We study how racial segregation shaped Manhattan landlords’ incentives, rent-setting practices, and maintenance decisions in the middle of the twentieth century. We construct a novel data set linking Census records to real estate transaction data, property images, and renovation records. We show that landlords operating in Black neighborhoods exercised substantial pricing power and faced weak incentives to incur the costs of maintenance or to limit overcrowding. Rents for Black tenants were more closely tied to household size than to measures of housing quality. Moreover, properties with Black tenants generated disproportionately high rental income relative to their market values. These short-term profit incentives had long-run consequences: properties housing Black residents in 1940 experienced fewer renovations and showed greater visual exterior deterioration by 1980. Landlord behavior thus perpetuated racial divides in economic conditions and housing quality.

We study how the organization of information production---and its response to economic and technological forces---affects informational efficiency, credit allocation, and borrower risk. Using U.S. administrative data linking mortgage applications to loan officers and subsequent loan performance, we show that underwriting facilitated by officers located close to the borrower increases approval rates without worsening ex-post performance or processing speed, but is not always deployed where it is most valuable, because lenders allocate loan-officer labor elastically with respect to local wages. These gains are especially large for observably riskier borrowers. We develop and estimate a model that combines a core information-production problem over latent borrower risk, an endogenous choice over local versus remote underwriting, and equilibrium in mortgage and labor markets. We find substantial baseline credit rationing---up to 15 percent in high-risk segments---with local officers eliminating roughly half of it while also reducing excessively risky approvals. A technology shock that raises the processing productivity of remote officers induces lenders to substitute away from local screening, lowering informational efficiency, increasing excessively risky approvals and expected defaults, and tightening rationing for marginal borrowers despite only modest reductions in interest rates.

Regulated lenders must explain denials and avoid disparate treatment in U.S. mortgages, yet flexible scoring can be opaque. We develop and study a less-discriminatory and transparent machine learning approach that embeds equalized-odds fairness in estimation and decomposes each decision into feature contributions suitable for adverse-action notices. Using HMDA applications, the model preserves performance while shrinking minority-nonminority error gaps and reweighting away from geographic proxies to core underwriting signals (debt-to-income, loan-to-value). Regression discontinuity design around underwriting cutoffs and lender-level tests confirm that the model lowers minority shortfalls at the margin and reduces within-lender disparities.