Font Size: 

Many  recent  papers  have  investigated  whether  data  from  internet  search engines such as Google can help improve nowcasts or short-term forecasts of macroeconomic variables. These papers construct variables based on Google searches  and  use  them  as  explanatory  variables  in  regression  models.  We add to  this  literature  by  nowcasting  using  dynamic  model  selection  (DMS) methods  which  allow  for  model  switching  between  time-varying  parameter regression models. This is potentially useful in an environment of coefficient instability  and  over-parameterization  such  as  can  arise  when  forecasting with Google variables. We extend the DMS methodology by allowing for the model  switching  to  be  controlled  by  the  Google  variables  through  what  we call Google model probabilities. That is, instead of using Google variables as regressors,  we  allow  them  to  determine  which  nowcasting  model  should  be used  at  each  point  in  time.  In  an  empirical  exercise  involving  nine  major monthly  US  macroeconomic  variables,  we  find  DMS  methods  to  provide large  improvements  in  nowcasting.  Our  use  of  Google  model  probabilities within DMS often performs better than conventional DMS.