more risk averse households used less labor for harvesting and more labor for applying forest inputs, while more loss averse households used more labor for harvesting.
Households with higher discount rates (i.e., stronger preference for income today) used less labor for applying inputs and spent less on forest inputs in response to forest certification.

Introduction
Forest degradation has steadily increased throughout much of the world . The cause of this continued degradation is complex and multifaceted but there is a growing realization that a key cause, especially in developing countries, is the insecurity of rights to ownership and use of forest resources FAO 2007;. Property rights to ownership and use of forest resources are often contested, overlapping or unenforced. This insecurity undermines sound forest management, for without secure rights forest holders have few incentives to invest in managing and protecting their forest resources. This realization has stimulated the recent trend in forest policy toward strengthening property rights for forest resources by transferring property rights from the state to communities and individuals, giving them defined rights to manage and extract forest resources FAO 2003;). In the most forested developing countries, this trend has resulted in a doubling of the percent of forest owned or controlled by indigenous and rural communities between 1985 and 2000 . By 2050, 40% of the world's forest is expected to be managed or owned by communities and individuals (FAO 2003). However, a puzzle remains unsolved-such reforms on property rights have not consistently led to the intended sustainable resource use and management, particularly in developing countries .
Despite the existence of this puzzle, a lack of attention has been given to understanding the heterogeneity in how people respond to property rights reforms depending on their individual preferences. In particular, given that forest management decisions need to be made by looking into the future and therefore inherently contain uncertainties (e.g., price uncertainty; uncertainty about future growth and quality of retained stands; uncertainty about property rights and expropriation; uncertainty associated with outbreaks of disease, pests, and forest fire and the occurrence of extreme weather events), the two key factors that would influence a forest management decisions are: 1) the household's preference for income today versus in the future (time preference), and (2) the household's attitude towards risk (risk preference).
In this paper, we examine how households' preferences over time (present vs. future income) and risk affect forest management responses to property rights reforms.  (Godoy et al. 1998;Godoy et al. 2001;Hagos et al. 2006). Our time preference experiment uses methods originally developed by  and Harrison and Lau (2002). The data are then used to estimate three parameters in a general time discounting model using nonlinear least-squares . Furthermore, our risk preference experiment design follows a recently developed methodology that expands the classic lottery experiment of Holt and Laury (2002) to allow for the estimation of a more flexible and richer description of a person's risk preference as described under prospect theory . To capture risk preferences, we use the data to estimate three parameters: the degree of risk aversion, the degree of loss aversion and a nonlinear probability weighting measure. The combined experiment and household survey data allow us to link behavior elicited in experiments to actual economic institutions and performance, which few studies have previously done .
To identify the effect of forest tenure reform and how risk and time preferences augment the effect, we use matching techniques to preprocess the data (Ho et al. 2007) and then use the preprocessed matched data in a difference-in-differences framework.
The strategy capitalizes on the exogenous variation across villages of the starting year of the reform, and the resulting variation in the year households received a forest certificate for their plots. Results show that risk and time preferences impact households' forest management responses to forest plot certification. Specifically, in response to forest certification more risk averse households reduced labor for harvesting more and increased labor for applying inputs more, while more loss averse households increased labor for harvesting more. As such, the results of this paper have implications for policymakers in China and elsewhere by informing them about how heterogeneity in households' preferences may impact the outcomes of property right reforms.
This paper proceeds as follows. The first section gives an overview of the forest tenure reform history in China, with an emphasis on the recent tenure reforms in Fujian, China. The next section provides an overview of the most relevant literature, followed by the hypotheses to be tested. An explanation of the data collection procedures and a description of the data follow. Then the empirical framework is outlined, followed by the results and a conclusion.  ).
The first major wave of reforms in China's collective forests began in 1981, and was aimed at transferring the responsibility of forest planting and management from the collective to households . By 1986, nearly 70% of collectively owned forest land had been transferred to rural household management ). In 1987, however, due the occurrence of unsustainable logging the government reverted a large portion of forest land under household management back to collective management .
By 1986, while 70% of the collectively owned forest land in China had been transferred to rural household management, in Fujian only 32% of the collective forest land had been distributed for household management . This low percentage of forest land under household management was due to the fact that Fujian had not fully participated in the first round of the tenure reforms in the 1980s. Instead, the provincial government in Fujian had implemented a shareholding system to keep forests under collective management while distributing "paper shares" of collective forests based on family population. In Fujian, forest land was not actually physically distributed, rather only dividends from the forest were distributed to households. At first Fujian's shareholding system was highly regarded by forest administrators for its ability to maintain forests under collective management but fifteen years after establishment of the system, two issues became increasingly evident ).
First, forestry's contribution to rural incomes was negligible in spite of the fact that forest land occupies more than 60% of the total provincial land area, and 80% of rural land area . Second, enforcing forest conservation had become increasingly difficult for local forest authorities due to lack of cooperation from farmers. For example, the severity of forest fire incidents grew over the course of the 1990s, and there is anecdotal evidence that many of the fires were caused by farmers   World Bank 2000). These common characteristics of the poor not only have traditionally been cited as reasons why the poor remain poor but they may also hinder the anticipated outcome of creating and strengthening forest property rights through the issuance of forest certificates for household forest plots. Therefore, it is of critical importance to understand the effect of time and risk preferences on household forest management responses to changes in forest property rights.

Property Rights, Individual Preferences, and Natural Resource Management
Economic theory predicts that if a natural resource is open access, individual extractors do not fully incorporate the resource cost associated with current extraction and thus the resource is overexploited (Gordon 1954;Hardin 1968). Moreover, without secure property rights, individuals lack long-term incentives to use their forest resources . In rural areas of poor countries, many forests are subject to open-access extraction even if the government has the property right for the forest because property rights are difficult and costly to enforce ). This lack of secure property rights is recognized as one of the key underlying causes of continued forest degradation in many parts of the world. In response to this recognition, many governments have begun to reform forest ownership policies by devolving resource management to the local level, giving individuals or communities rights to manage and extract the resources FAO 2003).
Research to date, however, presents conflicting conclusions regarding the impacts of tenure security reform on forest management decisions. For example,  and Holden et al. (2009)   .
This paper examines why tenure reforms may not work as intended from a microeconomics perspective, focusing on how risk and time preferences augment individual households' responses to forest property rights reforms. In forest management, households must make decisions about investments over a long time horizon. Furthermore, forest management decisions involve uncertainty over prices and about future growth and quality of retained stands and various production risks such as outbreaks of disease, pests, and forest fire; and the occurrence of extreme weather events (e.g., blizzards, flooding, earthquakes, etc.) as well as uncertainty about property rights and possible expropriation (Alvarez and Koskela 2004;. Specifically in China, the problems of pests and disease are extremely serious with increasing types, expanding affected areas, and shortening of intervals between attacks, as well as threats from forest fire (Kunshan et al. 1997; Yet to our knowledge, no previous study has directly examined how risk and time preferences affect household responses to property right reforms. This paper extends the literature on property rights reform by using a large-scale property rights reform to examine the heterogeneity of its impact due to risk and time preferences.
The most relevant set of previous work includes Godoy et al.'s (1998; studies in Bolivia on how tenure insecurity and rate of time preference affect forest resource harvesting and Hagos et al.'s (2006) study in Ethiopia on how tenure insecurity and time and risk preferences affect investment in land conservation. Godoy et al. (1998; use the duration of a household's residence in the village and the number of conflicts with abutters as proxies for tenure security, and find mixed results. The length of residence in the village was associated with a lower area of old-growth forest cleared but with a greater area of fallow forest cleared, whereas the opposite was true for conflict with abutters. Conflict was associated with a smaller area of fallow forest cleared but with a greater area of old-growth forest cut. Hagos et al. (2006) find that neither the degree of tenure security nor individual time and risk preferences explains the differences in land conservation and investment decisions.
Although these studies are informative, the measures of risk and time preferences and tenure security need to be improved to achieve stronger confidence in the estimates. Hagos et al. (2006) elicited risk and time preferences from households using hypothetical questions. Results using this method to elicit risk and time preferences may suffer from hypothetical bias, which means that people respond differently when the situation is hypothetical than when the situation is real ). By designing questions or experiments to elicit time and risk preferences that offer subjects real payoffs based on their choices, hypothetical bias can be reduced. (Smith et al. 1993;. Godoy et al. (1998) elicited time preferences by asking subjects if they would prefer one piece of candy now (at the midpoint of an interview) or two at the end of the interview. Although this method involves a real reward, the authors acknowledge that the choice of candy to measure time preference over a very short time may not capture with accuracy time preference or commitment for economic investments, which take place over a longer stretch of time, such as for forest resources. Godoy et al. (2001) elicited risk preferences using hypothetical questions but elicited time preferences using a series of choices with real monetary payoffs. In this paper, we use risk and time preference parameters elicited using economic experiments with real monetary payoffs; therefore the hypothetical bias is reduced.
In addition, the measures of tenure security also require improvement to achieve more reliable estimates. These studies use proxy variables such as number of conflicts with abutters and duration of residence in a village (Godoy et al. 1998;Godoy et al. 2001;Hagos and Holden 2006). These proxies are either subjective or indirect measures and may not accurately measure tenure security. In this paper, we utilize actual changes in forest property rights and use a more explicit and discrete measure of forest property right changes.
Furthermore, we estimate the joint effect of risk and time preferences on individual responses to changes in property rights, which none of these previous studies have done. Interacting the risk and time preferences with changes in forest property rights allows us to capture how risk and time preferences augment forest management decisions in response to changes in forest property rights.

