Presentation on theme: "Computational Economics Elan Pavlov. Central problem statement How to design mechanisms and algorithms which cause selfish players to act as if they are."— Presentation transcript:
Central problem statement How to design mechanisms and algorithms which cause selfish players to act as if they are interested in the global good. The problem is people will often be better off if they act selfishly. We need to align their choices to global good. Solve problems with consideration for their real complexities.
Computational economics Computational economics is a form of economics which is constrained by the demands of computer science. Classic economics often just proves that something can be done. Computational economics tries to actually do it.
New field The field started around 1999. Currently there are very few dedicated conferences for computational economics. The field is growing very fast.
Why is computational econ important? The Internet has spawned new problems such as: –Allocation of advertisements. –Auctions with budgetary constraints. –Multi-mechanisms run by different companies. –Strong emphasis on problems where the input is not known from the start.
New problems Auctions which could have been handled easily before are now harder to do. The standard economics approach don’t scale well. A need for quick decisions.
The field takes from Economics Problems. Definitions: players, equilibrium, strategy space Some worldview: Welfare, revenue. Some tools: –Game theory. –Auction theory. –Mechanism design. –Price of anarchy.
But also from computer science Mainly tools: –Online algorithms. –Approximation ratios. –Graph theory. –Reductions. –Optimizations. But also demands: –Polynomial time. –Hardness.
What we take from real life Relevant problems. Feedback. Motivation.
Various problems Keyword auctions. Dragon kill points. Cellphones for selfish users. Parking auctions.
A generalized min cost flow Source Target Advertiser 1Advertiser 2Advertiser 3Advertiser n Keyword 1 Keyword 2 Keyword 3 Keyword 4 Keyword m (capacity, cost, scaling) (n 1,0,1) (n 2,0,1)(n 3,0,1) (n n,0,1) (∞,-v 1,1/p 11 ) (a 1,0,1) (a m,0,1) (∞,-v 2,1/p 2m ) (Not all edges are drawn)
Benefits of transferring demand Better welfare (possibly twice as good) – happier advertisers. Exact values depend on advertisers. Better revenue (possibly much better). Happier users (on average). A triple win for the search engine, advertisers and users. Potential drawback: Loss of control for advertisers.
Other problems with ad auctions Estimating Click through rate. Buying ads. Selling by impression, click-through or conversion. What happens when there is competition for advertisers? How do you maximize revenue?
SDM Model 2 9 S B 20 12 5 1 4 2 3 4 6 7 8 1 3 Unit shipment cost Seller’s cost Buyer’s value M1M1 M2M2 M3M3 M4M4 Utility maximizing agents with quasi-linear utility function (value-payment)
New economies There are a lot of new social networks with their own economies. Second Life has created a (real-life) millionaire. WoW has an economy that is estimated to be larger than many third-world nations. You can buy hundred of items from Guild Wars on eBay.
Why are new economies interesting? These economies are different from classic economies in that: –They can be easily measured. No black economy or gray economy. –They can be easily manipulated (for example supply of land in Second Life). –They can be controlled and designed better. For example, money supply, ways in which people can create wealth etc.
Dragon kill points In MMORPG there are encounters which yield items. The harder the encounter the better the item is. Most hard encounters demand the participation of a large number of players to successfully complete the encounter (and get the treasure). Since the treasure is unsplitable the question arises of how to allocate the item.
An auction but how? The basic idea is to auction off the item to the participating players. The main challenge is a temporal one: Given a fixed set of players (the guild) the demand is fixed and since the supply can go up – the value of items decreases.
Temporal problems This means that players have to make a decision when to bid. The problem is also online. We have several results for different cases (depending on what distributional assumptions we make).
Algorithm Conceptually the algorithm works by renting out items. We look at the supply/demand per day and charge for that day based on that supply/demand. The price fluctuates (decreases) for each day. In practice items are not transferred between players.
Power for cell phones Power usage in cell phones is a function of the distance to the destination. Depending on the terrain the power usage is between distance 2 to distance 4. Why not use multi hops?
The problem People are selfish and the person who agrees to forward can have her cellphone battery depleted faster.
Our solution We offer a simple guarantee: “If you agree to forward for others your battery will always last at least as long as if you don’t”. Usually, it will last three times as long. Based on a simple system of debt. People will forward only if they owe the system. We also have debt forgiveness. (However we have a catch-22).
Auctions Auction are an efficient mechanism for determining value, but not all auctions are created equal! Current auction mechanisms allocate parking to the person with the highest value (willingness to pay).
Problem with current auctions Current mechanisms take no account of the duration of the requested parking. This can prevent several people from parking thereby leading to higher congestion. Worse, this model incentivizes squatting and extended parking.
The core idea The core idea is that for every parking lot there is a capacity and a currently used fraction of capacity. We also have estimates on the lower (P 0 ) and upper bounds of valuations (or more accurately r=log P max ). The price for any duration between [t 1,t 2 ] is ∫P 0 r capacity-usage if a drive is willing to pay the price then they can park.
Prices As demand approximates supply the price approximates the upper bound on value. Although the problem is both a packing problem and an online problem the total welfare approximates the maximal possible welfare. We might not be able to solve the problem but we can get within a factor 2 of the optimal.
Future areas for the field Broad questions: –How does computational economics interact with machine learning? –What happens to online algorithms with selfish users? –New kinds of prediction markets? –Markets and economies of online games. –Markets with intermediaries (such as eBay) –Auctions with probabilities.
In summary The field of computational economics is a new field which has many interesting directions in which to develop. The goal of the field is to understand/invent ways in which people can share resources for global good. The question is how to design, analyze and build mechanisms to enable sharing.
Q & (hopefully) A Sample question: How do you find true love? Answer: