Economics and Psychology of Social Norms and Strategic
Behavior
Lecture 1: Introduction
Behavioral economics (economics and psychology) improve economic analysis by incorporating
more realistic assumptions about human behavior. It is inspired by mooting evidence for descriptive
failures of the standard economic model.
Standard economic assumptions:
- People are rational
- People behave as planned
- People care only about themselves
Behavioral economic assumes:
- Bounded rationality
- Bounded will-power
- Bounded self-interest
There are 2 types of systems:
- System 1: based on intuition
- System 2: based on reasoning
Behavioral biases:
- Judgement and choice are often based on intuition, which come spontaneous a key
principle is also accessibility.
Scope of behavioral economics
- A broader perspective on human decision making
- A close interaction between theory and evidence
Objections
- Assumptions are only “as if”; theory should be judged by its predictions
- Errors cancel out in the aggregate
- Rational choice is a simple and general theory; the alternative (psychology) is chaos.
o But we must make our models as simple as possible, but not simpler (Einstein) and it
is better to be vaguely right, than to be precisely wrong (Keynes)
Features
- Rational choice theory is complemented with psychological elements and deviates from
rational choice in various ways.
- It is not a unified general theory yet
- Loss of parsimony and unity is more than compensated by increased relevance.
- It is not psychology it is a theory, asks questions and focusses on the outcome (not on the
process like in psychology).
Important things from chapter 1 of the book
Experiments
- We need participants in our experiments.
- Experimental session = refers to a particular instance in time and space when an experiment
was run.
- Subjects often need to do the same task many times. This is done to allow for learning.
, - Round = how far we are in the experiment.
- Experimental treatment = to distinguish between different versions of the experiment. There
can be 5 treatments for example.
- One treatment is called the baseline treatment which serves as a reference point for
comparison.
- Deception (lying to participants) is not allowed in economics.
- A basic requirement of most experiments is that they be single blind, meaning subjects are not
biased by knowing too much.
Heuristics and biases
- Simple heuristics make us smart. For example, a reasonable heuristic in repetitive situations is
‘Do what I did last time’. Anna, for instance, might always go grocery shopping in the same
store. Sensible though this may be, it can lead to a status quo bias in which she fails to change
her behavior ‘often enough’. Maybe she keeps going to the same store even though a cheaper
and better store opened nearer to her house. Such biases usually go by the name cognitive
biases.
- In exploring the connection between simple heuristics and cognitive bias it is useful to
introduce the notion of ecological rationality. This recognizes that a heuristic can be smart or
dumb depending on the environment in which it is used.
Cognitive bias is most likely to result when a heuristic is used out of context or in an environment
for which it was not designed. Experimental economics tries to unravel these biases.
Lecture 2: Interacting with others I
Game theory is used as a theoretical framework to analyse strategic interactions. It has a central role
in all domains of economic theory.
Nash equilibrium
- No player can attain a higher payoff by unilaterally choosing a different strategy (thus, by
keeping the other strategy fixed).
- Another explanation is that each player plays a best response to her or his beliefs about the
strategies of other players, and in equilibrium these beliefs are correct.
- Nash equilibria are universally applied to solve games and sometimes they are supplemented
with some refinements (Bayesian NE, subgame perfection).
- Do players play according to the NE? What do people do in case there are multiple equilibria?
Unfortunately, there is only little evidence from field data about these questions.
Beauty contest game
- Everyone submits a number between 0 and 100
- All numbers are collected
- Then the average of the numbers is taken (X)
- Then this is multiplied with 2/3 (2/3X)
- The one whose number is closest to this, wins 10 euros.
- The NE would be to submit 0 for all players.
- Results:
o We see some bumping around 0, but also a lot of players play strategies lower than 50
but higher than 0, and some players seem to play randomly (80, or 65, etc.).
o Results are similar across different types of experiments (different subjects, etc.)
o Results show spikes at 33, 22 and 0.
, o Even Game theorists do not play according to the NE
o The play moves more into the direction of the NE with some repetition.
o Readers who run a little experiment usually did best.
- What are the reasons that the NE poorly predicts this situation?
o The data are flawed
Results are a fluke
Subjects are not representative
Stakes are too low
Subjects are inexperienced
Etc.
o The theory is flawed
There are 3 key assumptions in the NE: payoffs represent utility, players are
rational, and rationality is common knowledge (I believe that others are
rational, others believe that other players are rational, etc.)
Behavioral game theory adjusts these assumptions:
Payoffs are not equal to utility; social preferences might play a role
Players can make mistakes; we will capture this in the Quantal
Response Equilibrium theory
Players do not assume that other players are always rational; we will
discuss this in the level-k model and the cognitive hierarchy model.
Level-k model
- A player forms a belief about the behavior of others, and then best-responds to that belief.
o Level-1 thinking: other players are thought to be level-0 (play randomly or non-
strategically) and you best-respond to that.
o Level-2 thinking: other players are thought to be level-1 (and best-respond to level-0),
and you best-respond to that.
o Level-k+1 thinking other players are thought to be level-k (and best-respond to level-
k-1) and you best-respond to that.
- When we apply this to the beauty contest game:
o Level-0 play randomly, so they choose 50 on average.
o Level-1 players best-respond by setting 2/3*50 = 33.3
o Level-2 players best-respond by setting 2/3*33.3 = 22.2
o Level-3 players best-respond by setting 2/3*22.2 = 14.8
o Level-infinity players best-respond by setting 0.
Level-k properties
- Accurately explains the peaks in the empirical distributions of the Beauty Contest.
- A questionable aspect of the level-k model is that higher levels do not necessarily make more
accurate predictions.
o This is because Level-k thinks all others are level-k-1 and nobody is a lower level.
o It could be the case that there is a distribution of level-0, level-1, level-2, etc.
- Moreover, how can we model this distribution of the different levels?
Cognitive Hierarchy model
- This model improves upon the previous level-k model by making 2 adjustments:
o Level-k now believes that other players are a mixture of lower levels.
o The distribution of levels is assumed to follow a Poisson distribution:
- Applying this to the beauty contest game:
, o
Cognitive hierarchy conclusions
- Advantages:
o Iterative strategic thinking is more in line with what people can do (and say they do)
o Higher levels typically do better than lower levels (a hierarchy of cognition)
o Easily applicable in many settings (in contrast to QRE)
o Describes behavior better than NE in many games (or at least as good as NE)
o Can inform us about equilibrium selection
What equilibrium is more likely to prevail?
Entry games
- Market capacity = c
- N potential entrants
- If they enter, profit is 1 if n(entrants) <= c, but -1 if n(entrants) > c