Hypotheses
Hypothesis (1) is based on the theory that increased tenure security gives households an incentive to invest in their forest resources; that is to increase labor allocation for applying inputs, to increase expenditure on inputs, and to delay harvest Hypothesis 1: The estimated forest certification effect (the conditional average difference in each forest management activity on plots for which a household has a forest certificate) will be positive when the dependent forest management variable is the value of labor used to apply inputs or expenditure on inputs and negative when it is the value of labor used for harvesting forest product.
until the optimal harvest time . A forest certificate increases a household's tenure security. Increased tenure security gives households confidence that if they invest in their plot (planting, maintenance, etc.) then they will be able to obtain the benefits from those efforts in the future. As such, households that receive a forest certificate for a plot will use more labor to apply inputs, spend more on inputs and will delay harvest until the optimal harvesting threshold is reached.
Hypothesis 2: Risk and time preferences augment households' responses to forest property right reforms.
Households making forest management decisions face many uncertainties such as those related to prices, growth and quality of retained stands, redistribution of forest land, outbreaks of disease, pest infestations, forest fire, and extreme weather events.
Furthermore, decisions about forest management often involve a long time horizon ). As such, households' risk and time preferences play an important role in their forest management decisions , and by extension will affect their responses to forest property right reforms.
To model risk preferences we use prospect theory because it allows for the estimation of a more flexible and richer description of a person's risk preferences than under expected utility theory. Most previous risk preference experiments conducted in the field are based on the expected utility theory notion of risk preferences but these models often fit experimental and field data less well than models with multiple components of risk preference . In expected utility theory, an individual's risk preferences are solely characterized by the concavity of the utility function and are classified as risk averse, risk neutral or risk seeking. In contrast, prospect theory allows for the possibility that an individual may be risk averse, risk neutral or risk seeking, depending on whether choices involve gains or losses and whether the probabilities of gains or losses are large or small . Under prospect theory, an individual's risk preferences are described by three measures: the degree of risk aversion, the degree of loss aversion, and a nonlinear probability weighting measure. We use these three parameters to represent a household's risk preferences. Hypotheses 2a thru 2f describe our hypotheses regarding each of these parameters.
Hypothesis 2a: A more risk averse household will allocate less labor for application of forest inputs, spend less on forest inputs and allocate more labor to harvesting.
Hypothesis 2b: A more risk averse household will exhibit a stronger behavioral response to forest certification (allocate more labor to application of forest inputs, spend more on forest inputs, and allocate less labor to harvesting).
With regard to a household's degree of risk aversion, we hypothesize that a more risk averse household will be less likely to assume the risks associated with forest production (such as potential loss of forest stock due to pests, disease, illegal logging, natural disaster, redistribution of property, etc.) therefore, a more risk averse household will allocate less labor for application of forest inputs, spend less on forest inputs, and allocate more labor to harvesting. This hypothesis is based on the theory that higher risk aversion decreases the optimal harvesting threshold, which has been the main conclusion in most studies dealing with forest management under production risk . However, it should be noted that some studies have found that the effect of risk aversion on the optimal rotation is ambiguous and depends on economic and biological parameters, as well as how risk is modeled .
Furthermore, we hypothesize that more risk averse households will exhibit a stronger behavioral response to forest certification. Assuming that a forest certificate reduces the risks associated with loss of forest stock due to redistribution of property, a more risk averse household that receives a forest certificate for a plot will respond to that reduction in risk by allocating more labor to application of forest inputs, spending more on forest inputs, and allocating less labor to harvesting forest products than risk neutral or risk-seeking households that receive a forest certificate.
Hypothesis 2c: A more loss averse household will allocate less labor for application of forest inputs, spend less on forest inputs, and allocate more labor to harvesting.
Hypothesis 2d: A more loss averse household that receives a forest certificate for a plot will allocate less labor for application of forest inputs, spend less on forest inputs, and allocate more labor to harvesting Loss aversion refers to an individual's tendency to strongly prefer avoiding losses to acquiring gains. Furthermore, people have a tendency to dramatically overweight losses relative to gains .
In forest management, households make management decisions involving potential losses and gains. Psychologically, losses may overshadow objectively commensurate gains in evaluation of prospects (Kahneman, Knetch and Thaler 1990). In general, individuals tend to be more reluctant to accept an uncertain gain over a more certain, albeit lower gain. As a result, households may not invest in forests or may harvest prior to the optimal harvesting threshold (when in actuality investments in the forest resource and delaying harvest until the optimal harvesting threshold would be beneficial). We therefore hypothesize that a more loss averse household will allocate less labor for application of forest inputs, spend less on forest inputs, and allocate more labor to harvesting.
To understand how loss aversion may impact a household's response to forest certification we consider the potential endowment effect of forest certification. As described by , an endowment effect "is produced, apparently instantaneously, by giving an individual property rights over a consumption good." As a result of the endowment effect, households may be more averse to loss of forest stock from the plot with a forest certificate than from a plot without a forest certificate. Therefore, we hypothesize that more loss averse households that receive a forest certificate for a plot will allocate less labor for application of forest inputs, spend less on forest inputs, and allocate more labor to harvesting because they will be more averse to potential loss of forest stock from a plot with a forest certificate than to loss from a plot without a forest certificate.
Hypothesis 2e: The effects of the probability weighting parameter on forest management are ambiguous.
Hypothesis 2f: The effects of the probability weighting parameter on forest management responses to receiving a forest certificate are ambiguous.
The probability weighting parameter indicates whether or not an individual puts excessive decision weight on small probabilities ). Since we do not know whether the actual probabilities that households may lose their forest stock to such events as pest infestation, disease, illegal logging, natural disaster, or redistribution of property rights are high or low, we cannot hypothesize about how a household's tendency to excessively weight small probabilities will affect its decision making process on its forest management decisions or on its responses to forest certification.
In addition to risk preferences, we examine how a household's time preference affects its forest management decisions and responses to receiving a forest certificate.
To represent each household's time preference we use a discount rate. There are several competing models for time discounting that have received a significant amount of attention in both experimental psychology (e.g., de Villiers and Herrnstein (1976), Ainslie and Haslam (1992), etc.) and behavioral economics (e.g., Laisbons (1997), , O'Donoghue and Rabin (1999). 3 The competing models were developed to account for observed behavioral regularities that are not consistent with the classic exponential discounting model. For example, the most common documented behavioral regularity is called "reversal of preferences." It occurs, for example, when a subject prefers $10 now rather than $12 a day later, but also prefers $12 in a year plus a day rather than $10 in a year. This type of preference is not consistent with exponential discounting but would be consistent with a rate of time preference that declines with time such as hyperbolic discounting. We use a hyperbolic discounting parameter because we find that the hyperbolic discounting functional form fits our data better than the exponential discounting functional form (constant discount rate). Similar to our findings, other studies have found that the hyperbolic discounting functional form fits field data better than the exponential discounting functional form (Rachlin, Raineri and Cross 1991;. Hypotheses 2g and 2h describe our expectations related to a household's discount rate.
Hypothesis 2g: Households with higher discount rates will allocate less labor to applying inputs, spend less on forest inputs, and allocate more labor to harvesting.
Hypothesis 2h: Households with higher discount rates that receive a forest certificate for a plot will allocate less labor for application of inputs, spend less on forest inputs and allocate more labor to harvesting.
Forest management decisions often have a long time horizon, making households' time preferences (i.e., preference between immediate income and future income) important in the decision making process. In forest management, the Faustmann model is best known for providing a benchmark model for determining optimal timber rotation age ). In the model, a forest owner's goal is to choose the rotation period that maximizes the net present value of the forest. In the infinite rotation model, the decision rule is to harvest when the marginal benefit of delaying (new growth) is equal to the marginal cost of delaying (lost interest on the timber revenue and on future stands). An increase in the interest rate will tend to shorten the optimal rotation length. As such, we hypothesize that households with higher discount rates (i.e., impatient) will shorten the optimal rotation length and allocate labor to harvesting more frequently. Furthermore, we hypothesize that households with stronger preference for consumption or income today (i.e., higher discount rate) will allocate less labor to applying inputs and spend less on forest inputs, as other shortterm return investment opportunities will be more attractive than the long-term returns from investing in forest resources. Additionally, households with a stronger preference for income today that receive a forest certificate for a plot will allocate less labor for application of inputs, spend less on forest inputs, and allocate more labor to harvesting than households with weaker preferences for income today that receive a forest certificate. To capture household forest management, we use the value of labor used for applying forest inputs to each plot, expenditure on inputs for each plot, and the value of labor used for harvesting from each plot. The expenditure on inputs includes expenditure on fertilizer, irrigation, animal or machinery rental fees, seeds and other forest inputs. The two labor-related outcome variables are based on the sum of the annual value of family and exchanged labor and the annual expenditure on hired labor for applying forest inputs and for harvesting forest products. The annual expenditure of hired labor is calculated based on responses to survey questions regarding the number of working days of hired labor and the wage per working day paid to hired labor for each forest management activity. For the annual value of family and exchanged labor, we sum the responses to the survey question regarding the number of working days of family and exchanged labor for each forest management activity. We then multiply the total number of family and exchanged labor working days times the average county wage paid to hired forest labor based our survey data, and use the resulting value as a proxy for the opportunity cost of a household's time. We recognize that an estimated shadow wage would be a more accurate measure of a household's opportunity cost of time spent laboring on its forest plot; however, the data necessary to estimate a shadow wage are not available  As a preview to more rigorous estimates of forest plot certification effects, we examine the descriptive statistics for the forest management variables by whether or not a household has received a forest certificate for its plot. Interestingly, we find that the change in the mean value of labor used for harvesting forest products and for applying forest inputs is statistically different at the 1% and 10% significance level, respectively, indicating that forest plot certification had an effect on households decisions regarding allocation of labor to their forest plot (

Risk Preference Data
To elicit a measure of risk preference, we follow the experimental design developed by  and later modified by , both of whom expand the classic Accept/Reject lottery experiments of Holt and Laury (2002) to incorporate prospect theory. We use cumulative prospect theory and a non-linear probability weighting measure extended from the one-parameter form of Drazen  axiomatically-derived weighting function . Following , we assume a utility function of the following form: where U(x,p; y,q) denotes the expected prospect value over binary prospects consisting of the outcomes x and y with the probability of p and q, respectively. The function v(x) denotes a power value function. 6 The parameter σ describes the curvature of an individual's value function. An individual's risk preferences are described as risk averse when σ > 0, risk neutral when σ = 0, and risk loving when σ < 0. The parameter λ describes the curvature of an individual's value function above zero relative to the curvature of the value function below zero. The higher the value of λ, the more loss averse the individual is. The parameter α is a non-linear probability weighting measure, which is extended from a model by . The probabilities are weighted by the function π(p). When α <1, π(p) has an inverted S-shape, indicating that an individual tends to overweight low probabilities and underweight high probabilities, as shown by . This model reduces to expected utility theory when α = 1 and λ=1.
In the experiment, participants were asked to choose between sets of lottery options. For example, Figure 1.1 illustrates one set of options that a subject was asked to choose between. In this example, Option A offers a 30% chance of receiving 20 yuan and a 70% chance of receiving 5 yuan. Option B offers a 10% chance of receiving 34 yuan and a 90% chance of receiving 2.5 yuan. A total of 35 choices, divided between three series were asked. The payoffs ranged from a loss of 10 yuan to a gain of 850 yuan, which is roughly half a months pay in rural China (CSY 2009). If a subject was illiterate (27% of our sample), then the enumerator read the choice to the subject and recorded the subject's answers on the record sheet. Monotonic switching was enforced, meaning that once the subject switched to option B they were not allowed to switch back to option A. 7 By enforcing monotonic switching, we eliminate the possibility of inconsistent choices within each series and also make the task more clear and concise for participants, as they only need to identify one switch point in each series. 8 Once the subject had completed the entire series of choices, one question was chosen randomly for payoff.
In our sample, the average derived values for α and λ are 0.73 and 6.02, respectively, and both are statistically different from 1 at the 1% significance level by t-test, implying that our experimental results reject expected utility theory in favor of prospect theory's inverted S-shaped probability weighting and loss aversion. We use the individual values for σ (degree of risk aversion), λ (degree of loss aversion) and α (nonlinear probability weighting measure) to represent the risk preferences of each household in our empirical model, which will be discussed in section 1.7. 11

Time Preference Data
Our time experiment design follows the methods originally developed by  and Harrison, Williams and Lau (2002). The data are then used to estimate three parameters-the conventional time discounting parameter (r), present-bias (β), and hyperbolicity of the discount function (θ)-in a general time discounting model using nonlinear least-squares, which allows us to test which discounting model fits the data best-exponential, hyperbolic, quasi-hyperbolic, or a more general form ).
In the time preference experiment subjects were asked to choose between, for example, a real monetary payoff today or a larger payoff six months from now. The hypothetical bias of earlier studies that aim to capture time preferences is addressed here because participants received a real monetary payment based on their choices.
Choices were always posed as a choice between a monetary payoff today versus a larger monetary payoff in the future. 12 To ensure the credibility of a future payment, subjects were told that the future payments would be delivered by China Post, which is the official postal service of the Peoples Republic of China, an agency with which rural households are very familiar and comfortable using for the delivery of money. Furthermore, we believed the credibility problem to be minimal because our participants were part of a panel survey and this was the second time that the household had been visited by a research team from Peking University. Repeat visits by our research team built trust with and provided reassurance to the participants.
Following the experimental design of , the subjects were asked a total of 75 questions divided into 15 series of 5 questions each. 13 A single series of questions is depicted in Figure 1.3. In this example, the subject was asked to choose Plan A or Plan B for each of the 5 questions. Plan A, the future payoff plan remained the same for each question in the series, while the immediate option increased as the subject moved down the column from 25 yuan to 125 yuan, at 1/6 increments of the future payoff. As in the risk experiment, monotonic switching within each series was also enforced here. The point at which an individual switches from choosing the more immediate reward to taking the delayed reward provides a bound on his or her discount rate. The discount rate indicates the rate that would make a person indifferent between the immediate and the delayed reward. An individual with a high discount rate has a preference for the present, whereas an individual with a low discount rate has a preference for the future.
We used 15 combinations of future payoff and time in the experiments; that is 15, 60 and 150 yuan with delays of 2 weeks, 3 months, and 6 months and 30 and 120 yuan with delays of 1 week, 2 months and 4 months. 14 The maximum payoff of 150 yuan is equal to roughly 2 to 3 days pay in rural China  Therefore, we focus on the estimates from the hyperbolic model and use those parameters to represent the time preference of each household in our empirical model, which will be discussed in section 1.7. Figure 1.4 depicts the distribution of the hyperbolic discounting parameter, from our experiment. Surprisingly, the figure shows that the hyperbolic time preference parameter was relatively low for the majority of our subject, indicating that they have a relatively weak preference for income today. In the hyperbolic discounting model, we find that on average a subject would be willing to trade 92 yuan today for 100 in 1 week, 74 yuan today for 100 yuan in 1 month and 32 yuan today for 100 yuan in 6 months.

Empirical Framework
Our objective is to identify how heterogeneity in households' time and risk preferences may impact the average effect of forest plot certification on household forest management. The ideal would be to compare forest management outcomes under the counterfactual of no forest certification. But plots cannot both receive a forest certificate and not receive a forest certificate, and so actual counterfactuals cannot be observed. Instead we need to estimate the value of this unobserved counterfactual's outcomes by obtaining a comparison group of plots that did not receive a forest certificate. The identification problem is that it is difficult to identify a reliable comparison group for those receiving a forest certificate because of nonrandom placement of forest plot certification and/or self-selection of households into forest plot certification. Without a carefully selected comparison group, we risk incorrectly attributing differences in measured forest management outcomes between those plots for which households received a forest certificate and plots for which households did not receive a forest certificate to forest plot certification when in fact differences may be due to initial differences in observed (e.g., education of the head of household) and unobserved characteristics (e.g., entrepreneurial ability) between the two groups .

