RESEARCH STATEMENT

Hanming Fang

Department of EconomicsYale University

Website: http://pantheon.yale.edu/~hf54

August 2006

During my days as a Ph.D. student at the University of Pennsylvania, I was trained to thinkabout economic problems through both the lens of economic theory and the empirical data. I wasalso instilled a strong belief that empirical research should be guided by appropriate theory andtheoretical research should be inspired by empirical observations. Such indoctrination has had aprofound influence on my research style and research interests. I consider myself as an appliedmicroeconomist with broad theoretical and empirical interests with a focus on Public andLabor Economics. My work tends to cross the boundaries of narrowly defined fields and disciplinesas I choose my projects based mainly on whether the ideas are promising instead of whether theyfit into my “area.” I have developed theoretical models in order to cast new light on a topic ofinterest, and when appropriate, have conducted empirical work to determine whether the theory isan adequate description of reality. So what unifies my research is not so much the research topics,rather it is the tight connection that I always attempt to maintain between theory and empiricalwork. I view the distinctions among applied microeconomics fields to be only artificial becausethere are no differences in their theoretical and empirical tools.

In this research statement, I attempt to categorize my research papers into several related topics.Within each topic, I first summarize my completed (both published and unpublished) research, andthen I describe my research plan for the next five years. All of my research papers are listed in“Reference” in the end.

In summarizing my research, I attempt to explain the questions of interest in the research topicsto which I contributed, describe the genesis of my interest in them, and try to put in context therelationship among the various papers I consider relevant to each topic. Some papers will appearin multiple topics. Table 1 below provides the crosswalk between research topics and papers.

1 Economics of Discrimination: Theory, Empirical Methods and

Evidence

1.1 Summary

This research topic originated in my 2000 University of Pennsylvania Ph.D. dissertation titled“Discrimination with Endogenous Group Choices and its Empirical Application.” The main moti-vation for this research project is the following observation: in many economic transactions, thebasis on which we are preferentially treated or discriminated against is not race or gender (traits

1

Research Topics Papers1. Economics of Discrimination:

Theory, Empirical Methods and Evidence [1, 12, 13, 14, 35, 36, 37]

2. Social Economics [1, 8, 9, 19, 26, 30, 32, 33]

3. Assessing the Impact of Welfare Reform of 1996on Single Mothers and Their Children [5, 6, 16, 23, 34]

4. Applications of Psychology in Economics [5, 7, 10, 16, 25, 34]

5. Multi-dimensional Private Information: Auctions,Public Goods and Insurance Market [2, 3, 11, 17, 21, 22, 24, 29]

6. Empirical Analysis of Asymmetric Informationand Dynamic Inefficiency in Markets [22, 38, 39, 40, 41, 42]

7. Other Topics [4, 20, 27, 28, 31]

Table 1: Papers in Each Research Topic: A Crosswalk.

Note: See “Reference” for the title of the papers. Papers numbered 1-22 are completed; and papers num-

bered 23 and higher are either actively in progress or in the immediate research agenda. All completed papers are

available from my website at http://pantheon.yale.edu/~hf54.

2

that are exogenously given and immutable), rather it is whether we belong to an endogenouslychosen group (such as whether we graduated from an Ivy League university or whether we belongto an exclusive golf club, etc.). To the extent that such group labels can be endogenously alteredwith some cost, my dissertation theoretically investigates whether discrimination based on suchmutable group memberships can persist. On the theoretical front, the main discovery is the ideaof “endogenous signaling instrument.” An endogenous signalling instrument is an activity (whichdefines one’s group label) that a priori does not satisfy Spence’s single-crossing property, but in adiscriminatory equilibrium, it does, and thus becomes a valid signaling instrument endogenously.I used this idea in my paper “Social Culture and Economic Performance” [1, AER 2001 ]to examine the efficiency rationale for providing preferential treatments to “elite” groups that aredefined by whether one undertakes some seemingly irrelevant activities. In that paper, I interpretthe connection between obtaining higher paying jobs and undertaking some seemingly irrelevantactivity as “social culture.” In the context of a society trying to adopt a new technology, I show thatby allowing the firms to give preferential treatment to workers based on some “cultural activity,”the society can partially overcome an informational free-riding problem, and allow the society toadopt new technologies that require workers’ unobservable skill investment to be productive. Asa result, social culture may affect the economic performance by altering the effective productiontechnology of the economy.

On the empirical front, I used the framework as the basis of a structural model of endogenouseducational choices and wage determination to estimate the contributions of ability signaling andproductivity enhancement in the college wage premium. This was my junior job market paper in2000, “Disentangling the College Wage Premium: Estimating a Model with Endoge-nous Education Choices,” [14, IER, forthcoming ]. Here, education choices, whether it iscollege or high school, become the relevant group labels. This is a parsimonious structural modelthat nests both ability signaling and human capital enhancement explanations of the college wagepremium. The model has empirical implications on college and high school wage distributions aswell as college enrollment rate. The key idea to distinguish ability signaling from productivityenhancement can be heuristically illustrated as follows. Imagine a population of agents with het-erogeneous abilities. In an ability signaling model, the college wage premium is the result of thedifference in the average abilities between college and high school graduates. To the extent thateducation choices are endogenous, the scope of ability signaling in generating wage differentials isrestricted by the equilibrating forces of education choices. On the other hand, the scope of produc-tivity enhancement in generating college premium is not restricted by the equilibrating process sinceit is an exogenous technology parameter in the education production function. This suggests that ingeneral a model that admits the forces of both productivity enhancement and ability signaling canbetter rationalize the observed college wage premium than a pure signaling model, given a familyof parametric models. In the extreme case that the observed college wage premium lies outside thelevel that can be rationalized in a pure ability signalling model, the data will actually call for somerole of productivity enhancement. I estimate the parametric structural parameters of the model us-ing Census data; and then disentangles the contribution of ability signaling to the observed collegewage premium through a counterfactual experiment. In that experiment, I examine the maximal

3

sustainable college wage premium in a hypothetical economy where college were not productivityenhancing; and I consider the difference between the college wage premium in the hypotheticaleconomy and the actual college wage premium as the contribution of productivity enhancementof college education. The model is estimated under various distributional parameterizations using1990 U.S. Census five percent Public Use Micro Sample. Under these parameterizations, I findthat college education enhances attendees’ productivity by about forty percent, and productivityenhancement accounts for close to two-thirds of the college wage premium.

The question of how information free riding problem in the labor market (which was the focus of[1]) may be alleviated or exacerbated by government policy is the focus of a joint paper with PeterNorman titled “Government-Mandated Discriminatory Policies: Theory and Evidence”[13, IER 2006 ]. In this paper we analyzed perverse incentive effects of government-mandateddiscriminatory policies. We study an economy with private and public sectors in which workersinvest in imperfectly observable skills that are important to the private sector but not to the publicsector. Government regulation allows native majority workers to be employed in the public sectorwith positive probability while excluding the minority from it. We show that even when the publicsector offers the highest wage rate, it is still possible that the discriminated group is, on average,economically more successful. The reason is that the preferential policy lowers the majority’sincentive to invest in imperfectly observable skills by exacerbating the informational free ridingproblem in the private sector labor market. The widening Chinese/Malay wage gap in Malaysiasince the adoption of its New Economic Policy in 1970, which is actually an economic policy thatexpanded and mandated discrimination against the Chinese and gave preferential treatment to theMalays, is consistent with our model, but difficult to rationalize with alternative explanations.

