ABSTRACT
Introduction Between 2001 and 2010, the number of internet users worldwide more than quadrupled from less than half a billion to almost two billion (www.internetworldstats.com\stats.htm). Approximately 85 percent of internet users have bought at least one product online, and as of August 2010 the largest and fastestgrowing social network platform, Facebook, connects more than 500 million active users. Roughly 250 million of them visit Facebook on any given day and spend, on average, more than 80 minutes per day on the website. This wide acceptance of the internet alters product development (Dahan and Hauser, 2002), and the persistent development of successful new products remains one of the most essential challenges for companies (Crawford and Di Benedetto, 2006). Yet, new product development still remains difficult and costly (Di Benedetto, 1999). The flop rates of newly launched products have remained high over the years, often surpassing 50 percent (Urban and Hauser, 1993). Hence, even small improvements in the new product development process can have a major effect on companies’ profits and competitive advantage if this flop rate is reduced. Therefore, new methods to improve product development are of high relevance for companies. Prediction markets, also called information markets (Hahn and Tetlock, 2006) or virtual stock markets (Spann and Skiera, 2003), are such a method. They attempt to connect a group of participants together in a virtual marketplace and enable them to trade shares of virtual stocks. In prediction markets, these stocks represent a bet on the outcome of future, uncertain events, and their value depends on the realization of these events (Forsythe et al., 1992; Spann and Skiera, 2003). For example, a stock may represent the predicted number of sold units of a new product (e.g., the iPhone 4G) in the first quarter after its market introduction. After the outcome of the specific event becomes known (i.e., the actual number of units sold), each share of virtual stock receives a specified cash dividend (e.g., $1 for each 1,000 product units sold). Participants in a prediction market use their own assessments about the expected event outcome and its corresponding cash dividend to derive an expected stock value and trade accordingly. For example, a participant’s expectation that 100,000 iPhones would sell during the first quarter after its market introduction corresponds to a cash dividend
of $100. If the current price of the corresponding stock is $95 (or $105), the stock appears undervalued (or overvalued) to this participant, so he or she should try to earn the anticipated profit of $5 by buying (or selling). The participant’s information thus affects the market price through his or her trading behavior. Such prediction markets initially were applied in the form of political stock markets (later called the Iowa Electronic Market) to predict the outcome of the 1988 US presidential election, with participation restricted to members of the University of Iowa community (for a more detailed description, see Berg et al. (2008) and Spann and Skiera (2003). In the ensuing two decades, prediction markets have achieved promising results for short-term forecasting tasks, such as political events (Berg et al., 2008; Forsythe et al., 1992), sports competitions (Luckner and Weinhardt, 2007; Servan-Schreiber et al., 2004; Spann and Skiera, 2009), business events (Elberse, 2007; Foutz and Jank, 2010; Gruca et al., 2003; LaComb et al., 2007; Spann and Skiera, 2003) and the identification of lead users or experts (Spann et al., 2009). The theoretical foundation for prediction markets is the market efficiency attained in a competitive market through the price mechanism, which Hayek (1945) considers the most efficient instrument for aggregating asymmetrically dispersed information possessed by various market participants. Prices in efficient markets always fully reflect the available information (Fama, 1970), so the prices of virtual stocks serve as good predictors (Spann and Skiera, 2003). The aim of this chapter is to discuss the application of prediction markets in new product development and to empirically determine factors that influence the forecasting error of prediction markets. For that reason, Section 2 discusses the possibilities for prediction markets to support the different stages of the new product development process. In Section 3 we describe an empirical study that uses prediction markets to forecast the success of new products, compare forecasting accuracy with those of expert judgments and analyze the factors that influence forecast accuracy. Section 4 summarizes the implications of the chapter.