Identification Strategy
In this study we use a two-step approach to reduce estimator bias caused by potential self-selection of households into forest plot certification. In the first step, we preprocess the data set with nonparametric matching methods so that the treated group (plots for which a household received a forest certificate) is as similar as possible to the control group (plots for which a household did not receive a forest certificate) to reduce estimator bias caused by potential self-selection of households into forest plot certification based on observed characteristics (Ho et al. 2007). The goal of matching is to create a data set that looks closer to one that would result from a randomized experiment. When we get close, we break the link between the treatment variable and the pretreatment controls, which makes the parametric form of the analysis model less relevant or irrelevant entirely. To break this link, we need the distribution of covariates to be the same within the matched treated and control groups.
Specifically, we divide all the plots into two groups: plots that received a forest certificate and plots that did not receive a forest certificate. We then use 1-to-1 nearest neighbor matching (without replacement) to match each plot that received a forest certificate ("treated plot") with a plot that did not receive a forest certificate ("control plot") based on the propensity score (the predicted probability of forest plot certification). The variables used to estimate the propensity score in a logistic regression include three household level variables (age of household head, household head's education level, and the household's total land holdings) and four plot level variables (distance from plot to home, distance from plot to the road, slope of the plot, and whether the plot's forest type is primarily bamboo). Once the propensity score is estimated, a comparison observation for each treated observation is created by choosing the "nearest neighbor", which is the untreated household with the closest propensity score. Control observations that are not matched are discarded. This reduced our sample to 134 plots owned by 69 households. Following Ho et al. (2007), we selected the matching method that produced the best covariate balance with each treated plot. As a result, in the preprocessed data set, the treatment variable is closer to being independent of other covariates, which helps us obtain more accurate causal effect estimates in the parametric model.
In the second step, using the preprocessed matched data we exploit plot-level variation in the year that households received a forest certificate for a plot in a difference-in-differences framework. The variation in the year that the household received a forest certificate is the result of exogenous variation across villages of the starting year of the reform. Using this framework, we can compare the before-after changes in forest management activities on those plots for which households received a forest certificate (the treatment group) to the before-after changes in forest management activities on those plots that households did not receive forest certificates (the control group). The difference-in-differences framework allows us to difference out any common trends between the treatment and the control group.
In summary, we use a two-step approach in which we preprocess the data using matching methods and then use that preprocessed data in a difference-in-differences framework in order to obtain more robust estimates of the forest plot certification effect on households' forest management and how that effect may vary depending on heterogeneity in time and risk preferences of each household.

Empirical Model
The base estimate of the forest certification effect is obtained from the differencein-differences estimation using the preprocessed data: where forest management ijt refers to each of the three forest management related dependent variables: the value of labor used for applying inputs (input labor ijt ); expenditure on forest inputs including chemical fertilizer, pesticide and seeds (inputs ijt ); and the value of labor used for harvesting (harvest labor ijt ) by household i on forest plot j at time t. All forest management variables are measured in yuan per hectare at the plot level. fcert ij is a dummy variable that is equal to one if household i had a forest certificate for plot j in any year. The coefficient on fcert ij controls for characteristics that may differ between plots that received forest certificates during the recent tenure reform and plots that did not. year2005 t and year2008 t are dummy variables that take the value one if the observation is for the year 2005 and 2008, respectively. The coefficients on year2005 t and year2008 t control for any systematic differences for years 2005 and 2008, respectively. AfterReform ijt is a dummy variable that takes the value one when household i has a forest certificate for plot j in a postreform year. The coefficient on AfterReform ijt is the estimated forest certification effect, which provides a measure of the conditional average difference in forest management activities on plots for which a household has a forest certificate.
To test our main hypothesis that time and risk preferences affect how households respond to property right reforms, we add the risk and time preference parameters and their interaction variables to equation (2) to capture the interaction effects between the risk and time preference variables and the change in forest certification status (AfterReform ijt ). Our main difference-in-differences model is: risk i is the risk aversion parameter; loss i is the loss aversion parameter; probweight i is a dummy variable that takes the value one if the probability weighting parameter is greater than one, indicating that individuals place excessive decision weight on small probabilities; and timepref i is the hyperbolic time discounting parameter for household i. The interaction terms (risk*AfterReform ijt , loss*AfterReform ijt , probweight *AfterReform ijt , and timepref*AfterReform ijt ) capture heterogeneity of the treatment effect due to households' risk and time preferences. For example, risk*AfterReform ijt picks up any differential patterns in changes in household forest management activities on plots that receive a forest certificate relative to plots that do not receive a forest certificate that may be correlated with the households' risk preferences. The interaction term timepref*AfterReform ijt, picks up any differential patterns in changes in household forest management activities on plots that receive a forest certificate relative to plots that do not receive a forest certificate that may be correlated with households' time preferences. X i is a vector of demographic controls, P ijt is a vector of plot characteristic controls, and V v is village fixed effects. Table 3 identifies each of the control variables used in this analysis. 15 For a better fit, we estimate a log transformation of equations (2) and (3) for each of the three forest management dependent variables. Table 1.5 summarizes our hypotheses from section 1.5 in terms of the sign of the estimated coefficients in equation (3).

Empirical Results
Overall we find that there is evidence that risk and time preferences impact households' forest management responses to forest plot certification (tables 1.6, 1.7 and 1.8; columns 3 and 4). 16

Impact on labor used for harvesting forest products
We hypothesized that the estimated certification effect (the conditional average difference in labor used for harvesting from plots with forest certificates) would be negative because increased tenure security from plot certification allows a household to have greater confidence towards future benefits, and hence delay harvest to allow the forest stock to grow larger. We do not find evidence of the hypothesized negative certification effect on labor used for harvesting (table 1.6). In all models the coefficient on AfterReform is negative but not statistically significant (columns 1-4).
When we allow the certification effect to vary with households' risk and time preferences, include both household and plot controls and village effects, and evaluate the estimate at the median values of the time and risk preferences parameters, the implied total certification effect on labor for harvesting is -2.64% but is not statistically significant (column 4). 17 Interestingly, when we allow the certification effect to vary with households' risk and time preferences, we find that the negative effect of certification on the value of labor allocated to harvest is larger for households that are more risk averse and smaller for those households that are more loss averse (column 3-4). Specifically, the interaction term between ln(risk) and AfterReform is -1.37%, suggesting that for a household with a risk parameter that is 10% higher (more risk averse), the certification effect on value of labor for harvesting is 14% less. And the coefficient on the interaction term between ln(loss aversion) and AfterReform is 1.34%, suggesting that for a household with a loss aversion parameter that is 10% higher (suggesting more loss averse), the certification effect on labor for harvesting is 13% more. This result implies that the intended effect of certification (reduce or delay harvest) is actually larger for more risk averse households and smaller for more loss averse households.
The certification effect did not vary statistically significantly with households' degree of time preference or their tendency to place excessive decision weight on small probabilities.
More generally, results indicate that households that are more risk averse or that tend to place excessive decision weight on small probabilities allocate more labor to harvesting forest products (table 1.6, columns 2-4).

Impact on expenditure and labor used for applying forest inputs
We find no evidence of a certification effect on either the expenditure on forest inputs or on labor used to apply forest inputs (tables 1.7 and 1.8). The implied total effect of certification is insignificant for both dependent variables and the signs are mixed.
However, when the estimation effect is allowed to vary with households' risk and time preferences, we find that for a household with a risk parameter that is 10% higher (more risk averse), the certification effect on labor used for applying inputs is 5.6% lower (table1. 7, columns 3 and 4). Also, we find that for a household with a time preference parameter that is 10% higher (stronger preference for income today), the forest certification effect on labor for applying inputs and expenditure on forest inputs is 9% and 14% lower, respectively (table 1.7 and 1.8, columns 3 and 4).
More generally, results indicate that households that are more risk averse tend to use less labor for applying inputs and have lower expenditure on forest inputs (tables 1.7 and 1.8, columns 2-4).

Robustness Checks
To check the robustness of our results, we run three additional variations of equation (3). First, we estimate the model using the number of days rather than the value of labor used for applying inputs and for harvesting (appendix tables 1.3 and 1.4). Second, we estimate the model using the exponential time discounting parameter instead of the hyperbolic time discounting parameter (appendix tables 1.5, 1.6 and 1.7). Third, we estimate the model using the number of years since the household received a forest certificate for a plot rather than the dummy variable, AfterReform ijt , that takes the value one when household i has a forest certificate for plot j in a postreform year (appendix tables 1.8, 1.9 and 1.10). We find that the results are robust to these alternative specifications with one exception. The exception is that when we estimate the model using the number of years since the household received a forest certificate for a plot rather than the dummy variable, the coefficient on the interaction variable between the years since the household received a forest certificate for a plot and the hyperbolic discounting parameter becomes insignificant in the full model.

Conclusion
Despite their potential importance, the heterogeneity in response to property rights reforms due to individual preferences has not been studied adequately. Progress is constrained by a lack of data. Measures of outcomes (such as forest investment, harvesting of timber, etc.) are difficult to come by and eliciting measurement of risk and time preferences is difficult (Frederick et al. 2002;. Furthermore, previous studies on tenure issues often use proxies to measure tenure security that are either subjective or indirect and may not accurately measure tenure security (Godoy et al. 1998;Godoy et al. 2001;Hagos and Holden 2006).
In this paper, we examined how preferences over time and risk affect household plot certification on forest management activities, a two step approach in which we preprocess the data using matching methods and then use that preprocessed data in a difference-in-differences framework in order to obtain more robust estimates of the forest plot certification effect on households' forest management and how that effect may vary depending on heterogeneity in the time and risk preferences of households.
Results suggest that more secure tenure as a result of forest certification affects households' forest management decisions. Although forest certification led to a decrease in labor allocated to harvesting as expected, surprisingly there was no evidence that forest plot certification led to an increase in labor used to apply forest inputs nor in forest input expenditure. The insignificant certification effect on labor used to apply forest inputs and forest input expenditure suggests that further research should examine whether or not households face credit constraints that prevent them from increasing investment on their forest plots in response to increased tenure security.
Results suggest that household preferences, particularly households' degree of risk aversion, affect the impacts of forest tenure reforms. According to our results, the negative impact of forest certification on labor allocated to harvesting was smaller for households that were more risk averse. This indicates that when households are risk averse, forest certification will be more likely to have the intended effect of households reducing or delaying forest product harvests. Furthermore, we find that the certification effect on labor for applying inputs is positive for households that are more risk averse.
The results indicate that households with a higher preference for income today that received a forest certificate used less labor for applying inputs and spent less on forest inputs than those with a lower preference for income today that received a forest certificate. Time preferences did not significantly augment labor for harvesting. The 7 Three examples were given in the instructions to help ensure that the subjects did not feel that they must make a switch within the series. In one example, the subject never switches to Option B. In another example, the subject switches at question 7 to Option B. And in a third example, the subject switches to Option B at question 1. front-end delay, we would lose information about how individuals treat choices between payoffs that are truly immediate versus payoffs that are not immediate.
Ideally, to address the credibility problem, while still having a way to capture the information about choices between immediate payoffs and future payoffs, an experimental design would include both questions with and without front-end delays.
Due to time constraints, in that participants may become exhausted with too many questions, we choose to only use questions without a front-end delay. 13 To see the entire set of payoff-time combinations that were used in the experiment and more details regarding the estimation of the time discounting parameters see Manuscript 2 and its appendix.
This paper investigates the correlation between poverty and individual preferences for time and risk. Specifically, we use field experiment data collected in Fujian, China to measure the time and risk preferences of 103 rural households combined with household survey data to examine the correlation between wealth and risk and time preferences. This relationship is important in understanding this 'vicious circle' because two key elements in many versions of this 'vicious circle' are that those living in poverty have both high levels of risk aversion and high rates of impatience Lipton 1968;Lumley 1997;Fafchamps 2003). For example, regarding time preferences, Irving Fisher wrote, A small income, other things being equal, tends to produce a high rate of impatience, partly from the thought that provision for the present is necessary both for the present itself and for the future as well, and partly from lack of foresight and self-control. (Fisher 1930, p.73) And with regard to risk preferences, Michael Lipton wrote, "The risk premium is an increasing function of risk and a decreasing function of assets." (Lipton 1968, p.335) In other words, the poorer a household, the more impatient they are and the more they seek to avoid risk. This makes it difficult, if not impossible, for these households to save and take the risks necessary to begin to accumulate capital. Therefore, the manner in which individuals discount the future and make decisions that involve risks are important for understanding behavior in developing countries.
The 'vicious circle of poverty' also has implications for the linkages between poverty and the environment. People everywhere consume water, food, energy and other natural resources in order to live, and these productive activities deplete the same natural resources upon which people depend. This is particularly true for poor communities in developing countries where livelihoods are often entirely dependent upon the local environment. When basic needs cannot be met with resources derived from the local environment, or when those resources are used in an unsustainable manner, subsistence communities expand into other areas to meet their needs, often drawing on those resources until they too are depleted. Thus the downward cycle continues (WCED 1987;UNCED 1993;World Bank 1996). Furthermore, since the poor are characterized as risk averse and impatient (meaning a short planning time horizon), they may be less likely to invest in conservation and new technologies to protect their natural resource base (Mink 1993;Perring 1996).
Since Binswagner's early use of experimental economics to capture risk preferences in India in the 1980s and Pender's work also in India in the late 1990s, economists have been examining the correlation between poverty, risk and time preferences. However, empirical evidence on whether individual time and risk preferences vary with wealth has been inconclusive (Binswanger 1980;Pender 1996;. 2 Furthermore, most studies have focused on correlations, with few aiming to identify the direction of causality . Is one impatient and risk averse because they are poor? Or is one inhibited from escaping poverty because they are risk averse and impatient? As such, there is a need to examine the direction of causality between wealth and risk and time preferences. In this paper, we address the potential endogeneity of wealth and begin to explore the direction of causality by using an instrumental variable for wealth. We believe that this paper has several contributions. First, we add to the empirical literature aimed at understanding the linkages between risk and time preferences and poverty. Second, this is one of the few papers to examine risk preferences of rural Chinese households Carlsson et al. 2009;Gong et al. 2010). 3 Third, to our knowledge, this is the first paper to measure time preferences of rural Chinese households using field experiments with real monetary rewards.
Overall, we find that on average participants exhibited risk aversion, moderate loss aversion and relatively low discount rates (i.e., patience). There is little evidence that wealth, measured by net worth per capita, affects risk and loss aversion. However, there is weak evidence that households with more forest land per capita are less loss averse. Also, we find weak evidence that net worth per capita has a negative significant effect on the discount rate, indicating that households with higher net worth per capita have lower discount rates (i.e., more patient).

