Brainstorm about Liquidity: how could we measure it?

[Tawnos]

Preamble: Liquidity is a measure of how easy it is for you to turn your commodity into gold. A highly liquid item is something that has a high sales volume; for instance, I'd say Heavy Savage Leather and heartblossom have a high 'indices' of liquidity. Items with low liquidity would be things like 377 crafted pvp armor or Darkmoon cards. They move, but they move very slowly, taking almost two weeks' worth of auctions to get a buyer (these specific examples may be a local problem on my medium population server - Drenden, but you get the idea).

The obvious way to measure this would be to track the volume of sales, but my understanding of the data available to us is that it's impossible to differentiate between when a good sells, gets canceled & re-listed, gets canceled outright, or expires. Never mind all the post-and-sell events that never even show up in the resolution of TUJ!

It'd be pretty amazing to be able to see, at a glance, the likelihood of a sale of any given item. We would know what to obtain and sell to make fast gold, assuming a positive profit margin. But i don't see any obvious way of getting around the fog of war imposed by the current data limitations. Any ideas on how we might go about quantifying such a thing?

The other question would be, if we DID index the change in number of auctions, how reliable would it be as a measure of volume? Like meaningless? Or 30% accurate?

[calianna]

Your understanding is essentially correct. TUJ basically looks at the AH every 40-50 minutes. It can see about how much time is left on an auction and see where it is 40-50 minutes later. Lots of things can happen though that look like a "sale". If the auction has 12 hours left and is gone on the next scan, did it sell? Did it get cancelled? Tough to know. Suppose an auction has 30 minutes left. It's obviously going to be gone at the next scan. It's assumed it expired. But maybe it sold. Suppose you list some red gems just after a scan has happened. All of them sell before the next scan. TUJ never even knows they existed.

Ultimately, you can't make any judgments about the "sell" data. It is going to be entirely dependent on the market. Glyphs on some servers have wicked cancel/repost cycles. Any data there would be worthless. Something like ore with high volumes but very little undercutting might be extremely accurate. Then again, it might be missing a lot of data that gets listed and sold between scans. So, best case you could make a reasonable guess to the accuracy of a single item's data, but nothing consistent across multiple servers or multiple items on a server.

[Gear]

Preamble: Liquidity is a measure of how easy it is for you to turn your commodity into gold. A highly liquid item is something that has a high sales volume; for instance, I'd say Heavy Savage Leather and heartblossom have a high 'indices' of liquidity. Items with low liquidity would be things like 377 crafted pvp armor or Darkmoon cards. They move, but they move very slowly, taking almost two weeks' worth of auctions to get a buyer (these specific examples may be a local problem on my medium population server - Drenden, but you get the idea).?

Liquidity does not just account for the ease of converting an item to gold but also the quickness in which the coverson occurs. So yes, single post 12 hr auctions with a guaranteed sell would be very liquid. While yor later examples would be illiquid

The obvious way to measure this would be to track the volume of sales, but my understanding of the data available to us is that it's impossible to differentiate between when a good sells, gets canceled & re-listed, gets canceled outright, or expires. Never mind all the post-and-sell events that never even show up in the resolution of TUJ!?

This statement is true but you could utilize MySales to determine your specific data. If you have a trustworth group of individuals you could also import their data into your data pool. Most of the TUJ data is not relevant to you anyway.

It'd be pretty amazing to be able to see, at a glance, the likelihood of a sale of any given item. We would know what to obtain and sell to make fast gold, assuming a positive profit margin. But i don't see any obvious way of getting around the fog of war imposed by the current data limitations. Any ideas on how we might go about quantifying such a thing??

See my very rudimentary solution above,

The other question would be, if we DID index the change in number of auctions, how reliable would it be as a measure of volume? Like meaningless? Or 30% accurate?

The data would be out of date as soon as you finished processing it. The final numbers would fall on the reliability scale somewhere between Lada and Yugo.

Sorry Mate,

Gear

[Annu]

Since I gathered so much information at this place, I will try to contribute with my thoughts here.

According to Calianna the main problem would be to figure out, which auctions were canceled or sold, right? Let's have a look at the possibilities why an Auction disappeared.

- Cancel/Repost situation: Here, we can assume that the person canceling the auctions will repost them very quickly. In this case, the auctions would reappear in the next scan at another price. Hence, you could simple count the number of auctions from a single user of a specific item and check if the same number appeared or increased (refill what's sold). There is a certain probability that the next TUJ scan is after canceling and before reposting, but since is mainly performed by Addon-Users we can assume the timespan to repost to be a few minutes which decreases this probability to a certain degree.

- Cancel without repost: This typically happens if the auction drops below the repost threshold or below the point the non-addon user considers it worth posting. Hence, if you count these items as sells, a filter for a minimum profit span should eliminate them.

