Author

Brennan Dyk

Date of Award

January 2013

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Economics & Finance

First Advisor

Cullen Goenner

Abstract

The unemployment rate is typically forecasted in the literature using initial claims for unemployment, a lagged value of unemployment, and an autoregressive component. This paper looks to improve upon existing models by adding Google search data to traditional models using initial claims or replacing initial claims with Google searches. One hypothesis is that Google searches may improve forecast accuracy due to employees knowing or getting a sense when they may become unemployed and be searching for jobs prior to filing for unemployment. Study of this is important to improve the accuracy of models and to provide more accurate out-of-sample forecasts for policymakers. If it showed a significant increase, it could perhaps diminish the frequency of surveys conducted by the Bureau of Labor Statistics. Several ARIMA models are used to determine whether or not the addition of specific Google searches can be more useful in an out-of-sample predictive model of unemployment. The results show that while Google was a good way for predicting the model in the past, it does not beat traditional models that use initial claims as an independent variable in predicting changes in the direction of the unemployment rate. In addition, models including regime switching do not improve the forecast accuracy compared with standard models.

Share

COinS