Previous Literature: Methods and Findings
Over the last three decades researchers have used a variety methods to measure both time and risk preferences. Binswagner (1980)  Since the 1980's several other researchers have followed Binswagner's Ordered Lottery Selection design to elicit risk preference measures from rural households in developing countries (e.g., Nielsen (2001) in Madagascar; Barr (2003) in Zimbabwe; Yesuf and Bluffstone (2009)  Similar to Binswagner (1980), Mosley and Verschoor (2005) and  found no significant correlation between wealth and risk aversion. Also, Mosley and Verschoor (2005) and  found that household income is not significantly correlated with risk aversion. However,  found that mean village income had a significant negative relationship with risk aversion, indicating that households living in wealthy villages are less risk averse. Yesuf and Bluffstone (2009) found a significant positive relationship between income and risk aversion, while Neilson (2001) and Wik (2004) found a significant negative relationship between income and risk aversion. 5 While many researchers have explored the correlation between poverty and risk aversion, few have examined the relationship between poverty and loss aversion . In expected utility theory, risk attitudes are solely described by the degree of risk aversion (the concavity of the utility function). Prospect theory allows for a broader description of risk attitudes by allowing for the possibility that individuals may be loss averse. Loss aversion refers to an individual's tendency to strongly prefer avoiding losses to acquiring gain, and describes the curvature of an individual's value function above zero relative to the curvature of the value function below zero . In China,  found that wealth is not significantly correlated with loss aversion. Similarly,  found in Vietnam that household income is not significantly correlated with loss aversion but that mean village income is highly correlated with loss aversion, indicating that households in poorer villages are more loss averse.

Time preference and poverty
Over the last three decades, researchers have also endeavored to measure time preferences by estimating a discount rate. Some discount rates have been derived from "real-world" behaviors while others have been derived from experimental elicitation  (2002).
As with the relationship between risk preferences and poverty, empirical finding related to the relationship between time preferences and poverty have also been mixed.
Again, as indicators of poverty, studies have used either wealth or income, or in some cases both. 7 Most studies use exponential discounting, however this model often does not fit experimental and field data well (Frederick et al. 2002). Pender (1996) found weak evidence in India that wealthier respondents had lower discount rates. Also, Neilsen (2001) in Madagascar and Yesuf and Bluffstone (2008) and Holden, Shiferaw, and Wik (1998) in Ethiopia both found that wealthier households had significantly lower discount rates, indicating that the poorer a household is, the more impatient they are. However, Bauer, Chytilová, and Morduch (2010) and Kirby et al. (2002) found that wealth was not correlated with the discount rate in India and Bolivia, respectively.
Findings in studies that use income rather than wealth as an indicator of poverty are also mixed.  in Vietnam found that both households with higher income and households that live in villages with higher mean incomes have significantly lower discount rates. Kirby et al. (2002) and Gunatilake, Wickramasinghe, and Abeygunawardena (2007) also find a negative significant relationship between income and the discount rate in Bolivia and Sri Lanka, respectively. However, Nielsen (2001), Anderson et al. (2004), and Bauer and Chytilová (2008) found that there was no statistically significant relationship between income and time preferences.

Survey Procedure and Data Description
The  A person can gain more face by having a beautiful house (Carlsson and Qin 2010). In our empirical work, we will use net worth per capita to measure household wealth.
Furthermore, we use both forest land per capita and sub-categories of net worth to check the robustness of our results.

Risk Experiment Design
To elicit a measure of risk preference, we follow the experimental design developed by , who expands the classic Accept/Reject Lotteries of Holt and Laury (2002) to incorporate prospect theory. Following,  we use cumulative prospect theory and a non-linear probability weighting measure extended from the one-parameter form of Drazen Prelec's axiomatically-derived weighting function ). We assume a utility function of the following form: U(x,p; y,q) denotes the expected prospect value over binary prospects consisting of the outcomes x and y with the probability of p and q, respectively. The function v(x) denotes a power value function. 9 The parameter σ describes the curvature of an individual's value function. An individual's risk preferences are described as risk averse when σ > 0, risk neutral when σ = 0, and risk loving when σ < 0. The parameter λ describes the curvature of an individual's value function above zero relative to the curvature of the value function below zero. The higher the value of λ, the more loss averse the individual is. The parameter α is a non-linear probability weighting measure, which is extended from a model by . The function π(p) weights the probabilities. When α <1, π(p) has an inverted S-shape, indicating that an individual tends to overweight low probabilities and underweight high probabilities . This model reduces to expected utility theory when α = 1 and λ=1.
In our experiment, participants were asked to choose between two sets of lottery options. For example, Figure 1.1 illustrates one set of options that a subject was asked to choose between. In this example, Option A offers a 30% chance of receiving 20 yuan and a 70% chance of receiving 5 yuan. Option B offers a 10% chance of receiving 34 yuan and a 90% chance of receiving 2.5 yuan. A total of 35 choices, divided between three series were asked. Monotonic switching was enforced, meaning that once the subject switched to option B they were not allowed to switch back to option A. 10 By enforcing monotonic switching, we eliminate the possibility of inconsistent choices within each series and also make the task more clear and concise for participants, as they only need to identify one switch point in each series. 11 Once the subject had completed the entire series of choices, one question was chosen randomly for payoff. The choices in the risk experiment were designed so that any combination of choices in the three series determine a particular combination of prospect theory parameter values. Table 2.3 shows the entire payoff matrix for the experiment. The payoffs range from a loss of 10 yuan to a gain of 850 yuan. 12 We use a relatively high maximum payoff of 850 yuan, which is roughly half a months pay in rural China. The average payoff in the risk experiment was 27 yuan (inclusive of the 10 yuan participation compensation), which is roughly half a single days wage in the survey area in 2008.
In the payoff matrix note that at first, the first column (Option A) dominates the second column (Option B) in terms of expected payoff and variance in the payoffs, but eventually, as the value of the high outcome in the second column increases, the expected value of the second column begins to dominate (table 2. Comparability of self-rated risk survey questions Dohmen et al. (2005) find that a general risk question can be used to predict actual risk-taking behavior, while  find that self-reported risk attitude does not predict risk aversion. Yesuf and Bluffstone (2009) find that risk aversion is lower when the rewards are hypothetical rather than real. To further investigate the comparability of self-rated risk questions to risk preference measures from field experiments with monetary rewards, we also asked participants two questions to allow the participants to self-rate their risk preferences. The first question was "How do you see yourself: Are you generally a person who is fully prepared to take risks or do you try to avoid taking risks?" Participants were asked to circle a number on a scale of 0 to 10, where the value 0 indicates that you are unwilling to take risks and the value 10 means that you are fully prepared to take risks. While the first question was asked about taking risks in general, the second question was more specific, asking about taking risks in investment, such as in agricultural investment. the experiments as the dependent variable, reveal that the self-rated risk aversion predicts the experiment risk aversion measure at a 5% significance level. However, the self-rated risk aversion in the context of taking risks in investment is not correlated with the experiment risk aversion measure.

Time Preference Experimental Design
There are several competing models for time discounting that have received a significant amount of attention in both experimental psychology (e.g., de Villiers and Herrnstein (1976), Ainslie and Haslam (1992), etc.) and behavioral economics (e.g., Laisbons (1997), , O'Donoghue and Rabin (1999). 14 The competing models were developed to account for observed behavioral regularities that are not consistent with the classic exponential discounting model. For example, the most common documented behavioral regularity is called "reversal of preferences". It occurs, for example, when a subject prefers $10 now rather than $12 in one day, but also prefers $12 in a year plus a day rather than $10 in a year. This type of preference is not consistent with exponential discounting but would be consistent with a rate of time preference that declines with time. There are a variety of specifications of discounting with this property of rates of time preference that decline with time, most notably hyperbolic discounting and quasi-hyperbolic discounting.
Our time experiment design follows procedures similar to those originally developed by  and Harrison, Lau and Williams (2002). The data are then used to estimate three parameters-the conventional time discounting parameter (r), present-bias (β), and hyperbolicity of the discount function (θ)-in a general time discounting model using nonlinear least-squares, which allows us to test which discounting model fits the data best-exponential, hyperbolic, quasi-hyperbolic, or a more general form Tanaka et al. 2010). 15 In the time preference experiment subjects were asked to choose between, for example, a real monetary payoff today or a larger payoff six months from now. The

Comparability of hypothetical time preference questions
Several experimenters have compared discount rates derived from questions with hypothetical and real rewards. Johnson and Bickel (2002), Madden et al. (2003), Hamoudi and Thomas (2006) did not find differences between hypothetical and real rewards in their experiments. However, Kirby and Marakovic (1996) and  found that hypothetical choices resulted in lower discount rates than real choices. To further examine the comparability of discount rates estimated from experiments with real monetary rewards to those with hypothetical rewards, we asked participants two hypothetical contextualized time preference questions. Specifically, we asked subjects: Imagine that you were given 10 yuan and told that you can use it to buy two types of tree seedling to plant on your forest plot. Imagine that your plot currently has no trees growing on it. Type 1 tree seedling costs 1 yuan and will grow into a tree that can be harvested and sold for 10 yuan in 5 years. Type 2 tree seedling costs 1 yuan each and will grow into a tree that can be harvested and sold for 30 yuan in 10 years. How much of the 10 yuan would you like to invest in Type 1 tree seedling? How much of the 10 yuan would you like to invest in Type 2 tree seedling?
The mean investment in Type 1 trees (faster growing, lower return species) was 4 yuan, and in type 2 trees (slower growing, higher return species) was 6 yuan. Figure   2.2 shows a box plot of the quartile level of the hyperbolic time discounting parameter given the level of investment in the Type 2 tree (slower growing, higher return species). The hypothetical parameter is a rather noisy parameter, although there is a similar pattern to the distribution of the time preference parameter (Figure 2.2). A regression with the hyperbolic discounting parameter as the independent variable and the investment in the type 2 tree (slower growing, higher return species) as the dependent variable indicates that investment in type 2 tree is negatively correlated with the experimental hyperbolic time discounting parameter, indicating that higher investment in the slower growing, higher return species is correlated with individuals who displayed a lower discount rate (i.e., more patient) in the time preference experiment. However, the coefficient on investment in type 1 trees in not significant (t=1.62).

Correlations
We begin our empirical analysis of the determinants of risk and time preferences by estimating ordinary least squares regressions for the risk aversion, loss aversion and hyperbolic time discounting parameter. In each regression, we include individual and household level characteristics. At the individual level, we control for the subjects gender, age and whether or not the subject has off farm employment. At the household level, we include net worth per capita (1000s of yuan), which is the variable of interest. We also control for household size, a dependency ratio that equals the number of children divided by the number of adults, the number of household members who work and the distance to both the post office and county seat.
Additionally, in the regressions where the dependent variable is the hyperbolic time discounting parameter we include two additional control variables: the subject's degree of risk aversion as measured from our risk decision-making task and the subject's earnings in the risk preference experiment. Participants were told that future payments in the time decision-making task would be delivered via China Post. If risk averse participants viewed receiving the future payments in the time decision-making task as "risky", then their risk aversion may impact their decisions in the time task; that is risk averse participants may choose the immediate reward, which they view as "safer". To control for this potential bias, we include risk aversion in the regressions where time preference is the dependent variable. If this was a potential source of bias in the time decision-making task, then we would expect to find a significant positive relationship between risk aversion and the time discounting parameter. In addition, earnings from the risk decision-making task were distributed to the participant prior to their participation in the time decision-making task. We might expect that higher earnings in the risk decision-making task might influence decisions in the time decision-making task. Individuals with higher earnings in the risk decision-making task may exhibit more patience in the time decision-making task (choosing the larger future reward more often than they would have if they had not just received a sum of money). To control for this potential bias, we include the earnings in the risk experiment in the regressions where the dependent variable is the time preference parameter. If this is a source of bias, we would expect to see a significant negative relationship between the two variables.
We also estimate each model with township fixed effects to control for unobservable factors that may be correlated with an individual's risk and time preferences, such as access to formal credit and insurance markets.

Correlations with risk preferences
Interestingly, we do not find a statistically significant correlation between any of the characteristics and risk aversion (table 2.4, columns 1 and 2). However, we do find a statistically significant negative correlation between net worth per capita and loss aversion (table 2.4, columns 3 and 4). This indicates that wealthy individuals are less loss averse. Also, we find a significant negative relationship between loss aversion and the dependency ratio, indicating that those from households with a higher dependency ratio (relatively more children than adults) tend to be less loss averse.

