Robert Litterman

Overview

American economist; former Chief Economist and head of the Quantitative Resources Group at Goldman Sachs; earlier career at the Federal Reserve Bank of Minneapolis (1982–1994) where he developed the Bayesian VAR approach to macroeconomic forecasting. Best known for the Minnesota prior (1986), for co-authoring the Black-Litterman portfolio model (1992), and for the Knez-Litterman-Scheinkman (1994) factor model of money market returns.

Key Contributions

• Minnesota prior (1986b): Original Bayesian prior specification for VARs. Prior standard deviations on lag $\ell$ of variable $j$ in equation $i$: $\sigma_{ij\ell} = \lambda/\ell$ (own lag) and $\lambda\theta(s_i/s_j)/\ell$ (cross lag); $\theta = 0.2$ default; $s_i$ from univariate AR residuals. Ridge-type estimator. Five-year real-time forecast comparison vs. DRI, Wharton, Chase: BVAR closest to actual 34.8% of 1,604 forecasts.
• BVAR address (1986a): ASA proceedings piece framing overparameterization as a problem of unrealistic implicit priors (exact zero vs. flat); introduced concept of hyperparameter for overall tightness; presented live 1985 GNP forecast.
• Minneapolis Fed 47-variable model (1984): Large-scale monthly BVAR with ~8,000 coefficients; 8 sectors; demonstrated feasibility of Bayesian estimation at scale.
• Specifying VARs for Macroeconomic Forecasting (1984): Staff Report 92 (FRB Minneapolis), published in the de Finetti festschrift (1986). Sequential four-step specification search for π₁ (cross-variable tightness), π₂ (random walk prior), π₃ (variable-specific weights), π₄ (time-varying parameters); log-det as fit criterion; Kalman filter updating. The canonical visual schematic of the Minnesota prior structure.
• Doan-Litterman-Sims (1984): Extended BVAR to allow time-varying parameters and richer prior structures; introduced conditional projection; survey of five years of development.

Related

• Minnesota Prior
• Sims-Zha Prior
• Vector Autoregression
• Christopher Sims