Market Value Vs Catalogue Value

To clarify, you can't really reverse engineer the machine learning, once the data sets approaches any reasonable size because of the shear exponential possibilities.
The system may know that you are from Female, from NY, age 42, bought stamps on Wednesdays more so than any other day of the week over a 3 year period, prefer keywords with SCV, and spend 9 average seconds at most inside auction lots reading listings you have no intention of buying.

I doubt that ebay gives a hoot about anybody's profile. Listings are the way they are based upon sellers sales or desire to pay up to gain more visiblity.

E-bay does use artificial intellegence for sure. Devin Wenig, the CEO goes on and on about it all the time. That and structured data. I agree that there is no way to reverse engineer the algorithm. I think you can figure out what some of the factors are through experimentation, but I don't think anyone will ever know it all.

The frustrating thing and the unethical thing is that E-bay misleads you into thinking that you are getting full exposure to a global market for the duration of your listing, even though you are not. My analysis of my traffic data revealed that most of my listings receive only 200-300 impressions a month, and of those, the click-through rate is 0.3%. 200 impressions is nothing. Basically over 30 days that means that it appears in search results 8 times.

Obviously in bookmarked searches it will still appear because the person has already seen it, to have bookmarked the search. But if you search for something in E-bay you have no real way of knowing if you are truly seeing everything that is there. I know this myself because I used to search all the Canada listings by newest listed, and I went through them, marking where I left off. Almost every time a listing I had never seen before would appear before the spot that I left off at, even though I had already been over the full section. If it is truly sorting by listing date that shouldn't happen, unless they are not showing me everything the first time.

I didn't finish my thought: 200 impressions means 7-8 times a day.

I am a systems architect, and there are clearly optimizations around response times for data pulls. For example a common action a user on ebay might perform is pick the catgory unused classic US stamps, and sort by date. When something like this is performed by multiple users, the data behind it can be server side cached with varying TTLs of the response data. If a listing was updated in between a TTL you can get varying results. I think there is less of a chance of nefarious behavior going on here, versus building a site that can handle millions of search queries without any perceivable slowness. If this ultimately means that some listings can appear missing, it is deemed acceptable by a site delivering millions of responses as the _intent_ for the request is to give you a search response a second after you click the button, not necessarily the unabridged data set. Call it a best effort fast search. This is why you may get even more responses if you use a very specific search queries.

Second the databases that have the listings are sharded (partitioned) also around performance goals. The querying engines may solicit the shards in a load balanced manner, in which cases the complete data sets are not always realized all times.

The last interesting piece here is ebay uses pagination techniques to deliver search results to provide "endless scrolling". Each subsequent load might be off a cached shard query or subsequent homerun to the data.

The point I am trying to make, is that general searching and subsequent results should not be considered a deterministic operation in the context of any public web services.

English please.

Data that the ebay system thinks it will need again in the near future is cached, or stored on the server. If a subsequent request needs that same data, it can be read from the cache instead of re-calculated, which saves time for the user and CPU cycles for ebay.

Re-sorting your search results by date or price, or even something as simple as going to the next or the previous page of the results, may draw its data from a cache instead of running an entirely new search query on the database.

The database, incidentally, is not stored on only one computer. It's stored in toto on computers in data centers all around the world. Keeping many copies of the database is necessary, just to serve the huge number of requests that ebay gets. That means there is also the problem of keeping data in sync across all those machines around the world. Almost every production-grade system will guarantee "eventual" consistentcy across machines some high percentage of the time, but the amount of time "eventual" takes may vary, and as a result, the database that is queried from one request to the next may actually have different contents (since it isn't fully synced with the others).

your mileage may vary on search results due to purposeful computer mumbo jumbo, and not evil intentions at ebay.

I don't doubt that what you are saying is correct. But all I can tell you is that prior to the replacement of Voyager search engine with the new Cassini search engine sellers did not have this happen to them. If you go to sellers forums and you read the comments, many many sellers from all different product categories have experienced year over year declines in sales of as much as 80%. This has been happening the entire time since Cassini was introduced and it's getting worse and worse.

So while your explanation may account for some of the instances I am talking about, it does not mean that E-bay is not manipulating search for their own benefit. There is simply too much evidence that points to this, from the sales declines across multiple product categories to the fact that many sellers seem to have a disproportionate number of customers from random geographical areas to the fact that on a regular basis, a multi-quantity listing that has had no sales at all for years, has 1 sale and then sells out within a week. It's just too weird and it happens too regularly to be sheer coincidence. Then there is the throttling that occurs on a regular basis where your sales are kept to within a narrow $ range every month, no matter how much you list or how you lower your prices. If you have a really great day you can count on the next 2-3 days to be dead, followed by another good day. If you've dropped anchor and have had a horrible month and your fees are about to become due and you don't have the money to pay them, then all of the sudden out of nowhere an item you have has listed for a long time will suddenly sell. This has happened to me at least 5 or 6 times. But the daily sales feast or famine pattern is so regular you can set your watch to it.

Wow, I was going to get something done, but so much has been written that I feel the need to comment on.

Starting with ikey:

His posting at the bottom of page 4 about dealers, about penny starts and about declining prices is dead on.

As to his question about consignments, they are less than half of what you see in stamps, but not insignificant,
however, there is no fixed model of how a consignment is handled - it all depends on what the consignor and consignee agree on.

mootermutt

"MV's have always reacted more quickly to economic conditions than CV's"

Of course catalogs react slower than the market - catalogs are supposed to reflect market changes not lead them (plus they have months long production lead times). In the case of Scott, they do not review prices for every country every year, which causes an even greater lag.

rismony:

"If I was a dealer particularly in this segment, I would be underbidding every single item up to 50% cat regardless."

You would go broke fast trying that.

This whole ebay thing is way off topic, but I will say, in general you should expect for the algorithm to trend results to where the clicks go and also base results on specific secretive sauce but gameable criteria then circle back around over time on deeper searches adding a more complete listing diversity. Price lowering is not necessarily the catalyst to make your listing pop, but rather a culmination of user and bidding activities across the spectrum.

I think this is why you see certain sellers push the limits on low quality listings at a penny. At the other end you have high end stamps that may sit for years.

In the end ebay wants the transaction to happen, that much you have to believe.

I have to think when your ebay fees are due, a sale is not automatically generated, but rather a buyer pulled the trigger for their own reasons. To suggest that a buyer pops out of the woodwork so you can pay your fees is nuts. Even if your listing pops to the top result, I have a hard time believing stamp buyers say, let me just get that without even a bit of further scrolling/searching.

"Even if your listing pops to the top result, I have a hard time believing stamp buyers say, let me just get that without even a bit of further scrolling/searching".

That totally depends on what the item in question is. I have certainly bought things as soon as I saw them and I suspect that many others have as well.

I know that I've bought items as soon as I saw them. Numerous times.