196H. Zarzycki et al.11.1IntroductionIn comparison to existing methods, more accurate forecasting methods can beobtained using a rule-based forecasting (RBF), a technique combining data extrapo-lation [7,13,14,25,26,43–45], time series [28,29,44,45], and elements of expertsystems [5–7,22,23,34,37,46]. The four most important methods of extrapolationwere used: linear regression, random walk, and Brown’s exponential smoothing, aswell as Holt’s exponential smoothing. In order to create rules some information fromthe literature, surveys, and knowledge of several experts was adapted [17,19–21,36,38,39]. The rules were calibrated using 80 time series. In contrast, the validationneeded another 40 series. In the opinion of the authors, RBF has been successfullyapplied by combining domain expertise with statistical methods. This has been con-firmed by many studies in the recent literature, where rule-based forecasting is afast-growing technology. It is worth mentioning a few examples from a very com-prehensive literature such as M. Adya, J.S. Armstrong, and F. Collopy [1–3,8,9],who publish in the International Journal of Forecasting, a magazine that inspiresother authors associated with the RBF methods. In this chapter time series of indexdata were preliminarily fuzzified [30,33] to check the proposed methods of detect-ing trends [18]. Trends identified in the sequence of literals are then used to developtrend prediction rules. Therefore fuzzy logic [12,13,16,35] was used to developlinguistic data input. Data for the study were quotations of the Nasdaq Compositeindex from the years 2006–2016. Figure11.1shows the data in an illustrative man-ner. Table11.1contains Nasdaq index data for a single trading day. Daily data are:opening, maximum, minimum, and closing values as well as the percentage changeFig. 11.1NASDAQ Composite index quotations from 2006 to 2016
