Student Comments – December 10, 2008

Brett Harrison

Optimizing Scrip Systems: Efficiency, Crashes, Hoarders, and Altruists

This paper builds upon a previous paper, one that we studied last class, and discusses some theoretical results about scrip systems. They introduce the concept of a heterogeneous network into their model, in which each user is assigned a random type. This type could be, for example, how altruistic that person is, i.e. how much utility they derive from actually doing the work as opposed to receiving work. The authors show that threshold strategies provide a Nash Equilibrium in the heterogeneous setting. Also, they provide definition for the concept of monetary crash, in which there is so much money in the money supply that everyone becomes to rich to have any desire to work, and hence the system fails. They show that social welfare increases with an increase in money supply up until (and obviously not including) the point where the money supply is so high that it causes a monetary crash. Another interesting result is that they show how you can deduce player types from the distribution of money. Finally, they show that greater numbers of altruists increases the social welfare up to the point that the system fails due to monetary crash. This last point seems to confirm some real-world situations; for example, in Bittorrent networks, much of the requests are fulfilled by altruists, and it seems clear that the existence of such altruists increases the welfare of the system, although it does indeed cause others to have less motivation to work.

Travis May

In this posting, I’d like to follow up on my proposal to use a true cash system instead of scrip system. Users would initially buy units of the scrip (using real cash to purchase it), the scrip would be exchangeable in the local community, and then users can sell their scrip back to the system.

This paper shows a number of problems with scrip systems that are not linked to cash. Like a cash system, there are recessions, spending gluts, and spending sprees that take place in the system, causing a suboptimal level of consistency to the ability to exchange scrip for services (or to earn scrip for services). The Washington babysitters’ coop is a telling example, in which the idea of scrip ended up causing a dysfunctional exchange system that did not properly align incentives. Under my proposal, each person would be offered cash, the value of which would be dependent on much broader systemic questions since there is a ready substitute for use of the money (i.e. purchasing items in the real economy).

My proposal, then, would be a fixed inflation-adjusted price level for scrip currency, at which anyone can buy or sell the currency at any time. Thus each scrip has a standard, invariable value that can be used in exchanges. Then, exchange prices should not be fixed, and should allow market matching based on willingness to pay, willingness to provide a service, and expected quality of the service. The value of the artificial currency in this case is primarily that it reduces transaction costs. The moral value of switching away from currency to scrip is no longer offered as a benefit in this system, but I would argue that it was largely a misperception in the first place – any system that requires effort in exchange for goods seems similarly dubious ethically and results in suboptimal behavior, such as hoarding. However, linking scrip to currency goes a long way towards properly aligning incentives for users to not engage in strategic behavior.

Haoqi Zhang

Optimizing Scrip Systems

The main contribution of the paper is the characterization of the effect of changing the money supply in a scrip system on agent behaviors and social welfare. In particular, the authors show that increasing the money supply maximizes the social welfare (up to a point). Furthermore, the authors extend the analysis from the previous paper on the effects of altruists and hoarders on the system. On the theoretical side, the authors offer a proof of Nash Equilibria in threshold strategy that is easy to compute.

I think the paper is a good step forward from the previous paper in that it is extending the model to allow for payoff heterogeneity and the analysis is more rigorous as well. I find it surprising how clean the model is as well as the strength of the results relating to changing the money supply.

It is not clear to me where the work goes from here, though. Do these results carry over to other settings? Can we capture more of the dynamics in the model?

Sagar Mehta

This paper extends the theoretical results of the previous paper to account for multiple types of agents (including hoarders and altruists). I found the results of the paper rather interesting. The monetary crash phenomenon as shown in Figure 4 is a surprising result to me. Would this make optimizing the money supply a very difficult thing to do? If altruists and hoarders are thought of as effectively increasing/decreasing the money supply, then would the socially optimal equilibrium suddenly break down if just one player switches from being a regular player to an altruist? Would this actually happen in real-life as well? One thing I had a problem with was assuming agents are extremely cognizant of what they are sharing and what requests they are fulfilling. In bitTorrent for example, you are always sharing files while downloading. If someone who normally plays a threshold strategy forgets to turn off his computer, could this be modeled as an altruist for some period of time?

I think another interesting type of agent, which I briefly touched on in my previous post, would be a malicious user. This could be someone who accepts requests, but doesn't follow through correctly in some way (thereby just taking the payment), or someone who seeks to infiltrate a filesharing system with buggy files.

Malvika Rao

Optimizing Scrip Systems: Efficiency, Crashes, Hoarders, and Altruists.

