David Emanuel Andersson

Railroads and Accessibility (THSR Edition) January 16, 2009

Posted by David in Economics, Life in Taiwan.
Tags: HSR, Oliver Shyr, Taiwan
add a comment

Over the past year, I’ve been cooperating with Oliver Shyr of National Cheng-kung University on a study of the effects of Taiwan’s high-speed rail (HSR) on urban property prices. Our goal is to compare the effects of HSR accessibility in all seven affected metropolitan regions. So far, the effects have been contradictory: substantial impacts in the north, no or negligible impacts in the south.

 

I’ve become quite skeptical about the HSR - the world’s most expensive build-operate-transfer project. The investment cost was US$15 billion, or about 5 percent of Taiwan’s GDP. In relative terms, this would correspond to spending the entire US bailout package (US$700 billion) on railroads. True,  it’s better than bailing out bankrupt firms, but still.

 

What follows is a summary of a future paper of ours. It concerns two regions: Hsinchu in the north and Tainan in the south. The empirical observations for the other regions are still being collected, but preliminary estimations indicate that Hsinchu is as representative of the north as Tainan is of the south. Central Taiwan (Taichung) so far seems to be in an intermediate position.

 

 

High-Speed Rail Accessibility and House Prices: Different Cities, Different Effects

David Emanuel Andersson

Oliver F. Shyr

 

Summary

 

A high-speed railway line (HSR) that was opened in early 2007 connects seven metropolitan areas on the west coast of Taiwan. We estimate and compare implicit prices of HSR station accessibility using hedonic price functions for the residential property markets in the Hsinchu and Tainan metropolitan areas. Control variables include three additional accessibility attributes as well as structural and neighborhood attributes. Both analyses use  the same initial specifications and the same functional forms:  four pre-specified and two Box-Cox-transformed hedonic price functions.

 

The estimated functions show that HSR accessibility is regionally differentiated. HSR accessibility has a substantial impact in the Hsinchu region. The estimated effect testifies to the economic importance of the Taipei-Hsinchu link for the Hsinchu Science Park. It has a greater impact on house prices than either the distance to Hsinchu’s traditional downtown area or the distance to the science park itself. All three distance effects are however statistically significant and economically non-negligible. The empirical results lend support to the notion that Hsinchu’s downtown and high-tech nodes have been transformed into outlying sub-centers in an enlarged metropolitan region. HSR accessibility can then be seen as a proxy for accessibility to Taipei’s central business district. Since the new high-speed rail link has resulted in a substantial reduction of the time distance between Taipei and Hsinchu, it seems that high-speed rail investments may contribute to urban enlargement processes.

 

In the southern region of Tainan, HSR accessibility has at most a minor effect on residential property prices. Estimated HSR accessibility effects in the Tainan region imply low or negligible distance elasticities. The effects are also less robust then in Hsinchu, and not statistically significant for all functional specifications.

 

The difference between Hsinchu and Tainan regarding HSR accessibility reflects two mutually reinforcing factors in the Tainan region: low intra-regional accessibility of its HSR station and an economic structure where knowledge-intensive services carry much less total and relative weight than in Hsinchu. In Taiwan, only the northern corridor from Taipei to Hsinchu exhibits consistently high land prices and positive population growth. Given the high investment cost of Taiwan’s high-speed rail line, a policy implication is that the investment may have been too costly from a southern Taiwanese perspective. The limited impact probably reflects the absence of a complementary restructuring of the southern economy. So far, the southern economy has shown no signs of the spontaneous innovative processes that are necessary for a self-generated economic transformation, nor is the region important as a center of scientific research or for the production of other knowledge services.

 

Table 1 presents the results for the Hsinchu and Tainan region of the best-performing functional forms; the log-linear and simple both-hand side Box-Cox models, after exclusion of one (Hsinchu) or two (Tainan) insignificant variables. The symbol λ refers to the Box-Cox transformation parameter, implying that it is very close to a log-linear function in the case of Hsinchu. The other estimated functions do not perform as well in terms of log likelihood or R squared as the ones shown in the table. However, almost all of the effects are quite robust across specifications, with the exceptions of HSR accessibility in Tainan, building height, and freeway accessibility (the latter is excluded from both the Hsinchu and Tainan equations that correspond to Table 1).

 

Table 1: Hedonic price functions for the Hsinchu and Tainan metropolitan areas (t-values in parentheses)

Variable	Hsinchu: log-linear	Hsinchu: BS Box-Cox	Tainan: log-linear	Tainan: BS Box-Cox
Constant	.76	.76	1.56	2.13
Floor area	.53** (19.6)	.53** (14.3)	.55** (26.9)	.65** (21.5)
Lot size	.44** (21.2)	.45** (13.5)	.47** (19.4)	.63** (14.3)
Age	-.04** (-5.7)	-.04** (-5.4)	-.12** (-19.7)	-.18** (-12.9)
Height	.59 (1.4)	.12** (3.5)
Shop/dwelling use	.13** (2.9)	.14** (2.9)	.14** (6.5)	.20** (6.6)
Street frontage	.11** (6.7)	.11** (6.3)	.19** (11.7)	.25** (10.6)
Road width	.05** (3.0)	.05** (3.0)	.10** (6.3)	.12** (6.5)
Commercial zone	.61** (13.5)	.61** (9.4)	.33** (6.5)	.40** (6.1)
Residential zone	.28** (16.3)	.28** (11.7)	.23** (5.9)	.25** (4.9)
Education	.16** (3.9)	.16** (3.9)	.45** (10.4)	.739** (7.7)
Distance to CBD	-.06** (-4.1)	-.06** (-4.0)	-.11** (-7.0)	-.17** (-6.7)
Distance to HSR	-.18** (-9.0)	-.18** (-6.9)	-.04* (-2.0)	-.05* (-1.9)
Distance to science park	-.09** (-5.5)	-.09** (-5.2)	-.08** (-3.6)	-.06** (-3.0)
λ		.001		.168
Adjusted R2	.85		.80
Log likelihood		-1375.7		-2453.7
N	846	846	1550	1550