An Empirical Analysis of House Price Bubble: Discussion

Results of long run regression analysis are presented in Table 2 below. These indicate that a substantial amount of the variation in the house price index (HPI), about 74.1%, is explained by the model. The table also shows that the log of GDP growth is statistically significant and has a positive (20.0417) effect on the house price index for Beijing. This implies that if the speed of GDP growth increases by 1%, house prices in Beijing will increase by 20%. GDP growth (proxied by income) in the model captures the ability of consumers to purchase houses (also known as housing affordability). This result supports Gallin’s study, where house price and income were linked in the long run via housing affordability; they tend to return to their long run equilibrium relationship, although in the short run such a relationship may drift due to speculation or other fundamentals such as easy credit.
Linneman and Megbolugbe analyzed the U.S. housing market and found that affordability is one of the more important determinants of housing expenditures. The authors reported a positive relationship between household income and house price, which our findings confirm. Gallin investigated the long-run relationship between income and house prices in the U.S. housing market and concluded that there is a positive relationship between income and house prices at the national level. However, the CPI coefficient in the long run model shows the inflation variable does not significantly contribute. This result is consistent with Ji and Wang’s study, where they compared the CPI and the PPI with HPI during the period from 2000 to 2010. Their analysis provided evidence that in the long run, CPI and HPI do not have a strict one-to-one correspondence. The authors pointed out that the CPI and HPI affect each other by creating “cost-driven upward pressure”, but add that this channel is not stable; therefore, there does not appear to be a reliable, significant link between CPI and HPI in the long run.
Previous studies have documented a negative relationship between interest rate and house price. This appears to be because most consumers cannot afford to pay cash for a house; thus, they will be forced to take out housing loans. Therefore, an increase in interest rate will increase borrowing cost, which in turn will decrease house demand. Electronic commerce

However, some researchers have also reported an opposite conclusion. The Yun, Wang and Seabrook study showed a positive relationship between house price and interest rate in the Hong Kong housing market, known as the “Gibson paradox,” as introduced by Keynes. The same conclusion also appears in Ayuso, Blanco and Restoy’s study of house prices in Spain and McQuinn and O’Reilly’s study in Ireland. In China’s
housing market, most researchers have found a positive relationship between house price and interest rate. Our research showed similar results in the long run model (the interest rate coefficient is 3.1275, which is statistically significant at the 10% level). This is likely because the interest rate variable not only negatively impacts house price by increasing the borrowing cost for buyers, but also positively affects house price via the growth of borrowing cost for developers.
Every year, thousands of people move to populous cities such as Beijing and Shanghai. The majority of these migrants are in the 20 to 40 years of age group. They include graduate students from universities and wealthy families looking for better educational opportunities for their children. Children cannot study and participate in the universities’ entrance examinations in Beijing unless they are registered as residents of Beijing. Therefore, wealthy families buy houses in the city in order to gain registered residence. Such population growth in the cities leads to greater demand in the Beijing housing market.

Table 2: Estimated Results of the Long Run Model (Equation 3 with annual data)

Dependent Variable: HPI
R-squared 0.835461
Adjusted R-squared 0.741439
S.E. of regression 2.393222
Sum squared residual 40.09258
Log likelihood -24.26497
F-statistic 8.885801
Probability (F-statistic) 0.007091
Durbin-Watson statistic 2.032336
Coefficient Std. Error t-Statistic Prob.
C 41.98271 19.74111 2.126664 0.0710
LOG(GDPGROWTH) 20.04172 7.246538 2.765696** 0.0279
DLOG(CPI) 44.16191 24.51272 1.801592 0.1146
IR 3.127505 1.373268 2.277418** 0.0569
SUPPLYCOST 0.001269 0.000296 4.288998* 0.0036