Uncertainty and the macroeconomy: evidence from an uncertainty composite indicator

ABSTRACT

This article proposes a uncertainty composite indicator (UCI) based on three distinct sources of uncertainty (namely financial, political, and macroeconomic) for the US economy on the period 1985–2015. For that, we use a dynamic factor model, summarizing efficiently six individual uncertainty proxies, namely two macroeconomic and financial uncertainty factors based on the unpredictability, a measure of (micro)economic uncertainty, the implied volatility index, the corporate bond spreads, and an index of economic policy uncertainty. We then compare the effects of uncertainty on economic activity when the UCI is used instead of individual uncertainty proxies in structural VAR models. The interest of our UCI is to synthesize theses effects within one measure of uncertainty. Overall, the UCI was able to account for the most important dynamics of uncertainty which play an important role in business cycles.

KEYWORDS:

• Uncertainty
• dynamic factor model
• economic activity

JEL CLASSIFICATION:

• C38
• C32
• E32.

Acknowledgements

Amélie Charles and Olivier Darné gratefully acknowledge financial support from the Région des Pays de la Loire (France) through the grant PANORisk.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental data

Supplemental data for this article can be accessed here.

Notes

¹ See Bloom (2014) and Bloom, Fernandez-Villaverde, and Schneider (forthcoming) for a comprehensive survey of the literature on uncertainty shocks.

² Most of studies analyzing the impact of uncertainty shocks have employed VAR models, whereas some studies have used Dynamic Stochastic General Equilibrium models, such as Bloom et al. (2012), Christiano, Motto, and Massimo Rostagno (2014), and Leduc and Liu (2016).

³ Bachmann, Elstner, and Sims (2013) compare the effects of four measures of uncertainty related to forecast disagreement, economic policy uncertainty, stock market volatility, and interest rate spreads in Germany and in the US. Born, Breuer, and Elstner (2014) quantify the contribution of various uncertainty shocks to the Great Recession and the slow recovery in the US.

⁴ Knight (1921) established a distinction between risk and true uncertainty. Risk refers to the possibility of a future outcome for which the probabilities of the different possible states of the world are known. Uncertainty refers to a future outcome that has unknown probabilities associated with the different possible states of the world. Note that some of what we call uncertainty may indeed be risk as defined by Knight (1921). Thus, we use different proxies for economic uncertainty, which can differ from Knightian uncertainty.

⁵ Other studies also use the DFM to construct uncertainty proxies but not a composite indicator based on uncertainty proxies as we do here. Jurado, Ludvingson, and Ng (2015), Ludvingson, Ma, and Ng (2015) and Henzel and Rengel (forthcoming) construct uncertainty measures based on common forecast errors (that they associate to uncertainty) on a large number of macroeconomic variables with a DFM. Chauvet, Senyuz, and Yoldas (2015) also construct a common factor on different measures of (realized and implied) volatility from a DFM.

⁶ See Barhoumi, Darné, and Ferrara (2013) for a survey on DFMs.

⁷ Doz, Giannone, and Reichlin (2011) also propose an alternative approach, the so-called two-step approach.

⁸ By analyzing the properties of the maximum likelihood estimator under several sources of misspecifications, such as an omitted serial correlation of the observations or a cross-sectional correlation of the idiosyncratic components, Doz, Giannone, and Reichlin (2012) show that these misspecifications do not affect the robustness of the common factors, particularly for fairly large N and T. More specifically, this estimator is a valid parametric alternative for the estimator resulting from a principal component analysis (PCA).

⁹ We would like to thank the authors to share their data. Others uncertainty measures have been proposed but on a shorter period or a quarterly frequency.

¹⁰ As an alternative to the VXO index, we could have used the newer VIX index, which was introduced by the CBOE on 22 September 2003. The VIX is obtained from the European style S&P500 index option prices and incorporates information from the volatility skew by using a broader range of strike prices than just at-the-money strike series as in the VXO. However, the daily data on VIX starts from 2 January 1990, which does not cover our full sample period, beginning in January 1986. The pre-1986 VXO data are calculated by Bloom (2009). See Whaley (2009) for a history of the VIX and a summary on its calculation.

¹¹ We have also applied the principal component analysis (PCA) as in Haddow et al. (2013) who construct an uncertainty index based on four indicators for the UK on the 1985–2013 period. This alternative common factor closely resembles that from the DFM, with a coefficient of correlation close to 0.99. We have also replicated the impact of uncertainty on economic activity under this alternative factor and the results are qualitatively similar. Nevertheless, we obtain slightly different results on the forecast error variance (FEV) decomposition as the UCI obtained from the DFM explains a higher fraction of the FEV than that from obtained from the PCA for most economic variables (see Supplement data).

¹² As in Bachmann, Elstner, and Sims (2013) and Jurado, Ludvingson, and Ng (2015), we do not detrend any variables using the Hodrick-Prescott filter, while Bloom (2009) did so for every series except the VXO index. Because the HP filter uses information over the entire sample, it is difficult to interpret the timing of an observation.