This study analyzes the determinants of vacant urban land price using the hedonic model with the data of 1,119 transaction cases obtained by eliminating outliers from 1,174 cases traded in Changwon from January 1, 2014 to December 31, 2017.

As the methodology, it uses not only a traditional regression, but also a spatial lag model and a spatial error model considering the spatial dependence of the neighboring land price. As for the dependent variable, it uses the land price per 3.3 squared meter area and per parcel price, and the independent variables include the squared value of land size, land size, the form of land, the number of road-contacting surfaces, the zoning, the slope, and accessibility variables such as distance to the city hall, highway IC, and highway. To control the regional effect, it also uses regional dummy variables such as urban dummy variable and development area dummy variable having rural area as the reference variable. The semi-log model is used for the analysis function because the dependent variable and the accessibility variables of the independent variable which have the biggest values need to be transformed into their logarithms. The rest of the independent variables are used as they are.

Empirical results show that spatial auto-correlation exists in all models. And from comparison for model appropriateness through adjusted R-squared, log-likelihood, AIC and SC value, and LM test, it turns out that the spatial model is more appropriate than the traditional regression, but which one between spatial lag and spatial error model is more appropriate differs depending on the analysis model. Land size and the squared value of land size prove to be insignificant having the coefficients of 0 in all models.

As for the form of land, a rectangular form has higher price than an indeterminate or an irregular form. This may be because a rectangular form of land provides higher utilization to the owner in consideration of building a new construction on it than an indeterminate form of land, thus resulting in higher per square meter price for the construction later on.

With regard to the road-contacting surface variables, the land of which the two sides are adjacent to the roads, or a corner has higher price than the land of only one side contacting the road, and the land contacting a wider road has higher price, which means the more road contacting surfaces and the wider the road, the higher the land price. This may also be because the land which has more road contacting sides and wider contacting road has better accessibility. This can reflect people’s preference for accessibility. However, the variable for road contact length is insignificant in all models, which implies that just the length of road contact may not contribute to the utility of the land itself, rather making it clear that the width of the contacting road and the shape of road contact are more important.

The variables related to zoning which restricts the floor area ratio and the use of the land (e.g., residential, commercial, and industrial area) where the reference variable is non-urban areas, prove to have higher to lower prices in the order of commercial, type 3 residential, semi-residential, and type 1, 2 residential areas. In other words, commercial areas which have the highest floor area ratio have the highest price, and type 1, 2 residential areas which have the lowest floor area ratio have the lowest land price. However, it seems strange that semi-residential areas show lower price than type 3 residential areas, even though the former has quite high floor area ratio compared to the latter and it also has the merit of construction of commercial buildings on it. The reason for this unexpected result can be that there are not many vacant land pieces in semi-residential areas in urban areas of Changwon, with most of them existing in rural areas. In fact, semi-residential areas in urban city of Changwon are mostly filled with commercial buildings and thus have high price.

Urban region dummy variable shows a significantly positive coefficient in all models at 1% level. This is quite natural because urban land is usually more expensive than rural land. Yet, development region dummy has higher price than urban region dummy. Although most of the urban development region samples are located in rural areas, they have higher price than just urban region. This may be because designation of development region is positively reflected in the price. This can suggest that the government may need to implement continuous urban development schemes to boost the land preference for lagged regions, thus ensuring even, or balanced development of the nation.

Degree of slope variables, with steep slope being the reference variable, show that flat land dummy has a significantly positive coefficient in all models, while gentle slope is insignificant. In other words, flat land is, as expected, more expensive than steep slope, whereas the price difference in between land of steep slope and that of gentle slope is unexpectedly not so big. This unexpected result may be because steep slope accounts for only a small portion of the sample, 10%, which may not provide enough explanatory power, because the land has a good view due to the locational characteristic - steep slope, or because there is development expectation for the land and thus it is traded as similar to gentle slope.

Among the accessibility variables, the distance from the city hall and the main road show significantly negative regression coefficients, indicating that the closer the city hall and the main road, the higher the land price. Accessibility to the city hall implies that although Changwon has been multi-polarized since its integration of Masan and Jinhae, most of its industrial bases are still located in old Changwon yielding high incomes thus raising the property values in there accordingly. However, the distance to and from IC appears to be insignificant in both the spatial lag and the spatial error model, while having a significantly negative coefficient in only the traditional regression model at 1% level.