Author’s Accepted Manuscript

MEASURING UNCERTAINTY IN THE STOCK

MARKET

Helena Chuliá, Montserrat Guillén, Jorge M. Uribe

www.elsevier.com

PII: S1059-0560(16)30267-2
DOI: http://dx.doi.org/10.1016/j.iref.2016.11.003
Reference: REVECO1319
To appear in: International Review of Economics and Finance
Received date: 18 November 2015
Revised date: 10 October 2016
Accepted date: 6 November 2016
Cite this article as: Helena Chuliá, Montserrat Guillén and Jorge M. Uribe,
MEASURING UNCERTAINTY IN THE STOCK MARKET, International
Review of Economics and Finance, http://dx.doi.org/10.1016/j.iref.2016.11.003
This is a PDF file of an unedited manuscript that has been accepted for
publication. As a service to our customers we are providing this early version of
the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting galley proof before it is published in its final citable form.
Please note that during the production process errors may be discovered which
could affect the content, and all legal disclaimers that apply to the journal pertain.
 MEASURING UNCERTAINTY IN THE STOCK MARKET

Helena Chuliáa*, Montserrat Guilléna, Jorge M. Uribeb

a
Riskcenter- IREA and Department of Econometrics, Universitat de Barcelona, Spain
b
Department of Economics, Universidad del Valle, Colombia

*Corresponding author: Universitat de Barcelona, Riskcenter- IREA and Department of

Econometrics, Av. Diagonal, 690, 08034 Barcelona, Spain. Tel.: +34 93 4021010.

[email protected] (H. Chuliá)

[email protected] (M. Guillén)

[email protected] (J.M. Uribe)

Abstract
We propose a daily index of time-varying stock market uncertainty. The index is
constructed after first removing the common variations in the series, based on recent
advances in the literature that emphasize the difference between risk (expected
variation) and uncertainty (unexpected variation). To this end, we draw on data from 25
portfolios sorted by size and book-to-market value. This strategy considerably reduces
information requirements and modeling design costs, compared to previous proposals.
We also compare our index with indicators of macro-uncertainty and estimate the
impact of an uncertainty shock on the dynamics of macroeconomic variables.

Key words: Uncertainty, Risk, Factor models, Stock market.

JEL Codes: E00, E44, G10, G14.
1. Introduction
Uncertainty and risk have been primary concerns in economics, and among scientists in
general, since the birth of modern science. Indeed, Bernstein (1998) goes as far as to
claim that the interest in measuring and mastering the two phenomena marks the
threshold separating modern times from the previous thousands of years of history.
In economics, Frank Knight was the first person to postulate the distinction between
uncertainty and risk on the grounds that the former cannot be described by means of a
probability measure while the latter can. According to both Knight (1921) and Keynes
(1921, 1939), economic agents inhabit an environment of pervasive uncertainty and,
therefore, there can be little hope of quantifying or forecasting economic variables, or of
even taking informed decisions that rely on quantitative measures of economic
dynamics (in other words, probabilities are incommensurable).
Today, the distinction between risk and uncertainty remains a lively topic for debate on
the academic agenda. Indeed, several recent studies have attempted to explain decision-
making under uncertainty, albeit oriented more towards the social conventions than
towards the development of rational calculations. Accordingly, in this branch of the
literature, there is a clear need to distinguish between the concepts, while measuring
what can be measured and not losing sight of what cannot be quantified in probabilistic
terms (Nelson and Katzenstein, 2014; Ganegoda and Evans, 2014; Taleb, 2007).
Although of obvious importance in its own right, this extreme Knightian differentiation
between risk and uncertainty leads to the impossibility of defining a probability space
and prevents us from using any variation of the Ergodic Theorem in empirical studies.
And this, in turn, leads to the impossibility of conducting any science at all (Hendry,
1980; Petersen, 1996) or, at least, the kind of social science based on ‘measurement’, as
has been fostered by the Cowles Commission for Research in Economics since its
foundation1.
However, confronted by this panorama, the profession has moved from this Knightian
extreme (fundamental) view of uncertainty and adopted a more promising approach to
the concept. In this new strand of the literature, uncertainty has generally been
assimilated to a time-varying conditional second moment of the series under study,
closely linked to underlying, time- varying, structural shocks, such as terrorist attacks,
political events, economic crises, wars and credit crunches. Yet, despite this, the
differentiation between risk and uncertainty in most instances is not properly dealt with.
Our contribution can be thought of as an attempt to measure the ‘known’ and part of the
‘unknown’, in the popular taxonomy of risk proposed by Gomery (1995). This author
differentiates between the ‘known’, the ‘unknown’ and the ‘unknowable’, and
highlights a traditional exaggerated focus on the former, while ignoring the other two
categories. That bias can lead to misconceptions about the world around us, because the
‘known’ constitutes only a very small fraction of what we see and face on our daily

