Diamond et al. (2019)

The economics of Housing & Homelessness

A Class Presentation of

The paper's motivation

What do you think about the way the paper motivates itself?

Certain states are considering repealing laws which limit a cities ability to pass/expand rent control

Paragraph 1:

There is a lot of economic research that highlights the negative consequences of rent control

Paragraph 2:

Maybe the fact that tenants cannot insure themselves against future rent increases is one reason we should consider rent control?

Paragraph 3:

#1

Rent Control Model

S
D

Rent Control

q^*
q^r

What Additional Insight Can We Gain

  • We find that between five and ten years after the law change, the beneficiaries of rent control are, on average, 3.5 percentage points more likely to still remain at their 1994 address relative to the control group

#2

Understanding the Magnitude

{\color{green}\checkmark}

Understanding how the Long-term Effects Differ from Short-term

{\color{green}\checkmark}
  • Rent Control Increase Gentrification
  • Landlords’ substitution toward owner-occupied and newly constructed rental housing not only lowered the supply of rental housing in the city, but also shifted the city’s housing supply toward less affordable types of housing

Applied Econometrics

Spillover Effects

  • The authors highlight that the two papers most similar to their work explore the impact that rent control has on never-controlled units

Why is the author emphasizing the manner in which the 1994 law was passed?

We exploit an unexpected 1994 law change that suddenly rent-controlled a subset of San Francisco buildings and their tenants, based on the year each building was built

Built Before 1979

Yes

No

Building with 5 or more units

Yes

No

Rent Control

(1979)

Rent Control

(1994)

Treated

Control

I am surprised that they include units up until 1990 in the control group

San Francisco

Pre-1994

San Francisco

Post-1994

Reduction in Supply of Housing

Decontrol studies focus on housing that remains under rent control

Identification 

Treatment is assigned in a single time period at the individual level

Intuitively, it feels wrong to control for individual level fixed effects

Local variation of treatment

Controls

Zip Code

Length of Residency

Year of Move-in

Check for Balance

Pre-treatment Differences

Not sure why they include this

This is likely why they keep buildings built up to 1990

Estimated Effects

Y_{izst} = \delta_{zt} + \alpha_i + \beta_tT_i + \gamma_{st} + \varepsilon_{it}

Regression Model

Individual Level Fixed Effects

Indicator for Moved

Rent Control Indicator

Date of Move in by Time Fixed Effects

Tenancy by Year Fixed Effects

Applied Econometrics

Nonparametric Identification

You should in practice be able to run a separate model for each treatment effect

Y_{i} \perp D_i \vert X_i

Conditional on Zip Code, Year of Move-in, and Length of Tenancy, Rent Control is as good as randomly assigned

\beta_1 = \mathbb{E}_x\big[\mathbb{E}[Y_i \vert X_i = x_j, D=1] - \mathbb{E}[Y_i \vert X_i = x_j, D=0]\big]

Pause for Concern?

Zip Code is quite large, right?

How different are these buildings in the treated and control groups? What are the monthly rents? What is the income of the tenants?

1
2

This is not difference-in-difference

3

"As further evidence of random assignment, we see no pre-trends leading up to time of treatment."

Pre-trends aren't meaningful in this context because the outcomes are of a different nature

(Magnitude of effects drop with census tract level fixed effects)

Results

Results

Heterogeneity

"The results above strongly suggest that while tenants value and take advantage of the protections offered by rent control, landlords actively take steps to reduce the burdens of the law, especially in those areas in which it would be most profitable to do."

Over the next few slides, do you agree with the following take-away?

Heterogeneity

"This result suggests that landlords actively try to remove tenants in those areas where rent control affords the most benefits, i.e. high price appreciation areas."

The regressions don't control for census level attributes

Parcel and Landlord Effects

This effect could also be explained by less overcrowding over time

Why is this effect delayed?

Why would you convert new buildings?

Do these results conflict with the tenant level regressions?

It's surprising that a paper on rent control doesn't make use of rent data and therefore doesn't show that the rent control measure is binding

1

Zip Codes seem too broad a control in this setting. Census tracts are available. I'm not sure why the authors don't include census tract fixed effects in their main specification (they show in one regression that census fixed effects make the treatment effect smaller)

2

The time-frame seems to broad

2

Treated

1900
1979

Control

1980
1990

Overview

Motivation

Bad

Misallocation of housing

Negative Spillover Effects

Physical Neglect

Good

Local insurance against price increase

In the medium to long term the beneficiaries of rent control are between 10 and 20% more likely to remain at their 1994 address

1

Effects of rent control on tenants are stronger for 

2

Minorities

Older Households

Longer tenancies

Effects of rent control on tenants are negative for 

3

Areas with rising housing prices

Results

What catches your attention?

Impacted landlords reduced the supply of available rental housing by 15%

4

What's missing from these results?

Migration Chains

"In addition, the conversion of existing rental properties to higher-end, owner-occupied condominium housing ultimately led to a housing stock increasingly directed towards higher income individuals. In this way, rent control contributed to the gentrification of San Francisco, contrary to the stated policy goal. Rent control appears to have increased income inequality in the city by both limiting displacement of minorities and attracting higher income residents.

Mast 2019

clarifying questions

This law capped annual nominal rent increases to 7%

Is this rent control?

Is the data this high quality?

The dataset provides the exact street address, the month and year in which the individual lived at that particular location, the name of the individual, and some demographic information including age and gender.

 

This captures only a subset of owners, right?

By comparing last names in Infutor to the listed owners of the property in DataQuick, we are able to distinguish owners from renters.

 

Sample Construction

Is 6 years of residence on avg relatively high compared to the broader U.S. in this time period?

Why does the main sample include units built in 1900? 

Can we see the age distribution of units?

Why not exclude owners?

Notes for Researchers

Since property can be owned in trusts, under a business name, or by a partner or spouse with a different last name, we expect to underclassify residents as owners.

Classifying Ownership

Presentation of Diamond et al. (2019)

By Patrick Power

Presentation of Diamond et al. (2019)

  • 149