Tag: nash equilibrium

What traffic can reveal about society

The last post was about Socially Optimal Solutions and the Nash equilibrium.

In this post, we examine traffic patterns under the same lens.

It appears bad traffic situations might be avoided if road users a) play by the rules or, b) play fair or, c) are trusted by other participants.

Scenario 1


In Scenario 1, the North-South traffic is waiting for the light to turn green.  Cars 1 and 2 are going South.  Cars 3 and 4 are going North.

Now, if the drivers of Cars 2 and 4 try to cheat by overtaking on the wrong side of the road, you could end up with Arrangement 2 (everyone is blocked).

This is what you would get at the Nash equilibrium assuming that there were no policemen at the intersection (the Nash equilibrium is one of many possible equilibriums that can be reached if people don’t cooperate, but just act in their own self-interest).

So, it is only if the drivers all respect the law,  without any trying to cheat, that they would end up with the socially optimal solution (Arrangement 1 rather than Arrangement 2).

Scenario 2


Here, the intersection does not have traffic lights.  The North-South traffic has stopped.  Cars are accumulating in the North-South road, and so Cars 1, 2, 3 and 4 are not going anywhere.  Now, 3 can choose to leave a gap for 5 to pass through.

The game here is between Cars 3 and 4.  If Car 3 feared that Car 4 might try to overtake by taking advantage of the gap, it would move forward and block the intersection.

So, in this case, an expectation of unfairness could lead to a solution that is not socially optimal.

And that would be the Nash equilibrium.

However, if people considered each other to be fair, there could be a socially optimal solution that is better than the Nash equilibrium.

Scenario 3


No traffic lights.  There are two lanes going in the same direction (South to North).  Traffic is stalled in the North-South direction (Cars 3, 4, 8 and 9 are going nowhere).  Cars 3 and 8 have the option of leaving a channel open for Car 5 to pass through.

What’s interesting here is that if Car 3 suspects that Car 8 might close the intersection, Car 3 would do well to close the intersection itself (preventing Arrangement 2).

Again, that would be the Nash equilibrium.  However, if everyone agreed and cooperated to keep the intersection open, you would get a much better result (a social equilibrium) than you would with the Nash equilibrium.

So, the socially optimal solution is obtained only if there is strong mutual trust between the participants in this traffic pattern game.


I find these traffic patterns interesting because they could be indicators of local ethics. When you visit a new place, you might be able to get an idea of how fair, trustworthy and rule-abiding the local people are, just by observing the traffic?

What game theory says about why gas stations are built next to each other

This lovely video tells you about the concept of a Socially Optimal Solution – a solution that works very well for society – and why such solutions are often not stable.

It also explains the concept of a Nash Equilibrium – a solution that is stable, but not always optimal for society.

It also tells you how aggressive competition can lead to results that are not optimal for society.

Quantum of Punishment – Or Should Draconian Punishment Ever be Used?

One of my friends – Nishanth Ulhas Nair – wrote up a set of suggestions on how one might go about solving the problem of violence against women in India, specifically rape (there have been some horrific occurrences lately in Delhi, Bombay and Bangalore).

Here is one of his suggestions that I particularly liked:

a) In schools, make students do the cleaning and if possible cooking/washing-dishes also. Boys should especially be made to do this. So that they learn to do work which are typically considered to be the job of a woman in India. Moreover this will reduce the class divide in India where there are some ‘inferior/poor’ people who will work for us. Why should there be a mess in schools/colleges? Students should cook their own food and clean their own plates.

But what I really want to talk about today is the counter-suggestion he had made:

The problem with death penalty is that the rapist may end up killing the victim because he feels it will be easier for him to get away with it (since death is anyway the worst penalty he can get, there is more incentive for him to kill her after the rape).

Stricter measures (for example, applying the death penalty to the crime) have been floated around a lot lately, but how do you decide what is an appropriate quantum of punishment?

I am going to see if it is possible to arrive at Nishanth’s conclusion using game theory.

We are going to start with a set of simplified assumptions:

a)  You have two players in this game – the law-enforcer and the law-breaker.

b)  The law-enforcer has two choices – a) to punish the law-breaker with death and b) to punish the law-breaker with a few years in jail.

c)  The law-breaker has two choices – a) to kill witnesses to better his/her chance of not getting caught and b) to not do anything beyond the primary objective of the crime.

d)  The aim of the law might be assumed to be to maximize the number of potential crimes prevented.

e)  The aim of the law-breaker might be assumed to be to minimize the punishment if any that might result from the crime.

