Detecting “Small” Mean Shifts in Time Series

Analyzing time series data for evidence of changes in market conditions is a problem central to development of a marketing information system. Examples are tracking field reports on competitors' prices and analyzing market share estimates computed from retail audits and consumer panel reports. These series may involve substantial random measurement error, and any changes, often small relative to this error, are difficult to detect promptly and accurately. One way to approach this problem is with techniques recently developed for handling time series with small stochastic mean shifts as well as random error. Two special procedures are compared in Monte Carlo analyses with a simpler data filtering and control-chart approach; the latter appears the most promising.