Housing Prices and Transportation Improvements Philip A. Viton

March 5, 2014

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

1 / 35

Introduction (I)

Planners propose improvements to the transportation system in order to improve peoples’accessibility. How can we tell whether such projects are successful or not?

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

2 / 35

Introduction (II)

In the case of the monocentric city, the answer is easy. Improved accessibility results in reduced (generalized) transportation costs, and as we’ve seen, reduced transport costs lead to increases in land prices (rents), except at the CBD. So in a monocentric city, a successful accessibility project will result in land prices increasing. But we suspect that many cities are not monocentric; we also have reason to believe (Small and Song) that individuals may not behave according to an optimization model.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

3 / 35

Introduction (III)

However, if improved accessibility is something that people value, then lots located near the improvements should become more valuable, as long as people need to travel to non-local destinations. Thus we suspect that independently of moncentricity or optimization, a successful project that improves accessibility should result in land prices rising near the source of the improvement. This would also include the CBD, unlike the monocentric model. We say that there is an accessibility premium for land. But is there? Does accessibility matter in the real world?

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

4 / 35

Prior Evidence

In the early days of the postwar US highway network (1950s, 1960s) studies showed that being located near major highways increased land values signi…cantly. Later studies (1970s, and some in the 1980s) suggested that the impact was much less, and sometimes zero. This makes sense: given that the existing network was highly accessible, later improvements could be claimed to add only a small amount to the existing high level of accessibility, so we would expect much smaller impacts, possibly even zero. Still, we would like to know what is happening now : if planners add accessibility today, will it increase land values, and if so, by how much.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

5 / 35

The Model (I) We can formulate our question using the language of our regression review. We are interested in what determines housing (land) prices. This is our independent variable. We suspect that land prices are determined by characteristics of the house (property); by neighborhood characteristics; and, crucially here, by a measure of the property’s accessibility to transportation. We will take the transportation improvement to be provision of a limited-access expressway. It then makes sense to use the property’s distance from the new highway as a measure of accessibility. In the case of a limited-access highway, this should presumably be the distance to the nearest access point (highway on-ramp). Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

6 / 35

The Model (II) So our model can be represented as:

price = f (distance, property characteristics, neighborhood characteristics and we will adopt a linear-in-parameters framework. The resulting population model is: price = β0 + βd distance + βpc property characteristics

+ βnc neighborhood characteristics where βpc and βnc are lists (vectors) of coe¢ cients applicable to the lists of property and neighborhood characteristics, respectively. The strategy of “explaining” the price of something by its characteristics is known as an hedonic model. Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

7 / 35

Study Context We will now discuss one empirical study of this issue, by Boarnet and Chalermpong (B+C), see last slide for a citation. Setting: Orange County, in southern California, see map on next slide. During the 1990s this area set up several Transportation Corridor Agencies. These were government agencies whose sole purpose was to oversee adding toll highways to the existing network. During the 1990s this resulted in: Foothills Transportation Corridor Backbone (FTCBB, southern portion of SR 241) : opened in 1993 San Joaquin Hills Transportation Corridor (SJHTC, SR 73): opened in 1993 Foothills Transportation Corridor (FTC, entire SR 241) : January 1999 Eastern Transportation Corridor (SR 133, SR 241, SR 261): February 1999 Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

8 / 35

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

9 / 35

Analyzing the Impact (I)

The fact that the new roads opened at various times during our observation period leads to a complication in the measurement of accessibility. We would not expect distance impacts to be the same for all property sales. For example, a property that sold in 1980 (well before any of these projects occurred) would probably show no impact of distance to something that did not then exist. On the other hand, a property near one of the new roads that sold in 2000 (say) would, if the project was a success, show a positive impact from being near the road.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

10 / 35

Analyzing the Impact (II)

B+C address this problem by de…ning a threshold year: this is the earliest year that the project is anticipated to in‡uence housing prices. Note that when this occurs is to some extent open to debate: it will not necessarily be the year that the project actually opened, as we would expect developers to anticipate improved accessibility, and adjust selling prices accordingly. So an accurate threshold year should re‡ect the time when developers anticipated that the project would become a reality. If for example a project became mired in litigation, then to the extent that this suggested that the project might not end up being done at all, the threshold year would be close to the year the project actually opened.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