This paper presents an interesting and informative follow-up to the

previous paper. I am skeptical about the idea that as a systems design

method we must inject money into scrip systems until just before a

monetary crash is reached. Isn't there a temporal component here? How

should this money be injected? All in one or two shots or slowly spread

out over a period of time? And would these two methods of injecting money

not yield different results in terms of the timing and extent of a crash?

It appears that as the effect of altruists and hoarders is opposite, it

might be beneficial to a system to have some representation of the two

groups. What should be the ratio of altruists to hoarders in order to

maintain a "good" equilibrium? What if a system contained only altruists

and hoarders? Would the altruists contribute forever and the hoarders

hoard forever?

Hao-Yuh Su

Optimizing Scrip Systems: Efficiency, Crashes, Hoarders, and Altruists

In this paper, the author gives an analysis of the scrip system with

heterogeneous agents, which is advanced from previous work.

Under this setting, the author shows the existence of the equilibrium

and the corresponding strategy that agents use. Also, it probes the

relation between the stability of the system and the number of altruists

and hoarders.

When the author is talking about monetary crash, he makes an

assumption that the price is fixed. I thinking that if the price can

be adjusted by agents, can such statement still be hold? That is,

for example, the author says that there will be no trading if every

agent fell itself rich. However, if bidder increases the price and is

willing to pay more scrip, will there still be no trading? In reality, the

price should be adjusted by supply and demand. I am wondering if

such assumption of fixed price is practicable. Another possible condition

is cooperation. Take the baby-sitting case for example. One may think

that he doesn't want a full service and want to buy the service with

several others with half or third of the price. In this case, the price is

adjusted equivalently speaking.

Nikhil Srivastava

The authors consider the optimization of scrip systems through the analysis of money distribution and supply in these networks. They demonstrate how it is possible to infer agent characteristics from an equilibrium distribution of money, using a generalization of the threshold strategies developed in our previous paper. They also demonstrate interesting characteristics in the economics of money supply, including "monetary crashes" when money is devalued sufficiently and the consequences of populations of altruists and hoarders to the thresholds for this occurrence. The paper seemed a strong continuation of their previous work, which did not formally address the problem of hoarding. My problem would be the opinion that the behavior of hoarders and altruists and their occurrence in the population would change as the money supply changed, with perhaps more hoarders being present when the value of money is high and more altruists when the value is low. This might skew the stabilizing dynamics they suggest in hoarders.

Victor Chan

Paper: Optimizing Scrip Systems: Efficiency, Crashes, Hoarders, and Altruists

The main contribution of the paper is showing how to find the distribution of money in scrip system, and how that number can be used to determine the number of users using each strategies. The paper also discusses how the amount of money in the system changes the efficiency of the system. This leads to a discussion of increasing the M in the system until a crash. This was interesting to me since in the real world, inflation would have kicked in and prices of the services would just increase. In the scrip system, it seems like the agents would stop working all together. Perhaps in the real world since some services are essential, people are willing to pay higher amounts of money for it, so naturally the price of the service will increase as a result of people not wanting to work. Another interesting point in the paper was about the altruists being bad for the system. I would agree that when there is scrip involved then altruists could potentially crash the money system if there are too many of them, however if the money system crashes and everyone else is still able to get the resources they want without paying, it is not necessarily a bad thing.

The paper also answered many of my questions from the previous paper by the authors. A project idea for this paper would be to extend the system so that the money expires. This was mentioned in class the other day, and I think this might help fix the problem where there are too many altruists and will likely eliminate hoarders.

Michael Aubourg

In this paper, they have examined the effect of the money supply on social welfare and showed that social welfare is maximized

by increasing the money supply until the monetary crash point of the system.

At this precise point, the money is sufficiently devalued that no agent is willing to perform anything.

This result remind me the case of the Yahoo Answers.

In this game, it is really easy too get more points. And points here correspond to what we call money.

Here are some ways to earn "money" :

-Begin participating on Yahoo! Answers : 100 points

-Have your answer selected as the best answer: 10 points

-Choose a best answer for your question : 3 points

-Answer a question: 2 points

-Log in to Yahoo! daily 1 point (!)

-Receive a "thumbs-up" rating on a best answer that you wrote (up to 50 thumbs-up are counted): 1 point per "thumbs-up"

so we see that it is easy to earn money in that system, which could lead to a too big amount of money in the system.

Hence, depending, on the "prices" of the "services", agents will or won't perform any services.

Which are the way to spend the money in Yahoo Answers ?

1) Asking a question. Cost : 5 points

2) Deleting an answer. Cost : 2 points.

First, the product "deletion of answer" is not really interesting. Thus, we can drop it. Obviously, the main product of Yahoo Answer is to ask a question.

Considering the 100 points we earn at the beginning, the point we just earned because we had to login, and the 2 points you can get very easily by providing an answer(maybe irrelevant),

it is really easy to get more and more money.

