economics
NAME REMOVED Econ 4400 Final Project
1
The Effects of Market Size on Average Price in World of Warcraft
Introduction
The World of Warcraft Auction House is a huge, active marketplace where players can buy and
sell a multitude of items. Because of the structure of the Auction House, and its freedom from
intervention (mostly), its economy is ripe for study. I say “Auction House”, however it may be more
appropriate to say “Auction Houses”. There are hundreds of different houses within World of Warcraft,
each only accessible by their respective populations, and faction. This division results in a large variation
of market size across houses. However, other than size, and server type, these houses and the items within
them are essentially the same. This structure of the Auction House offers an interesting opportunity to
explore the relationship between market size and item price.
Data
The data used in this analysis was gathered from a public database1 containing historical data
collected from The Undermine Journal.2 The Undermine Journal is a website that collects realm specific
auction house data through an in-game modification add-on used by players.3 This add-on scans and
collects information from the auction house such as quantity available and market price.
My sample contains 79,993 observations. These include observations for 8 items, across 120
United States servers, for the 13-week period from April 9, 2017 to July 1, 2017. These observations
include information about average daily price, average daily quantity, and population.
In World of Warcraft, there can be thousands of different items available for sale on the auction
house at any given time. To make the data in this analysis more manageable, I collected data for only a
limited number of items (see Table 2 below). In selecting this data, I first identified categories of items
which should always be available and in demand. These categories include: raw materials, item
modification, and consumables. I then randomly selected a few specific items from each category.
Table 1: Selected Items
Footnotes:
1. The existence of this database was made known to me by my classmate Samuel Adams. 2. https://newswire.theunderminejournal.com/phpMyAdmin/sql.php?db=newsstand&table=tblDBCI
tem&token=19fc32bd8b332cbce9bc012ae9bf4ee1&pos=0
3. https://theunderminejournal.com/
Category ID Name
Raw Materials
124105 Starlight Rose
123918 Leystone Ore
124113 Stonehide Leather
124437 Shal'dorei Silk
Consumables 127848 Flask of the Seventh Demon
133579 Lavish Suramar Feast
Gear
Modification
128550 Enchant Cloak - Binding of Intellect
130220 Quick Dawnlight
NAME REMOVED Econ 4400 Final Project
2
There are some notable weaknesses in the house population data for this sample. Because of the
large quantity of playable realms within World of Warcraft (well over 100), many contain a very small
population. While such small servers may have been interesting for my purposes, it is less than desirable
for the players of those realms. To combat issues arising from this, Blizzard combined many small and
large population realms into cross-server houses. These houses participate in a collective auction house
market. Thus, the populations reported in this sample are the sum of each realm within these houses,
rather than individual realm data. Even so, there remains a wide range of populations for these houses
(see Table 2 below).
Table 2: Population Statistics n Mean Std. err. Min Max
Population 120 300,016.26 12,194.695 77,379 838,176
There are also some difficulties with the price data in this sample. The values in this sample are
the average of all items in the auction house, not just those sold. This means that extremely overpriced
listings may be skewing some averages higher than is preferable. In addition, though I attempted to limit
items to only those which are readily available and in demand, some observations may see only one or
two listings for an item. It is easy to see how, in such situations, a seller may list an item that is
ridiculously overpriced in an attempt to tempt a foolish (or desperate) buyer. Such a situation seems to be
the case when considering the average price statistics in Table 3 (below). For example, consider the
statistics for the Cloak Enchant (item no. 128550). This maximum price (over 9,00,000) is simply absurd
and evidence for the existence of such complications.
Table 3: Average Price Statistics for Full Data Set
ITEM # Name n Mean Std. err. Min Max
123918 Leystone Ore 10016 1,268.48 7.8589 325 53,823
124105 Starlight Rose 10016 3,703.11 15.7759 866 27,562
124113 Stonehide Leather 10016 679.32 7.6754 162 44,849
124437 Shal’dorei Silk 10016 472.61 5.5551 84 34,376
127848 Flask of the Seventh Demon 10016 54,134.92 173.6969 19,421 195,238
128550 Enchant Cloak-Binding of Intellect 10016 333,069.59 1,751.1601 72,688 9,209,492
130220 Quick Dawnlight 10016 232,164.69 826.4470 52,336 872,315
133579 Lavish Suramar Feast 9881 146,551.81 568.0246 9,900 1,889,551
In an attempt to save the data from such absurdities, I cut the highest and lowest 1% of values for
each item. As shown in Table 4, this alteration had a great effect on the average price statistics. While
individual average values may still contain extreme outliers, the worst of the offending data has hopefully
been removed. In addition, this alteration has reduced my number of observations to 78,366.
