Monday, March 26, 2007

Virtual Economies, Virtual Experiments: Present and Future

Introduction

In the quest for expansion and profit, nothing can be so promising as an unexplored territory. It must come as a great disappointment to explorers that the last Earthly frontiers began to be exploited well over a century ago, and humanity still does not have the ability to properly extract resources from beyond the planet. Still, the meeting point between innovation and entertainment has provided a fresh avenue for wealth creation – and along with it, many questions of academic interest.

In particular, this intersection is the expansive worlds known as MMORPGs (Massively Multiplayer Online Role-Playing Games). These are fully 3D virtual spaces, with thousands to millions of participants; persistent state worlds on which a full life-cycle of activity continues even when a single user is not logged-in. Generally speaking, these games are characterized by fantasy settings, heavily inspired by J.R.R. Tolkien’s Middle Earth, complete with swords, armour, spells, monsters to be killed, and treasure to be had. These worlds have their own currencies, their own goods, unique production chains, different institutions and related transaction costs. As such, one might expect player-based economies to develop in a variety of manners, albeit with some consistency.

One of these consistencies is the existence of RMT (real-money trade), that is, the business of selling in-game currencies or other assets for physical currencies (most commonly U.S. Dollars, Euros, Yen, and Wan). While this may seem absurd on the surface, economics does not differentiate between the physical states of goods when determining what has value, rather value is dependent strictly on what someone is willing to pay. Some estimates have valued the RMT exchange trade to be worth close to USD 1 billion annually (Castronova, 2006), clearly no small business.

However, analysis of games is not the only area of interest. Adapting tools from MMORPG development processes with existing VR (virtual reality) technologies used for medical or military training, or rehabilitation can offer the ability to recontextualise laboratory experiments in a fashion that allows for greater parallelism. Fiore, et al. did just this in a 2007 study on governmental policy and willingness to pay for environmental protection. A comprehensive description of the technologies available and the manners in which they can be used is also provided by the authors.

Literature Review

Castronova’s seminal 2001 study, “Virtual Worlds: A First-Hand Account of Market and Society on the Cyberian Frontier,” inspired much of the still-sparse literature on the subject. He discusses his experiences in Sony Online Entertainment’s (SOE) game Everquest and describes various macroeconomic indicators. While there’s little theoretical work, it is useful as an introduction to the topic and its relevance in economic studies.

This is studied in more depth in Castronova’s 2002 paper, “On Virtual Economies.” Strong arguments are made for economic study of virtual economies, and the relevance of valuation of virtual goods is shown. Further theorizing is done, in which he develops a utility model for games of a general form, with the recognition that this is a unique setting in which people are willing to pay to be constrained. Models of challenge levels are introduced, the trade-off between various games is considered (including the “real-life game” of working and non-game leisure). A brief discussion of RMT introduces the concept and importance, the proper comparison to exchange rates and currency markets is made. This is an important point. Some future research into RMT requires an understanding that in-game currency is not simply another commodity to be fit into the consumer’s budget. Costs to players such as sunk start-up costs and switching costs are mentioned for the potential of future monopolization of the market. A brief discussion of policy concerns, including jurisdictions and tax revenues suggests these growing economies will come under closer scrutiny in the near future.

In “A Cost-Benefit Analysis of Real-Money Trade in the Products of Synthetic Economies,” (2006), Castronova seems to have lost some starry-eyed visions he had in earlier writings. He takes a critical look at RMT by performing a cost-benefit analysis of surplus divisions amongst four sectors: RMT businesses, players that use RMT services, players that do not, and corporations that run the games on which RMT operate. Making some assumptions about RMT elasticities of subscriptions and total costs, he finds that the socially efficient allocation is some level of RMT greater than zero, but below its current levels. A thorough discussion of economic impacts and policy responses is presented.