The direct consequence will be an increase of the total amount of money in the system.

Hence, the agents won't have any more incentive to provide service, and will just use it : they will ask questions.

Since it is the same for other people, the will also ask more questions, and answer less or no questions : This is the end of the system : the crash-point.

A solution to "balance" this system would be to do an analogy with an economic system, and to apply the Irving Fisher "equation of exchange" :

M.V=P.Q where

M is the total amount of money in circulation on average in an economy.

V is the velocity of money that is the average frequency with which a unit of money is spent.

P is the price level

Q is an index of expenditures.

-In this case, there is no reason for V to decrease when we get closer to the crash point.

-Then the left side will go even higher.

-It's hard to influence Q.

-So a possible solution could be to raise the price level, in order to match the increase of M.V.

Under that condition, we can save the balance of the system.

Concretely, it mean that the "prices" to ask a question should be readjusted periodically depending on the current measured M.

Peter Blair

This paper build upon the scrip model outlined the the paper that we read for last class. Here the authors attempt to characterize the different types of agents in a scrip system, in particular irrational agents such as hoarders and altruism and see how their strategies affect the equilibrium of the scrip market. The paper finds that altruist increase the money supply to the point where there is a system crash, which leads to lower social welfare. Hoarders serve an important role, and opposing role to that of altruist, by restricting the money supply, which helps to avert a market crash, but could have the effect of dimishing participation in the game, as mentioned in the capitol hill babysitting coop example. For the future work of this type of research, the authors mention identifying other types of irrational agents in order to model the effect of their irrational behavior on the scrip system. One example of an irrational agent is one who does not obey the threshold strategy i.e. an agent who spends until he has no money left in his scrip account -- this person is the opposite of a hoarder and I suspect would represent a significant fraction of users. One example of this type of user would be an individual who archives all of her email on gmail and only deletes messages when her inbox is full. In the case of a scrip in a P2P setting, we can imagine that fullfiling request of other agents incurs a cost to a user. So it's advantageous to be a requester up until the point that one's requesting privileges are non-existent because one has run out of scrips. Agents of this type effectively create hoarders out of agents who are altruist.

Nick Wells

Optimizing Scrip Systems: Efficiency, Crashes, Hoarders and Altruists

This paper presents an interesting overview on the design of scrip systems and the effects of their different components. Scrip is a substitute for money used within the framework of a specific system. The authors describe scrip systems as working fairly well up to a specific threshold value (of scrip created) after which the system crashes. Users of the system receive the most utility very close to the threshold value but if that is crossed, they receive zero utility. Behavior of the users can to some extent be seen as scrip creation or destruction. Altruists who satisfy all requests for little or no money in return are similar to scrip creation and hoarders who never make requests are similar to scrip destruction. In some systems, altruists are the ones performing the majority of requests. In most systems, there is also select group of people seek to hoard as much scrip as possible even irrationally. The behavior of the users in the system factors into its stability.

The authors provide useful recommendations to the designers of scrip systems. They recommend that to defuse altruists money needs to be removed from the system and to defuse hoarders, money needs to be added to the system. They also recommend that one keep adding scrip until the system is about to crash, but not over that.

My comment on the author's recommendations goes back to their initial anecdote about the Capitol Hill Baby Sitter Co-op. In the system, everyone tended to hoard all the scrip at the outset. The designers followed the authors recommendation and added scrip. Then, the users switched their behavior and began to spend scrip but were unable to find others to spend it on. How then should one prevent such a collapse from happening? It is possible that the designers simply went over their threshold amount, but intuitively it seems like this also has to do with simultaneous behavior paths of the users.

Reading this paper, I also couldn't help but wonder how this relates to the monetary system in the real world. Countries such as East Asian countries tend to save while countries like the United States tend to borrow. This international dynamic is not present in scrip systems.

Angela Ying

This paper discussed the existence and calculation of nontrivial Nash equilibria in an ordinary Scrip system. First, it proved that threshold strategies as a whole are stable, and then the paper discussed how to select the best and most stable equilibrium out of the possible threshold strategies, since many are possible. After these initial results, the authors went into a discussion about the fact that people have different utility functions and therefore have many different types of strategies, finding the Nash equilibrium given the types of people there are. An important result was the conclusion that the best strategy was the one requiring the least number of types. Finally, the paper demonstrated the existence of an optimal amount of money in the system, and discussed the effects of altruists and hoarders.