1 ‘Science is Measurement’ was the original motto of the Cowles Commission (though it would later
be changed in 1952 to ‘Theory and Measurement’). See Keuzenkamp (2004) and Bjerkholt (2014)
for details about the history and methodology of econometrics and the role of the Cowles
Commission and the Econometric Society in the transition of economics to a more formally based
science.

2
decisions. Nevertheless, there is still the ‘unknowable’, which is clearly beyond the
scope of this paper, since in this situation even the events defining the probability space
cannot be identified in advance as pointed out by Diebold et al. (2010).
In this paper we seek to make three specific contributions to the study of uncertainty.
First, we propose a new index for measuring stock market uncertainty on a daily basis
(or what we refer to as financial uncertainty). 2 The index considers the inherent
differentiation between uncertainty and the common variations between the series
(which we identify as risk). Recent advances in the field have identified the
methodological tools for performing the task using factor models (Jurado, Ludvigson
and Ng, 2015; henceforth JLN). These proposals, however, have tended to focus their
attention on the use of macroeconomic variables to construct their indexes, as opposed
to financial variables. Therefore, because of the low frequency of macroeconomic
series, the proposals lack a desirable property of traditional proxies of uncertainty based
on financial returns (such as VXO, VIX or credit-spreads): namely, practitioners and
policy makers cannot trace their dynamics in real time.
Our second contribution is to show how our financial uncertainty index can also serve
as an indicator of macroeconomic uncertainty. We examine the circumstances under
which our index might be thought to capture all the relevant information in the economy
as a whole. We exploit the fact that the information contained in hundreds, or even
thousands, of economic indicators can be encapsulated by just a few stock market
portfolio returns. This circumstance makes the construction of the index easier, in terms
of its information requirements, modeling design and computational costs, and it allows
us to provide a high frequency uncertainty measure. The construction of our index,
based on portfolio returns, for which there are significant and timely data, provides a
better basis for analyzing uncertainty compared to other situations, in which this kind of
information and frequency are absent. Therefore, the extension of the methodology
beyond the stock markets must be approached with caution, since there is little hope to
extract the uncertainty components of less timely data, in an accurate fashion.
Finally, we analyze the dynamic relationship between uncertainty and the series of
consumption, interest rates, production and stock market prices, among others. This
allows us to further our understanding of the role of (financial or macroeconomic)
uncertainty, and to determine the dynamics of the economy as a whole. Our empirical
model allows us to analyze the extent to which traditional monetary policy can be
trusted to manage situations of uncertainty. Thus, on the one hand, we document a
significant and negative relationship between uncertainty and real variables such as
production, employment and consumption; on the other, we find that the interest rate
tends to decrease after an uncertainty shock while uncertainty decreases following a fall
in the interest rate. However, this last effect only explains a small proportion of the total
variation in the forecasted uncertainty.
The rest of this paper is organized as follows. First, we review theoretical and empirical
studies of uncertainty. In section 3 we describe the methodology used to estimate the
uncertainty index. Our approach relies on generalized dynamic factor models and
stochastic volatility (SV) devices. In section 4 we present our data and in section 5 our
main results. We also relate our findings to macroeconomic dynamics by means of a
vector autoregressive (VAR) analysis. In the last section we conclude.