In a different vein, Fiore, et al., present a new formalization of experimental methodology to economists in their 2007 “Virtual Experiments and Environmental Policy.” They propose mixing VR systems and software toolkits developed to generate settings for MMORPGs with economics experiments, a methodology they refer to as VX – Virtual Experiments. The benefits discussed include a recontextualisation of the experiment to the subject, allowing greater flexibility and replicability, and removing concerns of parallelism. They describe an experiment they conducted in which they were analyzing choices made by subjects over policy changes to environmental protection schemas. They use Forest Service geographical data to reconstruct life-like forests with stochastic chances of fire dangers. The problem presented to subjects is whether or not to increase the level of fire-management methods to reduce the chances for fires, and elicit responses from a properly incentivized virtual property simulation. They do not present the results, perhaps because this is still a working paper.

Future Research

While Castronova seems to be focused on theorizing over utilities of players, a number of interesting economic questions are still present within virtual worlds. Clearly there is the possibility for an in-depth study of production chains and internal macroeconomic data. However, perhaps more interesting are studies involving the cross-over of the real world with the virtual one, that is to say, involving RMT. One important fact to note is that in almost all games players are split across 20 or more unique servers with completely non-linked economies, providing a great resource for controls over variables.

Inflation and RMT

A big debate in the developer and player communities is that of RMT causing inflation. Castronova (2006) seems to take this for granted. Though I am not aware of any research that shows this is the case, it is quite probable that it encourages or accelerates inflation, as it tends to keep money in the system rather than letting it flow out of the natural drains. It may seem that this is intuitively obvious. It bears remembering that the discussion at hand is not that of a commodity price versus a currency (where a drop in the commodity price clearly encourages higher demand), but rather an exchange rate between currencies. In real world economies there is no clear strong link between exchange rate fluctuations and local inflation even when measuring the import good sector of the CPI: “The relationship between consumer price inflation and the exchange rate tends to be quite weak across regimes” (Kara and Nelson, 2003). As such, it is clear proper research is needed.

In particular, a study of one MMORPG, Final Fantasy XI (FFXI) by Square-Enix, would be quite useful. FFXI has a number of things that differentiate it from other virtual worlds: an independent web site not associated with Square-Enix, www.ffxiah.com, reports all sales by all players on all servers that occur in the world, in almost real-time. Thus it is possible to acquire in-game price data quickly and easily. Furthermore, Square-Enix has lately taken an active role in removing RMT from FFXI. There are purges on a near monthly basis, resulting in deletion of several thousand accounts and the removal of billions of gil, the game currency, from the system. They publish rough numbers of the amount of gil removed. This combined with the RMT prices of gil on 3rd-party business web sites can provide a good estimate of both gil supply and, obviously, gil prices.

With this data, it becomes quite easy to conduct granger-causality tests on the hypotheses “A decrease in gil prices causes in-game inflation,” or “In-game inflation causes a decrease in gil prices.” The converse could also be tested: “An increase in gil prices causes in-game deflation.” These results would provide fodder for future studies, as well as actions by companies or the player-base.

Deflation

In a similar vein, the recent frequent removal of apparently large pieces of the money supply has left the FFXI economies in a tail-spin. Prices are consistently dropping and many players report increasing difficulties in selling their goods, as people hold on to their cash anticipating future deflationary pressures. In the meantime, certain fixed cost money drains required by the programmers of the game remain constant. This has put all but the richest of players in a bit of a pinch. Of course, it is entirely likely that the quantity of gil removed is not constant per server. It is also likely that each server has roughly the same levels of aggregate demand and supply for and of goods. If precise numbers about the money supply of each server and its reduction were obtainable from Square-Enix, one could conduct cross-server analysis of impulse response functions of the CPI to exogenous shocks in the money supply.

