experimentation in virtual worlds
TRANSCRIPT
Experimentation in virtual
worlds
Contents
1.Some words on my work in AVEA
2.Virtual words as a source for natural
experiments
3.A quick example
My work on game theory
• What is game theory?
• You and me, we both have to make a
choice.
• What you choose affects my decision
and vice versa.
• I’ve been studying large games, i.e.
games where the number of agents is
large
Large games
• I’ve been working on finding theoretically
sound ways to tackle this problem and
testing the findings using data from virtual
worlds.
In general, the larger
the game, the harder it
is to analyze.
Large games
• This is rather technical stuff, so let’s do
something else today to demonstrate
how the data from VWs can be used in
social sciences.
• Some virtual world oriented examples
can be found from the final report.
Virtual worlds are rich of
natural experiments
• Natural experiments are situations
where naturally isolated changes in
some variables allow us to control our
“other things equal” assumptions.
• In VWs the environment changes often
and creates a rich source of possible
natural experiments.
The credit crunch and
natural experiments
Let’s take a look at one of
them:
• After the credit crunch,
many economists have
criticized the so called
efficient market hypothesis
(EMH).
The credit crunch and
natural experiments• EMH: Prices reflect all
available public information
on assets (Fama, 1970).
• Many have argued that the
collateral for the subprime
loans was highly overvalued
and risk premiums were
systematically too low prior
to 2007.A free hair cut with a
loan!
How to test the EMH?
• EMH implies that changing production
from one good to any other shouldn’t
yield any profits if one does not have
any private information.
• Testing the hypothesis above can be
hard:
• Changing production from good A to good B
can be very costly .
• How can we control private information?
But how about Virtual
Worlds?• Let’s see if we can test the EMH in
virtual worlds using data from EVE
Online, an online role-playing game set
in space!
• The following data and graphs are from two
sources:
1) CCP’s Quarterly Economic News Letter 3/2009
2) www.eve-markets.net – a website using huge
quantities of user submitted data to monitor market
trends.
The setting
We shall compare the factor prices of Bane
Torpedoes (BT) and Wrath Cruise
Missiles (WCM).
Both are ammunition for space warfare sold
in high quantities in the ingame market.
WCM BT
The setting for the
experiment
• On June 2009 CCP Launched a massive
campaign called The Unholy Rage
against real money traders.
• Approx. 2% of player accounts were
banned.
• Nice source of exogenous
variation, since Wrath Cruise Missiles
are popular among farmers
An example of a natural
experiment
• On July alone the price of a WCM
dropped slightly less than 20% (from
little over 150k isk to about 125k isk).
• During the same time the price of a BT
dropped about 9% (from 320k to 290k).
• Is this in line with the EMH?
An example of a natural
experiment
• Both products take the same time to
produce, require the same skills and
their inputs are readily available in the
market.
• Thus changing from producing one to
another shouldn’t cost the producer
anything.
An example of a natural
experiment
• Economic theory predicts that the price
of a product should be the sum of the
prices of its factor inputs.
• =>The price of a torpedo should be the
cost of the materials used plus some
compensation for the skills, time and
effort the player has put into producing
it.
An example of a natural
experiment• As the skill
requirements, building time
and other manufacturing
requirements for both BTs
and WCMs are equal, the
EMH should imply that any
difference in their price
should be completely due
to differences in material
costs.15.6.2010: See also the last slide of this
presentation
The timeline
• Unholy rage on July.
• On both May and August the price
development of both items was again
relatively stable – an indication of
markets clearing.
• Were the markets able to correctly
adjust to Unholy rage?
Calculations for the two
warheads
Wrath Cruise Missile Bane Torpedo
May
Unit price 160 isk
Materials 122 isk
Mark-up 38 isk
August
Unit price 130 isk
Materials 108 isk
Mark-up 22 isk
May
Unit price 340 isk
Materials 302 isk
Mark-up 37 isk
August
Unit price 290 isk
Materials 273 isk
Mark-up 17 isk
Results
• In May the difference in mark-ups is
tolerable (2,5%)
• But still after one month from the
sudden drop in Wrath Cruise Missile
prices Bane Torpedoes earned a mark-
up that is over 22% smaller compared to
the WMC mark-up.
Results
• We also calculated that compared to
building Bane Torpedoes, disassembling
them and turning them into Wrath
Cruise Missiles would have paid more in
both periods.
• However, arbitrage is not possible: the
latter still pays less than building Wrath
Cruise Missiles straight from raw
materials.
Summa summarum
• The markets at EVE do not seem to
react efficiently to second order shocks
(i.e. the way how the price of WCMs
affects the price of BTs).
• Does this generalize to real markets?
Hard to say.
• The opposite result would have
generalized better.
What else can be done with
natural experiments?
• There is no ready-made formula. You
just have to keep your eyes open.
• For business purposes, estimating the
demand curve for some product is one
classic example.
• Another possibly profitable application
how good substitutes or complements
two goods are
What else can be done with
natural experiments?
• Testing scientific models and
estimating model parameters that are
otherwise hard to pin down.
• Especially instrumental variable
regression has become popular in
economics.
• A good first read: Angrist & Krueger
(2001)
Relevant literature
On large games:
Al-Najjar, N. 2008: Large games and the law of large
numbers. Games Econ. Behav., 1-34.
Tolvanen, J. 2010: Approximating Competitive Games with a
Large Number of Players. Master’s thesis, University of
Helsinki.
Tolvanen, J. & Soultanis, E. 2010: Corrigendum and some
further notes on “Large games and the law of large
numbers” [Games Econ. Behav. 64 (2008), 1-34], Games
Econ. Behav., in review.
Relevant literature
Efficient market hypothesis:
Fama, E. 1970: Efficient Capital Markets: A Review of Theory
and Empirical Work. Journal of Finance 2, 383-417.
Natural experiments:
Angrist, J. & Krueger, A. 2001: Instrumental Variables and
the Search for Identification: From Supply and Demand
to Natural Experiments. Journal of Economic
Perspectives 4, 69-85.
A Correction
• After my talk I got an insightful comment from a
member of the audience regarding my example:
Assuming that the manufacturers maximize the money
earned per unit of time and that they are credit
constrained, then the fact that manufacturing a BT
costs more than 2 times the cost of a WCM should imply
that the markup on a BT should be more than 2 times
the markup on a WCM, since producing one BT ties 2
times the amount of capital compared to a WCM. This
of course turns the setting up side down – the
manufacturers are still using public information
inefficiently but now they are producing too many
WCMs which is opposite to the previous case where they
were producing too few.