My most recent research interest on the topic of economics of discrimination centers on racialprofiling. In a joint paper with my student Shamena Anwar (now at the Heinz School of PublicPolicy at Carnegie Mellon) titled “An Alternative Test of Racial Prejudice in Motor VehicleSearches: Theory and Evidence” [12, AER 2006 ], we propose a simple model of trooperbehavior to design empirical tests for whether troopers of different races are monolithic in theirsearch behavior, and whether they exhibit relative racial prejudice in motor vehicle searches. Thetest we propose is a variant of the outcome test of Gary Becker. However, our proposed test ofrelative racial prejudice provides a partial solution to the well-known infra-marginality and omitted-variables problems associated with outcome tests. More specifically, the infra-marginality problemrefers to the problem that the “marginal” is typically not equal to “average.” While the rigoroususe of the outcome test relies on the comparison of “marginals,” typically only “averages” areobserved. Our proposed test relies on the comparisons of average search rates and search successrates against a given race of motorists by police officers of different races. Because our test relieson the comparison of search rates and search success rates against the same race of motorists, itis not subject to the infra-marginality problem as long as officers of different races face the samedistribution of motorists of that given race. We also propose resampling methods to deal withsituations for which this assumption is not valid. We call our test “rank order invariance test”because the test will reject the null hypothesis that the troopers of different races do not exhibitrelative racial prejudice when the ranking over officers of different races of the average search rates

4

and/or search success rates differ by the race of the motorists. When applying to a unique datasetfrom Florida, our tests soundly reject the hypothesis that troopers of different races are monolithicin their search behavior, but the tests fail to reject the hypothesis that troopers of different racesdo not exhibit relative racial prejudice.

1.2 Research Agenda

This will continue to be one of my primary research and teaching interests. In the immediatefuture, I will be exploring several projects in this topic.

Optimal Contract Design in Law Enforcement [35]. My research on racial profiling asreported in “An Alternative Test of Racial Profiling in Motor Vehicle Searches: Theoryand Evidence” [12, AER 2006 ] has led me to consider the “industrial organization” issuesof law enforcement. Imagine that a policy maker is interested in some objective function (which,for example, may be a combination of crime rates, or victimization rates, in the population andthe number of innocent drivers being searched). Assume that police wants to maximize their ownutility (which includes monetary payoffs and costs of search efforts, for example). Can we saysomething interesting about the optimal contract? This is an interesting principal-agent problemwhere the principal is the policy maker, and the agent is the police, but the feature that is novelto this application is that, here, the agent, namely, the police, is not completely in control of theoutcome that the principal cares about. For example, the principal cares about crime rate in thepopulation, but the crime rate itself is determined as the equilibrium outcome of the game playedbetween the police and the citizens.

Identification of Taste and Statistical Discriminations [36]. I am also interested in system-atically understanding how taste and statistical discrimination can be empirically distinguished.

“The Book Project:” Economics of Discrimination: Theory, Empirical Methods and

Evidence [37]. It is almost fifty years since the publication of Becker’s classic “The Economicsof Discrimination” (Chicago University Press 1957). The theoretical literature of discriminationhas developed tremendously in this time span. Theories of statistical discrimination, discriminationwith endogenous group choices, discrimination as a result of inter-group interactions all contributeto a better understanding of the disparate outcomes between groups. On the empirical side, manymethods such as outcome tests, audit methods and their combination have been developed andpolished. Ample empirical evidence has accumulated. In teaching this material in my Ph.D. publiceconomics class at Yale, I have found that students are very interested in this topic. I am planningto write a manuscript on this topic based on lecture notes I accumulated over years in teaching thistopic. A book outline has been prepared.

5

2 Social Economics

2.1 Summary

This project attempts to bring sociological concepts into economic analysis; and study therole of various important sociological constructs such as social norms, social status, culture andidentity in economic performance from a rational-choice perspective. My interest on this topic isheavily influenced by my Ph.D. advisors Andrew Postlewaite and George Mailath who have studiedthe effect of social norms in regulating non-market interactions. More recently I have also beeninterested in understanding the phenomenon of residential segregation, neighborhood emergenceand transition, as well as risk sharing properties of social networks.

My first attempt on this topic was an application of the discrimination with endogenous groupchoices model on social culture, reported in “Social Culture and Economic Performance” [1,AER 2001 ] which I described earlier.

With Glenn C. Loury, I have worked on a paper titled “Toward an Economic Theory ofCollective Identity” [9, in Social Economics of Poverty, Chapter 2 ]. “Identity” and “col-lective identity” are important sociological concepts that have not been incorporated into economicanalysis until recently by George Akerlof and Rachel Kranton. They proposed models of iden-tity where an individual’s identity directly enters his or her utility function. We take a differentapproach and ask the question of how such notion of identity-behavior mapping in Akerlof andKranton could be modelled in a rational-choice framework. In the process, we advance a novelchoice-theoretic model of “identity” based on the notions of categories and narratives. Identity, inour paper, is conceived as a matter of “reflexive perception” — how people understand themselves.Choosing an identity is equivalent to making a generalization about one’s past that highlights themost salient aspects of experience. When many individuals make a common choice in this regard,they embrace a collective identity which is dysfunctional if it is Pareto dominated by an alternativeself-classificatory schema. Using a simple multi-stage risk sharing game, we explore conditions un-der which dysfunctional collective identities might be expected to emerge. We show that differentcollective identities have different implications on the scope of risk sharing. A summary of thislonger book chapter also appeared in “Dysfunctional Identities Can be Rational” [8, AER

P&P 2005 ].My recent effort in this research topic is the collaborated research titled “Separate when

Equal? Racial Inequality and Segregation” [19, NBER Working Paper 11507 ] withPatrick Bayer and Robert McMillan. In this paper, we question the standard intuition whichsuggests that residential segregation in the United States will decline when racial inequality narrows.Instead, we hypothesize that the opposite will occur. We note that middle-class black neighborhoodsare in short supply in many U.S. metropolitan areas, forcing highly educated blacks either to livein predominantly white high-socioeconomic status (SES) neighborhoods or in more black lower-SES neighborhoods. Increases in the proportion of highly educated blacks in a metropolitan areamay then lead to the emergence of new middle-class black neighborhoods, causing increases inresidential segregation. We formalize this mechanism using a simple model of residential choicethat permits endogenous neighborhood formation. Our primary empirical analysis, based on across-

6

MSA evidence from the 2000 Census, indicates that this mechanism does indeed operate: as theproportion of highly educated blacks in an MSA increases, so the segregation of blacks at alleducation levels increases. Time-series evidence provides additional support for the hypothesis,showing that an increase in black educational attainment in a metropolitan area between 1990-2000significantly increases segregation. We discuss in length how other seemingly plausible mechanismsdo not fit the evidence. Our analysis has important implications for the evolution of both residentialsegregation and racial socioeconomic inequality, drawing attention to a negative feedback loop likelyto inhibit reductions in segregation and racial inequality over time.

2.2 Research Agenda

Collective Identities and Economic Incentives [26]. On the issue of collective identity,Glenn Loury and I are currently investigating the incentive effects of collective identity. Thestarting point of this analysis is the fact that in our previous papers [8, 9] we assumed that one’sexperience (in the risk-sharing example, one’s income realizations each period) is not affected by theadopted collective identity in the community. But if different collective identity implies differentscopes of risk sharing, and to the extent that individuals can choose effort that may affect theincome realizations, we may think that collective identities that are good at achieving ex post risksharing may be bad in providing ex ante effort incentives. We show that in a world of incompleteincome indicators (which was the reason for the role of collective identities in the first place), suchintuition is not necessarily true.

Endogenous Social Status [30]. In this joint work with Fali Huang at Singapore ManagementUniversity, we endogenize an individual’s social status as the sum of “social rewards” to this in-dividual chosen by other people. This differs from previous papers in this important social statusliterature where individual social status is typically determined by his/her position in the societalincome distribution. We show that there exists an equilibrium in which individuals will chooseappropriate amount of social rewards (or penalties) to others at levels that exactly compensatefor valuable externalities generated by others. As a result social optimum is realized in such anequilibrium. We also generalize the result to situations where social status is group specific ratherthan individual specific. We also compare this efficient equilibrium with other social status equi-librium and investigate how individual effort, group size and composition are affected. The paperalso sheds light on the relationship between social status and relative performance concerns.