11 / 35

Analyzing the Impact (III) B+C then de…ne two measures of distance to a project (road). Each project has its own a Threshold Year, as discussed above. Then for each house sale, we de…ne: DBefore = [Distance to project] [dummy for : sale occurred before threshold year] DAfter = [Distance to project] [dummy for : sale occurred after threshold year]

If the hypothesis that the project was successful in the sense of raising property values is true, then we would expect the coe¢ cient of DBefore to be zero and the coe¢ cient of DAfter to be negative (since after the project has been built, the further away a property is from the project, the lower its price).

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

12 / 35

Methodology (I)

The methodology in almost all property-values studies is essentially the same. Gather data on the prices at which properties sold. Match the sales against a database of housing and neighborhood characteristics. Compute the value of any additional variables that we are interest in (here, the two distance variables). Decide on the functional form of the regression. Run the regression and analyze the results.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

13 / 35

Methodology (II)

There is one caveat to be aware of here. There are two sources that one might use for property values. The …rst is the actual sales price for the property when it sold. The second is the property’s assessed value for tax purposes. This is often computed and published by the local property tax assessment o¢ ce, and is intended to represent a prediction of what the property would sell for, if it sold.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

14 / 35

Methodology (III)

Unless you have no other option, you should always prefer actual sales prices to assessed values. The reason is that if you run a regression with assessed value as the dependent variable, you are in e¤ect studying, not the determinants of the house price, but the decision rule that the assessor uses to construct the assessed value. While in some contexts this may be interesting (you might wonder if the assessor’s evaluations are racially biased, for example), it is not what you need in order to answer the question about the e¤ectiveness of the project in question.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

15 / 35

Data (I) B+C obtained data on all sales in Orange County between 1988 and February 2000. This was 367,841 sales. Note that they had access to actual sales prices, and did not use assessed valuations. After …ltering out missing data, data that seemed to have been entered incorrectly, and non-arms-length transactions, there were 275,185 “reliable” sales. They then focused on the impacts of two of the toll roads: the FTCBB (southern portion of SR 241), opened in 1993 the SJHTC (SR 73), also opened in 1993

These roads were chosen because they were the earliest segments to open and provided the longest before-and-after periods for any impacts to appear.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

16 / 35

Data (II) For the FTCBB, they selected all sales that were closer to the FTCBB than any other expressway. The SJHTC runs roughly parallel to Interstate-5, so a similar selection would probably not be able to isolate the impacts of the toll road. In this case they selected all sales of houses that were between 1125 feet and 2 miles (5280 2 = 10 560 ft) from an SJHTC on-ramp. In order to check that their data-munging made sense, they also looked at the impact of SR 22. This was a road that had no major improvements over the period in question. We would therefore expect to see no impacts. For this road, the authors constructed distance-impact variables using the same criteria and Threshold Year as with the SJHTC : year 1993, and sales of properties that were located between 1125 ft and 10,560 ft from an on-ramp. Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

17 / 35

Data (III) Data on sales could be matched to individual house characteristics via the County Tax Assessor’s database, since property tax assessments depend on property characteristics. The authors controlled for two neighborhood e¤ects: the crime rate, and a measure of the quality of the local public schools. Because their sales data was for di¤erent years (ranging from 1987 to 1999) they attempted to control for changes in the general housing market via a series of time dummys: Dt is equal to 1 if the sale occurred in year t and zero otherwise. To avoid the Dummy Variable Trap, one of the years had to be omitted. (If you know all but one of the values of the time dummys, then you automatically know the value of the last one). In this case, the authors omitted the dummy for 1999. Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

18 / 35

Data (IV)

House prices were measured in 1982–1984 dollars. (This adjusts for in‡ation over the period, and should always be done when you are looking at a time series of money values). Distances were measured in feet from (the center of) the property to the nearest project (ie either the FTCBB or the SJHTC) on-ramp. Constructing these distances is straightforward using a GIS program.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

