TS sproj baby names Jillian
This student project received credit for the time series ciourse because the candidates understood the time series techniques in the textbook. But the student project shows how one might try to over-fit a model. The popularity of names is affected by exogenous events, such as a movie start with a given name. For example, the higher popularity of the name Jullian in 1998-2003 probably reflects Jillian Reynold’s TV shows and hosting of Fox Sports Super Bowl games. The sudden popularity of this name after 1980 perhaps reflects the influence of Ann Jillian, another popular actress.
The R2 of the first three ARI models (5.04%, 5.02%, and 5.03%) are the same; the differences are random fluctuation. It is hard to find a rationale for higher order autoregressive processes. When choosing names for babies, parents are influenced by events of the past 1 to 3 years. Events that occurred earlier presumably have little effect. Since the R2 values are the same, we would choose the ARI(1,1) model.
The Box-Pierce Q statistic is not quite significant at the 10% level. But that is okay; we assume exogenous events affect the time series, and we don’t assume the residuals are pure white noise. The Durbin-Watson statistic is also reasonable; not exactly 2, but close enough.
This student project is well-written, and you can see the reasoning underlying the choices. Some time series have little exogenous influences, and we look for a close fit. Baby names are highly affected by exogenous items, especially the appearance of movie actors and actresses in widely seen films or TV shows. When exogenous influences are strong, we use simpler models, since the more complex models are picking up the exogenous influences, not autoregressive effects.