RESEARCH IN VIRTUAL ECONOMIES

Beginning with Bradley and Froomkin (2004) and Castronova (2006b), the argument has been made that virtual worlds would be interesting research tools. A virtual world contains a small human society, in the same way that a beehive contains a colony of bees.

The main advantage of controlled experiments with virtual worlds is the power of such a method to identify causation. Causation is the great bugbear of social science at the macro scale. Historical studies can only reveal the order in which effects occurred, not whether a prior effect made a posterior effect happen. Comparative studies can only reveal that two countries or provinces or villages had different experiences, not that characteristic X caused those differences. Cross-sectional data can only reveal patterns of correlation; ethnography reveals narratives and symbols. None of these methods show in a clear, replicable way that the presence or absence of factor X resulted in a change in factor Y This is precisely what a controlled virtual experiment does: it takes two societies that are alike in every way, and changes one factor in one of the societies. By comparing the experimental society to the controlled one, the researcher can observe the causal effect of the change.

Great questions of social science can be directly addressed by such a method. To what extent does taxation inhibit growth? How does inequality alter labor supply? Do the media affect culture? How does money affect political outcomes?

While virtual worlds would allow causal conclusions on such questions, the results are of questionable generality.

Like any test tube or petri dish, virtual worlds can shed light on some questions but not on others. As software for building virtual worlds becomes more accessible, there will be more reports of controlled experiments with populated artifact environments. In the course of this research, the best and worst properties of virtual worlds as research tools will come to light.

For now, researchers must usually rely on existing virtual worlds to perform research. Most published work in ‘pure’ economics of virtual worlds has come from Edward Castronova. Castronova (2001) explored the nature and magnitude of the virtual goods trade in the EverQuest online game. The more theoretical article ‘On virtual economies’ (Castronova, 2003) considers the consequences of a simple model of choice between time spent ‘in the real world’ and time spent playing a game.

Castronova (2004) examined the market for game accounts, one kind of RMT. Hedonic models of game account value showed that accounts with female avatars sell for 9 to 10 percent less than game accounts with male avatars, even though avatar gender is a purely cosmetic feature. Castronova (2006a) considers the problem of RMT, discussed above, and shows that the negative consequences of unsanctioned RMT mean that par­ticipants in that market create a negative externality that is born by non-participants. Depending on the assumptions made about the parameters in Castronova’s model, the externality can be quite substantial, on the order of several million dollars per 100 000 game subscribers per year.

Wang et al. (2013) investigate the sources of value of virtual goods from a number of online games, with a focus on the effect of social networks. Interestingly, and quite unex­pectedly, the authors find that social network intensity is negatively associated with the value of virtual goods. This might be because productive resources created by stronger social ties act as a partial substitute for virtual goods. The effect is mitigated, however, by the fact that the stronger social ties create greater demand for game time, which is a complement of virtual goods.

A curious feature of virtual economy research is the substantial number of insightful but unpublished manuscripts on the matter. Marks (2009) looks at auctions of virtual goods in the game World of Warcraft and finds that the so-called in-game auction house is a misnomer - in fact, some 98 percent of items posted sell at buy out prices, rather than being bid up in the style of a traditional auction. This finding suggests players’ discount rates in games might be extremely high, given how much more players are willing to pay to receive desired goods immediately.

The model produced in the working paper (Huhh, 2009) served as an important pre­cursor of Wang et al. (2013). Huhh showed that the availability of RMT, even if unsanc­tioned, may create positive network externalities enjoyed by both the users and the game producers. It is even possible that these externalities may overcome those discussed in Castronova (2006a).

One unpublished manuscript that we find particularly important is Jung, Lee, Yoo, and Brynjolfsson (2011) (hereafter JLYB). In an effort to understand the RMT market and its relationship with the game whose goods are being sold, JLYB model a player’s decision to spend time in a game primarily as a function of their productivity in the game. That is, they explicitly deal with the fact that game companies can set the rate at which virtual resources are rewarded to players, given their skill levels. This rate of reward, in turn, drives the player’s decision of time commitment to a game. JLYB show that players who commit substantial time (that is, who are higher skilled or who have a lower opportunity cost of playtime) become sellers in the RMT market, while those who have lower skill or higher opportunity cost become buyers. JLYB then derive the conditions of the RMT market as a derivatives market. Because this derivatives market affects the value of the game to both low and high-opportunity cost players, profit maximization of the game firm demands that they consider effects of in-game reward rates on both the value of the game and the value of the items in the derivatives market.

JLYB is important in that it is the first paper to reject the idea that a user’s preference for a game depends directly on the game’s challenge relative to the player’s skill. Instead, challenge only affects game demand indirectly, by regulating the rate at which the game’s content can be consumed. We think this is an important innovation, and its consequences for the structure of the game industry - especially the F2P sector - deserve renewed scru­tiny from virtual world economists. An industrial organizational analysis of the videog­ame industry using the JLYB framework seems a natural next step. Currently, models of duopoly using this framework are in development.

24.5