Macroeconomic Uncertainty

Definition

Macroeconomic uncertainty is the time-varying, economy-wide degree to which the future is unpredictable — formally, the conditional volatility of the unforecastable component of economic activity. Following Jurado-Ludvigson-Ng (2015), the $h$-step-ahead uncertainty of a series $y_{jt}$ is $U^y_{jt}(h)=\sqrt{\mathbb E[(y_{jt+h}-\mathbb E[y_{jt+h}\mid I_t])^2\mid I_t]}$, and aggregate macro uncertainty is the common component of these individual uncertainties across many series. The definition deliberately separates uncertainty from the volatility of a series: a series can be volatile yet forecastable, or calm yet unpredictable.

Key Ideas

• Uncertainty is residual, not raw, volatility. The forecastable part of a series carries no uncertainty; only the variance of the genuine forecast error does. Measuring uncertainty therefore requires first removing the conditional mean with the best available predictor.
• Purge the predictable part with factors. The conditional mean is estimated from a factor-augmented / diffusion-index model built on a large panel of macro and financial series, so the information set $I_t$ is far richer than a univariate model's — much apparent "uncertainty" in small models is just omitted predictability.
• Time-varying error volatility via stochastic volatility. The conditional variance of the forecast error is modeled with a stochastic-volatility process (for both the common factors and the idiosyncratic parts), producing an uncertainty series that moves over time.
• Aggregate to an index. Averaging (or extracting the common factor of) the individual $U^y_{jt}(h)$ across series gives an economy-wide uncertainty index at each horizon $h$; longer horizons are smoother and more persistent.
• Distinct from popular proxies. The resulting measure has large independent variation from VIX/implied volatility, cross-sectional dispersion of profits/returns/forecasts, and news-based indices — indicating those proxies capture much besides uncertainty. True uncertainty episodes are infrequent but large and persistent.

How It Works

1. Assemble a large panel of macro and financial series; extract common factors (as in the diffusion-index / dynamic factor model approach).
2. For each series, forecast $y_{jt+h}$ with a factor-augmented predictive model and form the forecast error $V^y_{jt+h}=y_{jt+h}-\mathbb E[y_{jt+h}\mid I_t]$.
3. Fit a stochastic-volatility model to the errors (of predictors and idiosyncratic components) to obtain the conditional error volatility $U^y_{jt}(h)$.
4. Aggregate across series into the macro uncertainty index $U_t(h)$; repeat for horizons $h=1,3,12,\ldots$

Why It Matters

• A clean benchmark for uncertainty-shock theories. Real-options, precautionary-savings, and financial-frictions models all assign a role to time-varying uncertainty; a measure that isolates the theoretical object lets those models be tested rather than proxied.
• Corrects the volatility-as-uncertainty conflation. By showing its index diverges from VIX and dispersion measures, it warns that stock-market volatility is a poor stand-in for economic uncertainty.
• Widely adopted. The JLN index (and its financial and real-activity variants) became a standard input in empirical macro-finance research on uncertainty and the business cycle.

Open Questions

• Whose information set? Uncertainty is defined relative to $I_t$; a richer forecasting model mechanically lowers measured uncertainty, so the number depends on the econometrician's (assumed) information.
• Specification dependence. The measure inherits the factor model's and stochastic-volatility model's assumptions; robustness to these choices is an ongoing concern.
• Filtered estimate. $U_t(h)$ is itself estimated with error that is rarely propagated into downstream regressions of activity on uncertainty.

Related

• Factor Model
• Dynamic Factor Model
• Stochastic Volatility
• Forecast Uncertainty Decomposition
• Volatility Forecast Evaluation
• Forecasting
• Kyle Jurado
• Sydney C. Ludvigson
• Serena Ng
• Jurado-Ludvigson-Ng (2015) Measuring Uncertainty