- Expired: I guess this should be the easiest case, as it can be filtered by dropping items below a certain timespan.

- Quickly reposted after sell: You would miss this one as a repost, so this would decrease your accuracy. However, a market with a high reposting frequency can be considered to be overloaded by goblins and not be the one you are actually searching for?

- Sold before the next TUJ scan: No way to detect, but if the systems works right, it should still detect a high sell rate as this will not happen to all auctions. Even at extremely high frequency, there will still be auctions that are put on shortly before the TUJ scan and show as sold on the next one.

To me the biggest flaw would be the repost during TUJ scan.

So let's assume that it takes 5 minute to repost your items (I would assume, this is actually quite long). Then there is a 5/50 or 10% chance that a repost of an item was during the TUJ scan. So the biggest problem would be if this is already too much uncertainty for such a system to work. I personally would assume that only a very low number of auctionators repost every hour. Thus, the likelihood of such an event is not too high and therefore the number of these events low enough, but I might be wrong.

You could decrease this uncertainty also by using only every n-th scan, but then you would also miss more auctions that are posted and sold in-between scans.

[Kathroman]

Looking at the huge discrepancy between even TUJ and Wowuction's sold data would to me, suggest that this is a situation that isn't going to be remedied anytime soon, based on available data. Unless Blizzard provides "better" data, or data more frequently, the only real solution is to find a way to account for sold data without needing it to be 100% accurate.

[Erorus]

Preamble: Liquidity is a measure of how easy it is for you to turn your commodity into gold. A highly liquid item is something that has a high sales volume; for instance, I'd say Heavy Savage Leather and heartblossom have a high 'indices' of liquidity. Items with low liquidity would be things like 377 crafted pvp armor or Darkmoon cards. They move, but they move very slowly, taking almost two weeks' worth of auctions to get a buyer (these specific examples may be a local problem on my medium population server - Drenden, but you get the idea).

The word you're looking for is "demand." Demand is trivial to estimate by any moderately-experienced AH player. There's only so many categories of items. People tend to want things from the current expansion. People tend not to spend lots of money on older items, except if they're collecting rare items. Crafting components tend to sell well, and in bulk. There's lots of general rules we can state about how the WoW market works. Perhaps someone should start a Goblin Rules of Acquisition?

The obvious way to measure this would be to track the volume of sales, but my understanding of the data available to us is that it's impossible to differentiate between when a good sells, gets canceled & re-listed, gets canceled outright, or expires. Never mind all the post-and-sell events that never even show up in the resolution of TUJ!

You can forget sales data, I've pretty much given up on it. There's just so much that falls through the hourly cracks that we'll never know. TUJ works because it's a good estimate of what you could find if you searched the AH for most things. You look for Heartblossom, and right now it's 4.5g each. Wait 10 minutes, and it might be 4g, or it might be 5g. But the cheapest stuff tends to last the shortest, and the most expensive stuff sits the longest, so the bottom 15% that we see is a good average estimate of the price at that time.

Once you try to go beyond the hourly resolution of available items, all you get are gaps in the data that you can't explain. You don't know what was sold. You can guess, but you don't truly know. You don't know how much was sold. Maybe nobody posted any Heartblossom this hour. Or maybe a farmer posted 30 stacks cheap and some scribe bought them all 10 minutes later. You have no idea, and that transaction wouldn't be tracked but could affect the prices of glyphs.

If you're stubborn and want to derive some sold data, here's what you do: Check out Blizzard's API. Download your realm's snapshot. Wait an hour, and download the next snapshot. Look at those two files. Tell me which auctions from the first snapshot were sold. Because that's all I get.

[calianna]

How often can an individual hit the Blizzard API, out of curiousity?

[Kathroman]

How often can an individual hit the Blizzard API, out of curiousity?

Pulling a single realm? Probably never. Even with 2 or 3 I think you'd still be fine. More than that, and I'm not sure what you're trying to do anymore

[Erorus]

How often can an individual hit the Blizzard API, out of curiousity?

Applications may make up to a total of 10,000 unauthenticated requests per day.
Applications may make up to a total of 50,000 authenticated requests per day.

TUJ is authenticated and makes about 15k requests per day on US and EU each. The actual json data file isn't counted since it's technically "static" but the API call to get the file's path is counted.

The "cost" of requests is fairly simple. Each API request costs 1 to 5 points, depending on how heavy the operation is and load. Requests that use cache headers cost less and those that use cache headers that result in a cache-hit cost only 1 point. Requests for static files, like for the auction house dumps, have a cost of 0 because from our point of view they are not part of the API.

[Stede]

Total volume is all you have. You could track materials prices, sellers, post volume per seller, apparent post frequency, number of sellers, and seller profiles to model / compensate for competition, but ultimately it's gonna be a lot of work with little value added.