Correlations with time preferences
We find a significant negative correlation between net worth per capita and the discount rate, indicating that poorer individuals have a higher discount rates (i.e., more impatient) (table 2.5). Additionally, we find a significant positive relationship between the discount rate and both age and off farm employment. Those individuals who are older or who have off farm employment tend to have higher discount rates (i.e., impatient). There is also weak evidence that the discount rate is positively correlated with years of education and the number of household members who work, indicating that more education and more workers in a household are associated with higher discount rates.
Consistent with previous studies (Holden et al. 1998;Nielsen 2001;Gunatilake et al. 2007;Yesuf and Bluffstone 2008), there is weak evidence that more risk averse individuals have higher discount rates. Earnings from the risk experiment were not significantly correlated with the time preference parameter, indicating that choices in the time preference decision-making task were not influenced by earnings in the previous risk preference decision-making task.

Instrumental Variable for Wealth
While we find that low levels of wealth are associated with higher levels of loss aversion and impatience, we cannot conclude that wealth causes these preferences because of the endogeneity of wealth. Unobservable or omitted variables that affect wealth may also affect a household's risk and time preferences, making the estimated coefficient on net worth per capita biased. For example, if an individual's profession is risky (such as mining), then the individual may be wealthier (assuming higher pay for higher risk work) and also being in such a risky environment may decrease their risk aversion (growing more comfortable with taking risks). In this case, the profession affects both the risk preferences and wealth, and we would falsely attribute decreases in risk aversion to increases in wealth, when in fact it was the individual growing more comfortable taking risks due to employment as a miner, which was causing a decline in their aversion to risk. In this case, the coefficient on wealth would be biased upward.
To address potential endogeneity of wealth and omitted variables bias, we use an instrumental variable approach. 19 Instrumental variable estimation solves the omitted variable problem by using only part of the variability in the endogenous variable that that is uncorrelated with the omitted variables to estimate the relationship between the endogenous regressor, wealth, and the dependent variable (Angrist and Krueger 2001). Households' rank in the village according to net worth per capita is highly correlated with net worth per capita but is not directly correlated with the risk and time preference parameters. Risk and time preferences could only influence household rank in net worth through households' net worth per capita. Furthermore, rank is not a variable over which a household has control and therefore is exogenous to the household. A household might decide that it wants to change its wealth rank in the village by increasing its net worth but a household's final rank in the village will depend on the decisions of other households over which this household has no control.  columns 3 and 4). This indicates that after addressing the endogeneity of wealth problem by using an instrumental variable approach, wealth no longer has a statistically significant relationship with risk or loss aversion. Our result that wealth is not correlated with risk aversion is consistent with the classic results of Binswanger (1980) and with the more recent results of  who also uses an instrumental variable approach to deal with the endogeneity problem. The result that loss aversion is not correlated with household wealth is also consistent with previous findings by  and  in China and Vietnam, respectively.
We find weak evidence that net worth per capita has a negative significant effect on the discount rate, indicating that households with higher net worth per capita have lower discount rates (i.e., more patient, table 2.7). When we use the instrumental variables approach in 2SLS, the effect of net worth per capita on the hyperbolic discounting parameter remains negative; however, the statistical significance of the coefficient on net worth per capita falls to the 10% level in the model without township fixed effects and the 15% level when we use township fixed effects. This result, although weaker, is consistent with the findings in earlier studies that do not address the endogeneity of wealth or income (Pender 1996;Holden et al.1998

Robustness Checks
As a robustness check, we examine alternative measures of wealth, including: 1) forest land area per capita, 2) house value per capita, 3) assets per capita and 4) liabilities per capita. House value, asset, and liabilities are sub-categories of the net worth per capita measure. We use forest land per capita rather than total land per capita or farm land per capita because we believe the former will be more indicative of a household's wealth than the later. This is because responsibility land, the main tenure type of agricultural land in China, has traditionally been allocated on the basis of the number of family members, the number of laborers in each family, or the desire and/or ability of the household to engage in agricultural production, resulting in little variation in agricultural land holdings across households (Brandt et al. 2002).
However, during the recent 2003 forest tenure reform the method of allocation was decided on by farmers through their voting representatives on village committees. This resulted in greater variance in forest land holdings per capita, making forest land area per capita a potentially better proxy for household wealth than agricultural land area per capita. 20 We find again that each of these alternative proxies for wealth is not correlated with risk aversion, and this result remains even after addressing the endogeneity of wealth with respective instrumental variables ( We also check the robustness of our results to the statistical significance of the individual hyperbolic discounting parameters. When we estimate the hyperbolic discounting parameter for each individual in the sample, only 41 subjects have an estimated parameter that is statistically significant between the 1% and 10% level. Using only those subjects that had a statistically significant hyperbolic discounting parameter, we estimate the regressions in tables 2.5 and 2.7 again. These results (table 2.11) are consistent with the main results. Again, we find a statistically significant negative relationship between net worth per capita and the discount rate when using OLS. However, when we use the instrumental variable approach in 2SLS and control for township fixed effects, the significance level falls to 12%, indicating that there is only weak evidence that an increase in net worth per capita makes individuals more patient (i.e., lower discount rate). This finding concurs with the classic assumption that households with low wealth have high discount rates (i.e., more impatient).

Conclusion
In this paper we investigate the correlation between poverty and individual preferences for time and risk. Specifically, we use field experiment data collected in Fujian, China to measure the time and risk preferences of 103 rural households combined with household survey data to examine the correlation between risk and time preferences and wealth. We find little evidence that wealth is correlated with risk aversion and loss aversion, however there is evidence that wealth is correlated with time preferences. Ordinary least squares regressions indicated that those who have a lower net worth per capita, lower house value per capita, or lower forest land per capita (poorer) tend to be more loss averse, as well as exhibit a higher discount rate (i.e., impatient). Those with lower assets per capita (poorer) also exhibit significantly more impatience.
To address the problem of endogeneity of wealth, we use households' net worth rank within their village as an instrumental variable for net worth. When we use the instrumental variable in 2SLS, we find that the wealth proxy variables no longer have a statistically significant effect on risk and loss aversion with the one exception that households with more forest land per capita are less loss averse. This suggests that forest land plays a significant role as a safety net for negative shocks. When a household experiences a negative shock, a household with more forest land per capita may be able to recover more quickly by harvesting from its forest land. Knowing that 119 they have this safety net, a household may be less averse to loss. We find weak evidence that net worth per capita has a negative significant effect on the discount rate, indicating that households with higher net worth per capita have lower discount rates (i.e., more patient). However, caution should be taken in concluding the direction of causality in a cross-sectional study like this one, as it depends on the validity of our choice of instrumental variable for wealth. To better identify the direction causality between poverty and time and risk preferences, future research on this topic should include collecting a panel data set that includes both household characteristics and time and risk preference experiment data for multiple years, so that changes in wealth and time and risk preferences overtime can be examined.  10 Three examples were given in the instructions to help ensure that the subjects did not feel that they must make a switch within the series. In one example, the subject never switches to option B. In another example, the subject switches to option B at question 7. And in a third example, the subject switches to option B at question 1. 12 Note that in the risk experiment is possible for the subject to lose up to 10 yuan.
However, it would be unethical to ask rural households who participate in the experiment to pay us if they incur the loss. To address this issue at the beginning of the game we announce that there will be participation compensation of 10 yuan (which is equivalent to the highest possible loss in the risk experiment).  follows a similar strategy to ensure the subjects are treated ethically. While Camerer (2000) suggests that losses that are in fact net gains may be treated differently from real losses, we believe that because the 10 yuan participation fee was pointed out at the very beginning of the experiment, 30-45 minutes later when the subject got to Series 3 the individual would not consider the 10 yuan as a windfall but rather earning they had made for participating, and therefore treat the possible losses in the experiment as true losses. 13 The risk preference parameter estimation methods are detailed in appendix 2.
14 For a critical review of time discounting and time preference see Frederick, Loewenstein, and O'Donoghue (2002). 15 Benhabib et al.'s (2007) model is detailed in appendix 2, as well as the time preference parameter estimation procedure. 16 Our design differs from the time preference experiments of  and Harrison et al. (2002) in that we do not frame the choices with a front-end delay. An example of using a front-end delay, is a choice between money one month from today and more money seven months from now, rather than asking participants to choose between money today and more money six months from now, as we did. A front-end delay is used in time preference experiments to control (at least partially) for the credibility problem. The credibility problem is that participants may not believe that they will receive future payments, and therefore; will be biased toward choosing the immediate payoff. However, in much of the behavior economics literature, a  18 The 15 combinations of future payoff and time as described in the text were used in 9 of the 10 villages. In the first village, we used the same payoffs but shorter timeframes. Specifically, in the first village we used payoffs of 15, 60 and 150 yuan with delays of 1 week, 1 month, and 3 months and 30 and 120 yuan with delays of 1 week, 2 weeks and 2 months. In the first village, 5 out of 10 households always choose the future payoff. We thought that this high degree of preference for the future amongst the households might be due to the timeframes being to short, and so in the remaining villages we increased the timeframes. 19 We use the Durbin-Wu-Hausmann (DWH) test as detailed in Cameron and Trivedi (2009)        0.07 0.12 0.14 0.16 Note: Absolute value of robust t-statistic in parentheses. ***=significant at 1% level, **=significant at 5% level, and *=significant at 10% level. The instrumental variable for net worth per capita in the IV-2SLS regressions is the households wealth rank in the village according to their net worth per capita. The first stage results of the IV-2SLS regressions are presented in appendix D, table 1. 0.23 0.29 Note: Absolute value of robust t-statistic in parentheses. ***=significant at 1% level, **=significant at 5% level, and *=significant at 10% level. The instrumental variable for net worth per capita in the IV-2SLS regression is the households wealth rank in the village according to their net worth per capita. The first stage results of the IV-2SLS regressions are presented in appendix D, table 2.  (1.57) Note: Absolute value of robust t-statistic in parentheses. ***=significant at 1% level, **=significant at 5% level, and *=significant at 10% level. n=103 for the regressions where the dependent variables is forest land per capita. n=96 for all other regressions. House value, assets and liabilities per capita are measured in 1000's of yuan. This table summarizes the results for our wealth variable of interest for 32 separate regressions. All regressions include the following controls: male, years education, age, has off farm work, household size, number of children/number of adults, number of household members who work, distance to post office, and distance to county seat. (1.24) (1.03) (0.56) Note: Absolute value of robust t-statistic in parentheses. ***=significant at 1% level, **=significant at 5% level, and *=significant at 10% level. n=103 for the regressions where the dependent variables is forest land per capita. n=96 for all other regressions. House value, assets and liabilities per capita are measured in 1000's of yuan. This table summarizes the results for our wealth variable of interest for 16 separate regressions. All regressions include the following controls: male, years education, age, has off farm work, household size, number of children/number of adults, number of household members who work, distance to post office, distance to county seat, risk aversion, and the amount earned in the risk experiment. Absolute value of robust t-statistic in parentheses. ***=significant at 1% level, **=significant at 5% level, and *=significant at 10% level. This set of regressions only includes those individuals whose hyperbolic discounting parameter when estimated was statistically significant at the 1%, 5% or 10% level. The instrumental variable for net worth per capita in the IV-2SLS regression is the households wealth rank in the village according to their net worth per capita. Note that in the risk experiment is possible for the subject to lose up to 10 yuan.
However, it would be unethical to ask rural households who participate in the experiment to pay us if they incur the loss. To address this issue at the beginning of the game we announce that there will be participation compensation of 10 yuan (which is equivalent to the highest possible loss in the risk experiment).  follows a similar strategy to ensure the subjects are treated ethically. While Camerer (2000) suggests that losses that are in fact net gains may be treated differently from real losses, we believe that because the 10 yuan participation fee was pointed out at the very beginning of the experiment, 30-45 minutes later when the subject got to Series 3 the individual would not consider the 10 yuan as a windfall but rather earning they had made for participating, and therefore treat the possible losses in the experiment as true losses.
In the payoff matrix (table 2.3) note that at first, the first column (Option A) dominates the second column (Option B) in terms of expected payoff and variance in the payoffs, but eventually, as the value of the high outcome in the second column increases, the expected value of the second column starts to dominate (table 2.

