Autoregressive conditional heteroscedasticity (ARCH) effects make the probability of large losses greater than standard meanvariance analysis for investor in shortterm. Capturing ARCH effects accurately is meaningful for the research of shortterm trend of housing price. This paper empirically research and compare the regressive model, GARCH model and AR model with the data of average housing price in China and four municipalities' housing price. The results show the ARCH effects in housing markets in Beijing, Shanghai, Tianjin except Chongqing. We also found out that the regressive model is a little better than GARCH model for shorttime forecast while the GARCH model is better in longterm forecast. Therefore, if it is hard to obtain relative data, GARCH model is more applicable.