Présentation au sujet: "Claude Montmarquette Les journées de l’économie Lyon, 20 novembre 2008"— Transcription de la présentation:
1Claude Montmarquette Les journées de l’économie Lyon, 20 novembre 2008 Parier sur l’économie expérimentale pour résoudre les problèmes actuelsClaude MontmarquetteLes journées de l’économieLyon, 20 novembre 2008
2Qu’est-ce que l’économie expérimentale ? Méthodologie crédible de recherche qui permet de recréer et d’étudier dans un environnement contrôlé en laboratoire :· L’importance de chaque motivation particulière (recherche du gain, besoin de réciprocité, réaction aux changements institutionnels,…) dans la prise de décision des agents.· Sous conditions de risque, d’incertitude ou d’équivalence certaine, permet de tester les hypothèses exactes postulées dans les modèles et d’isoler l’influence de certaines variables. On peut analyser et comprendre l’éventuelle différence qui existe entre les prédictions théoriques à l’équilibre et les résultats tant expérimentaux qu’observés dans la vie quotidienne.
3 Qu’est-ce que l’économie expérimentale ? (suite) · Rend possible la comparaison entre les environnements, les institutions et les politiques incitatives afin d'en évaluer l’efficacité relative. Cette approche est une plate-forme flexible permettant d’évaluer de nouvelles politiques et de nouveaux « designs » institutionnels sans avoir à subir les coûts sociaux et privés associés à leur mise en place.· Permet de tester les implications de certaines politiques sociales ou de décisions de gestion sans avoir à réaliser des projets coûteux qui sont plus souvent qu’autrement mis en place avec des paramètres considérés ex post comme ayant été mal choisis ou spécifiés.· L’économie expérimentale aide à la collecte de données empiriques pertinentes et fiables.
4Des distinctions….Expériences sur le terrain (field experiments): participation de différentes populations et permet de refléter les choix des individus dans leur milieu et contraintes naturellesExpériences naturelles: formidables si possibles; situation peu fréquente et permet peu de traitementsTrend actuel est de combiner le labo et le terrain
5Est-ce que les résultats obtenus sont transférables dans la réalité ? Plusieurs réponses :1. En économie expérimentale, les participants sont payés selon leurs décisions, comme dans la vraie vie. Si c’est le cas, pourquoi existerait-il des différences ?2. Plusieurs études allant de la réalité vers le laboratoire ou du laboratoire vers la réalité ont prouvé le caractère transférable des résultats.
6Aide à la solutions de problèmes actuels Notons d’entrer de jeu qu’il est impensable de recommander des politiques ou des solutions relativement aux problèmes étudiés sans comprendre les comportements des individus et leurs préférences. l’EE a consacré et continue à le faire beaucoup d’efforts à l’étude des comportements individuels, notamment relativement à leur attitude vis-à-vis le risque et vis-à-vis leur impatience à consommer.
7De quels problèmes peut-il s’agir? En principe, la limite des problèmes examinés est lié à l’imagination du chercheur à développer un protocole pertinent. Le défi à cet égard est de réussir à simplifier une situation complexe tout en maintenant la pertinence de l’analyse. L’expertise des analystes et les moyens technologiques disponibles repoussent continuellement les frontières. Historiquement, l’analyse expérimentale est passer de la validation de la théorie des jeux à des applications de politiques liées à la firme, au marché et à l’état.
8Exemples de problèmesRessources Naturelles et politique environnementale:Mise aux enchères des droits d’émissionMarchés concurrentiels d’énergie électriquePolitique industrielle et réglementaire:Affection des ressources en espaceDivulgation d’informationRègles fiscales et procédures de vérification
9Exemples de problèmes Investissement en éducation et en santé Politiques de financement de l’étatFraudes fiscalesMarché du travail et participationPolitiques industrielles
10A study sponsored by Human Resources Development Canada Will the Working Poor Invest in Human Capital? A Laboratory Experiment by Eckel, Johnson and Montmarquette SRDC Working Paper 02-01, February 2002A study sponsored byHuman Resources Development Canada
11Key Research QuestionGiven the right incentive, will the working poor save to invest in human capital?
