TOPIC: Applying bids to multi-item auctions

Ditto David. Immediate verification and no jack-wagon Z bidding $150 late to push out an earlier bidder.

And most importantly, Quarter house and "cutoff" house have a lesser chance of actually running.

That said, In practical terms (with full visibility on all bids for all involved), "cutoff" will be easier on the auction runner and come close to same amount due to PYPs basically being in the $100-120 range (at least for the data I’ve seen).

I’m not sure the same applies to trade goods - which are a significant portion of the auction.

Lequinian wrote: Ditto David. Immediate verification and no jack-wagon Z bidding $150 late to push out an earlier bidder.

And most importantly, Quarter house and lamda house have a lesser chance of actually running.

I think.you mean quarter and beta.
I believe lambda has the highest chance of funding.

Wade Schwendemann wrote:

Lequinian wrote: Ditto David. Immediate verification and no jack-wagon Z bidding $150 late to push out an earlier bidder.

And most importantly, Quarter house and lamda house have a lesser chance of actually running.

I think.you mean quarter and beta.
I believe lambda has the highest chance of funding.

Probably, it’s too early to keep the abstract nomenclature straight. Perhaps cutoff vs virtual assistant houses?

If quarter is where everything is a quarter
Cutoff is where every winning bid pays the same
And virtual assistant is where people can pay 1 bid increment less depending on timing

Then I think virtual assistant, by virtue of its possible higher later bids, results in the highest chance for funding.

This is a great abstract discussion. I remain unconvinced, but I am excited for Matthew's next auction attempt using his rule set. Perhaps after Starkhound funds, then Ralykam?

Also, it's worth noting that different people set their expectations differently in terms of minimum funding levels.

Matthew Hayward wrote:

kurtreznor wrote:

Wade Schwendemann wrote:

Matthew Hayward wrote:

Wade Schwendemann wrote: I dont know about the system that was proposed above, but my system is
A bids on item, takes the lead at 0.25
B is on item, takes the lead at 0.25
C bids on item, but at a max of $1

Since they bid first, A and B take the lead at $1 each, as they bid before C.

Basically, all winning bids are within one bid increment of one another.

If someone bids before you and has a lower max bid than you, they are the first ones to lose their items when your higher bid comes in.

If you want 10, and they want 5 of an item there are 12 of, shouldnt you have to pay more for the privilege of getting the number of items you want, and doing so later?

If you bid first with the higher max bid, you only pay their max bid for your items and get the number you want.

I'm not sure if I'm saying something very similar or something very different from what you're after. I am planning to try to figure out the difference between the final numbers for my recent auction using the method I described and what they would have been had I used your method.

I cannot guarantee I will get this done, but I am planning to try.

Hmm. What do you do in this scenario:

S1
3 identical items
A wants 2 at $105
B wants 2 at $100
C wants 1 at $1

And this one:

S2
3 identical items
A wants 2 at $105
B wants 1 at $100
C wants 1 at $1

In both scenarios my approach gives A two items, B one item, and both A and B pay $1.

I suspect deviating from my approach gives a large difference in the cost someone pays between S1 and S2.

I suspect this deviation would cause bidders to prefer my approach.

You are correct.

S1
A gets the 2 items they want, and pays $100
B gets 1 of the 2, and pays $100
C gets nothing

S2
A gets 2 items for $1
B gets 1 item for $1
C gets nothing

I believe that, while the result is the same with your method between the 2, the difference lies in the desires of the bidders.

Their desires need to be respected and accounted for in some way.

Hold on, i think i missed something. What method has S1 result in A getting two, B getting one, and all 3 sell for $1? Because that should be right out. If i am B, i would call bullshit. I wanted a 2nd item, and i bid more than the $1 selling price.

After you called bullshit, I'd explain to you that A bid more than you, and this is an auction, and ask why you're upset that someone who bid more than you in an auction won more items than you.

Then I would invite you to explain to me any way of running an auction whatsoever with these facts:

3 identical items:
A has bid 2 at $105
B has bid 2 at $100
C has bid 1 at $1

Where you get a better outcome as B than what I've proposed above, and to which both A and C would agree is fair also. (You should want A and C to agree it's fair too, because you could just as easily be A or C one day).

