Residential Marketing Duration: Film Studios as Neighborhood Sales Accelerators

Abstract

One measure of residential market liquidity is the speed at which houses sell. Most research about listing duration focuses on agent marketing activities, while external neighborhood changes that could influence a house sale are usually only examined in terms of their price impacts. This paper considers how opening a cultural industry facility influences the likelihood of sale and marketing duration for houses in surrounding neighborhoods. We employ a probit model to quantify the effect of opening large film studios in the Atlanta metropolitan area on the probability of sale and a hazard model to explore the influence on marketing duration in nearby neighborhoods. A difference-in-differences approach allows us to compare market results before and after the studio openings. We find that the likelihood of sale increases while marketing duration decreases for houses located within 1.5 mi of a large film studio once it opens. Thus, the film studios exert positive externalities on the surrounding residential neighborhoods. The positive response may be attributable to film industry workers attracted by convenience and industry buzz as well as other house buyers who are excited about being associated with the film industry or believe the presence of the studio generally improves the neighborhood.

Keywords:

• Housing market externality
• film production studios
• marketing duration

Notes

1 We calculate the residuals from a hedonic equation predicting list price based on the house and neighborhood characteristics listed in Table 1 as well as time (quarter and year of sale) and location (census tract) controls following Anglin et al. (2003). DOP is measured as the percentage deviation of the actual list price (LP) from the expected typical list price for a house with the house (H), neighborhood (N) and fixed conditions (F) of the subject house. DOP is calculated as log(LP) - E(log(LP); H, N, F).

2 We considered the possibility of announcement effects in advance of the opening date used in the analysis, but a specific announcement date is difficult to determine because the future opening of the studio may have been announced via different media and outlets across a range of time and we cannot accurately assess all of these. In addition, during the time that some film studios were known to be moving to the metropolitan area, news media reported multiple sites being considered and some sites were reported and then changed.

3 A simple probit DiD model can be presented as follows. Let the conditional probability that Sold = 1 be expressed as a function of the following:

βX+ β₁Post +β₂Treatment + β₁₂(Post x Treatment). We can write the nonlinear DiD model as follows:

P(Sold = 1|X)=F(βX + β₁Post + β₂Treatment + β₁₂(Post x Treatment)) and

P(Sold = 1|X, Treatment = 1, Post = 1)= F(βX + β₁ + β₂ +β₁₂)

P(Sold = 1|X, Treatment = 1, Post = 0)= F(βX+ β₂)

P(Sold = 1|X, Treatment = 0, Post = 1)= F(βX + β₁)

P(Sold = 1|X, Treatment = 0, Post = 0)= F(βX)

The parameter β₁ allows the linear index (and hence the P(Sold = 1|X)) to be different for all subjects in the posttreatment period compared to the pretreatment period. β₂ allows the linear index (and hence the P(Sold = 1|X)) to be different for treatment subjects compared to control subjects. β₁₂ allows the linear index to be different in the posttreatment period and hence the conditional probability that P(Sold = 1|X) to be different over and above the difference attributable to the nonlinearity of the model for subjects in the treatment group versus the control group. It is the additional difference in the differences that provides a measure of the treatment effect on the treated.

4 Data available at: huduser.gov/portal/datasets/usps.html.

5 Studio 1 opened in 2014. Built on open land south of metropolitan Atlanta, by 2017, it contained 361,000 sf of sound stages, 175,000 sf of office space, and 185,000 sf of workshop space. Studio 2 opened in 2015 after constructing a 200,000 sf studio and post-production facilities on a former military post near the city center. The first phase of Studio 3 included 130,000 sf of sound stages, 60,000 sf of office space and 50,000 sf of mill and flex space when it opened in 2016 on the site of a defunct mall. Studio 4 opened in 2017 with 200,000 sf of sound stages and 650,000 sf of warehouse, mill and office space.

6 For properties that are ultimately sold, days-on-market, DOM, is measured as the number of days from the listing date reported in the MLS to the date a sales contract is executed and the property is removed from the market. The DOM for houses that went off the market without a sale is calculated as the number of days between the initial listing date and the day the property went off the market. We limit to 365 days following Gilbukh and Goldsmith-Pinkham (2021). Review of data from CoreLogic, NAR, and Zillow indicate that the median and average number of days of the market was well below 365 days during this period.

7 The spatial definition of the treatment area is unknown a priori. In addition to our main model, we tested a range of distances for the treatment area. The marginal effect of a 0.5 mi buffer indicator suggests that houses within 0.5 mi of the future megastudio location are less likely to sell than houses in a control area 2 to 4 mi away. However, the marginal effect of the interaction variable suggests that houses within 0.5 mi of the megastudio had a higher probability of sale than baseline after the studio opened. Comparable results are found for further spatial buffers around megastudios until we examine houses between 1.5 and 2 mi. These houses experience a small, but statistically insignificant increase in the probability of sale after the studio opening relative to the 2 to 4 mi control group. Results are available upon request.

8 The log-rank test is a nonparametric test that allows between-group comparisons of survival estimates without making any assumptions about the shape of the survival curve or the distribution of survival times. The null hypothesis is there is no difference in the probability of an event at any point in time among the populations. The test statistic approximately follows a χ² distribution in large samples (Kleinbaum & Klein, 2012).

9 We do not report coefficients because coefficients measure the effect of a variable on the log-odds of the outcome, not on the probability. Furthermore, the interaction terms make the odds ratio interpretation even more challenging (Ai & Norton, 2003).

10 Allen et al. (2021), Benefield and Sirmans (2009), Brastow et al. (2012), Goodwin (2019), Johnson et al. (2007), Rutherford et al. (2018), and Waller and Jubran (2012).

11 Allen et al. (2021), Brastow et al. (2012), Rutherford et al. (2018), and Waller and Jubran (2012).

12 Log-rank tests indicate significantly different life curves for each buffer area pre versus post studio opening. Each of the four studio treatment areas experienced a significant decrease in marketing duration after studio opening.

13 If the hazard rate for sale changes over time, a Weibull distribution may be an appropriate function for marketing duration and is commonly used in real estate studies. We use the Weibull distribution to test our baseline results as a robustness check. We find the same significant relationships with marketing duration as with the Cox proportional hazards regression.

14 By using hazard estimation, we can consider both sold and unsold houses, thus avoiding problems of sample selection bias and censoring based on whether the house sells.

15 Allen et al. (2018), Anglin et al. (2003), Benefield and Sirmans (2009), Chen and Rutherford (2012), and Shi and Tapia (2016).