I wrote up a post yesterday, but I found I was unable to post it, except as a draft, since I lack the necessary karma. I thought it might be an interesting thing to discuss, however, since lots of folks here have deeper knowledge than I do about markets and game theory
I’ve been working recently for an auction house that deals in things like fine art, etc. I’ve noticed, by observing many auctions, that certain behaviors are pretty reliable, and I wonder if the system isn’t “game-able” to produce more desirable outcomes for the different parties involved.
I think Less Wrong readers might have some interesting insights into the situation. Hopefully, at the least, it’s an interesting thing to think about for a few minutes. Feel free to point out if this is well-worn territory; in fact, any feedback is welcome.
Structure of the game:
We have objects consigned with us. We have our experts evaluate the objects, and provide an estimate of their value, based on previous auction outcomes for similar objects and their own expertise. So, for instance, a piece of furniture may be estimated to bring a value of $400-$800, a particular painting might be estimated to bring $10,000-$20,000, and so forth. While not “arbitrary”, they are to some degree simply good guesses.
We publish a catalog before the auction, listing the items up for sale along with their estimated values. A minimum bid is set, usually half of the low estimate.
The auction proceeds following bidding increments, which vary from one price bracket to another. So, for instance, between $100 and $200, the bidding increments are $10 -- so, $100, $110, $120, etc. Between, say, $10,000 and $20,000 however, the bidding increments are $1000 -- $10,000; $11,000; $12,000 and so forth.
Regardless of what bracket the prices fall into, there are several tendencies that happen frequently:
-- Bidders will readily bid on items they want which still have an asking price of below the low estimate. They feel like they’re getting a bargain.
—Bidding will slow between the low estimate and the high estimate. Here, they’re really relying on the estimate for their idea of whether or not the deal is so good.
—Bidders become much more reticent about continuing to bid once the price reaches or exceeds the high estimate. -- “Bidding wars” are more likely to the degree to which bidders feel they still have room to get a “good deal”
It seems to me that it is advantageous (in terms of maximizing the final price paid for an item) to have MORE bidding increments between the low and high estimates than it is to have fewer. That is to say --
I would expect more bids on an item which has an estimate between $200-$400, where there are 20 bid increments between the low estimate and the high estimate, than I would expect on an item estimated to sell between $10,000 and $15,000, which only has five bidding increments between the low and the high.
Now, naturally, some of that has to do with the fact that the lower-priced item is affordable to more bidders. It’s also worth noting that increasing the bid increments makes sure that the auction itself doesn’t take forever to complete (the higher increments cause bidders to drop out faster, regardless of the number of increments.
So, all this in mind, it seems plausible to me that we could marginally improve the prices being paid for our larger value objects in one of two ways:
-- Increase the granularity of the bidding increments at higher values
—Provide low estimates that allow for a larger number of bidding increments (instead of saying, “the estimate is $15,000-$20,000” we could say, “the estimate is $10,000 - $20,000″)
It seems tricky to figure out whether the strategy works, though. After all, each of these objects is unique; it’s not like shares of stock or pork bellies or something, where you have a whole bunch of the same stuff and the market is setting a price.
My questions for you all then:
—Is my thinking on this subject sound?
—Do you think that the number of bidding increments available to bidders can affect their behavior in the way I’ve outlined (am I right?)
—Assume we implement one of the two strategies for maximizing the prices paid. Is there any reliable way to measure the outcome to see if it worked?
An experiment which would disprove my hypothesis regarding more bidding increments would be something like:
Run at least three auctions for the same or similar items with the same or similar bidders, one using normal estimates and bidding increments for a control, one where the low estimate was lowered to allow more increments, and one with the same estimates, but more granular increments. IF the price paid in each auction was roughly equivalent, THEN the hypothesis is disproven.
The problem with that is the nature of the property we auction—there’s only one of anything. Each auction lot is, in important ways, different from the others. There’s only one of this painting; only one of this desk. Even when two objects are similar, there are still often condition differences and so forth.
I’ll have to consult with some of the appraisers and see if there’s ever an exception to this rule.
But ok, that brings up another interesting question. Is there a way of simulating auction behavior? Has someone written a computer program to do this sort of thing? What kinds of assumptions do they make about the behaviors of individual agents?
Is there any reliable way to measure the outcome to see if it worked?
If we assume that the appraisals are disconnected from the winning bids*, then couldn’t one just see whether the ratio of sale:appraisal is increasing? If the appraisals are honest, then any jiggery-pokery should alter the ratio—eg. a successful manipulation will lead to people paying an average 93%, where they used to pay 90%.
that is, there is no feedback—the appraisers don’t look at recent sales and say, oh, I’ve been lowballing all my estimates! I’d better start raising them.