How Do Middle-Class Black Neighborhoods Emerge? [33] This project continues my col-laborated research with Patrick Bayer and Robert McMillan as reported in [19, NBER Working

Paper 11507 ]. In that paper we found that in 2000 there were about 143 Census tracts (outof a total of more than 49,000 tracts in the US) whose residents were more than 40% black andmore than 40% college-educated simultaneously. We call such neighborhoods middle-class blackneighborhoods. We know exactly where these 143 middle-class black neighborhoods were locatedin 2000. We argued in our previous research that such kind of middle-class black neighborhoods

7

are most desirable for highly-educated blacks, and the emergence of such neighborhoods plays animportant role in determining whether, in the long run as blacks catch up with whites in theirsocial economic status, we can expect to see more or less residential racial segregation.

In this project, we go a step further and ask the question of how these precious middle-classblack neighborhoods we saw in the 2000 Census emerge. To do this, we use the time series Censustract level data going back to 1970s compiled by Geolytics. We have extracted data from GeolyticsCensus time series data the tract level characteristics covering the period of 1970-2000. We willfirst trace back the characteristics of the current 143 middle-class black neighborhoods to see howthey evolved over time; we will then ask, back in 1990, 1980 and 1970, were there neighborhoodssimilar to those 143 back then? We will then investigate whether there are systematic determinantsof whether neighborhoods similar in their race/education composition will turn into a sustainablemiddle-class black neighborhoods, or disintegrate into low-SES black neighborhoods?

Racial Disparities in Risk Sharing Properties of Social Networks: Theory and Evidence[32]. This research, which is joint work with Andrea Moro and Giorgio Topa at the ResearchDepartment of the Federal Reserve Bank of New York, is motivated by the familiar images fromthe catastrophe caused by Hurricane Katrina where throngs of blacks, but rarely whites, wereseen in the shelters. Of course it is true that New Orleans had a very large percentage of blackpopulation, but there were still lots of whites, and especially poor whites. Katrina is a regionalcatastrophe, and it is a shock that many residents, unfortunately also all levels of governments,were not well prepared for. Our research question is, what explains the fact that whites, andeven those poor whites, are able to deal with this unforeseen negative shock better than blacks?We will focus on the social networks of individuals and families, and understand the sources ofthe racial disparities in the ability of their social networks to cope with negative shocks. Morespecifically my research hypothesis is that blacks were disproportionately represented in the shelterpopulation not only because they were poorer on average, but also because their social networkswere less diversified geographically. The question is what are the sources of this racial disparity inrisk sharing properties of social networks.

A simple theoretical model of residential location choices for different members of the family.In words, the model can be described as follows. Imagine that there are two possible residentiallocations, A and B; and consider the location decisions of a family of two people. Each individual’sincome will depend on his/her individual characteristics and the common characteristics of thelocation he/she chooses to live in. If the two members of the family live in the same location,the benefit is that they can tend to each other’s individual shock at a lower cost. For example, ifone member or his/her children are sick, the other family member can help. The cost of living inthe same location is that, if the location is hit with a negative shock such as a hurricane (as inNew Orleans) or experiences a decline of the main industry (as in Detroit), then there is no one tohelp them. The costs and benefits for the family members to live in different cities are exactly theopposite. The cost is that they cannot tend to each other’s daily negative shocks; but the benefitis that they can help each other out when one region, but not the other, is hit with a negativeshock. The theoretical part of the project aims to derive testable implications about how families

8

with different means will optimally choose their residential location patterns with these trade-offsin mind.

Empirically, we would like to, first of all, document the racial differences in social networks, interms of residential location and occupation. For this purpose we will be working with NationalSurvey of Families and Households (NSFH), Panel Study of Income Dynamics (PSID) and NationalLongitudinal Survey of Youth (1979 Cohort and their children supplement). We will also attemptto use recent regional data from the Federal government related to Katrina to do further analysison these issues. Second, we would like to test the implications of our theoretical model.

3 Assessing the Impact of Welfare Reform of 1996 on Single Moth-

ers and Their Children

3.1 Summary

This project originated with the puzzling observation that welfare reform of 1996, with its timelimits and work requirement, was advertised by politicians as being compassionate to the recipientsthemselves. This is puzzling because in standard economic models, eligibility restrictions such astime limits and work requirement will necessarily make a potential welfare recipient worse off.

With Dan Silverman in a paper titled “On the Compassion of Time-Limited WelfarePrograms” [5, JPubE 2004 ], we started out by proposing a simple model of agents with present-biased preferences that may provide a rationale for the claim. We first identify four types of outcomethat describe the behavior of a present-biased agent in the absence of time limits. We then showthat the behavioral consequences of time limits are contingent on which outcome characterizes theagent’s behavior in the absence of time limits. We show that under some conditions the impositionof time limits may improve the wellbeing of welfare recipients evaluated both in terms of long-run,time-consistent utility and the period-one self’s utility. This benefit of time limits may come eitherfrom allowing the welfare eligible to start working earlier than they otherwise would or, contraryto the intent of the reforms, from allowing them to postpone working.

We were not satisfied to merely have a possible theoretical rationale for the possible compas-sion of time-limited welfare reform; we want to know the empirical relevance of such theories. Forthat purpose, in a companion paper “Time Inconsistency and Welfare Program Participa-tion: Evidence from the NLSY” [16, CFDP 1465], we empirically implement a dynamicstructural model of labor supply and welfare program participation for agents with potentiallytime-inconsistent preferences. Using panel data on the choices of single women with children fromthe NLSY 1979, we provide estimates of the degree of time-inconsistency, and of its influence on thewelfare take-up decision. With these estimates, we conduct counterfactual experiments to quantifythe utility loss stemming from the inability to commit to future decisions, and the potential utilitygains from commitment mechanisms such as welfare time limits and work requirements.

My knowledge in the institutional details of the US welfare system was significantly boosted inmy joint work with Michael Keane on the paper “Assessing the Impact of Welfare Reformon Single Mothers” [6, Brookings Papers on Economic Activity 2004 ]. The main goal

9

of this paper is to ascertain the contributions of various components of welfare reform, and othercontemporaneous economic and policy changes, to the huge decline in welfare participation andincrease in work among single mothers from 1993-2002. To this end, we have constructed an exten-sive data set that characterizes changes in welfare policy, as well as other important determinantsof welfare participation and work, at the State level for the 1980-2002 period. Using these datatogether with the individual level data from the Current Population Survey, we develop a modelthat rather successfully explains both levels and changes in welfare and work participation ratesacross States, time, and various demographic groups, for the whole 1980-2002 period. We then usethe estimates of the model and simulation to obtain estimates of the contributions of various policyand economic variables to the labor supply of single mothers.

3.2 Research Agenda

Assessing the Impact of Welfare Reform on the Well-Being of Single Mothers and TheirChildren [23]. This project is a natural extension of my earlier work with Michael Keane asreported in “Assessing the Impact of Welfare Reform on Single Mothers” [6, Brookings

Papers on Economic Activity 2004 ]. Recall in that paper we attempted to disentangle thecontributions of various elements of the welfare reform and other economic and policy variables onthe labor supply/welfare participation decisions of single mothers. In this paper we will insteadfocus on the impact of welfare reform on the well-being of single mothers and their children. Forthis purpose we will first use Consumer Expenditure Survey data from 1983-2004, PSID data from1983-2004, Current Population Survey from 1980-2004 and the children’s sample of NLSY 1979to document the changes in the distribution of income, consumption, expenditure and children’scognitive measures. Then we will use the extensive policy and economic variables data I collectedearlier with Michael Keane to ask what elements are most responsible for the observed changes inthese measures of well-being for the single mothers and their children. This project is supportedby Smith Richardson Foundation in the form of a Domestic Public Policy Fellowship.