2 The daily index is available on www.ub.edu/rfa/uncertainty-index.

3
2. Related literature

2.1. Risk, uncertainty, economic decisions and policy intervention

The current paradigm for understanding uncertainty was developed within the
framework of irreversible investment, in which a firm’s future investment opportunities
are treated as real options and the importance of waiting until the uncertainty is resolved
is emphasized. Hence, aggregate uncertainty shocks3 are thought to be followed by a
reduction in investment, and possibly in labor, and, consequently, by a deterioration in
real activity (Bernanke, 1983; Bertola and Caballero, 1994; Abel and Eberly, 1996;
Leahy and Whited, 1996; Caballero and Pindyck, 1996; Bloom et al., 2007; Bachmann
and Bayer, 2013). Nevertheless, some studies point out that after the original worsening
of the variables, a rebound effect related to a ‘volatility over-shoot’ may be observed
(Bloom, 2009; Bloom et al., 2013). It is worth noting that these original impacts on the
macroeconomic variables may be amplified as a result of financial market frictions
(Arellano et al., 2012; Christiano et al., 2014; Gilchrist et al., 2014).
The study of uncertainty is not confined to the firm’s investment problem. For example,
Romer (1990) suggests that consumers may postpone their acquisition of durable goods
in episodes of increasing uncertainty. Ramey and Ramey (1995) and Aghion et al.
(2010) have studied the negative relationship between volatility and economic growth.
The effects of uncertainty on equity prices and other financial variables have also been
analyzed. In this stream, Bansal and Yaron (2004) provide a model in which markets
dislike uncertainty and worse long-run growth prospects reduce equity prices. In the
same line, Bekaert et al. (2009) find that uncertainty plays an important role in the term
structure dynamics and that it is the main force behind the counter-cyclical volatility of
asset returns.
Additionally, there has been a revival of interest in examining the relationship between
uncertainty and policy interventions. However, there is no clear consensus in this
resurgent research agenda. Some authors conclude that the optimal monetary policy
does not change significantly during episodes of crisis and that uncertainty about crises
has relatively little effect on policy transmission (Williams, 2012), but others report that
financial uncertainty plays a significant role in monetary policy transmission
mechanisms (Baum et al., 2013; Bekaert et al., 2013). Neither is it clear whether a
highly responsive or moderate monetary policy scheme is best when facing uncertainty.
For instance, Williams (2013), in the same spirit as Brainard (1967), forwards the
argument that, once uncertainty is recognized, some moderation in monetary policy
might well be optimal. In marked contrast (albeit under a different notion of
uncertainty), Fendoǧlu (2014) recommends a non-negligible response to uncertainty
shocks.

2.2. Empirical measures of uncertainty

Empirical studies have frequently relied on proxies of uncertainty, most of which have
the advantage of being directly observable. Such proxies include stock returns or their
implied/realized volatility (i.e., VIX or VXO), the cross-sectional dispersion of firms’

3Panousi and Papanikolaou (2012) explain possible sources of inefficiency in the investment
process arising from idiosyncratic uncertainty, under high-powered incentives and risk-averse
managers. Bachmann and Bayer (2013) also study the impact of idiosyncratic uncertainty shocks
on business cycles.

4
profits (Bloom, 2009), estimated time-varying productivity (Bloom et al., 2013), the
cross-sectional dispersion of survey-based forecasts (Dick et al., 2013; Bachmann et al.,
2013), credit spreads (Fendoǧlu, 2014), and the appearance of ‘uncertainty-related’ key
words in the media (Baker et al., 2016; Alexopoulos and Cohen, 2015).4
Although these uncertainty proxies have provided key insights to the comprehension of
uncertainty, and have been reliable starting points for the analysis of the economic
impacts of uncertainty on economic variables, most of them have come under criticism,
most notably from Scotti (2016) and JLN. On the one hand, volatility measures blend
uncertainty with other notions (such as risk and risk-aversion), owing to the fact that
they do not usually take the forecastable component of the variation into account before
calculating uncertainty. On the other, analysts’ forecasts are only available for a limited
number of series. Moreover, it is not entirely clear whether the responses drawn from
these surveys accurately capture the conditional expectations of the economy as a
whole. The disagreement reported in survey forecasts could be more of an expression of
different opinions than of real uncertainty (Diether et al., 2002) and even if forecasts are
unbiased, the disagreement in analysts’ point forecasts is not generally equivalent to
forecast error uncertainty (Lahiri and Sheng, 2010) 5 . Aimed at overcoming these
shortcomings, a new branch of the literature has emerged, which proposes measuring
uncertainty only after the forecastable component of the series has been removed
(Carriero et al., 20156; Gilchrist et al., 2014; JLN).
Our model takes into account the extraction of the contemporaneously forecastable
component of the variation before calculating uncertainty, which is important in order to
distinguish satisfactorily between uncertainty and risk. We also aim to construct
estimations of uncertainty by deliberately adopting an atheoretical approach, in the
same vein as JLN. Our study contributes to the existing literature by providing a daily
measurement of uncertainty. This is important, because it means the market can be
monitored in real time, while enabling the researcher to undertake event studies with
greater precision including uncertainty as a variable. The literature notes that
estimations of impacts extracted from event studies are much more precise and less
noisy as the frequency of the data increases (Fair, 2002; Bomfim, 2003; Chuliá et al.,
2010).
3. Methodology
The construction of our uncertainty index consists of two steps. First, we remove the
common component of the series under study and calculate their idiosyncratic variation.
To do this, we filter the original series using a generalized dynamic factor model
(GDFM). Second, we calculate the stochastic volatility of each residual in the previous
step using Markov chain Monte Carlo (MCMC) techniques. Then, we average the
series, obtaining a single index of uncertainty for the stock market, and possibly for the
economy as a whole. In sections 3.1 and 3.2 below, we explain each step in detail.