19 / 35

Results See supplementary handout for full results (Table 3 from the paper). For the FTCBB, the authors show two possible threshold year choices. I’ll discuss only the 1989 threshold: the results for 1990 are very similar. R 2 values are 0.41, 0.55 and 0.62 for the three roads in question. As noted in a previous handout, for our purposes this is pretty unimportant. We are interested in coe¢ cient values, and not in predicting house sales prices. For a developer on the other hand R 2 could be very important, and on that basis the R 2 values are not terribly impressive: they say that a lot of the variation in housing prices is explained by factors other than the included independent variables.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

20 / 35

Distance Impacts (I)

Here are the results for the distance variables: Corridor Threshold Year Variable DBefore DAfter

Philip A. Viton

FTCBB 1989 Coe¤ t-stat 0.41 0.87 0.87 3.79

SJHTC 1993 Coe¤ t-stat 1.90 1.47 4.49 3.68

CRP 4110 — Housing () & Transportation

SR 22 1993 Coe¤ t-stat 0.05 0.24 0.21 1.09

March 5, 2014

21 / 35

Distance Impacts (II)

Consider …rst the unobserved coe¢ cient (β-value) of DBefore in our population model. It is estimated by the b-value (“Coe¤") shown in the table opposite. Our hypothesis is that it should be zero, since before the threshold year, we expect that distance to the project would not have any e¤ect on prices. In other words, our hypothesis is that DBefore doesn’t matter in the determination of house prices. So we want to test the null hypothesis that the true coe¢ cient (β value) of DBefore is zero, in light of the estimated coe¢ cient (b value).

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

22 / 35

Distance Impacts (III) The test statistic for this hypothesis is the t-statistic (ratio of the estimated coe¢ cient to its estimated standard error); and under the null hypothesis it has a student’s-t distribution. However, given the large number of degrees of freedom, we can use the normal approximation, and our rule of thumb: reject if the t-statistic exceeds 2.0. In this case we see that none of the t-statistics exceed 2.0. So in all cases we fail to reject (ie we accept) the view that DBefore is irrelevant. For sales before the projects were perceived as likely to be implemented, distance to the nearest on-ramp doesn’t matter. This of course is exactly what we anticipated; but it does add plausibility to our modelling approach.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

23 / 35

Distance Impacts (IV) Now consider DAfter. If our view of an accessibility premium is right, then this variable should matter in the determination of house prices. We would therefore like to test whether the β coe¢ cient for this variable is zero or not. We can again test this using the t-test (t-statistic). In this case we reject the null hypothesis that the β-value for DAfter is zero in the case of both the FTCBB and the SJHTC. In both these projects (roads), distance to the road has an impact on selling prices. The point-estimates of the impacts are both negative. This means that as we go further away from an on-ramp, housing prices decline. This con…rms that accessibility has increased housing values.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

24 / 35

Distance Impacts (V)

The magnitude of the decline with distance di¤ers by road. For the FTCBB the decline is 87c/ per foot: for every foot you go away from an on-ramp, housing prices fall by 87c/. Equivalently, housing prices decline by 0.87 5280 = 4593. 60 dollars per mile of distance from the road. For the SJHTC the impact is much greater: the decline with distance is $4.49 per foot or 4.49 5280 = 23707.20 dollars per mile. Finally, note that for SR 22 the coe¢ cient of both DBefore and DAfter are statistically insigni…cant (we fail to reject the null hypothesis that the unknown β-coe¢ cients are really zero). Since this road was included as a control, this is what we would expect.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

25 / 35

Interpreting the Time Dummys (I) The time dummys can tell us something about the overall state of the housing market in Orange County. To see this, consider a speci…c house property, and imagine that it sold every year, and that its characteristics did not change. Then in year t its expected sales price would be: (ignoring the error term, whose mean is assumed to be zero): pt = β0 + βt Dt + E , where Dt is 1 because this is a year-t sale, all the other time dummys are zero, and E is everything else. So: pt = β 0 + β t + E Similarly, for the sale in the next year we have pt + 1 = β 0 + β t + 1 + E Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