Time Aggregation in Dynamic Choice Models when Agents Have Time InconsistentPreferences: Monte Carlo Evidence [34]. In the empirical research reported in “Time In-consistency and Welfare Program Participation: Evidence from the NLSY” [16, CFDP1465], we made the rather arbitrary assumption that a decision period lasts for a year. Similarassumptions about the length of the decision period are also made in all empirical studies of dy-namic choice models. These assumptions may be valid if we know exactly when and how frequentlyan agent can choose to make a move, but in most situations they are made for convenience, oras a result of data availability restrictions. The assumption about the length of a decision periodbecomes more important with hyperbolic-discounting decision makers because higher rate of dis-counting is assumed between now and the next period, but not between periods thereafter. In thisproject I plan to uses Monte Carlo simulations to investigate if standard methods would be ableto uncover the parameter that measures the length of a period an agent uses in decision making.More generally I would like to understand better the issue of time aggregation in applied dynamic

10

choice models.

4 Applications of Psychology in Economics

4.1 Summary

My interest in this topic initiates in my research on the compassion of time-limited welfareprograms described above. That project brought me to the realm of behavioral economics. Mygeneral view on this topic is that economists should not dismiss off-hand robust behavioral anoma-lies. Incorporating such behavioral biases into economic analysis does not at all mean the rejectionof rationality; rather, it implies rationality subject to more constraints, or rationality with respectto non-standard preferences or technologies.

My research in this vein includes two research papers described above with Dan Silverman ontime inconsistent preferences and welfare participation, namely, “On the Compassion of Time-Limited Welfare Programs” [5, JPubE 2004 ] and Inconsistency and Welfare ProgramParticipation: Evidence from the NLSY” [16, CFDP 1465].

We also jointly worked on a third paper titled “Distinguishing Between Cognitive Biases:Beliefs vs. Time Discounting in Welfare Program Participation” [10, in Behavioral

Public Finance, Chapter 3 ]. In that paper, we start to worry about how different formsof psychological biases can be empirically distinguished. We consider this to be an importantchallenge for behavioral economists as they introduce more and more cognitive biases into economicmodels. We present a simple model of welfare program participation that nests two well-documentedcognitive biases that economists have recently incorporated into their analyses: projection bias andpresent bias. We argue that agents with present bias and projection bias will exhibit differentattitude toward time limits and other welfare eligibility restrictions, both before and after suchrestrictions are implemented. To the extent that such attitudes can be accurately elicited andmeasured, we argue that we can use attitudinal data to distinguish present bias and projection biasmodels.

With Giuseppe Moscarini, I have examined the implications of overconfidence on firms’ optimalwage setting policies in a paper titled “Morale Hazard” [7, Journal of Monetary Economics

2005 ]. This paper provides a new and fully rational theory of fairness in wage setting, based onthe notion that workers may be overconfident about their own ability. (Self overconfidence of thepossession of desirable traits is indeed a very robust psychological finding.) In that model, weassume that a worker is uncertain about his own ability, which affects the marginal productivityof his effort. As a result, a worker’s effort responds to wage incentives both directly and indirectlythrough its effect on his perception of his own ability, which we call his “morale”. If the firm receivesnoisy and private information about worker’s ability through performance evaluations for example,then any wage offers by the firm play both the direct allocative role as well as a signaling role.Wage differentiation reveals who did and did not do well in the firm’s evaluation, thus affectingthe individual morale of each employee. The resulting decline in the effort of some workers, dueto their discouragement when receiving a lower wage (or a wage cut), may more than offset the

11

encouragement of the others, and be detrimental to the firm overall.

4.2 Research Agenda

Time Aggregation in Dynamic Choice Models when Agents Have Time InconsistentPreferences: Monte Carlo Evidence [34]. This project was briefly described above in Section3.

Is there a Mickey Mantle Effect? The Effect of Subjective Life Expectancy on HealthBehaviors [25]. In this joint work with Michael Keane, Ahmed Khwaja, Martin Salm and DanSilverman, we investigated the implications of optimism on the demand for health investment andhealth insurance. Beliefs about future health should play an important role in determining currenthealth decisions. In theory, the direction of the effect of differences in expectations for longevity onhealth investments is unclear. Greater optimism may make current health investments either moreor less appealing. OLS estimates using data from the Health and Retirement Study show that beliefsabout longevity are only weakly associated with current health decisions after conditioning a numberof other demographic and economic characteristics. Problems of reverse causality and omittedvariables imply that these OLS estimates may be misleading. An individual’s index of generaloptimism, measured as the tendency to answer questions about macroeconomic and policy eventsoptimistically, is not subject to the reverse causality problem. There is a statistically significantand relatively large positive association between general optimism and most health investments. Asan instrument for beliefs about future health, however, the index of optimism is weak. Therefore,IV estimates indicating that more optimistic beliefs about future health substantially increase thelikelihood of making most health investments are difficult to interpret. This finding does have aninteresting policy implication. Currently in the US and elsewhere, the government informationcampaign against smoking and drinking has mainly focused on the negative consequences, bothon health and on life in general, of such actions; our finding suggests a useful complementaryinformation campaign that emphasizes to the viewers that the medical technology has improved somuch that one can actually live a quite long life. Once viewers are convinced that they are likelyto live long, they would choose to smoke and drink less!

5 Multi-dimensional Private Information: Auctions, Public Goods

and Insurance Market

5.1 Summary

Issues of multi-dimensional private information appear in several of my research papers. Myfocus in this research topic is on how multidimensionality affects the equilibrium of standard mech-anism, and how it affects the optimal mechanisms, and how it affects the market equilibrium.

It started in my joint work with Sergio Parreiras (currently at UNC-Chapel Hill) when wewere both still Ph.D. students at the University of Pennsylvania in a project tilted “Auctions with

12

Financially Constrained Bidders.” Two publications came out of this project. The first is titled“Equilibrium of Affiliated Value Second Price Auctions with Financially ConstrainedBidders: The Two-Bidder Case” [2, Games and Economic Behavior 2002 ]. In thatpaper, we study affiliated value second price auctions with two financially constrained bidders. Inthis auction environment bidders have two dimensions of private information, their signal for thevalue of the object and their bidding budget. We prove the existence of a symmetric equilibriumunder quite general conditions and provide comparative static results. In the second paper is tilted“On the Failure of the Linkage Principle with Financially Constrained Bidders,” [3,Journal of Economic Theory 2003 ], we provide a class of examples of two-bidder commonvalue second price auctions in which bidders may be financially constrained and the seller hasaccess to information about the common value. We show that the seller’s expected revenue undera revelation policy may be lower than that under a concealing policy. The intuition for the failureof the linkage principle is as follows. In the presence of financial constraints, the bidders’ upwardresponse in their bids to the seller’s good signals is limited by their financial constraints, while theirdownward response to bad signals is not.

With Stephen Morris, I wrote a paper titled “Multidimensional Private Value Auctions”[11, Journal of Economic Theory 2006 ] in which we consider parametric examples of sym-metric two-bidder private value auctions in which each bidder observes her own private valuationas well as noisy signals about her opponent’s private valuation. Thus in our set up, bidders havemulti-dimensional private information in the sense that they both possess private information abouttheir own valuation for the object (which we call “valuation type” and is the same as the standardauction models) and private information about signals about their opponents’ value (which we call“information type” and it does not appear in standard auction models). This provides the simplestframework to examine the effect of higher-order beliefs on the equilibrium of standard auctions.We show that, in such environments,the revenue equivalence between the first and second priceauctions breaks down and there is no definite revenue ranking; while the second price auction isalways efficient allocatively, the first price auction may be inefficient; equilibria may fail to existfor the first price auction. We also show that auction mechanisms provide different incentives forbidders to acquire costly information about opponents’ valuation.

