Common sense would have it that the more accessible a private residential estate is, the higher its value would be. Research on the impact of accessibility on housing prices has used the Mass Rapid Transit (MRT) network as a proxy measure. From what we have sourced from the Internet, academic and commercial research published thus far has focused on two issues. One is how MRT lines affect property values spatially — that is, how private residential values change in relation to their distance from an MRT station. The other is how property prices change in the period straddling the pre- and post-MRT station announcement. Other academic research has used the MRT network as partial inputs to develop hedonic price indices or conduct town planning studies. Many of these studies rely heavily on regression-based techniques.
For this piece, we shall take a different approach and analyse the accessibility measures of major interchanges and end stations in our MRT networks. The study shall lean on the tools for basic network analysis used by sociologists. For this analysis, we will go further and correlate the various degrees of accessibility via MRT networks to rents and then see which measures of accessibility are stronger in expressing a relationship to rents (and, implicitly, capital values).
In this analysis, we will focus on two stages of accessibility — existing (excluding the Downtown Line 2) and future (the final state, when all the current and future MRT stations that have been publicly announced are completed). For rents, we are using non-landed properties located within a 500m radius of the MRT node. The terms “station” and “node” will be used interchangeably.
Before we begin, we will need to understand the importance of nodes: Not all nodes are equally important
Centrality analysis: This is to find out the most important nodes in a network. The three commonly-used measures are degree centrality, closeness centrality and betweenness centrality
Degree centrality
This is a measure of the degree of importance of a node and is determined by the number of nodes adjacent to it. Therefore, the more the number of adjacent nodes is to a particular node, the greater the degree of centrality. For this analysis, a node is taken as either an MRT interchange or the stations at the end of each line. For example, the Jurong East, Dhoby Ghaut and Tanah Merah stations are considered nodes because at these points, other lines are joined to them. Punggol and Pasir Ris are examples of end stations.
For the existing MRT network, in decreasing order of centrality, the stations and rents of non-landed private residential units are plotted in Chart 1a. It shows Serangoon as the node with the greatest degree of centrality, followed by Bishan and Dhoby Ghaut. Intuitively, this is true because in our existing network, these are the MRT stations with the greatest number of adjacent nodes.
Chart 1a
However, when we correlate the degree of centrality with the gross rents compiled in 3Q2015, visually, we find that although the relation between the two does not contradict the logic behind this construct, the fit is rather poor. Also, from a statistical standpoint, the relationship between rents and degree centrality is not significant. This could be owing to the fact that the logic behind measure of centrality penalises nodes such as City Hall/Raffles Place, which lose out to Serangoon in terms of the number of adjacent nodes (interchanges and end stations). Also, the Orchard MRT node is absent because under the present MRT network, this is not an interchange.
Now let us look at the second state, where all the publicly announced MRT stations that are either currently in operation or planned for the future are functioning. The picture changes quite dramatically, with Dhoby Ghaut having the greatest degree of centrality, followed by Outram Park and Paya Lebar. Serangoon is relegated to eighth position.
The interesting observation is that in both states, Jurong East, the much-touted regional centre with excellent links to public transport, does not quite rank high in this measure (see Chart 1b).
Chart 1b
By plotting the monthly rents against the ranked nodes, although we still get a poor fit, the correlation has improved. Nevertheless, it is still not significant to support the hypothesis that the greater the degree of centrality, the higher the rent. This suggests that this measure of centrality with regard to rents may not be applicable to our MRT network.
Closeness centrality
Central nodes are also important because they can reach the whole network more quickly than non-central nodes. It is a measure of how close a node is to other nodes. It is computed by taking the total number of nodes minus one and dividing that by the total number of shortest paths to each and every node in the network. To get the shortest path, we use the least number of interchanges transversed to get from the node in question to the destination node, and we do that for all nodes. The greater the number, the more central a node is.
Visually, the correlation of the closeness centrality with rents looks better than that for degree centrality. For the existing MRT network, the fit is better than that of the previous measure, namely degree centrality. Nevertheless, the statistical test for correlation between rents and closeness centrality is still not significant.
When we expand the MRT network to its final state, we have a slightly different picture. Bishan is no longer the node with the highest closeness centrality. Instead, the honour goes to Outram Park, followed by Orchard and Marina Bay. Unfortunately, the fit has deteriorated and even without the need to conduct any statistical test on the significance of the correlation, one can clearly see that it is almost non-existent. Therefore, as with the degree centrality case, the closeness centrality as a determinant for rents may not be applicable to our MRT network (see Chart 2b).
Chart 2a
Chart 2b
Direct routing
If we reorganise the analysis from that of nodes to the number of direct routes between major employment and shopping centres and also end stations, the ranking changes. To provide an illustration of what we mean by direct routes, say we arbitrarily start at the Pasir Ris MRT station and go from there to the City Hall/Raffles Place (a major employment centre) station using the minimum number of routes. In this case, because the Pasir Ris MRT station is linked directly to the destination by the East-West line, the number of routes is one. If we want to get to Orchard Road from Pasir Ris instead, the number of routes is two, as we have to change lines at City Hall.
Using the closeness centrality measure, the fit is much better than our previous two measures. Nevertheless, on the correlation front, the link between rents and the direct routes between major centres is still weak for three-bedroom units. For two-bedroom units, the relationship becomes significant only if one relaxes the level of significance from 5%to 10% (see Chart 3).
Ignoring statistics for the moment, if we are to use direct routing to determine rental to MRT connectivity ranking, we find the City Hall/ Raffles Place node come out on top of the pecking order. Again, Jurong East does not quite rank high; in fact, it is behind Woodlands.