NAME REMOVED Econ 4400 Final Project
3
Table 4: Average Price Statistics for Cut Data Set
ITEM # Name n Mean Std. err. Min Max
123918 Leystone Ore 9812 1,240.72 3.7618 580 2,639
124105 Starlight Rose 9812 3,653.16 13.6434 1,413 8,737
124113 Stonehide Leather 9812 641.79 2.8410 248 2,141
124437 Shal’dorei Silk 9812 452.62 1.6584 164 1,158
127848 Flask of the Seventh Demon 9812 53,753.77 158.6288 25,173 10,4623
128550 Enchant Cloak- Binding of Intellect 9812 328,257.91 1,378.2956 115,202 83,3821
130220 Quick Dawnlight 9812 230,496.89 759.3169 88,266 48,3755
133579 Lavish Suramar Feast 9683 144,761.88 467.0559 58,011 331,158
Identification Strategy
I’ve included additional variables to control for potential omitted variable bias, and better identify
the causal effect of interest. Average quantity could be related to average price and population. As
populations increase, more sellers will enter the house, increasing average quantity available, and driving
prices down. The purpose of this variable is to identify changes in average price due solely to the demand
side of the marketplace. That is, if supply is held constant, what sort of effect would an increase in
population (demand) have on average prices. I have also included a variable for “House” type. In general,
servers are divided into player v player (PvP) and player v environment (PvE) servers. Other than the
items which fall into the Raw Materials category, each item in my list is more suitable for end-game PvE
content. I think it is reasonable to assume that, due to the extra fun of slaughtering player enemies, more
players will select a PvP server, thus reducing populations in the PvE servers. In addition, the players
within a PvE server will be purchasing more PvE-related items (like those in my sample).
There are more potential omitted variables which I am unable control for. For example, this data
doesn’t take into account large, raiding guilds. Some players will choose servers specifically for the end-
game progression community, which may influence population sizes. In addition, if the servers within
these houses contain large guilds heavily involved in end-game progression, the demand for the items
within my sample will increase substantially, which may influence average prices. Although these
omitted variables may be influencing my data, I feel confident that a good estimated effect can be
identified.
Results
I ran 5 regressions in my analysis of house population on average prices (Table 5). The
dependent variable in all 5 regressions is the natural log of the average auction house prices. The
independent variables are: item, natural log of average quantity, natural log of house population,
and house type.
NAME REMOVED Econ 4400 Final Project
4
The first regression included only natural log of house population as an independent
variable. The results indicate that each 1% increase in house population is related to a 0.1934%
decrease in average auction house prices. These results are statistically significant. However, with the R2
value less than .001, more can be done to explain variation in the data.
The second regression adds variables to control for the item. Including these indicator variables
increases the R2 value to 0.9836. This drastic increase in explained variation makes sense when
considering the vast differences in average price for different items. By including this control,
the coefficient on population only measures the variation within each item. This regression also
changes the coefficient on house population to indicate a .2115% decrease in in average prices
with each 1% increase in population (this is significant).
The remaining regressions are the result of my attempts to control for potential omitted
variable bias. For the third regression, I added a variable to control for variation due to average
quantity. This result accounts for only slightly more variation in the data, but is statistically
significant. According to this regression, a 1% increase in average quantity is related to a
0.0691% decrease in average prices. Such an effect is small, yet statistically significant and
surprising. This regression holds the supply constant, therefore an increase in population should
indicate an increase in demand. However, in this regression, it appears that this is not the case.
The fourth regression includes a dummy variable for the type of servers that make up
each house (player vs player [PVP] servers =1, player versus environment [PVE] servers =0).
While the regression only slightly increases the R2 value, it is statistically significant and
influences the coefficients of the other variables. If the house is composed of player versus
player realms, one can expect to see a 2.4% reduction in average auction house price (all else
equal). In addition, coefficient on our treatment is reduced to .106% reduction in average prices
with each 1% increase in house population.
