Knowledge that Transforms

We identify a new and important property of affine term structure models: at long maturities, the slope of the term structure of risk premia becomes “almost observable”. Exploiting this property by fitting models to long-maturity data dramatically improves the precision of estimates of risk premia and expected short rates across the term structure. In a Monte Carlo experiment we show that using rates with maturities up to 25 years, rather than 10 years as is standard in the literature, reduces the standard deviation of estimation errors in risk premia, expected short rates and term premia by over 70%. Applying this approach to US Treasury data, we find that variation in expected short rates accounts for a much larger share of variation in yields (and variation in term premia a correspondingly smaller fraction) than in the widely used estimates reported by the Federal Reserve. Accounting for time-variation in yield volatility is key to successful extraction of the risk premium information in long rates. Our approach therefore departs from much of the empirical literature in two important respects: it incorporates long-maturity yields and allows for time-varying yield volatility, whereas most existing studies ignore long-term yields and rely on constant volatility models.

Using US Supreme Court’s narrow-decision on Shelby v. Holder as a shock that diluted voting rights of Black Americans in covered counties, we document the relationship between political voice and economic decision-making. We show fewer Black Americans buy homes after their de-facto disenfranchisement, reflected through lower mortgage applications. Among those who apply for mortgages, there is a flight to Black lenders. Leveraging individual-level survey data, we show confidence in government and financial institutions plays a significant role in driving these effects. Our findings demonstrate that the effects of disenfranchisement extend beyond the ballot box to fundamental economic decisions like homeownership.

We show that leading investor disagreement proxies have near-zero cross-correlation – making disagreement pricing inferences challenging. We develop a frictionless model that generates overpricing from latent disagreement, jointly rationalizes major proxies, and generates new testable predictions. Equilibrium interrelationships motivate a new disagreement measure, DIS, which is more predictive of returns cross-sectionally than major proxies and more consistently predictive across macroeconomic regimes. We provide evidence that the friction-based channel of Miller (1977) and a frictionless mechanism coexist and are empirically separable. DIS also provides a micro-founded, disagreement-based account of the IVOL puzzle, absorbing IVOL’s predictability while revealing its regime and friction dependence.

We examine how private foundations allocate charitable capital in response to recipient need. Using financial shocks to peer foundations (who fund the same recipients), we isolate exogenous changes in the marginal value of grants. Foundations respond to need by reallocating grants across recipients, particularly for scientific causes and longer donor–recipient relationships. However, their annual grantmaking budgets show little intertemporal flexibility, instead prioritizing capital preservation. Among governance characteristics, founder involvement predicts both cross-sectional responsiveness and intertemporal conservatism. Our findings highlight a tension: founders’ charitable intent supports effective allocation, but their inclination toward capital preservation can hinder impact maximization.

We develop a complete-market production economy in which heterogeneous beliefs about productivity growth affect both asset prices and real activity. Firms must hire workers before productivity is realized, so operating leverage increases the exposure of profits to productivity risk. Because the firm’s discount factor depends on a wealth-weighted aggregation of investor beliefs, shifts in optimism compress risk premia and raise employment, while reversals trigger contractions. This mechanism generates an equilibrium feedback between asset prices, hiring, and risk exposure. A taxonomy of belief systems shows that only extrapolative beliefs produce amplification and endogenous risk build-up. Disciplined by survey expectations, the model jointly matches key asset-pricing, trading-volume, and business-cycle moments.

Large language models are increasingly used in social sciences, but their training data can introduce lookahead bias and training leakage. A good chronologically consistent language model requires efficient use of training data to maintain accuracy despite time-restricted data. Here, we overcome this challenge by training a suite of chronologically consistent large language models, ChronoBERT and ChronoGPT, which incorporate only the text data that would have been available at each point in time. Despite this strict temporal constraint, our models achieve strong performance on natural language processing benchmarks, outperforming or matching widely used models (e.g., BERT), and remain competitive with larger open-weight models. Lookahead bias is model and application-specific because even if a chronologically consistent language model has poorer language comprehension, a regression or prediction model applied on top of the language model can compensate. In an asset pricing application predicting next-day stock returns from financial news embeddings, we find that ChronoBERT and ChronoGPT’s real-time outputs achieve Sharpe ratios comparable to a much larger and potentially biased Llama model. Our results demonstrate a scalable, practical framework to mitigate training leakage, ensuring more credible backtests and predictions across finance and other social science domains.