The work with Stephen Morris led to a collaboration with David J. Cooper at Case WesternReserve University on an experimental economics project titled “Understanding Overbiddingin Second-Price Auctions: An Experimental Investigation” [21, Cowles Foundation

Discussion Paper 1557 ]. It is well-known that overbidding is prevalent and persistent in labexperiments. Surprisingly economists have rather limited empirical understanding of why overbid-ding occurs in SPA. The key idea of our paper is the following. Suppose that bidders are eithergiven for free, or are allowed to purchase, noisy signals about their opponents’ value. Even thoughin standard models of SPA, such information about opponents’ value theoretically has no strategicuse, it provides us with a convenient instrument to change bidders’ perception about the “strength”(i.e. the value) of their opponent. This empirical relationship between the incidence and magni-tude of overbidding and bidders’ perception of the strength of their opponent provides the keyto understand whether overbidding in second price auctions are driven by “spite” motives or by

13

the “joy of winning.” Our experimental data show that bidders are much more likely to overbid,though less likely to submit large overbid, when they perceive their rivals to have similar valuesas their own. We argue that this empirical relationship is more consistent with a modified “joyof winning” hypothesis than with the “spite” hypothesis. However, neither of the non-standardpreference explanations are able to fully explain all aspects of the experimental data. We find clearevidence of learning both in avoiding costly overbidding and in subjects’ choices to purchase costlyinformation, thus lending support for the role of bounded rationality. We also find that bidderheterogeneity plays an important role in explaining their bidding behavior.

I have continued the investigation of the effect of multi-dimensional private information insettings other than auctions. The next topic of interest is the provision of multiple excludable publicgoods. This project is joint with Peter Norman and it resulted in a line of research on bundling thathas eventually gone beyond public goods. Before describing the details of the papers in this project,let me state up-front that mechanism design for multi-dimensional private information environmentis a very difficult problem whose complete solution will probably require new mathematical tools.Facing such a technical barrier, three different approaches, none of which is perfect, are taken inour research. The first approach is simply to do with simple parametric family of examples; thesecond approach takes the number of goods to infinity and appeal to law of large numbers; and thethird approach focuses on local deviations instead of global analysis of the optimal mechanism.

In “Optimal Provision of Multiple Excludable Public Goods” [17, Cowles Founda-

tion Discusion Paper 1441R], we took the first approach. We study the optimal provisionmechanism for multiple excludable public goods when agents’ valuations are private information.We provide some partial characterizations for the optimal mechanism that significantly reduce theset of mechanisms that we need to consider, but are unable to characterize the optimal mechanismfor general environments. However, we are able to completely characterize the optimal mechanismfor a parametric class of problems with binary valuations. We show in that parametric family ofexamples that the optimal mechanism involves bundling. Bundling alleviates the free riding prob-lem in large economies in two ways: first, it can increase the asymptotic provision probability ofsocially efficient public goods from zero to one; second, it decreases the extent of use exclusions.

In “Overcoming Participation Constraints” [18, Cowles Foundation Discussion Pa-

per 1511R], we took the second approach and ask whether any almost efficient mechanism isfeasible when the number of goods go to infinity. This paper is most closely related to a recentEconometrica paper by Matthew O. Jackson and Hugo Sonnenschein titled “Overcoming Incen-tive Constraints by Linking Decisions.” In that paper, Jackson and Sonnenschein showed that theutility costs associated with incentive constraints become negligible when the decision problem islinked with a large number of independent replicas of itself under the linking mechanism in whichagents must budget their representations of preferences so that the frequency of preferences acrossproblems mirrors the underlying distribution of preferences, and then arguing that agents’ incen-tives are to satisfy their budget by being as truthful as possible. The nice feature of Jackson andSonnenschein’s result is that the linking mechanism works under non-transferable utility environ-ments. Our paper complements Jackson and Sonnenschein’s analysis because it departs from theirin two dimensions. The first difference is that we restrict ourselves only to transferable utility

14

environments, which is obviously a shortcoming. The second difference, however, is a relaxation:we consider a large number of independent but unrelated issues. We provide regularity conditionsunder which a Groves mechanism amended with a veto game implements an efficient outcomewith probability arbitrarily close to one, and satisfies interim participation, incentive and resourceconstraints.

The paper “To Bundle or Not To Bundle” [15, Rand Journal of Economics, forth-

coming,] is about the profit-maximizing bundling decisions for a monopolist seller who sells finitenumber of private goods. The idea in this paper grew out of the public good provision research re-ported above where we first discovered the usefulness of the statistical concept called “peakedness.”We focus on the finite product case because for the infinite case research by Mark Armstrong hasshown that the monopolist can extract all the surplus by exploiting the laws of large numbers. Ourknowledge about whether bundle increases profits in the case of finite number of goods is limitedto a few numerical examples by Richard Schmalensee. In this paper, we obtain a rather intuitivecharacterization for when a multi-product monopolist should bundle and when it should sell thegoods separately in order to maximize its profits. To some extent this confirms (mainly) numericalresults in Schmalensee’s studies, namely, the higher the marginal cost and the lower is the meanvaluation, the less likely that bundling dominates separate sales. When limiting our comparison topure bundling and separate sales, we are able to highlight a clear intuition for what happens whentwo or more goods are sold as a bundle. The key effect driving all the results is that the variancein the relevant willingness to pay is reduced when goods are bundled. In our paper we provide apartial characterization for when this reduction in variance is beneficial for the monopolist, andwhen it is not.

Multi-dimensional private information also appears somewhat surprisingly in an empirical paper(joint with Michael Keane and Dan Silverman) titled “Sources of Advantageous Selection:Evidence from the Medigap Insurance Market” [22, NBER Working Paper 12289 ].This paper is closely related to the empirical implications of the classical theoretical models ofinsurance. These models assume that potential insurance buyers have one-dimensional privateinformation regarding their risk type, and they choose from a menu of contracts, specifying theprice and amount of coverage, the one best suited to their type. A testable empirical predictionof these models is the so-called “positive correlation property,” namely, there must be a positivecorrelation between insurance coverage and ex post realizations of loss. This property is implied bythe standard insurance model of one-dimensional private information with or without moral hazard.However, recent empirical papers fail to find empirical support for the positive correlation propertyin a variety of important insurance markets, including life insurance, automobile insurance andhealth insurance markets. These empirical findings have point toward the possibility of selectionbased on multi-dimensional private information, and the existing literature has focused on riskaversion as the leading suspect for other dimensions of private information that may affect thedemand for insurance and cause the positive correlation property to fail. However, so far theliterature has not provided direct evidence to support this conjecture. Our paper fills this gap.We study the advantageous selection phenomenon using data from the Medicare supplemental or“Medigap” insurance market. We show that the Medigap insurance market is characterized by

15

advantageous selection: conditional on the determinants of price, those who purchase supplementalinsurance tend to be healthier than those who do not. More importantly, we go beyond priorresearch to investigate several potential sources of this advantageous selection. We provide directevidence that risk aversion, can explain a substantial portion of the observed advantageous selectionin the Medigap market. But that is not the whole story. We find that additional important sourcesof advantageous selection include selection based on education, income, longevity expectations andfinancial planning horizons, along with other measures not typically included in economic models,such as cognition and numeracy. Once we condition on all these variables, we find that individualswith higher expected medical expenditures are indeed more likely to purchase Medigap insurance.

5.2 Research Agenda

Public Provision of Private Goods as Constrained Efficient Outcome [29]. In this paper,Peter Norman and I pursue a third strategy described above, namely, local deviations strategy, todeal with multi-dimensional mechanism design problem. Specifically, we show that public provisionof private goods may be justified on pure efficiency grounds in an environment where consumersconsume both public and private goods. The idea is that public provision of a bundle that consistsof a private and a public good can make it easier to extract revenue from the customers for reasonsfamiliar from the literature on multiproduct monopolies. We show that partial public provisionof the private good improves economic efficiency under a condition that is always fulfilled understochastic independence and satisfied for almost all distributions.