Investment Opportunities

As a result of actions taken by Square-Enix against RMT, FFXI has seen more frequent inflationary and deflationary cycles. Inevitable gil fountains continuously introduce money into the system, while the drains are perhaps insufficient to keep it at a stable level, so a certain level of inflation is natural. One avenue of research is to put together a basket of goods and track its performance over time. Good forecasting models could easily result in a net profit in in-game terms. The interesting question here is: “Is it possible to put together a basket of virtual goods such that, when compensating for in-game inflation, real-world inflation, and variable RMT exchange rates, a real-world positive profit can be earned?” Undoubtedly investment banks would be interested if the answer is yes. Having observed the prices of commodities and exchange rates over the past year, I estimate that if one had purchased USD 10.00 of gil in December of 2005 and not performed any in-game investment, selling the same quantity of gil today would gross USD 34.00 due to in-game deflation, for a return of 340% in just over a year.

Luxury Markets

FFXI has a rich manufacturing system. There are eight different sectors a player can pursue: Smithing, Goldsmithing, Alchemy, Clothcraft, Cooking, Leathercraft, Bonecraft, and Woodworking. Ability in the chosen craft is modeled by a skill number (out of 100, that signifying a master craftsman) representing a certain amount of practice in the trade. Each product is defined by its recipe of ingredients and a skill level. A player above the skill level of the chosen product can generally create the product easily. A player below the skill level of the recipe is either completely incapable of attempting it or risks a randomized chance of failure and loss of ingredients. One gains more ability by attempting the manufacture of goods above one’s current skill level. This is essentially a model of routinisation.

A player can practice all of the crafts up to a point. However, after a certain level of competence (a skill of 60, in this case), she must specialize. There are only 40 points total to allocate across skills over 60. The difficulty of this choice is compounded by the creation of “high quality,” or luxury goods. By being a certain number of skill points over the skill level of the chosen recipe, the crafter gains a specific percentage chance to produce high quality; the more points over the recipe’s skill level, the greater that chance – tests by the player-base indicate these percentages are either negligible, 10%, 30%, or 50% depending on the difference in skill levels between the crafter and the recipe. In some cases, the high quality manufacture results in a larger quantity of the item created for the same ingredients, generally in the consumables markets. In the durable goods markets, however, the results are luxury goods.

Expected Value theory informs us that a reasonable value to assign to an uncertain outcome is the probability-weighted sum of the potential outcomes. The luxury goods market is a perfect way to test that. With the current age of FFXI, it is reasonably safe to assume that all manufacturers of high-quality goods have reached the highest skill level of their chosen craft. As such, computing the probabilities of the outcomes is easy, as is finding the prices of the input and output goods. A brief review of one specific market in the game (the “Elemental Staves” market, consisting of eight non-substitutable items, each with normal- and high-quality outcomes, and requiring a high level of woodworking competence and a reasonable bankroll to attempt to enter) indicates that the expected value of the marginal good manufactured is exactly the cost of the inputs (with less than 0.25% forecast error)! While this seems to validate expected value theory, it raises at least one very interesting question: why is there no producer surplus rent being captured by the manufacturers to recoup the not-insignificant fixed cost of entering the market? Have players succumbed to the gambler’s fallacy, or is there an asymmetry of information, where one crafty manufacturer has learned of a way to improve their odds of the high-payoff result? Further research can help answer these questions and determine if this holds for many other popular luxury goods.

High-Stakes Trials and Replication

It is natural that experimentalists question whether results obtained by many studies scale up when the stakes are higher. MMORPGs can allow researchers with even frugal research budgets to conduct high-stakes experiments thanks to the relative value of in-game currencies and real world money. For instance, a typical payout to a subject in a real-world experiment may approach USD 20.00. At current gil prices, however, this can be converted to close to 1 million gil – a rather meaningful amount to most players, as the time required to earn that quantity well exceeds the time needed to earn the dollar-equivalent by most laboratory subjects (three to four hours at minimum wage, whereas the time required to earn 1 million gil can exceed 10 to 20 hours for most players). FFXI’s publicly visible random-number generator also provides a mechanism by which the experimenter can produce lottery outcomes. It might take several runs for the experimenter to gain the trust of the players (that payouts are real) and thus induce incentivised behaviour or study anything related to discount rates requiring future payouts, but once this is established it is relatively easy and cheap to replicate many experiments under high-stakes conditions.

