In Pursuit of Patterns: The Most Idiosyncratic Cities In China

China housing is not one market, but it is not 600 different markets either.

There are some clear patterns in how some cities behave similarly in their direction of the house price growth, and some exhibit very unique behaviour. A finer understanding of these patterns can help investors better assess the risk in city selection strategies for new investments or of an existing portfolio, including diversification considerations.

In our latest note “In Pursuit of Patterns: The Most Idiosyncratic Cities In China”, we quantitatively assess and visualize which cities are most unique and which most similar to each other in terms of their house price movements.

For investors, fund managers, developers and lenders – you can request a full copy of the note by contacting us, and here are some highlights:

  • Shenzhen, Wenzhou, Sanya and Haikou are among the most unique cities in terms of their house price growth correlations, among
    the 70 major cities, for the January 2011 to April 2018 period
  • However, Shenzhen only got such unique status over the recent four years, while the other three have been consistent throughout
  • There are no clear patterns by region except the North East cluster.
  • Tier-1 have become much less correlated to others in the last four years.

To visualize the results, we use a graph – a method that allows us to capture more information about the relationships between the cities:

A line (edge) connecting one city to another means that there is a non-zero correlation between the two cities. The darkness of the edges represents the relative strength of the correlation.

In this case, there are no cities negatively correlated -0.3 or more negative, and therefore the darker the colour of the edge, the stronger the (positive) correlation, ranging from 0.30 (between Tianjin and Xining) to 0.93 (between Nanjing and Hefei).

The distance (or a number of ‘hops’ between nods to get from one city to another) between the cities reflects the relative degree of similarity of the two cities in terms of their correlations with each other and other cities.

Shenzhen and Urumqi is an extreme example. Not only there is no correlation (the way we defined above) between them, but also the cities that they correlate with are very different and not so correlated with each other.

The proximity of city names on the graph suggests that they are more likely to be highly correlated. The larger the size of a city label the greater the total sum of all correlations for that city. The colours of the names of the cities reflect the classification by city-tiers: Tier-1, Tier-2 and Lower-Tier.

This note is part of the ‘In Pursuit of Patterns’ series by Real Estate Foresight – exploring key relationships and patterns in China housing markets via quantitative analysis, aimed at supporting clients in their longer-term and data-driven assessments of risks and city selection in China.