Korea is currently facing a severe population decline crisis, and household structure is also changing rapidly. It is anticipated that the future population and household structure will differ significantly from the present. This demographic shift nec...
Korea is currently facing a severe population decline crisis, and household structure is also changing rapidly. It is anticipated that the future population and household structure will differ significantly from the present. This demographic shift necessitates a systemic transformation across society, placing us at a critical juncture where decisions must be made on how to respond to these future changes.
These changes in population and households vary greatly by region, with significant differences—not only between the Seoul metropolitan area and non-metropolitan areas or between urban and rural regions but also among different cities and even within a single city. In Seoul, the capital city, the population has been in a long-term decline due to the outflow of the middle-aged and older population, despite the continuous inflow of the youth population. On the other hand, unlike the situation of continuous population decline, households in Seoul have shown a continuous increase driven by household fission. More recently, an accelerating outflow of households headed by individuals in their 30s and 40s is further reshaping Seoul’s household structure. These structural changes are unfolding in diverse ways across different districts within the single city of Seoul.
In this context of spatially differentiated demographic changes, it is important to forecast future changes by regions, and there is a growing demand for projection at smaller geographical scales. However, due to the complexity of the estimation process, projections are still predominantly produced only at the national/sub-national(Sido) level.
Therefore, this study defines small areas as districts—the basic units of local government—and develops and presents a deep learning-based model for future household projection applicable at this geographical level. Using Seoul as a case study, we present household projection results for the next 20 years. To utilize these results, we then estimate future changes in housing demand in Seoul and attempt to develop regional typology. The specific findings of this research are as follows.
In Chapter 3, a deep learning-based LSTM model was constructed to overcome the limitations of existing household projection methods used at the national/sub-national level. This model is designed to incorporate not only population and household changes at the small area level but also shifts in spatial distribution across regions, and we examine its applicability. Both univariate and multivariate LSTM models were designed for three different scenarios and evaluated their predictive power by comparing them with other household projection methods. The results showed that Scenario 3, which accounted for changes in the spatial distribution of housing area by regions, exhibited a trend most similar to the official future household projection by Statistics Korea.
In Chapter 4, to utilize the small area future household projection results, future housing demand was estimated and presented. Estimating housing demand is an inherently complex process, as it is shaped by a combination of the diverse characteristics and economic conditions of people who need to use housing. However, since housing is a good with low supply elasticity, if there is a mismatch between supply and demand, instability in the housing market may increase, so it is important to predict changes in housing demand from a long-term perspective. Furthermore, the geographical unit for estimating housing demand needs to be a spatial unit capable of responding to housing demand. Therefore, this chapter sets the estimation unit for housing demand at the district level and attempts to estimate housing demand by region.
The most influential variables on long-term housing demand at such a small geographical scale are demographic. Among these, it is essential to base the demand estimation on the household, which is the primary unit of housing consumption. Moreover, when analyzing housing, which has the characteristic of location fixity, it is necessary to consider locality and neighborhood effects. Accordingly, the housing demand estimation model presented in this chapter is a household-based Geographically Weighted Regression(GWR) model that accounts for spatial heterogeneity. Using this model, we estimated the housing demand in Seoul for the next 20 years.
In Chapter 5, based on the future housing demand for Seoul estimated in Chapter 4, regions with similar demand were spatially categorized, and their temporal variability was evaluated to examine future directions for housing policy. As geographical adjacency is particularly important when segmenting housing markets, this chapter attempts spatially constrained clustering. The analysis indicates that Seoul’s housing demand regions are expected to change dynamically in the short-term, showing differentiated patterns such as areas of rapid demand growth or sharp decline. However, in the long-term, as the patterns of housing demand change begin to stabilize, the demand region clusters are projected to maintain a more stable form. These findings imply the need for housing policies that are differentiated by time and region.
This study proposes a new model for projecting future household changes in small area and utilizes these projections to estimate future housing demand and create a demand-based regional typology. The findings from the case study are significant in that they empirically demonstrate the spatio-temporal changes in housing demand. Furthermore, the small area household projection results are expected to serve as a crucial basic dataset applicable to a wide range of fields in addition to housing market analysis.