26 / 35

Interpreting the Time Dummys (II)

Hence the (expected) 1-year change in the selling price is pt + 1

pt = β t + 1

βt

So the successive di¤erences in the coe¢ cients of the time dummys tell us the (expected) change in the selling price between the two years, holding everything else — all the observed independent variables — constant.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

27 / 35

Interpreting the Time Dummys (III) We can see how this works out using the FTCBB estimates. To do this, remember that the time dummy for 1999 was de…ned to be zero, to avoid the dummy variable trap. Then we get the following pattern: Period 1999–1998 1998–1997 1997–1996 1996–1995 1995-1994 1994-1993 1993-1992 1992-1991 1991-1990 1990-1989 1987-1988 Philip A. Viton

Price Change 15, 181 21, 308 7, 926 52, 722 12, 892 1, 213 8, 510 4, 785 6, 848 23, 574 52, 247

CRP 4110 — Housing () & Transportation

March 5, 2014

28 / 35

Interpreting the Time Dummys (IV)

These results are consistent with what we know about the Southern California housing market from other sources. There was a period of signi…cant prices increases until the end of the 1980s (represented here by the 1987–1988 change). This was followed by a decline in the early 1990s (1990-1994). And followed again by a recovery after the mid 1990s.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

29 / 35

[blank slide]

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

30 / 35

Some Anomalies (I)

Not all the empirical results are as straightforward as the distance impacts. Consider some other determinants of housing prices in the results: Corridor Threshold Year Variable Bedroom Age SAT Score

Philip A. Viton

FTCBB 1989 Coe¤ t-stat 9417.09 7.24 630.81 3.43 636.57 6.42

CRP 4110 — Housing () & Transportation

SJHTC 1993 Coe¤ t-stat 9953.85 3.54 13330.69 3.57 706.65 9.28

March 5, 2014

31 / 35

Some Anomalies (II) These results imply that for both projects, having more bedrooms in your house reduces property values. Surely this is not what we would expect. The authors attempt to explain this (p. 589, note 11) by saying that realtors agree that in Southern California house prices are more in‡uenced by dwelling size than by the number of bedrooms. Personally, I’m not convinced. If this explanation were true we might see a negative coe¢ cient estimate, but we would also expect a low t-statistic, so that we would not reject the hypothesis that the number of bedrooms is irrelevant. By contrast, here we see t-statistics well above 2.0, suggesting that there is a real e¤ect being measured.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

32 / 35

Some Anomalies (III) Next, look at Age, the age of the property, in years. The usual expectation is that the older the property (everything else held constant) the lower its property value. That is indeed what we …nd for the FTCBB houses; but for the SJHTC, the older the property the higher the sales price; and the e¤ect is signi…cantly non-zero (t-statistic above 2). Again, counter-intuitive. Finally, look at the coe¢ cient of SAT Scores, one of the two neighborhood e¤ects. The results here suggest that for the FTCBB houses, being located near a good school actually lowers property values. (For the SJHTC properties, the coe¢ cient has the expected positive sign). Note that we reject the null hypothesis that SAT scores are irrelevant. (We would be less concerned for the FTCBB properties if we saw a negative coe¢ cient but a low t-statistic). Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

33 / 35

Further Questions The authors assumed that the impact of the distance variables was linear (ie the two distance variables entered our population model linearly). One reason we might question this is that being relatively close to a major expressway could generate negative impacts, primarily noise. Because of this, property values could actually increase as we got further away from the road, at least for short distances. We might therefore think that the impact of increasing distance from a road is non-linear (positive up-close, but then negative from a short distance out). One standard way to accommodate this is to include both distance and distance-squared among the independent variables. It would be interesting to see if this would a¤ect the results. Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

34 / 35

References

Marlon G. Boarnet and Saksith Chalermpong. “New highways, house prices and urban development”. Housing Policy Debate, 12 (3) : 575–605, 2001.

Philip A. Viton

CRP 4110 — Housing () & Transportation

March 5, 2014

35 / 35