Akerlof’s Lemons Market Revisited: What if Sellers Have Multi-dimensional PrivateInformation? [24] The empirical results reported in “Sources of Advantageous Selection:Evidence from the Medigap Insurance Market” [22, NBER Working Paper 12289 ]point toward several interesting theoretical questions. The first question is the relationship be-tween selection based on multi-dimensional private information and the equilibrium size of themarket. Consider two insurance markets, say life insurance and annuity insurance markets. Inboth insurance markets suppose that individuals choose to buy more insurance if they have higherrisks (mortality risk for life insurance and longevity risk for annuity insurance) and if they havehigher values of another variable γ (I am deliberately vague about what γ is, but it is safe to assumeit denotes risk aversion). Suppose that in one market (say life insurance market) γ is negativelycorrelated with risk (i.e. mortality risk), but in another market (say annuity insurance market), γ

is positively correlated with risk (i.e. longevity risk). What can be said about the equilibrium sizeof the markets? Would it always be the case that the equilibrium market size will be higher in thefirst market where γ acts as a source of advantageous selection to alleviate the adverse selectiondue to selection based on risk? What about efficiency? These questions are important for a betterunderstanding of the implications of multi-dimensional private information on market equilibria.In this proposed research I intend to examine Akerlof’s lemons market which is a much simplersetting that allows me to abstract from the screenings by the sellers.

16

6 Empirical Analysis of Asymmetric Information and Dynamic

Inefficiency in Markets

The research on advantageous selection in Medigap insurance market reported in “Sources ofAdvantageous Selection: Evidence from the Medigap Insurance Market” [22, NBER

Working Paper ] further spurred a larger and new agenda of empirically analyzing the relevanceand importance of different forms of asymmetric information and dynamic inefficiency in a varietyof markets. Here I describe several projects that I will actively pursue in the next five years thatbelong to this research agenda.

Understanding the Dynamics of Advantageous Selection in Medigap Insurance Market[38]. In this joint work with Michael Keane and Dan Silverman, we continue our investigation ofthe selections in the Medigap insurance market. The starting point of the project is that we onlyexploited the cross-sectional variation in “Total Medical Expenditure” by Medigap status in [22]; wedid not examine the possible changes of Medigap coverage status over an individual’s lifetime. Thishas important implications because the price control we used (age polynomial, gender and Stateof residence) is no longer accurate if an individual’s Medigap coverage ever lapsed. In that case,insurance companies may subsequently impose both coverage and pricing rules different from thosewhich apply during the open enrollment period. Thus our Medigap price controls (age polynomial,gender and state of residence) do not reflect the prices faced by those who let their coverage lapse.In this case, our finding that expenditures for those with Medigap are about $4,000 less than forthose without may, to some extent, reflect the following possibility: those without Medigap areless healthy because their coverage previously lapsed and, moreover, are currently priced out ofMedigap due to their poor health.

It may be argued that if this mechanism is influencing our estimates its effect is consistent withour interpretation of advantageous selection. Consider the group of individuals without Medigapwho are unhealthy and priced out of Medigap because of a lapse in their coverage. One possibilityis that they never purchased a Medigap policy in the first place. This would be consistent withour interpretation of advantageous selection: less healthy individuals are less likely to purchaseMedigap during open enrollment (when everyone is approximately the same age and thus state andgender controls alone would be sufficient to control for Medigap pricing). The second possibility isthat those without Medigap did purchase a policy during open enrollment, but subsequently failedto renew. The standard adverse selection model with one dimensional private information aboutrisk suggests that less healthy people should be more likely to renew, just as they should be morelikely to enroll in the first place. If, in contrast, less healthy people are less likely to renew, this isarguably also a form of advantageous selection.

In this proposed research, we would like to investigate in detail the relative importance of openenrollment period and subsequent renewal in the overall advantageous selection we documentedearlier. Equally importantly, we would like to know whether the sources of advantageous selectionin the open enrollment period and in the renewals are different. We plan to use the rotating panelsavailable in MCBS and HRS to address these questions by following individuals overtime from when

17

they just become eligible for Medicare at age 65. Another goal of the project is to explain the agepattern of the magnitude of advantageous selection.

Adverse Selection or Moral Hazard? The Case of Commercial Bank Loans [39]. Thisproject is joint with Bin Liu and Li-An Zhou, both at Peking University in China. This projectis aimed to empirically distinguish adverse selection and moral hazard – two forms of asymmetricinformation – in the case of commercial bank loan market. The empirical strategy crucially relies onthe fact that for commercial bank loans, the lenders can use multiple instruments including interestrates, collateral, loan maturity and probability of granting loans etc. to screen borrowers of differenttype. Theoretical analysis showed that adverse selection and moral hazard models robustly predictdifferent relationships between collateral (incidence as well as amount), loan maturity and ex postperformance of the loans. More specifically, in adverse selection models, good borrowers are morelikely to asked to pledge collateral and will choose loans with shorter maturity in order to signaltheir high quality; in contrast, in moral hazard models, lower quality borrowers are more likelyto pledge collateral and when they do post collateral, there is a negative relationship between theprobability of success and the amount of collateral, and they will be offered loans with shortermaturity. We then empirically test these predictions using a unique data set of loan characteristicsand performance from a large Chinese commercial bank.

Dynamic Inefficiency in Health Insurance Market [40]. In this joint project with Alessan-dro Gavazza of Yale SOM, we attempt to empirically examine the dynamic inefficiency under anemployer-provided health insurance system. As is well known, most people in the United Statesreceive health insurance coverage through their employers. An important feature of health is thatit is a form of general human capital : a healthy worker is presumably more productive on her cur-rent job as well as other potential jobs. The standard Beckerian competitive labor market modelspredict that general human capital would be provided at an efficient level, paid for by the em-ployers. However, frictions in the labor market may lead to inefficiency in the provision of generalhuman capital. In this project, we first presents a theoretical model that predicts in equilibriuma positive relationship between the importance of industry-specific skills and the amount of healthinvestment. The reason for this relationship is that, firms in industries that value specific humancapital more will find it optimal to invest in specific human capital so as to lower the probabilityof worker/job separation, which in turn increases the incentives for investment in health, a formof general human capital. We then test the implications of the model using data from the MedicalExpenditure Panel Survey (MEPS) and Health and Retirement Study (HRS). In particular, weseek to investigate how the importance of industry-specific skills affect firms’ provision of healthcare and health insurance.

Risk, Risk Aversion and Moral Hazard: Estimating the Costs and Benefits of OfferingChoices in Health Insurance [41]. This planned research is joint with Liran Einav of StanfordUniversity, Amy Finkelstein of MIT and Mark Cullen of Yale School of Public Health. In thisproject, we plan to exploit two unique features of a unique administrative data set from Alcoa to

18

estimate the joint distribution of risk type and risk aversion in the population of Alcoa workers,and then use the estimates to evaluate the costs and benefits of offering choices in health insurance.First, in the pre-2004/Q1 period, Alcoa offered the same generous insurance plan to all workerswhich has no deductibles and has 5-10 dollars of copayments. Employees do not contribute to theinsurance premium), and moreover we observe all the health claims data. Second, post 2004/Q1,four insurance plan options are offered to a subset of workers according to their union contract statusin their working site, which can be thought of as more or less random. These insurance plans differin the employee’s premium contribution and deductible. The pre-2004/Q1 health claims data willallow us to directly estimate the risk type for each worker. Given the estimates of risk type frompre-2004/Q1 medical expenditure data, we can then use the choice over the four insurance optionsin the post-2004/Q1 regime to estimate risk aversion. Finally, we can use the post-2004/Q1 realizedexpenditure to estimate the elasticity to co-payment, which would provide us with an estimate ofmoral hazard (or price effect) in the demand for medical care. With these estimates, we can thenevaluate the welfare effects of a variety of counterfactual experiments.

