Regency B
Identifying Latent Effect in Time-Series Data with Applications to Problems in Economics and Veterinary Parasitology (303948)
Brandon Woosuk Park, US Food and Drug Administration*Anand N Vidyashankar, George Mason University
Keywords: Time-series data analysis, Latent effect, Non-stationary time-series, Seasonal effect, Time-series models
It is common in time-series data that a latent trend, representing a hidden structural component, prevails but is not easily identified by standard ARMA type modeling strategies. As a specific example, in veterinary parasitology, an increasing trend in drug resistance may be present which, however, may be hidden due to the presence of confounding seasonal effects; the presence of spatial patterns adds additional complications. These complications also arise in other applications; for instance, when investigating the time-series of (i) retail sales, (ii) number of new jobs created in a spatial-region, and (iii) amount of exercise and its impact on health benefits. While the cause for the trend may be different, statistical modeling can be used to identify this hidden trend in a variety of these applications. In this presentation, we describe a model-based approach to explore and identify the presence of non-stationarity and seasonal effects in the presence of latent trends. We establish statistical properties of the proposed methods and algorithms for computing the standard error of the model parameters.