Lessons Learned from Converting the Artificial Stock Market to Interval Arithmetic

This paper describes work undertaken converting the Artificial Stock Market (LeBaron et al., 1999; Johnson, 2002) to using interval arithmetic instead of floating point arithmetic, the latter having been shown in an earlier article to be the cause of changed behaviour in the ASM (Polhill et al., in press). Results of both a naive (potentially automatable) conversion and one involving a more in-depth analysis of the code are presented that suggest that interval arithmetic may not be the best approach to dealing with the issue of numeric representation in the ASM. We also find that there are good reasons to suspect that floating point errors are not a significant issue for the ASM.