| dc.description.abstract | This thesis introduces the application of real-time micro behaviour data in inflationnowcasts. Our study analyses if ARIMA models extended with Google search dataimproves the prediction of divisions of inflation compared to the high-performing simpleAR(1) process. This analysis addresses the issue of official inflation data containing alag of ten days. Real-time micro behaviour data can contain valuable information, whichprovides policymakers with a new tool to predict inflation in the present and near future.First, each division of inflation is assigned corresponding Google Indicators before in-samplemodel selection is performed using the Box-Jenkins Methodology. Then, comparisonsagainst ARIMA baselines are conducted to evaluate if Google search data improve modelselection. Further, out-of-sample predictions are performed for the improved divisionsfrom the preceding step. Finally, the nowcast performance for each division is comparedagainst the simple AR(1) process in terms of prediction error and ability to identify trendsand turning points.This thesis documents that Google search data improves model selection for six of twelvedivisions of inflation. These divisions consist of goods and are volatile compared tothe remaining six. Furthermore, four of six extended ARIMA models outperform thesimple AR(1) process in prediction error for the out-of-sample nowcasts. At the sametime, all divisions are improved in predicting trends and turning points. These findingssuggest that real-time micro behaviour data, represented by Google Trends, improve modelselection and nowcasts of some divisions compared to AR(1). However, when comparedto replicated and baseline ARIMA models, the only value of Google search data is inmodel selection. The improved performance is attributed to the properties of ARIMA. Toconclude, real-time data on micro behaviour are of value in model selection in inflationnowcasts. | en_US |