Can you do that? Is there any possible better outcome for B in this scenario than 1 token at $1 no matter how you run it?

If there is, I don't see it.

Ok, after reading this and more posts, i understand how your auction system works. And i agree that the results will be 'better' for the bidders than other fair systems...but it isnt obvious to anyone except the auctioneer.

The only way to keep people from questioning the results is to make the bids public. It shouldnt say who each bid is from, but you will need to publicly show all of them so bidders understand how they are losing even when their max hasnt been reached. Of course, then the risk is getting sniped because new bidders know what they need to beat. Maybe you could keep bids secret until auction is over, then release the top bid details so bidders can verify they were treated fairly?

David Zych wrote:

Matthew Hayward wrote: Example. there are 32 PyPs:

A bids $100 on 17
B bids $105 on 17
C bids $5 on 32
D bids $105 on 1
E bids $75 on 4, and $90 on 3, $100 on 2, and $110 on 1
F bids $1000 on 2
G bids $55 on 8

Who is winning what and at what price? And why?

Imagine a very faithful, very patient virtual assistant [...]

In my opinion, the appropriate auction result is the one that is consistent with this model. Of course there are implementation shortcuts available

It occurs to me that I have not yet actually provided an implementation shortcut, so here it is.

Optimized Lambda House implementation:

1. Sort received max bids from highest to lowest, breaking ties in favor of earliest submitted.
2. Allocate items to the highest bidders in order.
3. Set the price based on the first bid that doesn't win all of its desired tokens.
4. Any winner who bid later than the highest losing bid must pay one bid increment more.

That's it!

In the above example:
$1000 F x2 (wins)
$110 E x1 (wins)
$105 B x17 (wins)
$105 D x1 (wins)
$100 A x17 (wins 11, loses 6)
$100 E x2 (loses)
$90 E x3 (loses)
$75 E x4 (loses)
$55 G x8 (loses)
$5 C x32 (loses)

The highest losing bid is $100 by A, so A himself pays $100 (as would any other winner who bid earlier than A, though there aren't any in this example). The other winners bid later than A, and must therefore pay $105.

Result:
F pays $105 for 2
E pays $105 for 1
B pays $105 for 17
D pays $105 for 1
A pays $100 for 11

This is obviously much faster to evaluate, but will provably give the same result as the full virtual assistant model. To see this, simply observe that each and every winning price is exactly as high as it must be to avoid being outbid by the highest losing bid.

Note that a conscientious auctioneer will likely maintain his internal list in sorted price order for convenience, but will re-sort in chronological order when publishing status updates during the auction, in order to avoid telegraphing whose max bids were highest:

A pays $100 for 11
B pays $105 for 17
D pays $105 for 1
E pays $105 for 1
F pays $105 for 2

David Zych wrote:

David Zych wrote:

Matthew Hayward wrote: Example. there are 32 PyPs:

A bids $100 on 17
B bids $105 on 17
C bids $5 on 32
D bids $105 on 1
E bids $75 on 4, and $90 on 3, $100 on 2, and $110 on 1
F bids $1000 on 2
G bids $55 on 8

Who is winning what and at what price? And why?

Imagine a very faithful, very patient virtual assistant [...]

In my opinion, the appropriate auction result is the one that is consistent with this model. Of course there are implementation shortcuts available

It occurs to me that I have not yet actually provided an implementation shortcut, so here it is.

Optimized Lambda House implementation:

1. Sort received max bids from highest to lowest, breaking ties in favor of earliest submitted.
2. Allocate items to the highest bidders in order.
3. Set the price based on the first bid that doesn't win all of its desired tokens.
4. Any winner who bid later than the highest losing bid must pay one bid increment more.

That's it!

In the above example:
$1000 F x2 (wins)
$110 E x1 (wins)
$105 B x17 (wins)
$105 D x1 (wins)
$100 A x17 (wins 11, loses 6)
$100 E x2 (loses)
$90 E x3 (loses)
$75 E x4 (loses)
$55 G x8 (loses)
$5 C x32 (loses)

The highest losing bid is $100 by A, so A himself pays $100 (as would any other winner who bid earlier than A, though there aren't any in this example). The other winners bid later than A, and must therefore pay $105.

