Spatio-temporal local price dependence in a metropolitan housing market: evidence from 64,529 apartment transactions in Busan, Korea

Purpose This study aims to examine spatio-temporal local-price dependence in a metropolitan apartment market. It investigates whether focal transaction prices are conditionally associated with nearby prices observed in past, same-quarter and subsequent temporal positions, while avoiding causal claims about market memory, peer influence or forward-looking anchoring. Design/methodology/approach The analysis uses 64,529 geocoded apartment transactions in Busan, Korea. Sparse spatio-temporal weight matrices are constructed from projected coordinates and 14 quarterly periods. The empirical strategy compares a baseline hedonic model, a hedonic model with quarter and GU district fixed effects and a spatio-temporal SLX specification that includes past, same-quarter and subsequent local-price variables. Robustness checks examine zero-neighbour cases, influential observations, alternative spatial thresholds, residual Moran’s I and stronger inference procedures for the small number of quarter clusters. Findings The spatio-temporal SLX model substantially improves model fit relative to the baseline and GU fixed-effects hedonic specifications. The past, same-quarter and subsequent local-price variables are all positive and statistically significant. The results are robust to data-quality correction, zero-neighbour sensitivity checks, spatial thresholds from 500 to 2,000 m, wild cluster bootstrap inference and leave-one-quarter-out diagnostics. Residual Moran’s I is reduced but not eliminated, indicating that the model captures substantial local spatial dependence while leaving some residual spatial clustering. Research limitations/implications The study has three limitations. Firstly, the estimated relationships are conditional associations rather than causal effects, and unobserved neighbourhood shocks may affect both focal and nearby prices. Secondly, the subsequent local-price variable is an ex post local-trend proxy, not direct evidence of expectations or asking-price anchoring. Future research could combine transaction, listing and asking-price data. Thirdly, the quarterly structure may miss finer micro-dynamics. Future work could estimate submarket-specific models, compare pre- and post-shock periods or use spatio-temporal local-price variables as interpretable features in valuation models. Practical implications For appraisal practice, the results support a more temporally aware use of comparables. Nearby transactions should be assessed not only by geographic proximity but also by whether they occurred in past, same-quarter or subsequent periods. For market monitoring, the findings suggest that spatio-temporal local-price variables can help diagnose local price co-movement and persistent neighbourhood trends, although they should not be read as causal signals. For applied housing-market research, the spatio-temporal SLX approach offers an interpretable alternative to more complex simultaneous spatial models, especially when large transaction data sets make dense-matrix estimation impractical. Social implications The findings suggest that housing prices are embedded in temporally structured local market environments. Nearby prices observed in past, same-quarter and subsequent periods are conditionally associated with focal prices, but these associations should not be interpreted as causal behavioural mechanisms. For society and policy, the results show why local housing-market monitoring should consider neighbourhood-level price co-movement and persistent local trends, not only citywide averages. More spatially and temporally sensitive monitoring can help identify areas where price pressures are concentrated, supporting better-informed appraisal, public communication and housing-market oversight. Originality/value This paper contributes to applied housing-market analysis by providing a scalable sparse-matrix framework for decomposing local price dependence by temporal position in a large transaction data set. The revised interpretation emphasizes conditional spatio-temporal local-price associations rather than distinct behavioural mechanisms. This provides a more conservative and empirically defensible basis for analysing local price formation in dense metropolitan housing markets.