Chart 3
Number of stations
Thus far, all the above measures do not appear to show any convincing correlation between rents and MRT nodes or direct routes between major activity centres. As a final push (which, we must highlight, runs the risk of data-mining for the sake of discovering a pattern), we did get a significant correlation between rents and the number of stations between major activity centres. The fit is also the best amongst the various measures that we have thus far used.
We have used the closeness centrality measure as the basis of computation. As for the number of stations between the major activity centres, taking the Orchard MRT station as an example, we counted the total number of stations to get there from end points such as PasirRis, Woodlands, the Tuas Link and other activity centres such as City Hall/Raffles Place (see Chart 4).
Chart 4
As this measure gives the best results, it appears to suggest that tenants place more emphasis on choosing an abode that has the least number of stations between them and major interchanges or end stations. This also means the least number of start-stops, which directly translates into minimal travel time. As there is logic behind this correlation, the risk that this outcome is the result of pure data mining is lessened.
What is of interest is that a node like Paya Lebar is currently underperforming. This could be so because rents there are reflective of 3Q2015 conditions and as this hub is still in the process of development and gentrification, rents should pick up once developments such as Paya Lebar Central come to fruition.
Betweenness centrality
The betweenness centrality counts the number of shortest paths that passes through one node from other nodes. Nodes with high betweenness are important in communication and information diffusion. The betweenness centrality is computed as follows: We start of by counting the number of shortest paths from one node to another, then count the number of shortest paths between one node and another that passes through the subject node. For this measure, we compare two nodes, namely City Hall/Raffles Place and Jurong East.
To illustrate how the betweenness centrality is computed for Jurong East, we assume a commuter at Woodlands wants to get to Tuas Link. The total number of shortest paths between the two nodes is one (here, there are no two routes with the same number of paths).The total number of shortest paths that pass through Jurong East is also one. The division of the latter to the former gives one as the answer. By summing up all the ratios for every node pertaining to any pass through at JurongEast, we get the betweenness centrality for it. The higher the betweeness centrality, the more important the node is in terms of communication and information diffusion.
Computing the betweenness centrality manually is a very tedious process, so for this exercise, we restricted ourselves to two nodes for comparison purposes. As previously mentioned, these are City Hall/Raffles City (downtown) and Jurong East. These nodes were also chosen because of the hype surrounding how Jurong East would develop as an important regional centre. We wanted to use this measure to extract information on how Jurong East measures up to the downtown node in terms of it being able to fulfil its function of information infusion and dissemination.
For the City Hall/Raffles Place node, the betweenness centrality measure is 29, while that for Jurong East is 13, clearly showing that the former is far much more important than the latter. This finding is relevant because it appears to throw cold water on the enthusiastic belief that Jurong East, as a regional centre, can carry its own weight as a disseminator of information.
Owing to its much lower level of connectivity, this regional centre is going to find it difficult to come close to our traditional City Hall/Raffles Place node. Therefore, for Jurong East to rise up the ranks, much more has to be done to improve the direct paths between it and other major nodes and to make these paths the shortest ones that pass through theregional centre.
It does not mean that Jurong East cannot become an even more vibrant regional centre. But to turn that into reality may be a very expensive exercise as the capital expenditure needed to have additional MRT networks to improve accessibility to this centre will be gargantuan— something that can only be justified if massive pump-priming works are needed to boost the economy.
Conclusion
This brief analysis of centrality using our MRT network and social network analysis can be applied to quantify the various accessibility measures. However, once we correlate the nodes with the rents of non-landed properties located within a 500m radius of MRT stations, most of the relationships turn out weak or insignificant. Using rents from apartments that are located near stations that are currently operational, the only significant relationship is that of rents with the number of MRT stations between major interchanges and end stations.
In the analysis of movement from major nodes to major nodes, we find that the lower the number of MRT stations from one to the other, the higher the rents. This implies that tenants place emphasis on the minimal number of stop-go interruptions in their travel and not how well connected their MRT station is. This stop-go interruption can also be interpreted as travelling time.
This is enlightening because we have constantly been bombarded by a barrage of talk that if an MRT station is planned near a particular development, prices/rents will go up. It may, or it may not, but if every major housing estate is ultimately going to be served by an MRT station located within 400m of it, then the location factor becomes trivial.
It may not help boost rents much if a property is located at a node from which to get to other major nodes, one has to traverse many stations. A property with comparable physical attributes located further from other nodes (distance wise), but from where the traveller needs to traverse fewer stations to get to the other nodes should command higher rents. For example, two properties are located 8km from Orchard Road. If you need to make stops at 20 MRT stations to get from one property to Orchard Road, but just five stops to reach the same destination from the second property, the latter property will command higher rents.
Given that the traditional North-South and East-West MRT lines have less frequent start-stops, drawing from what we have discovered, could it be that properties located near stations along these lines will continue to command higher rents than those located near the upcoming MRT lines (which are populated with more stations per line)? If the answer is yes, then Orchard Road and City Hall/ Raffles Place nodes can still be expected to fetch the highest rents in future.
Also, properties located along the first MRT networks are also likely to command higher rents than comparable ones located along new stations. It is also interesting to note that the node with upside rental potential is Paya Lebar. Rents around this node are presently underperforming, probably because the area is still in the early stages of development and gentrification.
However, with the makeover of the district and improved accessibility when the final MRT lines join up, the probability of rents improving from the current baseline is high.
Start browsing for apartments/condos near City Hall / Raffles Place MRT station
Alan Cheong is head of research and consultancy at Savills Singapore. He can be reached at alan.cheong@savills.com.sg.
This article appeared in The Edge Property Pullout, Issue 709 (December 28, 2015) of The Edge Singapore.