Result:
F pays $105 for 2
E pays $105 for 1
B pays $105 for 17
D pays $105 for 1
A pays $100 for 11

This is obviously much faster to evaluate, but will provably give the same result as the full virtual assistant model. To see this, simply observe that each and every winning price is exactly as high as it must be to avoid being outbid by the highest losing bid.

Note that a conscientious auctioneer will likely maintain his internal list in sorted price order for convenience, but will re-sort in chronological order when publishing status updates during the auction, in order to avoid telegraphing whose max bids were highest:

A pays $100 for 11
B pays $105 for 17
D pays $105 for 1
E pays $105 for 1
F pays $105 for 2

This describes how I do auctions. The sorting and presentation can take a lot of time and one reason I have worked towards more fixed priced offerings.

If bids are exposed then folks might tend to incrementally increase their bids. Example Day 1 I bid $0.25 and do mutiple raises of bid increments up to what I want to spend or I am winning and I will snipe at the end. This will tremendously increase the time to sort and present.

David Zych wrote:

David Zych wrote:

Matthew Hayward wrote: Example. there are 32 PyPs:

A bids $100 on 17
B bids $105 on 17
C bids $5 on 32
D bids $105 on 1
E bids $75 on 4, and $90 on 3, $100 on 2, and $110 on 1
F bids $1000 on 2
G bids $55 on 8

Who is winning what and at what price? And why?

Imagine a very faithful, very patient virtual assistant [...]

In my opinion, the appropriate auction result is the one that is consistent with this model. Of course there are implementation shortcuts available

It occurs to me that I have not yet actually provided an implementation shortcut, so here it is.

Optimized Lambda House implementation:

1. Sort received max bids from highest to lowest, breaking ties in favor of earliest submitted.
2. Allocate items to the highest bidders in order.
3. Set the price based on the first bid that doesn't win all of its desired tokens.
4. Any winner who bid later than the highest losing bid must pay one bid increment more.

That's it!

In the above example:
$1000 F x2 (wins)
$110 E x1 (wins)
$105 B x17 (wins)
$105 D x1 (wins)
$100 A x17 (wins 11, loses 6)
$100 E x2 (loses)
$90 E x3 (loses)
$75 E x4 (loses)
$55 G x8 (loses)
$5 C x32 (loses)

The highest losing bid is $100 by A, so A himself pays $100 (as would any other winner who bid earlier than A, though there aren't any in this example). The other winners bid later than A, and must therefore pay $105.

Result:
F pays $105 for 2
E pays $105 for 1
B pays $105 for 17
D pays $105 for 1
A pays $100 for 11

This is obviously much faster to evaluate, but will provably give the same result as the full virtual assistant model. To see this, simply observe that each and every winning price is exactly as high as it must be to avoid being outbid by the highest losing bid.

Note that a conscientious auctioneer will likely maintain his internal list in sorted price order for convenience, but will re-sort in chronological order when publishing status updates during the auction, in order to avoid telegraphing whose max bids were highest:

A pays $100 for 11
B pays $105 for 17
D pays $105 for 1
E pays $105 for 1
F pays $105 for 2

This is literally Beta houses system, except sometimes it charges the winning bidders more.

Having converged on this - I’m glad to say as a bidder I’d rather pay less than pay more for the same thing.

Thanks for the spirited discussion!

I agree lambda house is more likely to fund (because sometime it charges more for the same winning quantities).

I agree that many people feel there is a some kind of transparency problem - I think that just familiarity with other systems - but it’s a matter of opinion.

It will be interesting to see if I can get one of these off the ground.

I’m not going to run one until the 2020 can be included.

Matthew Hayward wrote: I agree that many people feel there is a some kind of transparency problem - I think that just familiarity with other systems - but it’s a matter of opinion.

It will be interesting to see if I can get one of these off the ground.

I agree with both of these points.

I would likely bid in your auction once one begins.

Necro-ing this thread (which was a really fun debate!) to reflect on the intervening year's worth of observing real auctions...

While I still maintain that the Lambda House model is theoretically sound, I have come to the conclusion that it has a serious practical flaw: far too many people (including some of its own "employees" i.e. auctioneers) don't actually understand it, and don't realize that they don't understand it.