Time preference parameter estimation
For our time preference experiment, we use a general model proposed by , which allows us to test exponential, hyperbolic, quasihyperbolic, and a more general form. Benhabib et al.'s (2007) model assigns a value of reward y at time t according to yD(y,t) where: The conventional time discounting parameter is r. The present-bias parameter is β, and hyperbolicity of the discount function is described by θ. The model reduces to exponential discounting when β=1 and θ=1. When β=1 and θ=2, the model reduces to hyperbolic discounting. When θ=1 and β is free the model reduces to quasi-hyperbolic discounting. When θ >2 and β is free, the model is "hyper-hyperbolic", meaning that, for example, the weight on future rewards drops even more steeply than in the hyperbolic model. By using this specification, we can compare the three models at once.
The probability of choosing immediate reward x over the delayed reward y in t days is denoted by P(x>(y,t)). We use a logistic function to describe this probabilistic relation as follows: The variable µ is a response sensitivity or noise parameter. We estimate the parameters r, β, θ, and µ in the above logistic function. For each subject, there are thirty observations, one observation for just before the switching point and one observation for just after the switching point for each of the fifteen series of questions.
For example if a subject choose to receive 150 yuan in 6 months over 75 yuan today (Plan A) and switched to Plan B when the payoff today increased to 100 yuan, then the dependent variable for the first response is 1 and the dependent variable for the second response is 0. The complete set of discounting choices is presented in table 1.
We estimated the above logistic function using non-linear least squares. In addition, to estimating the full model above, we estimated the model with restrictions for exponential discounting, hyperbolic discounting, and quasi-hyperbolic discounting.
Appendix Table 2  10.20 9.15 Note: Absolute value of robust t-statistic in parentheses. ***=significant at 1% level, **=significant at 5% level, and *=significant at 10% level.  Binswanger (1980) -India -Tropical area, characterized by high climatic risk for agriculture -240 participants -Given list of 8 choices; each with 50% probability (coin toss) but low payoff decreased and high payoff increased as moved down the list -Included one safe option where heads or tails resulted in Rs. 50 -There was a sequence of games over time and higher levels of payoffs -Photographs of sums of money to be received indicated by coins placed in each field were given several weeks prior to the experiment to help illiterate people understand -Objective was to determine whether differences in behavior between farmers of different wealth levels are the consequence of different attitudes toward risk or of different constraint sets such as limitations on credit or on access to modern inputs -Experimental measures of risk aversion indicate that at higher payoffs virtually all individuals are moderately risk averse with little variation according to personal characteristics -Wealth tends to reduce risk aversion slightly, but its effect is not statistically significant Nielsen (2001) -Toliara province of Madagascar -70 households across 6 villages -Follow design of Binswanger (1980) -Each participated in 4 experiments; 2 time preferences experiments (both hypothetical payments); 2 risk preference experiments (1 hypothetical and 1 with real payoffs) -Presented with series of 6 binary choices between two payoffs with 0.5 probability -One experiment involved only gains, the other involved gains and losses -Finds a linkage between asset poverty, time discounting and environmental degradation in the form of deforestation (and slash-and-burn agriculture) -Finds empirical linkage between willingness to take risks and willingness to delay Barr (2003) -Zimbabwe -678 subjects across 23 villages -Follow design of Binswanger (1980) -Presented with six gambles; each yields high or low payoff determined by guessing which researcher's hand contained a blue rather than yellow ball -Risk-pooling introduced by giving subjects the next days choice list and allowing them to form groups in which all winning would be shared equally between group members -Finds that more extrinsic commitment is associated with more risk pooling but that more information is associated with less risk pooling -In 4 of 5 villages networks of risk pooling contracts during the experiment and the networks existing in real life were significantly correlated Mette Wik et al. (2004) -Northern Zambia -110 participants across 6 villages -Follow design of Binswanger (1980) -Given choice between set of 6 games each with 50% probability of winning -Played 7 games during the first visit and 6 more during a second visit two weeks later (varied payoff levels between games) -Paid randomly on several games -Wealth indicator variables are found to be significant, and partial relative risk aversion decreases as wealth increases -Females are found to be more risk averse than males Humphrey (2004) -Uganda -Two regions: Sironko township in Sironko District and Bufumbo subcounty in Mbale District -Presented 12 pair-wise decisions between risky lotteries -One question was randomly chosen for payment -Also asked two hypothetical valuation tasks in terms of disease control decisions -Find that risk preferences of east Ugandan farmers exhibit systematic and predictable deviations from expected utility maximization; including: violations of the independence and transitivity axioms of expected utility theory, and reference-dependent preferences  Mosley and Verschoor (2005) -Uganda(205 participants) -Ethiopia(100 participants) -India(227 participants) -Participants presented with various pairs of lotteries; one 'risky' with a higher expected value but riskier than the other -Paid randomly on 1 choice -Additionally, asked two hypothetical questions to elicit certainty equivalents -Examines all of the linkages in the 'vicious circle of poverty' -Finds that there is very little relationship between risk aversion and the income measure of poverty but there are strong and significant linkages from low return on assets, to asset levels, to ability to diversify and manage risk, to income poverty -It may be forward-looking perceptions of vulnerability to risk on behalf of themselves and their families best explain their attitudes of risk aversion, and thus help determine their investment and diversification decisions, capacity to manage risk, and ultimately whether they remain in poverty Hamoudi and Thomas (2006) -Mexico -1,253 participants in 11 rural communities in the states of Guanajuato and Michoacan -Use modified design of Binswanger (1980) -Use 6 questions with 50/50 probability including one safe choice -With riskiest choice could win 540 pesos or lose 20 pesos (if lose, loss taken from show-up fee) -Choice presented in a circle, increasing risk as moved clockwise, high payoff would increase while low payoff would decrease -Paid randomly on 1 out of 5 preference tasks -Examine the relationship between inter-generational transfers and attitudes towards risk -Finds that inter-generational transfers are associated with attitudes toward risk (although associations are mostly weak or insignificant) -Risk attitudes measured were correlated with actual behaviors  -   Yesuf and Bluffstone (2009) -State of Amhara in the highlands of Ethiopia -262 farmers in seven local areas, in five counties and two zones -Use design of Binswagner (1980) but frame choice sets to reflect real farming decisions -Use six farming systems, all having similar costs but different output levels depending on 50% probability of good or bad harvest (based on coin toss) -Use 5 experiment sets with 6 choices each; sets 2 to 5 derived by scaling up amounts of set 1 by 5, 10, 20 and 30 ETB; set 5 was hypothetical -After experiment with only gain-gain choices those who had made enough earnings were asked to participate in experiment with gain-loss choices  (2002) -Series of 10 lottery-choices (Option A and Option B), where probability of higher payoff increased as participant moved down the list and Option B was more "risky" than Option A since its payoffs (CNY 35 and CNY 5) are more variable than the payoffs for Option A (CNY 20 and CNY 16) -One question chosen randomly for payoff -Participants exhibit substantial risk aversion -Risk aversion affects input intensity differently for market-oriented versus subsistence farmers -Risk aversion related with increasing use of pesticides by market-oriented producers but a reduction of pesticide use by subsistence farmers -Market producers are more concerned with stabilizing income, while subsistence producers are more concerned with stabilizing production Tanaka, Camerer, and Nguyen (2010) -Vietnam -180 participants across 4 villages in the south and 4 in the north -Use modified design of Holt and Laury (2002) -Use 3 series of paired lotteries, 35 choices in total -Choice between lotteries (A and B) -Probabilities stay the same in each series but payoffs increase as move down rows in the lottery B column -Monotonic switching is enforced -One question chosen randomly for payoff -Results indicate that mean village income is related to risk and time preferences -See Appendix B, Table 1 and 2 for ,more details Appendix Note: Also creates an index of perceived vulnerability as a better measure of poverty -Income per capita -OLS -Binary logistic regression (when risk aversion is a RA1-6, where RAi = 1 for participants who state a preference for a risky lottery less than i times) -Only in Ethiopia are any of the risk aversion measures correlated with income, and only RA2 at the 10% level -In Uganda per capita wealth is correlated with three RA measures but not with any Arrow-Pratt measures -The vulnerability index (when substituted for wealth and income in the regressions) is significantly correlated for all RA risk aversion measures, so it may be subjective rather than objective factors that drive attitudes towards risk (more vulnerable, more risk averse in Uganda; more vulnerable, less risk averse in India)  -Four provinces in China: Henan, Shandong, Hebei and Anhui -320 participants -Use wealth per capita, where wealth is defined as the value of durable goods per capita -OLS -Wealthier respondents were less risk averse (significant at 10% level) -Wealth did not have a statistically significant impact on loss aversion or the probability weighting parameter  (1991) -Each participated in 3 experiments (6 variations of experiments used) -Presented with a series of 8 to 10 binary choices between a specified amount of rice to be received at a particular date and alternative amount to be received at some other date -Each choice presented on a separate card -One card randomly selected for payment -Time frames ranged from 7, 12, 19 and 24 months; reference point was 1 month, 13 months or 25 months -Follow-up experiments conducted in 1991 -Find that minimum discount rates in all experiments were higher than the maximum interest rates paid by most respondents -Use experiment data and credit market data to test three models of credit markets : (1) the permanent income model, (2) upward sloping credit supply to individual borrowers, and (3) constrained credit due to imperfect enforcement -Rejects the permanent income model -Discount rate data are consistent with (2) and (3), while the credit market data are consistent with a combination of (2) and (3) Godoy et al. (1998) -Chimane Amerindian households in 18 villages in the Bolivian rainforest -209 participants -Twenty minutes into an interview asked participants "We realize you may be getting tired from answering questions. We would like to give you a rest. Would you like to have one candy now or two candies at the end of the interview?" -If participant said no, then asked "One now or three at the end?" -Then delivered candy at the appropriate time -The average impatience of the household heads was associated with less deforestation Holden, Shiferaw, and Wik (1998) -Indonesia(41 participants) -Zambia (86 participants) -Ethiopia (120 participants) -Each participant asked "If you were told you have the choice between an amount of money today (PV) and the amount (FV) in one year, how large would the amount PV have to be for you to prefer it instead of FV in one year?" -Question was repeatedly asked lowering the PV until a cut-off point was identified -In Indonesia and Ethiopia used cash value and in Zambia used both cash and maize; however, questions were hypothetical -Discount rates found to be very high -Market imperfections (credit and insurance markets) led to variation in discount rate -Poverty in assets, or cash liquidity constraints, was leading to or correlated with higher rates of time preference -In Zambia estimates of risk preferences were also estimated; more risk averse people tended to have lower discount rates Godoy, Kirby, and Wilkie (2001) -Bolivian lowlands -443 participants across 42 villages -Each asked 9 questions about a small reward today or larger reward at a specified delay (7 to 162 days) -Carried out experiment half way through field work to ensure delivery of future reward at specified time -Rates of time preference had a small economic and statistical effect on the use of natural resources (old-growth forest, fallow forests, fish, and game) Nielsen (2001) -Toliara province of Madagascar -70 households across 6 villages -Each participated in 4 experiments; 2 time preferences experiments (both hypothetical payments); 2 risk preference experiments (1 hypothetical and 1 with real payoffs) -Presented with series of 6 binary choices between payoff today and 1 year from now -One experiment involved only gains, the other involved gains and losses -Finds a linkage between asset poverty, time discounting and environmental degradation in the form of deforestation (and slash-and-burn agriculture) -Finds empirical linkage between willingness to take risks and willingness to delay  Godoy et al. (2001) -Patient farmers clear more forest than impatient farmers -Similar coefficients on impatience for clearance of oldgrowth and secondary-growth areas but only significant for secondary growth Kirby et al. (2002) -Beni, Bolivia -154 Tsimane' Amerindians from 53 households across 2 villages along the River Maniqui in the tropical rainforest -Participants given a list of 8 choices between X today and X+Y in the future -Future time frames ranged from 7 to 157 days -Participant were also given a list of 8 choices between a smaller number of candy today and a larger numbers of candy in the future -Conducted the experiments quarterly over the course of 1 year -Discount rates increased with age, decreased with educational levels and literacy, and tended to decrease as recent income rose -Discount rates were not associated with wealth, nutritional status, or moderate drug use -Low but reliable correlations between discount rates across quarters, suggesting that a person's discount rate is a somewhat stable characteristic Anderson et al.
-Vietnam -Two villages in the region of Hanoi city, one considered a rural commune (Thach Ban) and the other considered a urban commune (Quynh Mai) -Asked respondents to imagine that they had the opportunity to receive a loan form a local NGO and that they had the choice of paying back the loan immediately or postponing the payment to a later date, at which time they would have to pay a larger amount -9 questions -Future times included: 1 day, 3 months or 1 year -Hypothetical question -Trade-offs between today and tomorrow are different from trade-offs between any other 24-hour period -Examines correlations between discount rate and household characteristics -Find no relationship between income or gender and discount rate, an inverse correlation between age and discount rate -Find that those living in rural area have significantly higher discount rates Casse et al. (2005) -Toliara province of Madagascar -74 participants across 6 villages (part of larger sample of 240 households across 20 villages) -Each participant asked to choose among six hypothetical options -Options were between for example "X payment now or X+Y payment for one year later?" -High rates of time preference found Hamoudi and Thomas (2006) -Mexcio -1,253 participants in 11 rural communities in the states of Guanajuato and Michoacan -Use 10 questions; "Receive X today or X+Y in the future (1 and 2 months; 3 years?" -Those subjects who opted for future payoff were given contact information, a postcard to tell them if they moved and a written pledge that the surveyor would return on the specified date with the specified amount -Examine the relationship between intergenerational transfers and time preference -Male adults who are more patient are more likely to support parents -Both mothers and fathers who are more patient appear to invest more in their children -Time preference measures collected were correlated with actual behaviors -First conducted a survey to calculate the value of non-timber forest products (NTFP) collected by the household in the previous year -Then asked hypothetical stated preference survey question -If the Forest Department (FD) told them that they could not collect any NTFP for 1 year and that they would be compensated for the NTFP they did not harvest but that the payment would be delayed X months due to administrative problems. How much would the FD have to pay you if payment was made exactly X months from the due date?
-Investigate impact of time preference on NTFP harvesting, using a simultaneous question model -Villagers discount future consumption at an average rate of 24%, which is above existing market rate of interest for bank loans (18.5%) -Individuals with a higher rate of time preference harvest more forest resources

