3 Theoretical Framework
5.2 Geographic Distribution
How is house price growth distributed geographically? Figure 6 plots the growth in house prices for each municipality in Denmark for different periods. Panel (a) plots the increase
23It is possible to rationalize the observed empirical patterns in alternative models of the housing market.
The positive coefficients can be due to costly learning, where households learn about the state of the housing market from prices, or due to search and matching as in Head et al. (2014).
from 1998Q1 up to 2004Q1 (pre-boom period); panel (b) plots the increase in prices from 2004Q1 to the municipality-specific peaks; and panel (c) plots the decrease in prices from the peak to trough. Note that values in panel (c) denote decreases in prices. The largest price increases in the pre-boom period occurred in the eastern regions of Denmark (specifically in the area around Copenhagen), while smallest price changes occurred in western Denmark, where house prices declined in some municipalities and growth was generally low. During the boom, prices increased substantially in all local markets – the average increase was 54 percent from 2003Q3 to the peak in house prices. The largest price increases are again found in the area around Copenhagen, on the east coast of Denmark. In contrast to the pre-boom period however, the largest increases were not concentrated only in eastern Denmark. Two other high growth areas of the boom are located around the cities of Odense and Aarhus, the second and third largest cities in Denmark; areas that were not among the fastest growing municipalities prior to 2003. During the bust, all municipalities experienced substantial decreases in prices – the smallest decrease was 14 percent from peak to trough, and the average decrease was 31 percent. Price declines were concentrated in areas that previously experienced substantial price growth, in particular the area around Copenhagen.
Table 4 presents evidence on differences in observable characteristics between munici- palities. Specifically, I present results for municipality characteristics from 2002, before the boom occurred. All municipalities are divided into four groups based on price appreciation during the boom in columns (1)-(4), and the last column shows the results of a t-test of differences between the first and fourth column. Between 2004Q1 and the peak prices ap- preciated by 40 percent on average for the 23 municipalities with the lowest appreciation in column 1, and by 71 percent for the municipalities with the largest appreciation in column 4. Note that even the group of municipalities with the lowest appreciation rate experienced a substantial increase in prices during the housing boom. However, most of these gains were reversed during the boom. For the municipalities with the largest appreciation (column 4), a 69% increase followed by a 39% decline meant that house prices increased by just 3% between 2004 and the trough. Average income, the average income of renters below
the age of 35 (young renters), population density and income inequality (as measured by the ratio of the 90th income percentile and the 50th income percentile for each munici- pality, the P90/P50 ratio) were all significantly larger in high-growth areas compared to low-growth areas. The boom was larger in magnitude in municipalities with a higher initial price level, higher levels of income, higher levels of inequality, and in municipalities with more constrained housing and higher population density. Taken together, this suggests that the boom was mainly a city phenomenon, and that consistent with the theory of Glaeser et al. (2008), house price growth was higher in areas with less elastic housing supply, only to decline more when prices adjusted.
Table 5 presents the results of a regression of municipality-level growth in house prices during the boom separately on several variables measured in 2002. Price increases during the boom is positively and significantly correlated with higher levels of income, whether measured by the average income, income for renters below the age of 35, or buyer income. The P90 / P50 ratio, a measure of income inequality, is also positively correlated with house price appreciation, showing that more unequal municipalities experienced a larger increase in prices. Neither the unemployment rate nor the population size in 2002 are significantly related to house price growth. Moreover, non-income related variables such as housing supply, the pre-boom increase, the square meter price, and population density are all positively correlated with house price appreciation. The square meter price and density explain 39 and 30 percent of the variation in house price growth across municipalities respectively, and buyer income explains 20 percent.
Table 6 further characterizes where the boom occurred. Column (1) shows the result when price increases during the boom are regressed on all the above variables related to income and labor markets in 2002. While the signs remain positive for income and the P90 / P50 ratio, none of the income variables are significant. The level of unemployment in 2002 is the only significant variable. Column (2) shows that municipalities with higher initial house prices experienced higher house price growth during the boom. This positive correlation remains in column (3), where all variables are included. In addition to the square
meter price, column (3) shows that log population and pre-boom house price growth are negatively correlated with boom growth. Overall, the results suggest that about half of the variation in house prices can be explained by observed characteristics of municipalities.