12Objectives of the experiment Laboratory experiment can be used as a complementary approach to generate valuable information for the design of social experimentsSRDC wanted to shed light on the behaviour and preferences of the working poor with respect to saving for learning activities before launching the learn$ave demonstration project
13Three research questions Will the working poor invest in various assets?Are these subjects willing to delay consumption for substantial returns?How do these subjects view risky choices?
14Experimental Instruments Two instruments:Information questions (43)SocioeconomicBehaviouralAttitudinalCompensated questions (64)
15Compensated Questions - 64 Investment PreferencesCash v. Investment choicesTime PreferencesCash v. Cash laterRisk PreferencesCash v. Risky cash
16Sample Compensation Question From the Experiment You must choose A or B: Choice A: $100 one week from today Choice B: $400 in your own training or education
21What Have We Learned ?In general, the working poor are risk averse and impatientNevertheless, many can be induced to invest in their own education44 percent accepted offer analogous to learn$ave (3 to 1 match)Overall, own educational expenses was preferred to family member’s education and retirement savingsnot true for non-labour force participantsSome (16%) couldn’t be induced to invest in any asset even when return approached 500%
22What Have We Learned ?The more patient people are, the more likely they are to invest in their own educationThe more risk-averse subjects are, the less likely they are to invest in their own education.Savings programs may benefit from higher take-up rates if theyOffer high returnsStress absolute returnsAllow short term savings horizons
23Fostering Adult Education: A Laboratory Experiment on the efficient use of loans, grants and savings incentives by Jonshon, Montmarquette and Eckel SRDC Working Paper 03-09, December 2003A study sponsored byCanada Student Loans Directorate and Applied Research BranchHuman Resources Development Canada
24Object of the experiment To address a particular set of specific policy issues:How do various types of learning subsidies (grants and loans) affect the participation rates in adult education?Would the availability of incentives for part-time studies discourage full-time studies?What is the extent of windfall gain resulting from different levels and types of financial incentives?What are the “barriers” to participation in adult education?Lack of informationLack of timeLoan aversionFear of FailurePreference for the presentLack of readiness to learnAnd to examine actual choices by individualsThis yields an unusually rich analysis of the decision to invest in Human Capital:Controlling for demographic characteristics such as age, sex, family structure, income anWe can examine the role of risk attitudes and time preference and other attitude on decisions
25The Experiment Focus of the full study is on four sets of measures: 1. Experimental preference measuresconsumption over timerisky choice alternatives2. Survey measures: demographics and attitudes3. Numeracy Assessment4. Willingness to invest in post-secondary educationGrantsLoans (regular and income-sensitive repayment – ISR)Matched-savings grantsWord on protocol:No computers, or complicated devices. We used dice and bingo balls for random draws.The controlled environment were classrooms at 4 YMCA’s in Montreal
26Survey measures Demographics Labor market and educational status Age, gender, incomeLabor market and educational statusAttitudinal measuresPlanning, debtBarriers to educationSkills, dispositional, situational
27Example of risk aversion decision Choice A$ for sureChoice B80% chance for $175 and20% chance for $0Risk averse subject selects A.E(A)<E(B)120 < 140
29Example of Time Preference Decision Choice A$65 todayChoice B$130 one year from today
30Cash vs. Investment Choice Cash alternative made the choice of investment costly to the subjectResults used to calculate elasticities of demand for education with different types of subsidyThrough their choices, subject reveal their preferences for education when financed by:GrantsLoansISR loansMatched savings
32Take up Rates for $1,000 in Educational Financing
33Proportion of urban participants that chose education financing over $100 cash
34Determinants of choosing $1000 Grant Over Cash (Ordered Probit, 801 observations) Labour Force attachmentImmigrants, disabledWillingness to save (decision)Positive attitude with respect to Education and Labor MarketMathematical CompetencyPSE experienceAgeEmployee with education supplementmarriedChildren (older)HS equivalency
35Labor Market Information Session How does information influenceKnowledge?Attitudes?Investment?