Yesuf and
Bluffstone (2008) -State of Amhara in the highlands of Ethiopia -262 farmers in seven local areas, in five counties and two zones -Four experiment sets; each with choice between X amount today or Y amount in the future (3,6, and 12 months); amounts were either ETB 15 or 40 ($1.76 and $4.70) -Each choice set presented on a card and recorded on the card, after 28 cards completed one was chosen at random for payment -Find that median discount rate for each set of experiments is high (more than double the average interest rate on outstanding debt) -Discount rate varied systematically with wealth (physical asset) and risk preferences Bauer and Chytilová (2008) -Rural population in Mukono district, southern Uganda -910 participants, 10 villages -Asked "Would you prefer Ush 200,000 today or Ush 250,000 in one year?" -Asked 5 questions, each time increasing the future payment -Hypothetical survey questions -Examine causal impact of education on subject discount rates using instrumental variables (varying school frequency in different villages and number of school-age years that overlap with the era of Idi Amin) -Find that for men education has significant impact on discount rate Bauer, Chytilová, and Morduch (2010) -Rural population of Karnataka in south India -573 participants, 9 villages, 2 taluks (Honavar and Haliyal) -35 people selected in each village by random walk (90% of invited participated) -Asked "Would you rather consume Rs. 250 tomorrow or X+Y in t months?" -Asked 2 sets of 5 questions each -In one set t=3 and in the other t=15 months -Shifted future question exactly 1 year to avoid any seasonality -In current question included 1 day time delay to lower credibility and higher transaction costs associated with future payments -Real monetary rewards with stakes as large as a week's wage (min Rs. 250,max Rs. 375) -Integrate experimental measures of time discounting and risk aversion with survey data on financial activity to identify time inconsistencies between current and future questions -Identify 1/3 of population exhibits choices consistent with hyperbolic discounting (discount future more heavily when asked a series of questions about consumption now vs. in 3 months, relative to discounting in similar questions about consumption in 12 vs. 15 months) -Women with hyperbolic preferences save less at home, save less in total levels and are more likely to borrow generally but to do so through microcredit institutions specifically  -Vietnam -180 participants across 4 villages in the south and 4 in the north -Subjects are asked to make 75 choices between smaller rewards delivered today and larger rewards delivered at a specified time in the future -Future times include: 3 days, 1 week, 2 weeks, 1, 2 and 3 months -Payment varied between 30,000 to 300,000 dong (15 days wage in rural north) -Enforced monotonic switching within question sets -A single question was selected at random for payment -Before experiment selected trusted agent to deliver the future payments   Bauer and Chytilová (2008) Mukono district, southern Uganda -Use profession as a proxy of income -Self-employed farmers and non-farm workers (drivers, shopkeepers vs. employed individuals (teachers, employees of public bodies or NGOS) and students -Examines average discount rates across profession groups -OLS, clustering at village level -Individuals facing less income pressures discount less when looking at average discount rates across profession groups -From OLS, some evidence that employed females discount more, however no other significant impact of profession on time preference Bauer, Chytilová, and Morduch (2010)   To further examine the source of increased wealth, we also examine the effect of the reform on household forest use. Results suggest that forest certification increased bamboo revenue, while obtaining a new plot (without a forest certificate) increased non-timber forest product revenue, although these results are statistically weak. Overall this paper provides weak evidence that forest tenure reform garners potential for improving poor rural households' livelihoods in China.

Introduction
Many people living in or near forests in developing countries are poor . For example, in China,  observed that many counties with abundant forest also were categorized as being severely poor. In India, approximately 275 million people live in or near forests and depend on them for their income. These people are disproportionally 'tribal' ethnic minorities, who are among the poorest and most vulnerable people in India World Bank 2006). Similar observations have also been made in Cambodia , Vietnam (Muller et al. 2006) and Brazil . Overall there are hundreds of millions of people who depend on forests for their livelihood Calibre Consultants and Statistical Services Centre 2000). 1 The correlation between people living in poverty and their dependence on forest resources, combined with the continued deforestation in the world, has stimulated a call from international institutions, NGOs, and community organizations, for pro-poor forestry policies in the last decade FAO 2003;FAO 2009). Amongst the various pro-poor forestry polices that have been recommended, one that has received notable attention and gained momentum in implementation is forest tenure reform. Property rights to ownership and use of forest resources are often contested, overlapping or unenforced, leaving households with insecure ownership and use rights to forest resources. This insecurity undermines sound forest management, for without secure rights forest holders have few incentives to invest in managing and protecting their forest resources. These realizations have stimulated the recent trend in forest policy toward strengthening property rights for forest resources by transferring property rights from the state to communities and individuals, giving them defined rights to manage and use forest resources FAO 2003;.
In this paper, we assess the impact of forest tenure reforms on household wealth in Results provide statistically weak evidence that the forest tenure reforms have had a positive effect on household wealth in our study area. Specifically, increased tenure security in the form of a forest certificate increased net worth per capita. This positive forest certification effect on wealth was larger in magnitude when the forest certificate was on a plot that a household had already been managing than when the forest certificate accompanied a new plot that a household received as a result of the reform.
To gain insight into the mechanism through which tenure reform leads to increased wealth, we also examine the effect of the tenure reform on households' forest use.
Specifically, we examine changes in household revenue per capita from the sale of non-timber forest product (NTFP) and total revenue from the sale of bamboo, as households were relatively more engaged in these two forest income generating activities than in the sale of timber. The results suggest that forest certification had a positive effect on total bamboo revenue, while it did not have a significant effect on NTFP revenue. Obtaining a new plot (without a forest certificate) resulted in an increase in NTFP revenue per capita, while it did not have a significant effect on bamboo revenue. This paper proceeds as follows. The first section gives an overview of the forest tenure reform history in China, with an emphasis on the recent tenure reforms in Fujian, China. The next section explains the data collection procedures, gives a description of the data and reports a preliminary examination of the impact of the forest tenure reform on households' livelihoods using descriptive statistics. Then the empirical framework is outlined, followed by the results and a concluding section. The first major wave of reforms in China's collective forests began in 1981, and was aimed at transferring the responsibility of forest planting and management from the collective to households . By 1986, nearly 70% of the collectively owned forest land had been transferred to rural household management . In 1987, however, due to unsustainable logging the government reverted a large portion of forest land under household management back to collective management .

China's Forest
By 1986, while 70% of the collectively owned forest land in China had been transferred to rural household management, in Fujian only 32% of the collective forest land had been distributed for household management . This low percentage of forest land under household management was due to the fact that Fujian had not fully participated in the first round of the tenure reforms in the 1980s. Instead, the provincial government in Fujian had implemented a shareholding system to keep forests under collective management while distributing "paper shares" of collective forests based on family population. In Fujian, forest land was not actually physically distributed, rather only dividends from the forest were distributed to households.
At first, Fujian's shareholding system was highly regarded by forest administrators for its ability to maintain forests under collective management but fifteen years after establishment of the system, two issues became increasingly evident . First, forestry's contribution to rural incomes was negligible in spite of the fact that forest land occupies more than 60% of the total provincial land area and 80% of rural land area . Second, enforcing forest conservation had become increasingly difficult for local forest authorities due to lack of cooperation from farmers. For example, the severity of forest fire incidents grew over the course of the 1990s, and there is anecdotal evidence that many of the fires were caused by farmers  We construct a balanced panel data set by using only those households that were included in both survey years, so that we have pre-and post-reform data for every household in the analysis. Ten households, that had no forest land area, as well as no forest income, in any of the survey years are excluded from the analysis. Seven households for which there was missing data that was essential to this analysis are also discarded. This results in a sample size of 87 households.
In 2000, average household size in the sample was 4.9 people ( forest tenure reform has not had an effect on the growth in net worth per capita.

Forest Use
As an extension to the analysis of the affect of the forest tenure reform on household wealth, we also examine the affect of the reform on household forest use.
Households use their forest resources as both an income source (sale of bamboo, timber and non-timber forest products (NTFP)) and to meet their personal needs for forest products (  . We examine the mean NTFP revenue per capita and the total bamboo revenue for the same three discrete measures of household forest tenure reform status, as we examined for wealth above. Interestingly, when we examine the change in bamboo revenue per capita based on whether or not the household received at least one forest certificate, those households that got a forest certificate for at least one plot between 2000 and 2008 experienced a 209% increase in their NTFP revenue, while those that did not experienced a much higher increase of 426% in their NTFP revenue (  ). When we examine the trend for those that got at least one new plot or one forest certificate, we see that those households that did also have a lower average total bamboo revenue in each year between 2005 and 2008 than those that did not ( figure   3.4, panel c).
While these differences in both NTFP and bamboo revenue are noticeable, there is no statistically significant difference in the change (from 2000 and 2008) in the means between households grouped according to these discrete measures of household forest tenure status. This suggests that the forest tenure reform has had no effect on household revenue from NTFP and bamboo.

Measuring household forest tenure reform status
Although our descriptive analysis suggests that the forest tenure reform has not had an impact on wealth, NTFP revenue per capita, or total bamboo revenue, it may be that the discrete measures used to capture household forest tenure status do not fully capture the magnitude of changes in a household's forest plot tenure structure. For instance, with a discrete measure of whether or not a household received a new plot, a household that received a new plot with an area of 5 hectares would be categorized in the same way as a household that received a new plot with an area of 0.01 hectares.
However, it is likely that acquiring a new plot with an area of 5 hectares would have a greater impact on a household's wealth and forest product revenue than a plot with an area of only 0.01 hectares. Therefore, in our empirical analysis each plot in each year is identified as belonging to one of the following four categories: 1) pre-FTR plots without a forest certificate; 2) pre-FTR plots with a forest certificate; 3) new plots without a forest certificate; and 4) new plots with a forest certificate. In each year, a given household's total forest area will be distributed amongst one or more of these four categories (table 3.4).

Empirical Strategy
Our main objective is to identify how the forest tenure reform affected household wealth. In order to do so, we must address the concern that changes in household wealth could be due to factors other than the changes in household forest plot tenure structure. For example, changes in household wealth between 2000 and 2008 could be due to unobservable time-invariant variables (e.g., household's entrepreneurial drive or location factors that affect forest productivity) or unobservable variables that change over time (e.g., increased forest productivity due to favorable weather conditions). Furthermore, changes in a household's wealth that had an increase in plot area with a forest certificate and/or a new plot relative to those that did not could be due to initial differences in observed (e.g., education of the head of household) and unobserved characteristics (e.g., entrepreneurial ability) between the two. Without controlling for this we risk incorrectly attributing differences in wealth between those households that experienced a change in their forest plot tenure structure to those that did not experience a change in its forest plot tenure structure, when in fact they are due to initial differences between the two groups.
To address these concerns, we use three years of balanced household panel data in a fixed effect model, which allows us to control for time-invariant observable and unobservable variables. The limitation of approach is that it does not allow us to control for time-variant unobservable variables or for possible self-selection of households into acquiring a forest certificate or a new plot during the reform.
As an extension to the analysis, we also examine the effect of the reform on households' bamboo and NTFP revenue. To do so, we must again address the concern that changes could be due to factors other than the changes in household forest plot tenure structure, and additionally we must address possible selection bias (i.e., factors that are inherently different about those households that engage in NTFP or bamboo sales and those that do not). For example, on average only 26% of households engaged in the sale of bamboo. Therefore, our dependent variable (bamboo revenue) is censored (i.e. a positive outcome is not observed for many households) and ordinary least squares estimation will produce biased parameter estimates. To address this issue, we use Heckman's two-stage estimation procedure for panel data that uses the Inverse Mill's Ratio to take into account selection bias . In the first stage, we use a probit model to estimate the likelihood of a household engaging in sales of each forest product. The estimated parameters are then used to calculate an Inverse Mill's Ratio for each forest product and year. We then include the Inverse Mill's Ratio as an explanatory variable in the fixed effects estimations to capture the selection effect.

Empirical Model
The base estimate (model 1) of the forest tenure reform effect is obtained from the ordinary least squares estimation: where net worth it is the net worth per capita (yuan) of household i in year t. household size; head of household's education level and age; and the number of household members who work, as well as household total area of pre-FTR plots without forest certificates and total area of crop production (area of production multiplied by the number of harvests) in each year. 5 In models (4), (5) and (6), we add township, village and household fixed effects, respectively.
As an extension to the analysis of the effect of the forest tenure reform on household wealth, we also estimate the effect of the reform on household NTFP and bamboo revenue. To estimate the effect of the forest tenure reform on household NTFP and bamboo revenue, we use Heckman's two-stage estimation procedure for panel data, which uses the Inverse Mill's Ratio to take into account selection bias   (3) thru (5) above for the effect of the forest tenure reform on wealth and NTFP revenue we control for the total area of crop production in each year. However, we do not have the total area of crop production data for each year from 2000 to 2008, and so instead here we control for the total area of crop production in the base year (2000).

