This study examines the impact of living floor on the price of an apartment with the data sample of 3,249 real transaction cases of apartments in Sejong city. To complement the shortcomings of a traditional regression that it may not reflect the impact of the history of transactions of neighboring apartments on the price of the apartment of interest, this study adopts a spatio-temporal model and a STAR model as its main methodology. Along with them, a traditional regression was also conducted for the purpose of comparison analysis.

The explanatory variables include, as structural variables, top floor level, living floor, squared living floor, living floor dummy, relative living floor level (living floor/top floor level), age, squared age, size, the number of total units, parking capacity, independent unit dummy, and brand-name apartment dummy, and as proximity variables, distance to/from the integrated Government Building, distance to/from Daejeon city hall, and BTR dummy. To control the price change of an apartment during the study period, time fixed effect dummy was used. Also, an additional analysis was carried out to this end based on log values of the real prices which were adjusted with apartment price index released by Kookmin Bank.

For the empirical models, 3 different types of models were used such as one with living floor dummy which combines each 3 floors up as one floor dummy, another with relative living floor(living floor level/top floor level), and the other with both living floor and the squared living floor. An additional model which includes regional dummy was also analyzed in this research. The purpose for using the living floor dummy which combines 3 floors up as one floor dummy is to see how the height of living floor affects the price. Even the same height of living floor may have different influence on the price if the total floor level differs. To consider this effect of the total floor level, the variable of relative living floor, namely, living floor/top floor level, was used in the second model on the list. The third model using both living floor and the squared living floor was aimed to examine whether the living floor has curvilinear effect on the price.
The results of this empirical analysis are as follows:

The STAR model turns out to be more appropriate than the other models employed in this study. The coefficients of a space dependence variable in both the spatio-temporal model(ρ) and the STAR model(ψ) have positive signs at the 1% significance level, which proves the existence of the impact of the transaction history of neighboring apartments on the price of an apartment of interest. Here, the value of ρ is higher than ψ, which may be because the spatio-temporal model does not consider temporal causality, thus overly reflecting the space dependence.

The regression which uses living floor dummy shows that the higher the living floor goes up, the higher the price turns except for the dummy of 19-21 floor level. In the model using both the variables of living floor and the squared living floor, the living floor showed a positive coefficient at the 1% significance level, whereas the squared living floor was insignificant. This result is different from that of most previous studies that the price goes up as the height of living floor rises to a certain point and then decreases above the inflection point. It seems that this odd result of mine may need some explanation. This may be because most home buyers in Sejong city are relatively young compared to the average age of home buyers nationwide and rather prefer to live on the high floor of an apartment since Sejong city is a newly constructed town with mostly high-rise apartments.

In the meantime, in the model considering regional dummy, both living floor and the squared living floor appeared to be significant at the 1% significance level. However, the aforementioned inflection point turned out to be 40 while the top level was actually 35 in the study sample. This can be interpreted as: the price keeps rising with the height of living floor.

Yet, the coefficient of the relative living floor (living floor/top floor level) was bigger than that of the living floor, which implies that home buyers value the relative height of living floor rather than the height of living floor itself.

Total floor level showed negative coefficients at the 1% significance level in all models regardless of the living floor variables used. This result also proves the above finding that the home buyers may not favor high-rise apartment building itself even if they prefer high living floor.

Size, parking capacity and stair type apartment dummy showed positive coefficients in most models, as expected, at the 1% significance level. The number of apartment units showed positive coefficients only in STAR model at the 10% significance level.

The proximity variables such as the distance to and from the Integrated Government Building and Daejeon city hall were also not different from the expectation in that they showed negative coefficients at the 1% significance level thus reflecting the buyers’ preference for the proximity to those sites.

BTR dummy and brand name dummy also showed significantly positive coefficients at the 1% level. It proved that the apartment with a BTR station nearby or with a high-end brand name was valued higher than its counterpar