| Summary: | 碩士 === 東吳大學 === 經濟學系 === 101 === In recent years, due to rising house prices, resulting in high prices in metropolitan area. In this thesis, the relationship between macroeconomic variables and real estate prices with the ups and downs in Taiwan will be considered. The three lags period of macroeconomic independent variables will be used to predict the current dependent variable (Change in real estate prices in Taiwan). Periods of dependent variable are set from June 1998 to September 2012, a total of 172 observations; independent variables are from March 1998 to June 2012. In this paper, the traditional regression, logistic regression, genetic algorithm, back propagation neural network and generalized regression neural network models are used to forecast the change in real estate prices in Taiwan.
The empirical analysis shows that five models for forecasting performance results are follows: BPN (96.88%) best, followed by OLS and GA (87.50%), and the final are LR (84.38%) and GRNN (62.50 %) respectively. As performance of prediction are concerned, we also applied ROC model to analyze and showed that BPN has best predictive ability, OLS, LR and GALR model also has a good performance, however the predictive ability of the GRNN (smoothing parameter value is set to 1) should be further be improved; In the meantime, we also applied DA direction performance test for these models.
In conclusion, In addition to GRNN (smoothing parameter value is set to 1), the accuracy prediction rate of the models in this papers are more than 84.38%, and are shown statistical significant, it also show that a variety of measurement methods mentioned in this article has a good performance.
|
|---|