36Labour Market Information Treatment InitialexperimentMoreresearch?No further actionNoYesGood general understanding of labour market or received educational compensationNo further actionScreenRelatively poor understandingof labour marketRandom assignmentTreatment: LMI sessionComparison: No actionFollow-up experiment
37What we hope to learn Overall, Is there evidence of Debt Aversion? Are certain types of students prone to Debt Aversion?
38Determinants of choosing more education after the LMI session Probability of choosing more education for the young participants goes up by 15 percentage points or by 33%!From 42% to 57%
39What have we learned?Experimentally measured individual characteristics, such as time preference and risk preferences, can explain variability in the decision making process as much as demographic and social characteristics.Overall, participants were sensitive to different levels of incentives and different forms of financingLMI interventions can make a difference
40The Canada Millennium Scholarship Foundation 2008 Willingness to Borrow: Using lab experiments to examine debt aversion among Canadian high school studentsThe Canada Millennium Scholarship Foundation2008
41Research QuestionsDoes the willingness to borrow vary significantly among types of students?It is believed that students or potential students belonging to low SES families, Aboriginal families or first generation students’ families are less likely to be willing to borrow (doubt benefits of PSE, low likelihood of success).How big a problem is debt aversion among these populations?Are there other socio-economic groups that are more likely to be less willing to borrow?Individual-choice experiments conducted with target populations can provide especially meaningful results for policy design. To give an example from our own work, in December 2000, SRDC and CIRANO conducted a laboratory experiment with 256 participants to examine the behaviour and preferences of the working poor with respect to saving for learning activities. Analysis of the experimental results showed that the relative generosity of grants, and individual characteristics such as time preferences and attitude toward risk, play a significant role in the decision to invest in one’s own education. This particular laboratory experiment was used to determine point estimates of take- up rates for different levels of matching funds offered to potential savers (Eckel, Johnson, and Montmarquette, 2005). The results where then used to calibrate the offer to individuals who participated to a large-scale demonstration project to test the appeal and benefits of Individuals Development Accounts.In a follow-up study, the same set of researchers conducted a larger experiment for the Canada Student Loans (CSL) Directorate of Human Resources Development Canada that was designed to investigate the impact of different types of subsidy for postsecondary education (Johnson, Montmarquette, and Eckel, 2003). The primary objective of the CSL study was to investigate what types of government assistance best serve the policy objective of increasing human capital investment among adults from different socioeconomic backgrounds. The focus of the study was on “barriers” to education, factors that may prevent adults from investing, including access to credit, opportunity costs, time constraints, lack of information, fear of failure, debt aversion, and readiness to learn. A secondary purpose was to examine whether experimental measures of preferences may prove to be superior to survey measures for use in policy calibration.The CSL study used field experiments on a Canada-wide stratified sample of 900 adults aged 18–55, representing both rural and urban areas. The design included a series of choices among amounts of money and grants, matching grants, or loans with various parameters, earmarked for education; experimental measures of risk attitudes and time preferences; survey information on demographics; as well as several psychological measures of preferences and attitudes.The studies proposed herein build on the approach used in these two recent projects. Instead of using survey information alone, we combine experimental, incentive-based decisions with survey measures of attitudes toward debt to determine the roles of demographics, preferences, and attitudes in the decision to take up postsecondary education financing. This study will be one among very few that use experimental methods in the field to provide input into the structure and calibration of a specific government policy.
