B1368
Title: Detection of intervention effects on time series
Authors: Anish Ganguli - Walmart (India) [presenting]
Abstract: Intervention effects is a very popular term in the study of Time Series Analysis. There are a lot of factors which intervene with sales, and as a result, the pattern of the sales might get impacted a lot which eventually affects the forecasted sales numbers and, finally, the business decisions. The major problem with these intervention effects is to identify whether the effect is significant or not and also, if it is significant, how can we adjust for the impact or include its impact in the modelling/forecasting process? The objective is to review the approaches by which we can understand if the effects are significant or not in terms of impacting our Time Series sales data. To understand the significance of these effects, two approaches have been considered, viz. Autoregressive Distributed Lags (ARDL) and Google's Causal Impact (CI). These two approaches are helpful in determining if the intervention effect in consideration is significantly impacting our Time Series or not.
