With a social network adjacency matrix constructed from the Facebook Social Connectedness Index (SCI), this paper examines whether social learning facilitates climate risk perception updates to inform climate adaptation. We find that Hurricanes Harvey and Irma‐induced regional flooding increased flood insurance policies nationwide to the extent of each county’s social network proximity to the flooded areas, with a corresponding update in climate risk perception. Social learning resulted in an additional 250,000 policies in flooded counties and 81,000 policies in unflooded counties over 3 years. We find evidence of the salience effect but no support for adverse selection or over‐insurance.

The demand for voluntary individual lifetime annuities is low, as merely 10% of soon-to-be retired Canadians care to buy such contracts. To assess the reasons why, we design a stated-preference experiment in which we vary characteristics of annuity contracts to estimate individuals’ sensitivity to an annuity’s money’s worth (that is, the value-to-cost ratio). Using different measures of longevity risk and survival expectations, we investigate how knowledge of annuity products and mortality risk misperceptions affect the take-up and the sensitivity of the demand for annuities. We find that annuities are objectively actuarially neutral in general (meaning that annuity premiums are equal to their expected payment), and can appear to offer great value for the money given an individual’s subjective mortality risk. We also find that demand is somewhat price-inelastic so that lowering the price of annuities could increase demand by at most 2 percentage points for a base of 10%. Lack of knowledge of annuities explains another 0.8 percentage points. We find limited additional interest for deferred annuities compared to immediate annuities, although respondents are less sensitive to deferred annuity prices.

We report evidence from the largest study of racial price differentials in the U.S. housing market, constructing a panel of 40 million repeat-sales transactions. We find that price premiums facing Black and Hispanic homebuyers are ubiquitous and systematically higher in neighborhoods with a larger share of non-white residents. We find that non-white buyers purchase at a premium from sellers from a different group. Consistent with predictions from theoretical models (Becker, 1957), we find higher premiums in supply-constrained markets. Leveraging exogenous variation in racial segregation, we find that segregation increases price premiums paid by Black homebuyers.

Persons of color are more likely to live in and move into high flood risk areas. African American and Hispanic individuals also tend to pay significantly more than whites for equivalent housing and there is significant spatial heterogeneity in these differentials. I measure the impact of flood zoning on racial housing price differentials using a panel of 26M repeat-sales transactions across the United States and flood maps from 1999, 2011 and 2022. I estimate price differentials by combining a repeat-sales model with plausibly exogenous changes in flood zone status over time. I find that, while persons of color pay over 3% more than white buyers for equivalent housing outside flood zones, these premiums are reduced to approximately 1% inside flood zones. Where flood risk is most salient, premiums for Black and Hispanic buyers to live in “safe” areas outside flood zones are highest, reaching close to 5%.

This study investigates the impact of flood insurance on neighborhood recovery following flooding events. Utilizing a shift-share instrumental variable approach, we assess the influence of flood insurance claims on housing prices in the aftermath of Hurricane Harvey in the Houston area. Our analysis reveals that flood insurance claims at the census-tract level have a significant protective effect on housing prices. For every $1,000 increase in claims per single-family household in a census-tract, housing prices increase by 1.4%, which is equivalent to $3,080 given the median housing value of $220,000 in the sample. We find that while the median flooded homeowner in affected tracts experiences a 1.7% decrease in housing values after a flood, homeowners in neighborhoods at the 75th percentile of flood insurance claims see no such decrease. Additionally, we observe substantial positive spillover effects of flood insurance claims on the prices of nearby uninsured homes. We further explore potential mechanisms driving these outcomes and uncover suggestive evidence that, post-Harvey, homes listed for sale in well-insured tracts are less likely to be foreclosed properties, more likely to have done home remodeling, and tend to command higher listing prices.

Modeling the dynamic transmission of climate disaster information during these events is vital for effective disaster preparation, response, and recovery. This study uses Hurricane Ida as a case study and Twitter (currently branded as X) as a social sensor to construct a disaster information network that captures climate disaster information transmission across regions. Analyzing the co-movement of region-specific hurricane-related tweet trends that reflect unobservable disaster information in each region, we decompose the forecast error variance and covariance matrix to develop a weighted and directed information network adjacency matrix. We identify 79 sub-state regions with high flood risk but low hurricane information indegree. While most disaster information exchange occurs within each region, cross-region exchange closely aligns with snapshots of the Facebook friendship, SafeGraph mobility, and pre-existing migration networks. Notably, it diverges from the baseline Twitter communication network, highlighting the vital role of social-emotional connections in disseminating disaster information compared to other types of information. Disaster location and local population size predict disaster information indegree and outdegree, and the pre-existing migration network mediates the spread of disaster information. The disaster information network exhibits a higher proportion of unexplained variations compared to other social networks, underscoring the importance of collecting supplementary real-time data through social sensing.