42Proposed Sample 1400 12th graders and CEGEP students Manitoba, Ontario and Quebec and SaskatchewanAboriginalsRural/UrbanLow and High SES
44Protocol Info packets delivered to selected schools Parental Consent Parental SurveyStudents (pre-session) web surveyIn-school Session ($20)Practice DecisionsExperimental DecisionsNumeracy AssessmentPayoff
45Student Survey Educational ambitions Expectations with regards to ambitionsPerceived obstacles to pursuing PSEFinancial means at student’s disposalDebt aversionExperience with debtEducational background and experiencesParent’s education and economic statusInter-temporal orientation (planning ability)Attitudes towards riskAspiration levelEngagement while in high schoolPerceptions of labour market conditionsPerceptions of the cost of, and returns to, PSE
46Parental Survey Expectation and aspirations for children Education IncomeFamily size
47Numeracy AssessmentMeasures how participants use math in every day lifeMost compact way to control for differences in ability among students or schoolsMarked inter-student variance that will interact with how they respond to experimental decisionsThere is also a more important link - numeracy skill is the single most important determinant of both high school completion and PSE participation rates
48Experimental Measures Time PreferencesRisk PreferencesEducation Choices
49Time Preferences Binary Decisions organized in increasing reward 6 rates4 Front End Delays2 investment or Wait times48 Decisions
50Time Preferences Earlier Payoff Rate of Return Later Payment $75 0.05 One MonthOne Yeartomorrow0.0575.3178.75One Week0.175.6382.500.276.25903 Months0.578.13112.50181.25150287.50225
51Risk Preferences All Graphical Representations Two Basic Measures Holt/Laury10 binary decisionsEckel Grossman1 decsion chosen from SIX 50/50 gambles(Binary Version of Eckel Grossman)
53cash v. Education financing Education ChoicesBasic Design:cash v. Education financingUse these decisions to distinguish pricing from form of financingControl forSize of cash alternativePrice of subsidy per $1 edu financingAbsolute value of edu subsidy
54Types of Edu Financing Grants: $500 - $4000 Loans: $1000 - $4000 Income Contingent LoansHybrids (loans + Grants) $800 - $4000Cash Alternatives: $25 - $700
55Aspiration levels and Educational Choices: an Experimental Study Lionel PageLouis Lévy-GarbouaClaude Montmarquette
56A sociological explanation for differences in educational choices Sociologists (Boudon 1973) also invoke differences in aspiration levels among social classes: children from upper classes have higher aspirations than children from lower classes with identical abilitiesAspiration levels are reference-dependent and the natural reference for children is their parents’ levelReaching a given level of education may be perceived as a failure in upper classes and a success in lower classes
58Prospect theoryReference points play a central role in prospect theory (Kahneman and Tversky 1979)The same outcome is framed or perceived as a GAIN if the reference is low, and as a LOSS if the reference is highPeople are risk averse in the domain of gains and tend to be risk seeking in the domain of lossesMoreover, people are averse to lossesPage (2005a, 2005b) has, shown that the impact of aspiration levels on educational outcomes can be modeled with the notion of reference point from prospect theory.
59Why an experiment?On real-life data, it is difficult to control for many factors (e.g., abilities) and for the context of decision; and it is often impossible to observe causal variablesIn our experiment, we observe and manipulate the reference point; and we are able to measure task-specific abilities so as to control for this important factor econometricallyWe simulate experimentally the simplest schooling system in a context-free setting and compare the “human investments” of our experimental subjects in a GAIN treatment and in a LOSS treatment
60The experiment is made of two treatments. In one treatment, the outcomes are displayed as gains, framing a low reference point.In the other treatment, the outcomes are presented as losses, framing a high reference point.According to prospect theory, the framing of the monetary outcomes as losses should have two effects:(i) The participants should be more likely to choose to continue at stages 9 and 12.(ii) The participants should exert more effort to perform the task.
61Experimental Design15 stages grouped in 3 levels. Each stage involves solving a given number of anagrams. The first level contains the stages 1 to 9, the second level the stages 10 to 12 and the third level the stages 13 to 15.At the end of each level, a participant must have solved two thirds of the anagrams to be allowed to pass to the next level.The difficulty of the level increases according to the following criteria:• The number of anagrams per stage increases with the level with a constant time limit of 8 minutes per stage. Specifically:– 6 anagrams per stage for level 1,– 9 anagrams per stage for level 2 and– 12 anagrams per stage for level 3• The length of anagrams increases on average. The structure of the experiment is represented in Figure 1. At the end of each level, the participant fails or passes, and correspondingly there are two possible outcomes in terms of monetary payments.
