Home » 2013 Categories » 2013 - Technical presentations » Government Investment in Local Public Goods

Government Investment in Local Public Goods

by Michael Rebuck

Prior to the 2008 Beijing Olympics, the Beijing metropolitan government strategically placed enormous investments to increase green space and to improve public transit in southern Beijing. Siqi Zheng and Matthew E. Kahn, The authors of the paper entitled “Does Government Investment in Local Public Goods Spur Gentrification? Evidence from Beijing?” define gentrification as, “taking place when a geographical area undergoes an increase in its quality of private-sector economic activity as shown by rising local home prices, new housing construction and new restaurant openings”.  They find that the investments made by the Beijing municipal government do indeed cause local gentrification. In these areas, richer people have moved in and have continued to spur the gentrification process.  Public investment combines with private investment to synergistically transform the local area.

Their paper simultaneously studies the effects of new public transit (and its access to a central business) and new investments in green space. Several papers have previously examined the effects of these two investments, and the work of Zheng and Kahn contributes to these findings. To test the gentrification claims, Zheng and Kahn utilized the results of three pieces of data to give a more complete analysis of the gentrification. The first piece of evidence comes from real estate prices. Secondly, they study the geographical patterns of new residential projects and restaurant openings. The third piece of evidence focuses on demographic changes by zone. Equation 1 is the following:


This equation uses hedonic regression to examine whether the local infrastructure improvements are capitalized in land price and residential property price. Hedonic regression splits the researched item into its constituent components and uses estimates of their contributory value.  The basic premise is that the price of something is related to its characteristics, and each of those characteristics has some measurable value. In this particular example the unit of analysis is a residential property project j located in zone z in quarter t. (αz, Φt) are used to account for zone and quarter fixed effects and Xjz is used to account for project-specific attributes that are time-invariant. The subscript sb symbolizes the different types of subways (old subway lines, new subway lines, unbuilt subway lines). To account for the suburbanization effect the CBD price gradient is allowed to vary over time. In this case, t counts the number of quarters since 2006Q1 and there is a linear time trend (a x t). Distance to Subway varies over time for a given location. For example, when a new subway line is built, this value will shrink. Distance to Olympics is time-invariant but increases in value as the construction of the Olympic Park nears completion.

The authors hypothesize that the CBD price gradient will be negative, meaning that the further an area is from access to the CBD, the cheaper the land will be. The CBD price gradient will be smaller when the proximity to a nearby old subway stop is included, and perhaps will even become insignificant. The price gradient is also expected to be negative with regard to distance to the Olympics. These are but a few observations that can be intuitively made through an examination of this model. However, there are several extensions of this equation that can be performed. One example, with respect to new subway construction, is that one could determine whether a shift in price happens when construction starts or when construction ends (or both). Equation 2 is the following:


In this example, count regression models are used to study the spatial distribution of new housing supply and new restaurant openings. This is used to determine what areas are attractive to real estate developers. The unit of analysis in this particular case is zone/ quarter for residential projects and zone/ year for restaurants. An increase in density in either is seen as a sign of gentrification. Both real estate developers and restaurants have a strong incentive to locate houses/ stores in areas where there are customers. Negative binomial regressions are used because there are only two possible outcomes, more or less housing/stores.

In the above regression equation, time-fixed effects are accounted for with Φt. The gradient with respect to the distance to the CBD varies by quadrant and changes over time (again the suburbanization effect). b2t  is a linear function of t and therefore b2t=b2×t. Xz is used to account for zone-specific attributes that are time-invariant.  The anticipated results should be similar to problem 1. The difference is that we are now measuring for housing units/ restaurants offered rather than measuring for the price of land/ housing. A negative price gradient with regard to distance from the subway is expected. A negative price gradient with regards to distance to Olympics is also expected. Equation 3 is the following:


The third equation focuses on demographic changes by zone. In the Beijing example it was difficult to obtain micro household data with geographic identifiers. These data constraints deny the authors the opportunity to perform micro-level regressions and they are also unable to control for some household demographic attributes. The evidence therefore should only been viewed as “suggestive.” However, the authors were able to acquire zone-level average annual household income (incomez) and the household head’s years of schooling (eduz). By using a geographic unit (zones), it is possible to undergo parsimonious regressions (simple regressions) to test whether the zones close to the Olympic Park and/ or the new subway stops have shifted towards higher-income residents with higher levels of education.

In the above equation, the dependent variable Yzt is the zone-level average of annual household income (incomez) and household head’s years of schooling (eduz). Quadrant-fixed effects are accounted for by 4 . Y2010 is 1 for the year 2010 and c1 is the average income/schooling growth rate. In this example, it is calculated from years 2007 and 2010 (the years the data is collected). Zheng and Kahn test whether the spatial gradients with respect to the distance to CBD, distance to the closest old subway stop, and distance to the Olympics have changed by including the interaction terms of the distance variables and Y2010 (ex: c3 x Y2010 x Distance to CBDz). In a different example, any two times (for which data has been collected) can be used in order to examine a change over the difference in duration. No interaction term is included this time for Distance to New Subwayzt because it is time-variant. The effect becomes smaller when new subway stop is opened nearby).

The authors obtained their data from a variety of sources. When studying the spatial distribution of new housing construction the micro transaction data they used was obtained through a private relationship with the Beijing Municipal Housing Authority. The time period used is only from the first quarter in 2006 to the fourth quarter in 2008, but this could no be avoided because transaction data prior to 2006 could not be converted into electronic form. Additionally, the authors had to identify whether the housing developed was a state-owned enterprise (SOE) or an auction type, because SOE developers may be able to obtain inside information on urban planning details. Slightly more than 30% of real-estate developers are SOE and they have been shown to buy land leaseholds at slightly higher prices. For data on restaurant cuisine and patronage, the authors constructed their own indicators by using the most famous food guide and review website www.dianping.com. The authors identified 33 chain restaurants (like McDonald’s and Starbucks) that target rich Chinese urban customers. The authors then noted the restaurants’ location and opening date information.

The authors’ research indicates gentrification was caused by the Beijing government’s investments in local public goods. Local home prices increased, developers increased their construction and more restaurants of higher quality opened nearby. Demographic data suggests that high-income and more highly educated households are attracted to areas where government investment in local public goods has been made. These three pieces of evidence support the assertion that government investment and private sector investment serve as complements that can gentrify previously underdeveloped areas.

In conclusion, government infrastructure projects have dramatic effects on local markets. Decisions about where to place these projects have enormous consequences on the surrounding communities. The models introduced in this paper do not have to be used exclusively for subway projects or Olympic facilities. Any number of projects can be substituted with some minor adjustments, and their effects do not always have to have a negative price gradient. For example, the construction of a dump or of a nuclear facility could be reasonably expected to have a very high price gradient. I especially enjoyed this paper because it tested its hypothesis in a variety of ways. It used price data, count data, and demographic data. Through the use of regression analysis the model gives an in-depth look into the issue of gentrification.

The full document can be found here

Works Cited

Zheng, Siqi, and Matthew E. Kahn. “Does Government Investment in Local Public Goods Spur Gentrification? Evidence from Beijing.” Real Estate Economics 41.1 (2013): 1-28. Web.

Leave a comment

Your email address will not be published. Required fields are marked *