Homelessness increased dramatically from the 1980s to the 2010s. Concurrent with this was a large increase in housing prices for both owner occupied and rental units. Previous work has shown that both personal and economic factors are strongly correlated with homelessness, but due to self-acknowledged data limitations and lack of natural experiments, few have explored causal roles for either. I will build on this work and test for a causal effect of rents on homelessness. The results will provide new insight into the causes of homelessness and help policy better target interventions to prevent and reduce homelessness.

                I use homelessness data from the 1990 and 2000 Census counts and the 2010 HUD Point-in-Time counts. I combine these with metropolitan area (MSA) characteristics (including rents) from the 1990 and 2000 Decennial Censuses and the 2010 Census ACS; climate data from the National Oceanic and Atmospheric Administration; and de-institutionalization data from the Bureau of Justice Statistics. I combine all of these data into MSAs that are geographically consistent over time.

                Using these MSA-level data for 1990, 2000, and 2010, I first estimate the following fixed effects model, which controls for unobserved characteristics (such as generosity or leniency) of MSAs that are fixed over time.

ln(homelessness_it) =  alpha + beta*ln(rent_it) + gamma*ln(population_it) + lambda*X_it + tau_t + mu_i + epsilon_it

i indexes MSAs, and t indexes each of the three periods. X is a vector of MSA characteristics, and tau and mu capture period and MSA fixed effects, respectively. Homelessness and rent are the key variables of interest. Ideally, rent would measure the cost of the minimal shelter possible. To approximate this, I measure rents at the bottom of the rent distribution in each MSA. Measures of homelessness include sheltered, unsheltered, and total.

                I also estimate instrumental variables (IV) models as robustness checks using a recently developed instrument for rents that measures the share of land around the center of the MSA that is available to build on (Saiz 2010). The idea is that more geographically constrained MSAs have lower housing supply elasticities and thus greater levels and greater increases in rents.

                Using the IV approach, I estimate models separately for 1990, 2000, and 2010 to see if the effect of rent on homelessness has changed over time. Second, I exploit data available only in 2010 to test for effects of rents for various sub-groups of the homeless: family type; length of shelter stay; previous living situation; and a rich set of demographics that includes age, gender, ethnicity, mental illness, substance abuse, veteran status, diagnosed with HIV/AIDS, and victim of domestic violence. Third, I estimate the effect of (instrumented) changes in rents from 1990-2010 on changes in homelessness from 1990-2010. This model with changes is more likely to satisfy the exclusion restriction than the first IV model, as any potential threat would also have to change over time.

                These results will lend empirical support for or against existing theories about the causes of homelessness and thus help better target homelessness policies.