68Econometrics Analyses Table 3Choices: Probit regressionsChoice stage 9Choice stage 12Choice both stages(1)(2)(3)(4)(5)(6)LF0.439 (1.42)0.693 (1.77)*0.649 (1.65)0.701 (1.53)0.436 (1.88)0.544 (2.02)**Male0.785 (2.01)**0.817 (1.73)*0.627 (2.32)**Not French native(2.37)**(0.35)(-1.53)Play scrabble(1.45)0.114 (0.75)*(-0.95)***Ability a(2.99)***(1.88)*(-4.11)***Risk aversion b(2.56)**1.126 (2.31)**(-3.63)***Dummy level 12(0.085)***Constant0.896 (4.53)3.092 (3.66)***(0.23)1.334 (1.17)0.601 (3.83)2.450 (4.32)***Observations1091084443153151Absolute value of z statistics in parenthesis. Significant : *10%, **5%, ***1%a Ability is measured with the mean time individual required to solve one anagram at the previous level.b Dummy equal to 1 if the participant chooe an uncertain lottery in a hypothetical choice
69Aspirations and Performances Proposition 1: Framing (LF) matters to continue educationProposition 2: In LF participants should exert more effort
74DiscussionAspiration levels may play a major role in educational choices causing social inequalities in educational outcomesGender differential effect in LF not expected. If Emma if from a poor family, she would consider her outcome as positive if stopping at any intermediate level of education. If Ben is from a high social background, stopping at any intermediate level would be consider a failure
76On TableMales from LF represent 55% of participants reaching the highest level vs 25% from chance aloneMales represent 78% of the highest achievers while they represent 55% of participantsCould the concentration of males in higher levels of education be due to the highest rate of success of males with high aspiration levels?
77ConclusionWe find that to frame outcomes as gains or losses in our experiment significantly changes the choices of the participants. Participants in the loss framing treatment chose more often to continue further in the stages of the experiment than participants in the gain framing treatment.Concerning the effect of aspiration levels, the prediction stemming from prospect theory are only validated for males.The framing of outcomes as losses, which was expected to increase the motivation of the participants, does so, but only for males.
78Individual Responsibility in the Funding of Collective Goods Louis Levy-Garboua (TEAM, University of Paris I)Claude Montmarquette (CIRANO, University of Montreal)Marie-Claire Villeval (CNRS)
791. Motivation A specific example: Public health insurance How to increase individual responsibility in voluntary contributions to funding collective goods?Aim 1: Comparing the efficiency of taxation and rationing systems with respect to the private supply of public goods and the funding of deficitsAim 2: Analyzing the effectiveness of individualizing the deficit handling by taxation or by rationingA specific example: Public health insurance
80A laboratory experiment A 2-stage experiment with a 2x2 designVoluntary contributions to a common pool set by members of a group serve to compensate for the losses incurred by hit membersIn case of a shortage of the common pool, 4 possible deficitmanagement modes: taxation / rationinguniform/ individualized
812. Theory A two-stage collective goods game Stage 1: Voluntary contribution to a common pool intended to compensate for the losses suffered by group members randomly afflicted in stage 2Stage 2: Random selection of the victims and determination of the payoffs. Treatment of the possible deficit.N =12; Number of victims: S =4 ; Probability of a lossIndividual endowment: Y = 100Individual contribution:: loss suffered by k, i.i.d.: total losses in the group
82Uniform taxation Individual tax = 1/N (deficit) Taxation involves a deadweight lossgi = 0 is a Nash equilibrium if
83Individualized taxation The tax is individualized according to giTaxation involves a deadweight lossNash equilibrium: gi = L/N. Unique if all players are assumed similar.Nash equilibrium = Optimum
84Uniform rationingIn case of a deficit, compensation is partial => payoff becomes uncertain. All the victims receive the same compensationgi = 0 is a Nash equilibrium
85Individualized rationing A victim’s compensation in stage 2 depends on his individual contribution in stage 12 conditions:(i) A victim cannot be compensated for more than his loss(ii) The total amount of compensations is always covered by the total amount of contributionswhere ci (0<ci<1) is the rate of compensationand with
86u.c.The Nash equilibrium is positive but below the optimum
87To sum up Optimum Equilibrium Uniform Taxation L/N 0 (provided not too large)Individualized Taxation L/N L/N (if )Uniform Rationing L/NIndividualized Rationing L/N gi>0