The fifth, and final, regression in this analysis does not include the control for average
quantity. I chose to run this regression for two reasons. First, as discovered in regression 4, when
holding average quantity constant, an increase in population leads to a decrease in average
prices. By relaxing the control on average quantity, the regression indicates that average prices
decrease by twice as much when not controlling for quantity. Therefore, while the increase in
population doesn’t lead to an increase in average prices as one would expect, it does stymie the
fall in prices due to increased supply. Along these same lines, lifting the control on average
quantity allows us to consider both supply and demand with changing populations, and may be
more useful in some situations. The result, is that when not controlling for quantity, each 1%
increase in population size relates to a .205% decrease in auction house prices. This coefficient
now contains the effect of the changing supply and demand sides of the market. The larger
reason, is suspected multicollinearity. Table 6 shows a regression of population on average
quantity (controlling for item). This regression indicates a possible multicollinearity issue with
population, item and quantity (VIF=10). This issue may be the cause of the negative relationship
between population and average price when average quantity is held constant.
NAME REMOVED Econ 4400 Final Project
5
Table 5: Regressions
Dependent Variable: Natural Log of Auction House Prices
(1) (2) (3) (4) (5)
Intercept 11.8277
(.3126)
10.7905
(.0402)
10.2053
(.0414)
10.1247
(.0422)
10.7334
(.0409)
Leystone Ore - -1.06 (.0048) -1.0815
(.0047)
-1.0817
(.0048)
-1.06
(.0048)
Starlight Rose - 0 (base) 0 (base) 0 (base) 0 (base)
Stonehide Leather - -1.7552
(.0048)
-1.792
(.0048)
-1.7923
(.0048)
-1.7552
(.0048)
Shal'dorei Silk - -2.0867
(.0048)
-2.1220
(.0048)
-2.1225
(.0048)
-2.0867
(.0048)
Flask of the Seventh - 2.7103 (.0048) 2.4578
(.0071)
2.4553
(.0071)
2.7103
(.0048)
Enchant Cloak - 4.4789 (.0048) 4.0341
(.0104)
4.0299
(.0104)
4.4789
(.0048)
Quick Dawnlight - 4.1557 (.0048) 3.7901
(.0095)
3.7563
(.0095)
4.1557
(.0048)
Lavish Feast - 3.6968 (.0048) 3.347
(.0087)
3.3434
(.0087)
3.6966
(.0048)
Average Quantity
(LN)
- - -.0691
(.0014)
-.0699
(.0014)
-
House Population
(LN)
-0.1934
(.0249)
-.2115 (.0032) -.1137
(.0037)
-.1061
(.0038)
-.2064
(.0033)
PVP Server - - - -.0243
(.0025)
-.0185
(.0026)
R2 0.000768 0.9836174 0.98409 .984109 .983628
R2 Adjusted 0.000756 0.983616 0.984075 .984107 .983627
NAME REMOVED Econ 4400 Final Project
6
Table 6: Regression Testing for Multicollinearity
Intercept -8.455
(.0985)
Leystone
Ore
-.3112
(.0118)
Starlight
Rose
0 (base)
Stonehide
Leather
-.5306
(.0118)
Shal'dorei
Silk
-.5118
(.0118)
Flask of
the
Seventh
-3.6487
(.0118)
Enchant
Cloak
-6.4274
(.0118)
Quick
Dawnlight
-5.7156
(.0118)
Lavish
Feast
-5.0548
(.0118)
House
Population
(LN)
1.413
(.0078)
R2 .908813
R2
Adjusted
.908803
Conclusion
With effects measured in fractions of percentages, it can be difficult to determine if such
effects would be noticeable for players within the game. However, when placed in context, it can
be seen that these effects are not at all trivial. Consider two players. Player Low is part of a low
population house with 287,821.565 players (one standard deviation below the mean). Player
High participates in a high population house with 312,240.955 players (one standard deviation
above the mean). The change in population from Player Low to Player High is a percentage
increase of 8.13%. Because each 1% increase in House population leads to a .2064 % decrease in
average prices (not including a control for quantity), the average prices can be expected to
NAME REMOVED Econ 4400 Final Project
7
decrease by 1.678%. This effect may be small if these players are selling Leystone Ore, where
average prices will decrease by 21.29 gold (all else equal). However, if selling a more valuable
item like the cloak enchant, the decrease in average prices is expected to be 5,558.91 gold.
This analysis shows that increased house population within World of Warcraft has a
negative effect on average prices when controlling for item and server type. When controlling for
average quantity, the prices are still expected to decrease with increasing population, however,
this may be due to a multicollinearity issue between item, population, and quantity. The results
of this analysis indicate that as markets increase in size, prices can be expected to decrease.