Future Experiments

A research project with a much longer-term goal involves direct experiments with MMORPG players. It is easy to imagine academics having some role in development of future games, or perhaps having the research budget to develop and host a game directly. As a fantasy world created for the enjoyment of others, one could easily construct a world that contains, for example, (at least) two different nationalities, into which the players are randomly- or self-sorted when starting to play. One of these nationalities would have few institutions (much as the games of today do, any government qua government is almost exclusively related to plot-lines and has little or no effect on the game world) and have a relatively laissez-faire approach to its citizens. The other would be a polar extreme: heavily institutionalized, perhaps even changing rules and laws every other month – this is exclusively how the player observes it, in actuality the changing laws reflect a new experiment being conducted.

For example, consider the possibility that a certain rare token (a saleable commodity) could be found occasionally by killing monsters (a dominant mechanism for obtaining commodities in these games). This token would be sufficiently rare that the average player should expect to receive no more than 6-8 of these during a week of play. Once a player has accumulated a large quantity of tokens, say, 75, it could be possible to redeem them for a highly-valued and desirable (but non-saleable) item, for instance, a fancy new hat that increases power in some manner. In the control group (the laissez-faire nationality) a market for tokens would proceed as normal. In the treatment group (heavy institutions) the following program would be offered: a herald for the kingdom would announce an exciting investment opportunity in the kingdom’s foreign trading company. All players are welcome to contribute as many tokens to the kingdom’s efforts as they like. At the end of each week, the kingdom’s accountants tally the donations, the investments are made, and the tokens are doubled (“200% returns guaranteed by the kingdom’s treasure coffers!”). Each citizen of the kingdom is then eligible to claim her per-capita share of the total proceeds. By now it should be clear this is a public goods game, with the benefits of having interested participants, properly incentivized, a large sample size, no laboratory bias, a control group, and immediate and obvious parallelism. As above, changing the structure of the institutions observed by the players is in reality conducting another experiment.

Following the lead of Fiore, et al., it’s also possible to recreate the experiment in a slightly more real-time manner, perhaps avoiding some potential criticisms about its parallelizability. In this case, one would create a world on a much more vast scale. There is a rich history of online games where a player is an empire-builder, rather than simply controlling a single avatar. Much like the game “Solar Realms Elite,” one would be able to purchase land and develop it, and play some role in policy-making not just for their own territories, but for the world as a whole. Policy interactions and development strategies would require interface to a rich climate modeling system, where probabilistic weather events could destroy developments, and hence, income earning possibilities. As a persistent game, the players would have a much greater stake in maintaining a sufficient level of global well-being than in the one-shot experiment conducted in Fiore, et al.. Furthermore, it’s easy to systematically vary a variable of interest by allotting players to different world in the same virtual universe, each world in competition with the others. It’s simply a matter of imagination to saliently convert an economic question into a game eagerly played by thousands.

References

Castronova, Edward (2001a), “Virtual Worlds: A First-Hand Account of Market and Society on the Cyberian Frontier,” CESifo Working Paper No. 618, December.

Castronova, Edward (2002), “On Virtual Economies,” CESifo Working Paper No. 752, July.

Castronova, Edward (2006), “A Cost-Benefit Analysis of Real-Money Trade in the Products of Synthetic Economies,” Info, Vol. 8 No. 6.

Kara, Amit and Edward Nelson, “The Exchange Rate and Inflation in the UK,” Centre for Economic Policy Research, Bank of England, Discussion Paper No. 3783, 2003, December.

Fiore, Stephen M., Glenn W. Harrison, Charles E. Hughes and Elisabet Ruström (2007), “Virtual Experiments and Environmental Policy,” RFF/EPA Conference Frontiers of Environmental Economics, Washington.