Social Learning and Herding: Evidence from A Natural Randomized Field Experiment[42]. This project is joint with Hongbin Cai and Yuyu Chen, both at Peking University. Inthis project we plan to exploit a natural randomized field experiment conducted by a large chainrestaurants in Beijing, China, to assess the relevance and importance of social learning. Therestaurant chain has 17 sites in Beijing, with each site serving about 1,000 groups of diners on anaverage day. They adopted a practice of displaying a card with the “Top Five Dishes According tothe Number of Plates Sold Last Week” randomly in some sites and within the selected site randomlyon some tables. For infrequent diners with little information about the 35-40 dishes on their menu,the choices of other diners in the past could be informative in making their own decisions. The datafrom the restaurant provides us with an almost idea natural randomized field experiment to examineinformational herding. We are also planning further field experiments in order to distinguish the“saliency effect” from the “informational herding effect.” More specifically, “saliency effect” refersto the phenomenon that the dishes mentioned on the card may be more popular simply becausethey are more prominently mentioned than other dishes on the menu, which has nothing to do withinformational herding. We can distinguish the “saliency effect” from the “informational herdingeffect” by an experimental design of randomly displaying cards with dishes without mentioningtheir past sales. We can even provide an “anti-herding but salient” treatment by displaying cardsthat mention “The following dishes did not sell well in the past week. Do you want to give them atry?”

7 Other Research Interests

I now describe several research papers that are at this moment only my nascent efforts intonew research topics, and as such I am not yet ready to call them one of my main areas of researchinterest. I hope that some of them will eventually become part of a larger research agenda as mycareer progresses.

19

7.1 Issues Related to Development Economics

I am interested in the interaction between development economics and public economics. AsI emphasized earlier I view the distinction between these fields to be artificial. Individuals indeveloping economies make calculations similar to those in developed world, and the differenceis in their institutional constraints. As such data from developing economies provide important,sometimes vital, sources of information for us to understand the effects of different institutionalarrangements. I have worked on two research papers using micro level data from China, one tostudy the determinants and effects of “Entertainment and Travel Costs” expenditure item amongChinese firms, and the other to study the insurance role of Rotating Savings and Credit Association(ROSCAS).

“Corruption.” I worked with Hongbin Cai of UCLA and Peking University and Colin Xu ofthe World Bank on a paper titled “Eat, Drink, Firms and Government: An Investigationof Corruption From Entertainment and Travel Costs of Chinese Firms” [20, NBER

Working Paper 11592 ]. As some background, Entertainment and Travel Costs (ETC) is anexpenditure item in standard accounting books of firms in China that amount to about 20% oftotal wage bills in a sample of 3470 Chinese firms. Using a detailed dataset of these firms, we ana-lyze the composition of ETC and effects of ETC on firm performance. We develop a simple modelof managerial decisions on the amount of entertainment expenditures to spend on strengtheningrelational capital with suppliers and clients, bribing government officials, and private consumption.This model allows us to identify components of ETC by examining how they should respond todifferent environmental variables. We find strong evidence that firms’ ETC compromise a mix thatincludes expenditures on government officials both as “grease money” and “protection money,”expenditures to build relational capital with suppliers and clients, and managerial private con-sumption. Overall, ETC have significantly negative effects on firm performance, but their negativeeffects can be much less pronounced when their marginal returns are higher, particularly, undersevere government expropriation and when the quality of government service is very poor. Thisresearch has been profiled in mainstream print media in Singapore (The Strait Times) and in HongKong (South China Morning Post).

Rotating Savings and Credit Associations (ROSCAS) [31]. ROSCAS are an importantinformal financial institution in developing countries. In a typical ROSCA, a group of individuals,mostly connected in certain social network, commit to contributing a given sum of money at each ofequally-spaced dates to form a pot, and then the pot will be allotted through certain mechanism toeach of members in turn. There are two prevalent mechanisms to determine the order in which theseindividuals access the pot: random lottery and bidding. The seminal analysis by Timothy Besley,Stephen Coate and Glenn Loury modelled the role of ROSCAS as a Pareto improving institution inthe context of saving to purchase expensive durable goods. The subsequent literature has extendedtheir model in several important directions, but none has examined the role of ROSCAS in thepresence of formal credit markets. Indeed in a unique data set of a large number of bidding ROSCAS

20

from China (City of Wenzhou, Zhejiang Province), Rongzhu Ke (currently a Ph.D. student at MIT)and I found two interesting features. First, many of the ROSCAS participants actually have accessto formal credit market and in fact some of them are borrowing money from the formal creditmarket in order to fulfil their obligations in the ROSCAS. Second, the implicit interests for theseROSCAS typically are non-monotonic with respect to time, which is at odds with the predictionsfrom Besley, Coate and Loury’s model. These empirical observations motivate our theoreticalinvestigation of a model of ROSCAS in the presence of formal credit markets and focus on its roleof mutual insurance.

7.2 Issues Related to Education

I am also interested in two issues in education [27, 28]. My interest in these issues are newand not much formal progress has been made yet. The first issue is about college major choices.I am interested in documenting its change over time and understanding how much of the changesin college wage premium, Black-White wage gap, and gender wage gaps could be attributed tothe changes in the major choices of the college students, and the changes in the mapping fromcollege majors and occupations. Second, I am interested in understanding the labor market returnsto non-cognitive skills acquired through extracurricular activities. I plan to use nationally repre-sentative post-secondary education data sets, NLS-72, High School and Beyond-80, NELS 88 andBaccalaureate and Beyond-93 to examine these issues.

References

[1] Fang, Hanming. “Social Culture and Economic Performance.” American Economic Review,September 2001: 934-947.

[2] Fang, Hanming and Sergio O. Parreiras. “Equilibrium of Affiliated Value Second Price Auc-tions with Financially Constrained Bidders: The Two-Bidder Case.” Games and EconomicBehavior, May 2002, 215-236.

[3] Fang, Hanming and Sergio O. Parreiras. “On the Failure of the Linkage Principle with Finan-cially Constrained Bidders.” Journal of Economic Theory, June 2003, 374-392.

[4] Fang, Hanming. “Lottery versus All-Pay Auction Models of Lobbying.” Public Choice, Septem-ber 2002, 351–371.

[5] Fang, Hanming and Dan Silverman. “On the Compassion of the Time-Limited Welfare Pro-grams.” Journal of Public Economics, Volume 88, Issues 7-8 , July 2004, 1445-1470.

[6] Fang, Hanming and Michael P. Keane. “Assessing the Impact of Welfare Reform on SingleMothers.” Brookings Papers on Economic Activity, 2004, Volume 1, 1-116.

[7] Fang, Hanming and Giuseppe Moscarini. “Morale Hazard.” Journal of Monetary Economics,May 2005, Vol. 52, 749-777.

21

[8] Fang, Hanming and Glenn C. Loury. “ ‘Dysfunctional Identities’ Can Be Rational.” AmericanEconomic Review Papers and Proceedings, May 2005: 104-111.

[9] Fang, Hanming and Glenn C. Loury. “Toward an Economic Theory of Dysfunctional Identity,”in The Social Economics of Poverty: On Identities, Groups, Communities and Networks,Edited by Christopher B. Barrett, London: Routledge, 2005, pp. 12-55.

[10] Fang, Hanming and Dan Silverman. “Distinguishing Between Cognitive Biases: Belief vs. TimeDiscounting in Welfare Program Participation,” in Behavioral Public Finance, edited by JoelSlemrod and Edward McCaffery, Russell Sage Foundation, pp. 47-81.