4 Other studies, such as Shi et al. (2016), have used linguistics-based sentiment scores of the news
releases to study the dynamics of stock volatility.
5 Bachmann et al. (2013) and Scotti (2016) acknowledge these problems and address them by
using additional proxies for uncertainty. Nevertheless, as noted by JLN, these studies focus on
variation in outcomes around subjective survey expectations.
6 These authors do not address the problem of measuring uncertainty directly, but still they use a
closely related methodological approach to the one employed in this strand of the literature.

5
3.1. Idiosyncratic component extraction
Following Bai and Ng (2008), let be the number of cross-sectional units and be the
number of time series observations. For and , the dynamic factor
model (DFM) can be defined as:
( ) , (1)

where ( ) ( ) is a vector of dynamic factor loadings of order

. When is finite, we refer to it as a DFM. In contrast, a GDFM allows to be infinite.
Stock and Watson (2002, 2011) provide examples of the former and Forni and Reichlin
(1998) and Forni et al. (2000) introduce the latter. In any case, the (dynamic) factors
evolve according to:
( ) , (2)

where are errors. The dimension of , denoted , is the same as that of and it
refers to the number of dynamic or primitive factors (Bai and Ng, 2007).
The model stated in (2) can be rewritten in static form, simply by redefining the vector
of factors to contain the dynamic factors and their lags, and the matrix of loads
accordingly, as:

( ) ( )( ) ( ), (3)

where ( ) and ( ) . Clearly, and are not separately

identifiable. For any arbitrary ( ) invertible matrix , ,
where and , the factor model is observationally equivalent to
. Therefore restrictions are required to uniquely fix and (Bai and
Wang, 2012). Note that the estimation of the factors by principal components (PC) or
singular value decomposition (SVD) imposes the normalization that and is
diagonal, which are sufficient to guarantee identification (up to a column sign
variation).
The GDFM is a generalization of the DFM because it allows a richer dynamic structure
for the factors. It places smaller weights on variables with larger idiosyncratic
(uncertainty) components. So that the idiosyncratic ‘error’ contained in the linear
combination is minimized. In this way we ensure that the uncertainty component is
purged from risk-related variations.

Our first step enables us to estimate the idiosyncratic variation of the series
̂ , where ̂ ( ) . This component is primarily related to uncertainty, whereas
the common variation (i.e., the variance of ̂ ) can be referred to as risk.
3.2. Conditional volatility estimation

Once we recover the series of filtered returns, , a SV model is specified on an

7
individual level, for each , as:

7 In what follows we omit the cross-sectional subscript to simplify the notation.

6
 (4)
( ) , (5)