So let's do something easier!

I have yet to run an auction myself, but if I were going to right now, here are the bidding rules I think I would use. Please feel free to use this if you like it (that's why I'm posting it).

---

How bidding works:

• Bids should be sent to me via PM.
• Feel free to bid on any number of items for which there are multiples available.
• For the auctioneer's sanity, all bids must be divisible by $0.25 (per item), bids above $10 divisible by $1, and bids above $50 divisible by $5. If you try to bid $117.42, I will round it down to $115.
• Items will be won by the highest bidder(s), breaking ties in favor of the first bid received (which rewards early bidding).
• The price per item for all winners is equal to the highest bid that does NOT win all of its desired items (or the reserve price if no bid goes unfilled), so you can submit your top bid initially and not overpay.

Example: there are 5x Zyzzlvarian Ducks.
A bids $95 on 5x.
B bids $105 on 1x.
C bids $140 on 1x and $105 on 3x (wanting 4x total).
D bids $120 on 2x.
The winners are C,D,D,B,C for $105 each (because C still wanted more items at that price).

---

This produces almost the same outcomes as the optimized Lambda House implementation, except that D is not charged an extra "bid increment" for having bid later than C; the fact that D bid higher is sufficient. Which means the auctioneer is comparatively giving up a little bit of money (pathological worst case: one bid increment per item in the auction), but I suspect also suffering less headache since the chronology of bids now only matters in an exact tie.

It's also similar to what Matthew ended up using in truedungeon.com/forum?view=topic&catid=584&id=251008 , except that I've added back an "auctioneer's sanity" provision in order to regain what I perceive as the upside of discrete bid increments.

Edit: for anyone who might be curious, "Zyzzlvarian Duck" is my homage to this 2002 MIT Mystery Hunt puzzle.

This was a lot of fun!

Here's hoping it takes off again, as I have a light day at work today (famous last words).

David Zych wrote: Necro-ing this thread (which was a really fun debate!) to reflect on the intervening year's worth of observing real auctions...

While I still maintain that the Lambda House model is theoretically sound, I have come to the conclusion that it has a serious practical flaw: far too many people (including some of its own "employees" i.e. auctioneers) don't actually understand it, and don't realize that they don't understand it.

So let's do something easier!

I have yet to run an auction myself, but if I were going to right now, here are the bidding rules I think I would use. Please feel free to use this if you like it (that's why I'm posting it).

---

How bidding works:

• Bids should be sent to me via PM.
• Feel free to bid on any number of items for which there are multiples available.
• For the auctioneer's sanity, all bids must be divisible by $0.25 (per item), bids above $10 divisible by $1, and bids above $50 divisible by $5. If you try to bid $117.42, I will round it down to $115.
• Items will be won by the highest bidder(s), breaking ties in favor of the first bid received (which rewards early bidding).
• The price per item for all winners is equal to the highest bid that does NOT win all of its desired items (or the reserve price if no bid goes unfilled), so you can submit your top bid initially and not overpay.

Example: there are 5x Zyzzlvarian Ducks.
A bids $95 on 5x.
B bids $105 on 1x.
C bids $140 on 1x and $105 on 3x (wanting 4x total).
D bids $120 on 2x.
The winners are C,D,D,B,C for $105 each (because C still wanted more items at that price).

---

This produces almost the same outcomes as the optimized Lambda House implementation, except that D is not charged an extra "bid increment" for having bid later than C; the fact that D bid higher is sufficient. Which means the auctioneer is comparatively giving up a little bit of money (pathological worst case: one bid increment per item in the auction), but I suspect also suffering less headache since the chronology of bids now only matters in an exact tie.

It's also similar to what Matthew ended up using in truedungeon.com/forum?view=topic&catid=584&id=251008 , except that I've added back an "auctioneer's sanity" provision in order to regain what I perceive as the upside of discrete bid increments.

I respectfully think bid increments are dumb. Use a google sheet and let people pay what they want, then if they’re willing to pay 12.80 per 1k gold or 194.20 for a wish ring the auction has the extra 4.20 on the wish ring and 0.30 per 1k. I also have never run an auction so $0.02.