Empirical Results
Overall we find statistically weak evidence that the forest tenure reform has increased household wealth (table 3.5). 7 Specifically, forest certification of a plot that a household had already been managing prior to the reform had the most consistent and largest positive effect on household net worth per capita. In all models the coefficient on preFTRplot_FC it is positive (table 3.5, row 1). In models (4) and (5) that included township and village fixed effects, respectively, the coefficients on preFTRplot_FC it are statistically significant at the 10% level, while in the remaining models, (1) to (3) and (6), they are statistically significant at the 15% level. The coefficient on preFTRplot_FC it in model (6), which includes household fixed effects, suggests that for a one hectare per capita of pre-FTR land area that receives a forest certificate the effect is an increase in net worth per capita of 5,650 yuan. Households on average received forest certificates for 0.14 hectares of their forest land, his implies that stronger property rights in the form of a forest certificate have increased household wealth on average by 42% (i.e., on average 5.2% per year).
Receiving a new plot with a forest certificate also had a positive effect, however the evidence is weaker. In all models the coefficients on newplot_FC it is positive (table 3.5, row 2) but has a lower magnitude (328 to 820 yuan) than the coefficients on preFTRplot_FC it (2923 to 5650 yuan). In models (3)

Impact on non-timber forest product sales
The results suggest that forest certification of a forest plot had no effect on household NTFP revenue per capita, as the coefficients on preFTRplot_FC it or newplot_FC it in each model (except for the coefficient on newplot_FC it in model 1) are not statistically significant (table 3.6, rows 1 and 2). 8 (1) and (5) and at least at the 20% level in models (4) and (6). The coefficient on newplot_noFC it in model (6), which includes household fixed effects, suggests that for a one hectare area per capita increase in forest land (without a forest certificate) the effect is an increase in household NTFP revenue per capita of 1,474 yuan.

Impact on bamboo sales
The results suggest that forest certification of a plot that a household had already been managing prior to the reform has a positive effect on total bamboo revenue, while receiving a new plot with a forest certificate has a negative effect on total bamboo revenue (table 3.7). However, the evidence in both cases is statistically weak.
In all models the coefficient on preFTRplot_FC it is positive, however it is only statistically significant in model (3) and (6)  and nut trees and bamboo plantations over the last two decades, as they are considered more profitable than conventional timber plantations and have a less burdensome more transparent taxation system ).

Conclusion
Over the last decade there has been a call from international institutions, NGO's, and community organizations, for pro-poor forestry policies FAO 2003;FAO 2009). Amongst the various propoor forestry policies that have been recommended one that has received notable attention and gained momentum in implementation is forest tenure reform FAO 2003;. The hope is that when communities and individuals receive stronger rights to ownership and use of forest resources, those rights will give them an incentive to invest in managing and protecting those resources, and in doing so will also allow poor, rural households to improve their livelihoods.
In this paper, we examined the impact of forest tenure reform on household wealth in Fujian Province, where a large-scale reform of forest land tenure began in 2003.
Empirically, we used a balanced household panel data set among 87 households. We examined the effect of the reform on net worth per capita as a measure of wealth. Then as an extension we also examined the effect of the reform on total bamboo sales.
Changes in household forest tenure structure were captured by three variables: the total area of pre-FTR plots with a forest certificate; the total area of new plots with a forest certificate, and the total area of new plots without a forest certificate. To identify the effect of the reform on net worth per capita, NTFP revenue per capita, and total bamboo revenue, we used a fixed effects model. Additionally, to identify the effect of the reform on total bamboo sales, we used a two-step Heckman selection approach for panel data to take into account the effect of selection into engaging in the sale of NTFP or bamboo.
Results suggest that more secure tenure, resulting from the distribution of forest certificates, increased household wealth, although the evidence is statistically weak.
This positive forest certification effect was stronger on plots that households had been managing prior to the recent reform than it was on new plots households received during the reform. After identifying a positive effect of the forest tenure reform on wealth, we examined its effect on households' revenue from the sale of forest products. Households were relatively more engaged in the sale of NTFP and bamboo than in the sale of timber. Therefore, we examined changes in household revenue per capita from the sale of NTFP and in total revenue from the sale of bamboo. The results suggest that forest certification of a plot that a household had already been managing prior to the reform had a positive effect on total bamboo revenue but no significant effect on NTFP revenue. Conversely, receiving a new plot without a forest certificate had no significant effect on bamboo revenue, while it had a positive effect on NTFP revenue.
While the reason behind the differing effects of receiving a forest certificate or a new plot is a question for future analysis, a potential hypothesis is that the differing effects may be due to differences in forest stock quantity, quality or type on new plots relative to plots that households had already been managing. Testing this hypothesis would require an analysis at the plot level; however, we do not have bamboo revenue data at the plot level to support such an extension of this analysis.
This paper provides statistically weak evidence that the forest tenure reform has had a positive effect on household wealth in our study area. While the sample is very small, relative to the number of households affected by the forest tenure reform in China and China is a large diverse country, this paper does suggest that the forest tenure reform garners potential for improving poor rural households' livelihoods. In particular, since only 30% of all forest plots had forest certificates, expanding such certification could potentially increase household wealth. It is likely that with forest plot certification, tenure security will be enhanced. And increased tenure security will stimulate households' investment in their forest plots, improving their livelihoods.       (China Statistical Yearbook, 2009). Household characteristics control variables in models (3) to (5) include the following variables for the year 2000: household size, head of households' education level and age, and the number of household members who work, as well as the total area of pre-FTR plots without a FC and the total area of crop production in each year. Model (6) includes only the total area of pre-FTR plots without a FC and the total area of crop production in each year as control variables. Robust t-statistics are in parentheses. * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level. Source: Authors' data. Table 3.6 Effects of the forest tenure reform on non-timber forest product revenue Dependent Variable: Non-timber forest product revenue (yuan/capita) Model: (1)  (China Statistical Yearbook, 2009). Household characteristics control variables in models (3) to (5) include the following variables for the year 2000: household size, head of household's education level and age, and the number of household members who work, as well as the total area of pre-FTR plots without a FC and the total area of crop production in each year. Model (6) includes only the total area of pre-FTR plots without a FC and the total area of crop production in each year as control variables. Robust t-statistics are in parentheses. * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level. Source: Authors' data.    (4) and (5) add township and village fixed effects, respectively, to model (2). Model (6) adds household fixed effects to model (2) and does not include the household characteristic controls from the base year (2000), the dummy indicating if a household had at least one new plot in any year, nor the dummy indicating if a household had at least one plot with a forest certificate in any year. All models also control for the inverse mills ratio for each year from 2000 to 2008. Values for the years 2001 to 2008 are adjusted for inflation using the rural consumer price index for Fujian Province, China (China Statistical Yearbook, 2009). Forest certificate abbreviated to FC. Forest tenure reform abbreviated to FTR. Robust t-statistics are in parentheses. * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level. Source: Authors' data.   (China Statistical Yearbook, 2009). Robust t-statistics are in parentheses. * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level. Source: Authors' data.  (China Statistical Yearbook, 2009). Robust t-statistics are in parentheses. * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level. Source: Authors' data.

CONCLUSIONS AND RECOMMENDATIONS FOR FUTURE RESEARCH
The underlying motivation for this dissertation research was to examine if property rights matter for forest management. Economic theory predicts that improved tenure security in the form of strengthened property rights will give households an incentive to invest in their forest resources, which will stimulate income, ultimately increasing households' wealth. However, in the past tenure reforms have not always led to these intended effects. While researchers have empirically examined the impacts of tenure reforms, they have not examined potential heterogeneity in responses due to differences in households' risk and time preferences. Furthermore, these reforms are often implemented in areas where the poverty rate is high. Those living in poverty are assumed to have both high discount rates and high levels of risk aversion, which make them less likely to make investments. Such characteristics may also hinder the intended effects of forest tenure reforms.
In this dissertation, I examined these issues in the context of rural Fujian, China, where a large-scale reform of forest property rights began in 2003 in areas where the poverty rate is still high. To explore these issues, I used panel household survey data and risk and time preference data collected using field experiments with real monetary rewards in empirical models that aimed to alleviate potential biases due to selfselection into receiving a forest certificate for a plot. The four main hypotheses examined included: H1) Forest property right reforms affect how individuals manage their forest resources; H2) Time and risk preferences affect forest management and therefore also augment individual forest management responses to forest property rights reforms; H3) Time and risk preferences differ across individuals and are correlated with wealth; and H4) Household wealth increases as a result of the forest tenure reforms.
In manuscript 1, I tested H1 and H2. In the base difference-in-differences estimation that does not allow for heterogeneity in time and risk preferences, I found that on average there was no significant forest certification effect on forest management. This suggests that on average that reform is not working as intended.
However, it should be noted that this insignificant effect may be due to the possibility that households are still in a transition phase. Since forest management requires long term investments and our data is at most 5 years post-reform, it may be that not enough time has passed since households received their forest certificates to discern the reforms intended effects of increased investment in forest resources. In the future, researchers should collect additional post-reform data to see if the reform has its intended effects over a loner period of time.
Interestingly, when I allowed for heterogeneity in forest certification effect due to households' time and risk preferences, I found that the average overall forest certification effect had been masking a variety of responses that were occurring but which varied depending on households' time and risk preferences. As expected we found that in response to receiving a forest certificate households that were more risk averse used less labor for harvesting and more labor for applying inputs than those that were risk neutral or risk seeking. This supports the hypothesis that households believe that the forest certificate gives them greater assurance that if they invest or delay harvest then they will be able to get their returns in the future. More generally, we found that those households that were risk averse, used less labor for applying inputs, spent less on forest inputs, and used more labor for harvesting. This suggests that in areas where many households are risk averse, policymakers may want to couple forest tenure reforms with other programs and policy instruments to reduce households' risks (e.g., pest, disease, forest fire). Such coupling may stimulate investment in forest resources generally and also in response to forest tenure reforms. Future research should examine how households perceive the risks specifically associated with forest investment and what the levels of actual risks are in order to inform policymakers.
Such information would aid policymakers in identifying which risks need to me dealt with and to design programs or instruments (e.g., programs to reduce threat of pests and fire; insurance programs; encourage the formation of voluntary cooperatives within village as a risk-sharing mechanism) that specifically help to mitigate those risks.
In Manuscript 1, I also found that more loss averse households used more labor for harvesting in response to receiving a forest certificate. This suggests that loss aversion affects harvesting responses to receiving a forest certificate based on the manifestation of loss aversion in an endowment effect of the forest certificate. Receiving a forest certificate may have an endowment effect in that once a household receives a forest certificate for a plot, it becomes more painful for the household to experience a loss of forest stock from that plot than from other forest plots without forest certificates, and therefore loss averse households would harvest more in response to getting a forest certificate for a plot. This suggests that future research should examine households' demand for insurance that hedges against the risk of loss of forest stock from plots with forest certificates.
Lastly, in response to receiving a forest certificate those with higher discount rates used less labor for applying input and spent less on forest inputs than those with lower discount rates. And in Manuscript 2, I found statistically weak evidence that the poorer a household was the higher their discount rate (i.e. more impatient) was. Combined, these findings suggests that forest tenure reforms should be coupled with programs to reduce poverty and to allow for and encourage borrowing. As poverty is alleviated, households' discount rates may fall, making them more likely to be able to invest, whether from their own accumulated savings or by borrowing. While a component of China's forest tenure reform has been the establishment of a loan program that allows households to obtain a loan using their forest certificated plot as collateral, in 2008 only 1 of the 104 surveyed households had used their forest certificate as collateral for a loan. Further research should investigate why households are not taking advantage of this credit opportunity.
This research used risk and time preference field experiments designed in a generic context. The case can be made that preferences measured in a generic context may not translate well to preferences in forest management decisions. I did find that these preferences had some effects on forest management and responses to receiving a forest certificate. While this suggests that these generic risk and time preferences are relevant for forest management decision, it would be valuable in future research to design and implement risk and time preference field experiments that are contextualized in a forest decision-making problem in order to examine if context matters.
In manuscript 3, I found statistically weak evidence that improved tenure security in the form of a forest certificate increased net worth per capita by 42% between 2000 and 2008. Furthermore, I found that there was also statistically weak evidence that forest certification increased bamboo revenue, while obtaining a new plot (without a forest certificate) increased non-timber forest product revenue. This suggests that even in this early post-reform time period, the forest tenure reform in China appears to be improving households' wealth. Given the long-term time horizon of forest investment, it may be that wealth will increase more over the longer term. In the future, researchers should collect additional post-reform data to see if the reform has a more significant impact on households' wealth over a longer period of time.
The overall goal of this research was to understand how heterogeneity in time and risk preferences affected responses to forest tenure reforms. I have found that these preferences matter for forest management and responses to forest tenure reforms. This suggests that policymakers who are going forward with a tenure reform should consider the particular context of the reform and consider coupling the reform with appropriate programs and instruments to alleviate poverty and to help households' to deal with risks and make long-term investments to further stimulate the intended effects of the reform-increased investment in forest resources and improved livelihoods.