STA457 Assignment
Instructions:
1. Download the "British Columbia" Google Flu Trends and Flu Test data using
GFluTrends<-read_excel("case_study_1_fluwatch.xlsx",sheet="Google Flu
Trends", skip = 1)
fluWatch<-read_excel("case_study_1_fluwatch.xlsx", sheet="FluWatch-BC",
skip = 2)
tim<-timeSequence(from = "2003-09-28", to = "2015-08-09", by = "week")
tim1<-timeSequence(from = "2003-09-07", to = "2015-08-23", by = "week")
GFT<- timeSeries(GFluTrends[,"British Columbia"], charvec = tim)
fluTest<- timeSeries(fluWatch[,"FluTest"], charvec = tim1)
2. Split the dataset into two groups: training sample and test sample (as shown in TimeSeries-Forecasting-Examples.html);
3. Follow the steps and techniques in Time-Series-Forecasting-Examples.html to answer
the following questions.
A. Transfer function noise modeling (10 pts)
1) Write the mathematical equation of the fitted transfer function noise model—using
Google Flu Trends as the explanatory variables to predict Flu Tests.
2) Is the fitted model adequate? Explain your answer.
Support your answers with R codes and empirical evidences.
B. Forecast evaluation
Consider ARIMA, TFN, NN, and NNX models. (see Time-Series-Forecasting-Examples.html
for more details)
1) Calculate the forecast performance in the training sample, and answer which model
performs best in each forecast measure.
2) Calculate the forecast performance in the test sample for the following lead time
i. h=1
ii. h = 4
iii. h = 8
iv. h = 50
and answer which model performs best in each forecast measure.
Support your answers with R codes and empirical evidences.