Flash Estimates of Household Final Consumption Expenditure in East Java

Abstrak

The purpose of this study is to construct a flash estimates model for household final consumption expenditure in East Java. There were 153 indicators that were used, covering Big Data and non Big Data indicators. Indicators from Big Data were the Google Trends indicators. The Google Trends categories were selected based on the highest correlation coefficient. While the non Big Data indicators were the total deposit, consumption credit, consumer lending rate, term deposit rate (1, 3, and 6 months), withdrawal (outflow), and Consumer Price Index (CPI). This study examines several models: ARIMA, ADL, Elastic Net, Support Vector Machine, Random Forest and Ensemble. The models were examined using various scenarios of out-of-sample and estimation periods (t+10 days, t+20 days and t+30 days). The prediction performance of models were evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). The evaluation results showed that the Elastic Net and the Ensemble were the best model for any scenarios. Both models had good performance since t+10 days.