GTX-A 정차역 소재 시군구의 토지시장 가격 변화에 관한 연구: 시계열 예측모형 비교를 중심으로

초록

This study empirically examines land price changes associated with accessibility improvements following the opening of the Unjeong Jungang–Seoul Station section of GTX-A (Metropolitan Area Express Railway Line A) across municipal districts containing GTX-A stations with different positions in the urban hierarchy. Monthly land transaction data comprising 171,678 observations collected between January 2005 and December 2024 (240 months) were analyzed for five municipal districts containing GTX-A stations: Paju (Unjeong Jungang Station), Goyang Ilsanseo-gu (Kintex Station), Goyang Deogyang-gu (Daegok Station), Seoul Eunpyeong-gu (Yeonsinnae Station), and Seoul Jung-gu (Seoul Station). Four time-series forecasting models— ARIMA, SARIMAX, Prophet, and LSTM—were applied and compared. To isolate changes in locational value associated with accessibility improvements, land prices were deflated using the Consumer Price Index (CPI; 2020 = 100), thereby controlling for inflationary effects. Accessibility variables reflecting travel-time changes before and after the opening of GTX-A were incorporated into the forecasting models, and the predictive performance of the extended models was compared with that of the baseline models. The results indicate that the effects of accessibility improvement differed according to urban functional characteristics and hierarchical position. The municipal district containing Daegok Station, a major multimodal transportation hub where metropolitan, urban, and conventional railways intersect, exhibited the largest short-term increase in land prices (+32.0%), whereas the municipal district containing Kintex Station showed a relatively modest increase (+4.5%), possibly reflecting the structural characteristics of a mature first-generation new town. In several districts, the extended LSTM models exhibited severe overfitting after the inclusion of accessibility variables. This result is interpreted as a consequence of reduced model generalization caused by distribution shifts in exogenous variables between the training and validation periods, rather than evidence of nonlinear market responses. Overall, the findings suggest that land market responses to metropolitan railway investment are influenced not only by accessibility improvements but also by pre-existing urban structure, functional characteristics, and local market conditions. These findings provide empirical evidence supporting the need for spatially differentiated land-use and urban policy strategies in response to large-scale metropolitan railway investment.

키워드