An analysis of auction houses for prognostic shill bidding

[kirispupis]

Recently, since I suspected several auction houses of not only shill bidding, but doing so while knowing what the max bid is, I performed a quick analysis of my activity with four auction houses. Two were on Biddr and two use their own websites and work along a "max offer by a certain time" model. Of these, three I suspected of shill bidding while one I did not.

First, some quick notes. These bids were placed roughly within the last year and all coins are for my own purposes. I am not a reseller.

I will start with the one that did not, which I will call "House S". This is a Biddr house and I removed all live bids since I obviously would win any such coins at my max. These include proxy and pre bids.

Here is a graph showing the distribution for coins I won.

I believe this pretty much confirms my suspicions that House S does not engage in shill bidding.

Now we'll move on to House A, which is also on Biddr. My sample size is small here because I quickly shifted to live bids only because I felt something funny was going on.

Perhaps the sku was made up by the auction being more competitive, since I only won 69% of my bids as compared to 78% for house S, but 55% of my winning bids were within two increments of my max, as compared to 19% at House S.

Let's move on to a non-Biddr model in House L. Note that the data here includes both "pre-bids" and live bids. I shifted to more live bids as I suspected the shill bidding, but if they do have such a system they're capable of adjusting the last second.

So, this does appear a bit more suspicious. I won 63% of my bids, so the auction is more competitive than the others, but a whopping 83% of my coins were within two increments.

Let's finally look at House R. This used to be among my favorites, but I've reduced my bids considerably due to expected shill bidding and because I found I could obtain most coins cheaper elsewhere.

This was easily the most competitive auction since I won only 49% of my bids. Of those, 78% were within two increments.

Here's a (stupid) chart comparing the four.

The real issue here is the lack of data. However, House L in particular seems odd. It does seem logical that the more competitive a house is, the more likely I'll pay closer to my max. However, House R is actually more competitive than House L, but has a lower "hit" rate.

Both House L and House R do also appear interesting because the lines slope upwards instead of downwards. Indeed, House A - who I suspected more than any other - actually appears defensible here. What is unknown is at what point should lines expect to slope upwards given increases in competitiveness. More data would be necessary to settle this.

Finally, let's take a look at how often either bidders or houses were able to "guess" my max bid.

Here, you can see why I was most worried about House A and House R. Roughly a third of the time they hit my exact max, which seems high - especially given the same size of House R. Interestingly, House L, whose line above was most anomalous, is roughly in line with House S here. The data may imply that, in the absence of the knowledge of one's max bid, roughly 10% of coins should go for one's max. At the very least, if there is prognostic shill bidding going on, different houses use different techniques.

Feel free to add your own data to mine. Also, note that I have no training in statistics, so feel free to use this data for a more thorough analysis.

My summary is I cannot for sure say that my data proves any house of "prognostic shill bidding", but they do qualify my own apprehensions to be very careful with these three houses.

 

[Kaleun96]

Appreciate you tabulating the data, even if as you say it may not be conclusive, it's refreshing to see some data in this discussion!

Auction House L - is it possible, assuming they are shillbidding, that they purposefully bid one increment below the max so as not to arouse suspicion? Including max bids and within 1 increment bids, that would put L at 63%, R at 59%, A at 44%, and S at 15%.

That being said, I feel auction houses L and R are perhaps the most competitive (less so for R these days) so that might go someway to explaining the pattern. Anyway, here's my data for "L". The ones I've won at my max I've felt are fair prices so for my data I don't think it's too suspicious.
At max: 4
1 Increment: 1
2 Increments: 4
>2 Increments: 1

[Ancient Coin Hunter]

Very interesting and thank you. I suppose that for those management teams seeking multiple valuation returns, e.g. 5x-6x of initial bids that the house starts with versus prices realized, that the temptation to engage in shill bidding, at least to move prices up 3-4 x the opening bid, is too great to avoid. We need more data to say anything conclusive but this is an excellent start. Certainly the web and online app technology has made it way easier to conduct shill bidding than in the days of mail order and phone bidding like the times up though 1995-1996. Let's see how generative A.I. impacts the auction houses and their ability to deal with individual bidder experiences!

[kirispupis]

39 minutes ago, Kaleun96 said:

Auction House L - is it possible, assuming they are shillbidding, that they purposefully bid one increment below the max so as not to arouse suspicion? Including max bids and within 1 increment bids, that would put L at 63%, R at 59%, A at 44%, and S at 15%.

I do believe that if an auction house engages in prognostic shill bidding (which is far worse IMHO than "standard" shill bidding), then this can most effectively be carried out in the software. It wouldn't be difficult to add some code to randomly go within 1-4 increments of a max bid so as to not arouse suspicion. Many of these bids in my experience occur in the last second which may suggest something automated.

That being said, the goal of tabulating my data was to improve my opinion on whether prognostic shill bidding is taking place, especially in auctions L and R, and I remain undecided. These are very competitive auctions, so the data can't provide any specific conclusions.

[Heliodromus]

If we're limiting discussion to fixed end time auctions, then I think it's useful to separate the cases where there was a large price gap between top two bidders (winner + a potentially suspicious, likely automated, underbidder) and rest of bidders, and other cases where there was at least a 3rd bidder not too far behind. There might be shenanigans in both cases, but it would seem more likely in case where there is no close 3rd bidder, and this is where I'd expect any signs of impropriety would be most obvious. In my case that is what I did with my "L" wins over last year or so - ignore the "close 3rd bidder" ones and focus on the rest, which as noted looked 100% (4/4) like bot bids.

biddr.com proxy bids are probably the gold standard for what a non-manipulated auction should look like, and how often one wins at close to max vs below. There might still be shill bidding in these (e.g. proxy bids submitted by auction house/consignor), but it should be unrelated to proxy bid maximums which are truly hidden. I suppose a comparison between that and all (not filtered by 3rd bidder) outcomes in a fixed end time auction would be interesting, ideally a range of fixed end time auctions on different private platforms.

 

Edited December 19, 2023 by Heliodromus