| Summary: | 碩士 === 國立臺北大學 === 統計學系 === 107 === Along with the popularity of internet, people start to look for information by In- ternet. Search volumes gradually become the factor to affect vegetable prices. We construct different models from two aspects. In aspect of data, we take into account the information of Search Volume Index from Google Trends to improve the ac- curacy of model. In aspect of model, besides single models, we also construct and hybrid model. We construct the hybrid model by time series model, the statistical model usually used in economy, and a non-linear machine learning model, support vector regression to combine the advantages of linear and non-linear model. Last, we compare the forecasting performances among those models. The results show that the models with Google SVI have better prediction performance. This im- plies the models with SVI enhance the forecast ability. Besides, the hybrid model has the most prediction performance among four models in both In-Sample and Out-Of-Sample .
|
|---|