The Payoff Function of the Law-Breaker

The law-breaker’s payoff is the punishment, and so is a negative number.

It is set to -6 (maximum punishment) if the law-breaker does not kill the victim after the crime (which could be a burglary or a rape) but the punishment for the crime is the death penalty (to act as a deterrent).

It is only -2 if the law-breaker does not kill the victim, and the punishment is not draconian.

The law-breaker’s payoff is only -3 if (s)he kills the victim assuming the chances of detection decrease by 50% as a result of the murder.

Since the crime is compounded by the killing, the payoff is the same no matter whether the penalty for the original crime is harsh or not.

The Payoff Function of the Law-Enforcer

The law-enforcer’s model is simpler.  It is equal to the number of innocent lives saved.  It is 1 if the victim is not killed.  It is 0 if the victim is killed after the crime.


Now to start analyzing the game, you write the assumptions down in a bi-matrix (normal-form game) as follows:

the law (enforcer)
awards a few years in jail
the law (enforcer)
awards the death penalty
the law breaker
chooses not to kill the victim
-2, 1
the law breaker
chooses to kill the victim
-3zero -3, zero
Normal form or payoff matrix of a 2-player, 2-strategy game

The first of the two numbers in the matrix represents the payoff to the law-breaker.

The second of the two numbers in the matrix represents the payoff to the law-enforcer (or society).

Strictly Dominated Strategy

By examining the matrix, it is possible to see that there is no strictly dominated strategy for the law breaker.

A strictly dominated strategy is one that will benefit one of the players more than all his/her other strategies no matter what the other player does.  No strictly dominated strategy exists for the law-breaker in this particular game.

That is because if a draconian punishment strategy is used by the law-enforcer, killing the victim appears to be a better strategy for the law-breaker.  If the death penalty is not used for a lesser crime, then not killing the victim appears to be a better strategy for the law-breaker.

Nash Equilibrium

A Nash Equilibrium consists of a set of strategies for both players that are the best possible for each of the strategies an opponent might choose.

It turns out that this system contains two Nash Equilibrium points.

In this formulation, the strategy pairs that can yield a Nash Equilibrium are:

1)  the law-enforcer does not award the death penalty + the law-breaker does not kill the victim

2)  the law-enforcer awards the death penalty + the law-breaker kills the victim


So, what this analysis suggests is that if the death penalty is awarded for crimes like rape, there will be a strong motive for perpetrators to kill their victims.  Conversely, if the penalty for crimes like rape is less than the penalty for murder, there will be a strong motive for perpetrators not to compound lesser crimes with murder.

What is also interesting to note is that if law-enforcement is not very effective at identifying perpetrators without a victim’s assistance (if there were no DNA matching technology or if the police force were ineffective), a criminal would have a good incentive to kill his/her victims [the result would be obtained if you change the -3 to -1 in the law-breaker’s payoff function].  Ineffectiveness of policing would reduce the negative payoffs for a rob+murder strategy to the point where murder to compound the crime might become an appealing alternative to a law-breaker.

The Alternative Scenario

I had heard from a friend from another country that in his country, thieves would be killed by anyone who caught them (passers by would tie them up and kill them summarily and without trial – by ramming pins into their heads).  I can imagine that this strategy might cause law-breakers to do everything possible to hide the crime – including murdering anyone who might hinder their escape or later identify them to their captors.

So, in a way, the matrix justifies having an effective police force at public expense.


So, we’ve tried to show using some simple game theory that Nishanth’s intuitions about penalizing rapists with the death penalty are possibly right on target.

However, we have chosen a model that is very simplistic and not a very accurate fit for the problem.  The model chosen is of a static game with complete information, but that is a bit of a simplification.

If you liked this article, you might like some of our earlier writings that attempt to analyze game theoretic models of social media customer service.


We learnt a lot about the subject of game theory from a book that is absolutely-required-reading for anyone with the faintest interest in economics and game theory – “Game Theory for Applied Economists” by Robert Gibbons.