An Agent-Based Competitive Product Diffusion Model for the Estimation and Sensitivity Analysis of Social Network Structure and Purchase Time Distribution

Keeheon Lee, Shintae Kim, Chang Ouk Kim, Taeho Park
2013 Journal of Artificial Societies and Social Simulation  
To maximise the possibility of success for a new product and minimise the risk and opportunity cost of a failed product, firms must understand the diffusion dynamics of competing products. The diffusion dynamics of competing products emerge from the aggregation of consumers' decisions. At the individual level, a consumer's decision consists of "which product to buy among the available products" and "when to buy a product". Individual product choices are affected by local and global social
more » ... ctions among consumers. It would be helpful for firms to be able to determine the characteristics of the relevant social network for their target market and how changes in this social network influence their market shares. In addition, determining the distribution of product purchase times of consumers and how their variation affects market shares are interesting issues for firms. In this study, therefore, we propose an agent-based simulation model that generates the market share paths (market shares over time) of competing products. We apply the model to estimate the social network and purchase time distribution of the Korean netbook market. Our observation is that Korean netbook consumers tend to buy a product without hesitation, and their social network is rather regular but sparse. We also conduct sensitivity analyses with respect to the social network and the purchase time distribution. 1.6 We choose to use small-world network model for the social network of consumer-agents. Rewiring a small fraction of connections in a ring lattice form, where every node is linked to its neighbours with a fixed degree of connectivity, results in a small-world network (Watts and Strogatz 1998) . In this network, nodes are highly clustered, and the information transfer time between nodes is short. It is well known from empirical evidence that the small-world network represents many types of real social networks (Alkemade and Castaldi 2005). Normally, the rewiring probability of a small-world network is within the range of 0.01 and 0.1. As illustrated in Figure 1 , a high rewiring probability leads to rapid product information transfer between clusters of nodes, causing the global network effect to increase. The degree of connectivity normally has a value between 4 and 20 (Kim et al. 2011) . When the number of network connections is zero, individuals choose products based only on their own product evaluations, without interactions with neighbours to obtain product evaluation information. As shown in Figure 1 , a high degree of connectivity leads to rapid product information transfer within clustered ties, causing the local network effect to increase. Therefore, we can observe local and global network effects by varying the rewiring probability and the degree of connectivity from zero to positive values.
doi:10.18564/jasss.2080 fatcat:f4ywztkkvjerhaxibhjaprc5zu