[11] Fang, Hanming and Stephen Morris. “Multidimensional Private Value Auctions.” Journal ofEconomic Theory, January 2006, Vol. 126, 1-30.

[12] Anwar, Shamena and Hanming Fang. “An Alternative Test of Racial Prejudice in MotorVehicle Searches: Theory and Evidence.” American Economic Review, March 2006, Vol. 96,No. 1, 127-151.

[13] Fang, Hanming and Peter Norman. “Government-Mandated Discriminatory Policies: Theoryand Evidence.” International Economic Review, March 2006.

[14] Fang, Hanming. “Disentangling the College Wage Premium: Estimating a Model with En-dogenous Education Choices.” Forthcoming, International Economic Review.

[15] Fang, Hanming and Peter Norman. “To Bundle or Not To Bundle.” Forthcoming, Rand Journalof Economics.

[16] Fang, Hanming and Dan Silverman. “Time Inconsistency and Welfare Program Participation:Evidence from the NLSY.” January 2006, under journal review. [Cowles Foundation DiscussionPaper 1465]

[17] Fang, Hanming and Peter Norman. “Optimal Provision of Multiple Excludable Public Goods.”April 2006, under journal review. [Cowles Foundation Discussion Paper 1441R]

[18] Fang, Hanming and Peter Norman. “Overcoming Participation Constraints.” Cowles Founda-tion Discussion Paper 1511R, Revised, February 2006, under journal review. [Cowles Founda-tion Discussion 1511R]

[19] Bayer, Patrick, Hanming Fang and Robert McMillan. “Separate When Equal? Racial Inequal-ity and Residential Segregation.” October 2005, under journal review. [NBER Working Paper11507]

[20] Cai, Hongbin, Hanming Fang and Colin Xu. “Eat, Drink, Firms and Government: An Investi-gation of Corruption from Entertainment and Travel Costs of Chinese Firms.” Revised April2006, under journal review. [NBER Working Paper 11592]

22

[21] Cooper, David J. and Hanming Fang. “Understanding Overbidding in Second Price Auctions:An Experimental Study.” Revised, March 2006, under journal review. [Cowles FoundationDiscussion Paper 1557]

[22] Fang, Hanming, Michael Keane and Dan Silverman. “Sources of Advantageous Selection: Evi-dence from the Medigap Insurance Market.” April 2006, under journal review. [NBER WorkingPaper 12289]

[23] Fang, Hanming and Shing-Yi Wang. “Assessing the Impact of Welfare Reform on the Well-Being of Single Mothers and Their Children.”

[24] Fang, Hanming. “Akerlof’s Lemons Market Revisited: What if Sellers Have Multi-dimensionalPrivate Information?”

[25] Fang, Hanming, Michael Keane, Ahmed Khwaja, Martin Salm and Dan Silverman. “Is therea Mickey Mantle Effect? The Effect of Subjective Life Expectancy on Health Behaviors.”

[26] Fang, Hanming and Glenn C. Loury. “Collective Identity and Economic Incentives.”

[27] Fang, Hanming, Michael Keane and Andrew Postlewaite. “Estimating the Returns to Non-cognitive Skill Investments.”

[28] Fang, Hanming. “Understanding the Contributions of Changing College Major Choices on theCollege Wage Premium, Black-White and Gender Wage Gaps.”

[29] Fang, Hanming and Peter Norman. “An Efficiency Rationale for the Public Provision of PrivateGoods.”

[30] Fang, Hanming and Fali Huang. “Endogenous Social Status.”

[31] Fang, Hanming and Rongzhu Ke. “The Insurance Role of ROSCAS in the Presence of theCredit Market: Theory and Evidence.”

[32] Fang, Hanming, Andrea Moro and Giorgio Topa. “Racial Differences in the Risk SharingProperties of Social Networks: Theory and Evidence.”

[33] Fang, Hanming, Patrick Bayer and Robert McMillan. “How Do Middle Class Black Neighbor-hoods Emerge?”

[34] Fang, Hanming. “Time Aggregation in Dynamic Choice Models when Agents Have Time In-consistent Preferences: Monte Carlo Evidence.”

[35] Fang, Hanming. “Optimal Contract Design in Law Enforcement.”

[36] Fang, Hanming. “Identification of Taste and Statistical Discriminations.”

[37] Fang, Hanming. Economics of Discrimination: Theory, Empirical Methods and Evidence.[“The Book Project”]

23

[38] Fang, Hanming, Michael P. Keane and Dan Silverman. “Understanding the Dynamics of Ad-vantageous Selection in Medigap Insurance Market.”

[39] Fang, Hanming, Bin Liu and Li-An Zhou. “Adverse Selection or Moral Hazard? The Case ofCommercial Bank Loans.”

[40] Fang, Hanming and Alessandro Gavazza. “Dynamic Inefficiency in Health Insurance Market.”

[41] Cullen, Mark, Liran Einav, Amy Finkelstein and Hanming Fang. “Risk, Risk Aversion andMoral Hazard: Estimating the Costs and Benefits of Offering Choices in Health Insurance.”

[42] Cai, Hongbin, Yuyu Chen and Hanming Fang. “Social Learning and Herding: Evidence fromA Natural Randomized Field Experiment.”

24

TEACHING STATEMENT

HANMING FANG

August 2006

I taught my first course “Mathematics for Economists” at the University of Pennsyl-

vania for incoming PhD students in the summer of 1998. After coming to Yale University at

2000, the main course I teach is “Public Economics,” both at the second-year PhD and

undergraduate level. At the undergraduate level, I used a standard textbook “Economics of

the Public Sector” by Joseph Stiglitz combined with a course reading packet that consists of

materials from New York Times and other newspaper and magazine so that the students can

see the theoretical concepts in action in the daily life. The PhD level public economics class

is taught from reading journal articles and unpublished working papers. It is organized as a

topics course where I cover the classical public economics topics including optimal taxation,

externalities and public goods provision, as well as topics that are more pertinent to my

own research area such as theoretical models and empirical methods of discrimination, wel-

fare and social security reforms, and social economics and market imperfections. The most

attractive feature of my PhD level public economics class is a nice balance between theory

and empirical research. For every theory topic covered, I always try my best to expose the

students to relevant empirical evidence, either from experimental data or from field data.

I emphasize the in-depth reading of classical papers for each topic and ask the students to

critically present recent researches in each area. My view is that the ability to critically read

research papers is the key step for second-year graduate students to transition to indepen-

dent and creative researchers; and my sense is that the presentation requirement helps them

in that direction.

For three years I also taught “Introduction to Probability and Statistics” for un-

dergraduate students in mathematics and economics majors at Yale. This course is the first

of a one-year sequence of probability, statistics and econometrics designed for students who

are likely to pursue graduate degrees in economics and other disciplines. I attempted with

some success to incorporate real-world applications of relatively dry probability formulas and

statistical theorems in law, politics and medicine. The students from this class are provided

with a very solid foundation in probability and statistics for their subsequent econometrics

class. I have also developed and taught a rather popular undergraduate seminar “Topics

in Behavioral Economics” that attempts to expose students to the recent developments

1

in the interface of psychology and economics. The seminar was targeted toward seniors in

economics majors but it also attracted several students from cognitive psychology majors.

I have been actively involved in advising Ph.D. students in applied microeconomics. I

served as the main advisor for Shamena Anwar, who wrote a dissertation about discrimi-

nation and is currently at Heinz School of Public Policy at Carnegie Mellon University. I

served on the dissertation committees of Elena Krasnokutskaya (2003), Caterina Calsamiglia

(2005), Mainak Sarkar (2005), Panle Jia (2006), and expect to be on the committees of Xia

Li, Lauren Lax, Kaj Thomsson and William Bunting. I also worked with a lot of under-

graduate students, in the capacity of both their senior essay advisor and ROME faculty

advisor.

2