The Brain as a Hierarchical Organization

1. The Brain as a Hierarchical Organization Isabelle BrocasUSC and CEPR Juan D. CarrilloUSC and CEPR March 2006

2. Evidence from Neuroscience:which brain system is activated when Modeling techniques of Micro theory:agency & incentive theory,organizational design, etc. The brain is and should be modeled as a multi-agent organization What is “Neuroeconomic Theory”?

3. Some topics of interests for neuroscientists • Information processing • Resource allocation • Competition and control • Monitoring • Self-management • Discounting • Preferential access to information, etc. • Remarkably similar to the topics of research on the • “theory of the firm”

4. Objective of this research • Understand behaviors difficult to reconcile with traditional • theories (just as recent behavioral economics literature): • Guilt • Self-imposed rules • Mistaken consumption, etc. • Provide “micro-microfoundations” for characteristics • traditionally considered exogenous: • Discounting • Risk-aversion, etc. • Open the black-box of decision-making just as the modern theory of the firm opened the black-box of the firm • Revisit the individual decision-making paradigm • (not decision-theory but game-theory approach)

5. This paper Incorporate in a model of the brain two findings that have received support in neuro-experiments : • Conflict in the brain between • Forward-looking system (cortical system)capable of intertemporal tradeoffs • Myopic system (limbic system)interested only in immediate gratification [McClure et al. (2004)] • Restricted cognitive access within brain to information [Berns et al. (1997), Whalen et al. (1998)] “The heart has its reasons which reason knows nothing of” (Blaise Pascal)

6. An important caveat • Assumptions based on neuroscience evidence: • Conflict between myopic and forward-looking • Asymmetric information • Hierarchical order and control • Modeling choices for which there is no evidence for or against (yet!): • Vertical hierarchy between • Forward-looking = “planner” • Myopic = “doer” • Private information possessed by myopic

7. Related literature • Hyperbolic discounting with incomplete information(Carrillo-Mariotti, Brocas-Carrillo, Benabou-Tirole, Amador-Werning-Angeletos) Main Differences: • Conflict within (rather than between) periods • Asym. info within (rather than between) periods • Other dual-self theories (Thaler-Shefrin, Fudenberg-Levine, Loewenstein-O’Donoghue, Benhabib-Bisin, Bernheim-Rangel) Main Differences: • Asym. Info (rather than full info.) within periods • Constraints (rather than costs) in decision-making • (Two linked activities)

8. The model 2 periods of consumption and labor and Utility “Principal” Pcortical system “Agent 1” A1 limbic system at date 1 “Agent 2” A2 limbic system at date 2 where u’ > 0, u” < 0 and θt is valuation at date t

9. Consumption is non-negative: Labor is non-negative and bounded: 1 unit of labor  1 unit of income  1 unit of consumption Perfect capital markets with interest rate r > 0 Intertemporal budget constraint: [Note: no individual rationality constraint]

10. Atchooses his preferred pair … but P can restrain At’s choices, and we allow any conceivable rule / restriction such as: • “consumption cannot exceed labor” • “labor must be at least x ” • “utility must be the same for both agents”, etc. • P deals with A1and A2sequentially Principal P date 1 date 2 t Agent 1 A1 Agent 2 A2

11. For each θt , P imposes on Ata specific pair : Benchmark: conflict under full information P knows the valuation θt of At , so P solves:

12. Consumption at t increases with θt (valuation at t ) • Labor at 1 is maximum (positive interest rate on savings) • Labor at 2 is adjusted to meet budget constraint • Positive relation consumption at 1+2 and labor at 1+2“work more in your lifetime to consume more in your lifetime”(due to intertemporal budget constraint) • No relation between consumption at 1 and labor at 1

13. Conflict under asymmetric information At knows his valuation θt P only knows that θt i.i.d. F(θt)  P cannot impose restrictions that depend on valuation θt Note: Because constraint (no access to θt ) rather than exogenous cost of imposing choices (e.g. c(θt - θt’) when θt ≠ θt’ ): • No presupposed tradeoff • No preconceived idea of which restriction P willimpose