I wrote up a post yesterday, but I found I was unable to post it, except as a draft, since I lack the necessary karma. I thought it might be an interesting thing to discuss, however, since lots of folks here have deeper knowledge than I do about markets and game theory
I’ve been working recently for an auction house that deals in things like fine art, etc. I’ve noticed, by observing many auctions, that certain behaviors are pretty reliable, and I wonder if the system isn’t “game-able” to produce more desirable outcomes for the different parties involved.
I think Less Wrong readers might have some interesting insights into the situation. Hopefully, at the least, it’s an interesting thing to think about for a few minutes. Feel free to point out if this is well-worn territory; in fact, any feedback is welcome.
Structure of the game:
We have objects consigned with us. We have our experts evaluate the objects, and provide an estimate of their value, based on previous auction outcomes for similar objects and their own expertise. So, for instance, a piece of furniture may be estimated to bring a value of $400-$800, a particular painting might be estimated to bring $10,000-$20,000, and so forth. While not “arbitrary”, they are to some degree simply good guesses.
We publish a catalog before the auction, listing the items up for sale along with their estimated values. A minimum bid is set, usually half of the low estimate.
The auction proceeds following bidding increments, which vary from one price bracket to another. So, for instance, between $100 and $200, the bidding increments are $10 -- so, $100, $110, $120, etc. Between, say, $10,000 and $20,000 however, the bidding increments are $1000 -- $10,000; $11,000; $12,000 and so forth.
Regardless of what bracket the prices fall into, there are several tendencies that happen frequently:
-- Bidders will readily bid on items they want which still have an asking price of below the low estimate. They feel like they’re getting a bargain. —Bidding will slow between the low estimate and the high estimate. Here, they’re really relying on the estimate for their idea of whether or not the deal is so good. —Bidders become much more reticent about continuing to bid once the price reaches or exceeds the high estimate.
-- “Bidding wars” are more likely to the degree to which bidders feel they still have room to get a “good deal”
It seems to me that it is advantageous (in terms of maximizing the final price paid for an item) to have MORE bidding increments between the low and high estimates than it is to have fewer. That is to say --
I would expect more bids on an item which has an estimate between $200-$400, where there are 20 bid increments between the low estimate and the high estimate, than I would expect on an item estimated to sell between $10,000 and $15,000, which only has five bidding increments between the low and the high.
Now, naturally, some of that has to do with the fact that the lower-priced item is affordable to more bidders. It’s also worth noting that increasing the bid increments makes sure that the auction itself doesn’t take forever to complete (the higher increments cause bidders to drop out faster, regardless of the number of increments.
So, all this in mind, it seems plausible to me that we could marginally improve the prices being paid for our larger value objects in one of two ways:
-- Increase the granularity of the bidding increments at higher values —Provide low estimates that allow for a larger number of bidding increments (instead of saying, “the estimate is $15,000-$20,000” we could say, “the estimate is $10,000 - $20,000″)
It seems tricky to figure out whether the strategy works, though. After all, each of these objects is unique; it’s not like shares of stock or pork bellies or something, where you have a whole bunch of the same stuff and the market is setting a price.
My questions for you all then: —Is my thinking on this subject sound? —Do you think that the number of bidding increments available to bidders can affect their behavior in the way I’ve outlined (am I right?) —Assume we implement one of the two strategies for maximizing the prices paid. Is there any reliable way to measure the outcome to see if it worked?
Make sure you’re asking yourself, “what experiment would disprove my hypothesis?” You have several hypotheses in there which might not be optimal.
An experiment which would disprove my hypothesis regarding more bidding increments would be something like:
Run at least three auctions for the same or similar items with the same or similar bidders, one using normal estimates and bidding increments for a control, one where the low estimate was lowered to allow more increments, and one with the same estimates, but more granular increments. IF the price paid in each auction was roughly equivalent, THEN the hypothesis is disproven.
The problem with that is the nature of the property we auction—there’s only one of anything. Each auction lot is, in important ways, different from the others. There’s only one of this painting; only one of this desk. Even when two objects are similar, there are still often condition differences and so forth.
I’ll have to consult with some of the appraisers and see if there’s ever an exception to this rule.
But ok, that brings up another interesting question. Is there a way of simulating auction behavior? Has someone written a computer program to do this sort of thing? What kinds of assumptions do they make about the behaviors of individual agents?
Do you have a large body of data? It’s possible a statistician would be capable of devising appropriate measures to test your hypothesis.
If we assume that the appraisals are disconnected from the winning bids*, then couldn’t one just see whether the ratio of sale:appraisal is increasing? If the appraisals are honest, then any jiggery-pokery should alter the ratio—eg. a successful manipulation will lead to people paying an average 93%, where they used to pay 90%.
that is, there is no feedback—the appraisers don’t look at recent sales and say, oh, I’ve been lowballing all my estimates! I’d better start raising them.