I thought that a strength of this paper was its acknowledgement of the many types of people in a scrip system. Monday's papers assumed that there was only a "standard" user with the same utility function and then hoarders and altruists as irregular types. In addition, it clearly distinguished between hoarders and altruists, which clarified a lot of me from the previous paper. However, I was still a little confused between the two, because there are not many scrip systems where you can choose to either be a hoarder or an altruist - if you an altruist in a P2P system with scrip, you will automatically get paid regardless of if you want it. So in most systems, you are the same type of person, with differnet payoffs depending on the system.

For the future, I think it would be interesting to examine the possibility of people changing types - from perhaps being an altruist to a standard user and vice versa.

Rory Kulz

The Krugman anecdote is really great; it made the idea perfectly clear.

Section 4 is also quite interesting, a nice structure framework for

turning the model-choice problem against itself; many of the papers we

have examined over the course make so many assumptions about priors or

common priors or certain distributions, it was nice to take a look at

a reversal of, well, what can we infer from observation?

One idea for expanding on more types of irrational or abnormal agents

beyond altruists and hoarders might be to look at discriminatory

practices. Especially on peer-to-peer systems, I know that I would

typically discriminate against low bandwidth type agents, i.e. users

having a file I need but on, say, a 56K connection. This is a very

important type of "irrational" behavior that occurs in these sorts of

systems that I think has been too ignored in the analysis. This cannot

be fully rolled into just the service probabilities unless

discrimination happens uniformly across the system in similar fashion.

Lastly, I'm not coming to this paper from much of an economics

background, but can I just say the monetary crash stuff is really

cool?

Ziyad Aljarboua

This paper is a continuation on Monday's paper. It presents an efficient scrip system and an empirical study of systems. It shows the effect of money supply, hoarders and altruists on the system.

Just like in the Capitol Hll Baby Sitting Coop, where few or too many coupon supply might cause a rescission or crash respectively, the amount of scrip in circulation in a given system has a direct impact on the effectiveness of the system. This suggests that there is an optimal amount of scrip supply that will make the system most efficient.It was shown that social welfare is maximized by increasing the money supply to the point that the system is near a monetary crash. Agents with no funds cannot purchase services, thus impacting the system the most. So, system can be optimized by minimizing number of agents with no funds, or in other words, increasing money supply while maintaining it below monetary crash limit.

It is shown that characteristics of agents can be inferred from equilibrium distribution of money in the system. In this setting, having few altruists may improve the social welfare since altruists represent money in the system. while having too many might damage the social welfare because they lead to a monetary crash. On the other hand, Hoarders decrease social welfare as they decrease money supply in the system but promote system stability by preventing monetary crashes.

In conclusion, this paper ,written by same authors of Monday's paper, came to the same finding: agents follow a threshold strategy whereby they stop providing services. The only suggestion here is that the system designers should also follow the same strategy: flood the system with money while not causing a monetary crash (stay below threshold). I wonder what is the real meaning of increasing money supply in an online system? for the scrip to have an effect on the system, it must have a value. So, drastically increasing money supply in an online system must have a cost associated with it that is not discussed here.

Moreover, this paper does not examine manipulation or agent's priorities. It assumes that all agents have the same goal and motivation. One way this system could be breached is when agents exchange system scrip with real money.

Subhash Arja

The authors of this paper explore the theoretical aspects of "scrip" systems. "Scrip" referes to currency that is used in artificial societies such as second life or p2p networks. The look to find some equilibrium point in the system such that people are not hoarding money but at the same are not over spending. Overall, I found the paper very interesting and a nice foundation project for future work. The authors had to limit themselves to agents who were payoff heterogenous. I didn't think that this oversimplified the problem, but provided a small enough scope to obtain meaningful results.

I would be interested on future research in this area focuses specificly on examining content distribution networks like BitTorrent. It would be interesting to see how this model applied onto such a real system to find how many people need to share a file for everyone else on the network to benefit.

Xiaolu Yu

This paper shows that maximum entropy and mononicity of the best reply function are the two keys in determining the existence and properties of equilibria. In another word, agent i's best response function is monotone if all the other agents play threshold strategy. This proves the existence of equilibria in pure strategies and these equilibria are computable with efficient algorithms. Maximum entropy also provides tools for studying the types of a population from the distribution of wealth.

The contribution of this paper to system designers is that in order to optimize social welfare, money supply should be kept adding into the system until the system is about to experience a monetary crash.

As many protocols for P2P systems, the threshold strategies in scrip systems have the feature that agents stop providing service once they have received a certain amount of services. Attackers can exploit this opportunity by ensuring that a peer maintains a good reputation or supply of money without satisfying any requests the peer makes, then that peer will no longer provide service for others. Thus the system is potentially vulnerable to such attacks.

The authors point out some interesting area for future work, including the impacts of other types of irrational behaviors, the validity of predictions made in this paper on real functioning scrip systems, and the effects of a dynamic price system.