14. Trivial. No restrictions (except budget balance) because at date 2 no conflict between P and A2 Optimal rule at date 2: P vs. A2 sunk Principal P date 2 t Agent 2 A2 A2 chooses

15. Optimal rule at date 1: P vs. A1 Principal P date 1 t Agent 1 A1

16. P offers the following menu of consumption and labor pairs : • A1 picks the pair he prefers • Note that the pairs are designed such that • Different valuations  different choices • Higher valuation  more consumption and more labor [the result is reminiscent of the mechanism design literature]

17. Some conclusions • Endogenous emergence of (second-best) self-imposed rule:“work more today if you want to consume more today”. • Behavior has “a feeling of guilt” without assuming it. • Current reward (leisure) tracks current earning (one-day-at-a-time, narrow bracketing effect). • No consumption smoothing, with testable implications.Distribution of consumption over life cycle depends on: • Source of income (endowment vs. current labor) • Period-to-period access to labor

18. Asym. info. vs. Full info. “similar to” Positive discounting vs. No discounting Time-preference rates • Consumption shifts to first period • Labor shifts to second period • Increase in consumption greater for high valuations • Decrease in labor greater for low valuations

19. But there are also differences: distribution F(θt) from which valuations are drawn affects consumption. Some conclusions • “Micro-microfoundations” for intertemporal discounting. • Testable (?) differences: given current valuation, consumption is smaller if individual usually likes the good a lot.

20. “Incentive salience” and “visceral factors” Neuroscience: “incentive salience” [Berridge, 2003] • One system mediates motivation to seek pleasure (wanting) • A different system mediates the feeling of pleasure (liking) Stimulus of 1st system  more work for same reward Social Psychology: “visceral factors” [Loewenstein, 1996]Emotions (fear, anger) and drives (hunger)  discrepancy optimal vs. realized (out-of-control) behavior

21. “liking” Principal P date 1 date 2 t Agent 1 A1 Agent 2 A2 “wanting at 1” “wanting at 2” At is tempted to overconsume (biased motivation) P does not integrate At ’s salience

22. Optimal rule at date 1: Pvs. A1very similar to previous case • P offers a menu of consumption and labor pairs • Higherθ1 higher c1 and higher n1

23. P imposes only two constraints: Consumption cap Budget balance A2 chooses: If θ2 < θ* : unconstrained optimal pair given his bias If θ2 > θ* : same pair as an agent with valuation θ* Optimal rule at date 2: P vs.A2 sunk Principal P date 2 t Agent 2 A2

24. Some conclusions • Optimality requires a simple, non-intrusive rule-of-thumb:“do what you want as long as you don’t abuse” • Stronger bias  tighter control Note 1: not aware of similar result in any mechanism design problem in which P can use two tools Note 2: nice reinterpretation as parent / offspring relation

25. Intuition. Three ways to satisfy “incentive compatibility”: • Freedom: choose whatever pair you prefer • Pooling: everyone is offered the same pair • Self-selection: specific monotonic relation between c and n Standard mechanism design: 3 (or 3 and 2 if only 3 not feasible) Our problem: 1 is costly only for strong conflict 2 is inefficient 3 implies wasted resources (excessive work, bad also for P) Optimality  1 for low conflict and 2 for high conflict

26. What’s next? • Test of behavioral implications • Period-to-period labor opportunities affect consumption • How much the good is usually liked affects consumption • More realistic and comprehensive models of the brain. For this, we need many more neuroeconomic experiments to guide theoretical models (and viceversa): • Is the hierarchy of the myopic vs. forward looking “vertical”? • Is the superior information possessed by the myopic system? • Are systems with restricted access to knowledge aware of their informational deficit? • Does the forward looking system discount the future? • When are salient incentives more likely to operate?