883. Experimental design Regate software 24 sessions (12 in BUL-C3E at CIRANO, Montreal, and 12 at GATE, Lyon)288 participants from undergraduate classes in engineering and business schools50 repetitions90 minutesA test of risk aversion at the end of the session (Can.$ 5 or €2 for sure or 50% chance of winning $11 or €5 and 50% chance of 0)Average earnings: 35 Can.$ (23 €)
90ConclusionWith respect to the relative efficiency of the diverse deficit coverageinstitutions, the experimental results are compliant to the theoretical modelUniform rationing is the worst system.Uniform taxation, while encouraging free-riding just as much, is not muchmore efficient since it imposes upon the community an extra tax burden.Individualized taxation is the best deficit coverage model since- it gives individuals a sense of responsibility- it eliminates the sucker aversionIf taxation encourages cooperation (Andreoni, 1993), this is true forindividualized taxation but not for uniform taxation
91Cathleen Johnson, David Masclet, Claude Montmarquette The effects of perfect monitoring of matched income on tax compliance: An experimental investigationCathleen Johnson,David Masclet,Claude Montmarquette
92Issues Tax evasion is still an open question There is more voluntary compliance than game theoretic models predictThere are more successful audits than principle agent models predictEmpirical evidence offers contradictory evidence on the effects of audit rates
93MotivationTypically, taxes are held for some time by businesses and paid to the government on a periodic basisIt is now possible for taxing authorities to receive sales taxes directly through financial institutions when payments are electronic
94MotivationThe IRS (1996) reports that income underreporting is the largest simple source of tax evasion. 72% in 1988Would the implementation of an automated collection scheme increase tax revenue?
95NoteMust consider that individuals may react differently to an substantial increase in audit rates:Those who are relatively more risk averse will comply to maximize expected income.Less risk averse will underreport even more to maintain current level of income
96The Basic ExperimentSubjects are instructed to play an unspecified number of periodsIn each period SsReceive income (10-110)Report incomePay taxes on reported incomeExperience an audit with some probabilityHave complete history (private info)
97Income Two sources of income each period Total = A + B 3 types of income distributionPlayer type and amount of income is private informationSource ASource B80%20%50%
98Auditing Participants pay 40% tax on reported income 20% probability of Audit on income for bottom half on income distribution10% probability of Audit on income in top half of income distributionPenalty: unpaid tax + 50% and automatic audit on previous two periods.
99Before examining a change in monitoring… “basic” income and reporting0: A + B (48)
100A change in monitoring (I) “basic” income and reportingAnnouncementThe implementation of perfect monitoring of Source A incomeI: A + B (21)“A” will be perfectly revealed (6)As promised (21)
101A change in monitoring II “basic” income and reportingAnnouncementThe implementation of perfect monitoring of Source A incomeII: A + B (21)“A” will be perfectly revealedYou can trade 6 A for 5 B (6)As promised (21)
102A change in monitoring 12 sessions of 12 Ss each All sessions implemented the change in monitoring (two treatments)6 sessions allowed for Ss to transfer income from source A to source B (II)
103Descriptive resultsBefore announcement (basic phase), observed that audit rates did affect compliance.Higher income, lower compliance rateOverall compliance ≈ 70%
104Figure 1 : The reporting rates through time and segments
105Observations Tax revenues increased for 80% monitored Tax revenues decreased for every other group % total decreaseAnnouncement period:Tax revenues decrease when individuals don’t see have an opportunity to transfer incomeRemain the same when opportunity to shift to Souce B income (treatment II)
106Final thoughts Do we think this is what will happen in real life? Other changes must happen in conjunction with this monitoring system or it may not workTransition individuals to bank accountsReduce other costs of electronic paymentsTax decreasePublic goods aspectAbout the difficulties of reducing fiscal fraud
107Conclusion Générale L’EE aide a la compréhension des problèmes Elle souligne des pistes de solutionsElle permet d’influer sur les décideurs. Ces derniers ne sont jamais faciles à convaincre sur des bases théoriques, mais ils sont plus sensibles aux faits empiriques.Pariez sur l’EE pour faire avancer les idées est un bon choix