Analyzing Dynamic Connectedness in Korean Housing Markets

ABSTRACT

This study investigates regional housing market connectedness among the 16 first-tier administrative divisions in Korea and 25 districts in Seoul, the capital city. Time-varying parameter vector autoregressive model is used to capture time-varying nature of Diebold and Yilmaz (2014) connectedness network. Rapid increases in connectedness during the sample period are mostly associated with housing booms rather than downturns. The connectedness cycles for the whole country and for Seoul seem to diverge after the global financial crisis. During the 2006 and 2018 connectedness surge episodes, when housing booms were driven by the Seoul metropolitan area, Seoul and the surrounding Gyeonggi province had a strong influence on the whole country network. However, their impact was much weaker in 2010–2011 when the housing boom arose outside Seoul. The influence of Gangnam-3 districts in Seoul’s connectedness network is low overall, but tends to lead the total connectedness index by a few months.

KEY WORDS:

• Diebold–Yilmaz connectedness index
• Korean housing market
• time-varying parameter vector autoregression

JEL:

• E44
• E58
• R31

Acknowledgments

We are grateful to Hyunju Kang and other seminar participants at the Korea Capital Market Institute and Bank of Korea, Euna Cho, two anonymous referees and KB Financial Corporation. This work was supported by INHA UNIVERSITY Research Grant.

Supplementary material

Supplemental data for this article can be accessed on the publisher’s website.

Notes

1. An overview of the general economic significance of housing markets in Korea is provided by Kim (2004).

2. See Kilian and Lütkepohl (2017) for a more detailed explanation.

3. Considering we use monthly data, we choose the VAR lag length to be at least three.

4. Although we do not report the results in this paper, the key results remain unchanged if we extend the forecast horizon to 12 months.

5. Because the value of the forecast error variance decomposition can be affected by the number of variables in the VAR model, the magnitude of the total connectedness index itself may not indicate the strength of connectedness. For example, at their peaks, connectedness reaches about 0.77 for the whole country and 0.94 for Seoul. However, one first-tier administrative division in the whole country has 15 other divisions to be connected, while one district in Seoul has 24 other districts to be connected. The connectedness index per connection is therefore about 0.05 (≈ 0.77/15) for the whole country and 0.04 (≈ 0.94/24) for Seoul.

6. We do not report volatility connectedness with these alternative VAR lags in the paper, but the results are in general similar to those for return connectedness.

Additional information

Funding

This work was supported by the Inha University.