WoW Econometrics and Data Collection Methodology
Author: Roboticus
One of the promising things about analyzing online gaming economies is that the data are abundant and already in digital format. Extracting the relevant data, however, has been much more difficult than I first expected. But good news! Over the last few weeks, I’ve cobbled together an approach that captures the entire set of all auctions extant on a given server’s Auction House. These data include additional information, such as seller’s identity, time available for bids, number of identical goods up for auction, buyout price, and current highest bid.
The approach for extracting these data can also be repeated over set intervals to construct pooled data (combines cross-sectional data with time-series data).
Once pooled, the data set could be analyzed for auction results, tracking what was purchased, how bidding wars developed, and even to construct a picture of how the economy shifts in real time with exogenous factors (such as changing game mechanics or content updates).
Additionally, the process for pooling these data could be quickly replicated on a multitude of server, creating the rather interesting prospect of analyzing parallel economies. While our real world is set in its singular existence, and therefore a system-wide anomaly or macro outlier or war can throw off an otherwise balanced analysis, the gaming world has several hundred separate servers, each with their own economy, but following an identical set of rules. Presumably, the surfeit of economies would allow cross-sectional studies that could account for economy-wide outliers.
The Data Extraction Method
The World of Warcraft is designed to allow third-party software authors to create add-ons that work as separate modules within the game (like an excel plugin, for instance). Software authors have access to Blizzard’s API via published LUA interfaces. For the purposes of creating my data sets, I have used the add-on Auctioneer, a tool that quickly scans all auctions available on the Auction House. While the add-on is intended for in-game users to quickly post their auctions at appropriate market prices and to know when auctions are undervalued (and therefore, offer potential profit taking), it also creates an exhaustive data set that, with concerted effort, can be exported and manipulated for the purposes of econometric analysis.
The data created by Auctioneer are stored as a text file which is rewritten with new information from each scan. By running a scan and then moving the data file out of the Auctioneer folder before rerunning the scan, it is possible to create a series of market-scans that can be pooled into a 3-dimensional time-series and cross-sectional data set. Once the scans are complete, however, there is still the onerous task of transforming the data out of the obscure and poorly documented data structure used by Auctioneer into something resembling a more traditional data set. An additional problem is that the data set is extremely large and unwieldy with traditional computing power. To address both of these problems, I created a custom online PHP program that runs a series of transformations to de-serialize the data and morph it into rows and columns and then output it as a file readable by Excel.
My next step will be to develop this procedure into a desktop application with the hopes that I won’t cause my web host to block my site due to excessive data processing. Additionally, creating a desktop application will bypass some of the processor-intensive steps that have a tendency to heat up my small apartment.
The final step, of course, is to analyze these data with traditional econometric programs such as GRETL. Check back for my results as they start to come out.
Tags: Auction House, Auctioneer, data collection, Econometrics
November 17th, 2008 at 8:32 am
As a designer working in the industry (not blizzard mind you) I’m interested in seeing what data you manage to pull – while blizzard is watching the economy like a hawk, those lessons learned regarding balancing and player interactions are, as always, held hostage in the vault.
In any case, this is already turning into an intriguing look at one particular aspect of a large virtual economy. seeing these trends as they’ll fluctuate and trying to make some sense of what happens will certainly be an interesting process.
Thanks for all your hard work so far,
Martin
November 17th, 2008 at 10:15 am
Good to hear you’re finding it interesting, Martin. Due to the expansion just hitting, I have been spending plenty of time in the game rather than writing, but I have been recording Auction House scans every night to start to get a time-series data set going. Hopefully I’ll have a post up this Tuesday.
December 9th, 2008 at 4:03 am
Hi Roboticus
that is interesting design! actually now I am also working on the modelling of MMORPGs. Auction house activity is one of the variables in my econometric model. I am very interested in your research. so have you published any academic article so far? maybe we can talk more over it via email. you can contact me at shicong8332@hotmail.com if you are keen on exchanging some ideas:)
cheers
Kristy
December 9th, 2008 at 11:27 am
Hi Kristy, thanks for your response! I have not published any academic articles yet, but I’m at the point where I have enough data and tools to start looking at what I want to publish. I’d be very curious to hear more about what you’re working on. It sounds like your model is more broad than just the Auction House? How are you measuring these other factors (and, i suppose, what are they)? I sent you an email with my contact info, but feel free to post here, as I’m sure others would be interested in what you’re doing.
December 9th, 2008 at 11:05 pm
Hi, Justin
nice to hear from you. I employed Cobb-Douglas function to evaluate the input and output in game on individual perspective. I don’t know whether you are interested in this technical perspective… if you are, we can exchange more:)
btw, I am at Nanyang Technological University. where are you?
best,
Kristy