where and are independent standard normal innovations for all and belonging
to { }. The non-observable process ( ) appearing in equation 5 is
the time-varying volatility with initial state distribution | ( ( )).
This centered parameterization of the model should be contrasted with the uncentered
reparameterization provided by Kastner and Frühwirth-Schnatter (2014):
̃
( ) (6)
̃ ̃ ( ). (7)
Whether the first or the second parameterization is preferred for estimation purposes
generally depends on the value of the ‘true’ parameters (Kastner and Frühwirth-
Schnatter, 2014). Nevertheless, both of them have intractable likelihoods and, therefore,
MCMC sampling techniques are required for Bayesian estimation.
Kastner and Frühwirth-Schnatter (2014) provide a strategy for overcoming the problem
of efficiency loss due to an incorrect selection among the representations in applied
problems. They propose interweaving (4)-(5) and (6)-(7) using the ancillarity-
sufficiency interweaving strategy (ASIS) as introduced by Yu and Meng (2011). Their
results indicate that this strategy provides a robustly efficient sampler that always
outperforms the more efficient parameterization with respect to all parameters, at little
extra cost in terms of design and computation. We follow their advice to estimate the
volatilities of the idiosyncratic shocks.
Once the idiosyncratic stochastic volatility measures have been constructed, we are able
to estimate the uncertainty index in the stock market as the simple average of the
individual volatilities:
∑
. (8)

This scheme corresponds to the equally weighted average, with ∑ → ( ),

where ⁄ . Alternatives, such as using the first PC to aggregate the series of
variances, are possible but have no grounding in econometric theory to guarantee their
consistency in the estimation process (Jurado et al., 2013; JLN). Unlike the previously
referenced studies by JLN, here we only use information from portfolio returns
organized by different factor criteria; thus, there is no ex ante reason to weight each
portfolio return using different loads. In principle, any firm might belong to any
portfolio, and all of them are equally important in the estimation of the aggregate shock.
Hence, it is natural to favor the equally-weighted scheme over other asymmetric
alternatives, but note that the asymmetric scheme would be more appropriate when
macro-variables are blended with financial or other kind of variables.
4. Data
In our empirical exercise we use 25 portfolios of stocks belonging to the NYSE,
AMEX, and NASDAQ, sorted according to size and their book-to-market value, as

7
provided by Kenneth French on his website8. Those portfolios have been widely used in
the literature examining multi-factor asset pricing models (Cochrane, 2005), and can be
seen as a good summary of whole market dynamics. Moreover, Sentana (2004) justifies
the use of portfolios for extracting the subjacent factors by proving that many portfolios
converge to the factors as the number of assets increases. Clearly this does not rule out
the fact that other possibilities might be explored in future research, such as the use of
less well-known portfolios constructed on an industry sector basis, or using different
factors to organize the series.
Our data set spans from 1 July 1926 to 30 September 2014, which gives a total of
23,321 observations. More details on the portfolio formation are provided in Davis,
Fama and French (2000) and on Kenneth French’s web page.
In section 5.3 we estimate a vector autoregressive (VAR) model. The data for this
exercise were taken from the web page of the Federal Reserve Saint Louis (FRED:
http://research.stlouisfed.org/). Specifically, we use the Industrial Production Index; the
total number of employees in the non-farm sector; Real Personal Consumption
Expenditures in 2009 prices; the Personal Consumption Expenditures Price Index; the
New Orders Index known as NAPM-NOI; Average Weekly Hours of Production and
Nonsupervisory Employees for the Manufacturing sector (the all-sector index is not
available from the beginning of our sample); Effective Federal Funds Rate; M2 Money
Stock in billions of dollars and Standard and Poor’s 500 index. Each series was taken
seasonally adjusted where necessary, and the sample spans from February 1959 to
September 2014, which is the longest period possible using these series.
5. Results
In this section we present our uncertainty index (section 5.1); we compare it with some
of the main macro-uncertainty indicators (section 5.2); we analyze the relationship
between our proposal and some real and financial variables, including policy variables
(section 5.3); and, we perform several robustness exercises (section 5.4).
5.1. Uncertainty index
We estimate the GDFM using six static factors and one dynamic factor, which are
optimal following the criteria proposed by Bai and Ng (2002) and Bai and Ng (2007),
respectively. Based on these estimates we construct the uncertainty index by
aggregating the conditional volatilities of the idiosyncratic residual series as explained
in section 3.
The daily uncertainty index is presented in Figure 1, together with the recession dates in
the United States, as indicated by the NBER on its web site. The index peaks coincide
with well-documented episodes of uncertainty in the financial markets and the real
economy, including the Great Depression, the recession of 1937-38 in the US, Black
Monday in October 1987, the bursting of the dot-com bubble and the Great Recession
2007-2009.
Recession dates, such as August 1929 to March 1933, May 1937 to June 1938 and
December 2007 to June 2009, clearly correlate with the amount of uncertainty in the
market, although interestingly, not all recessionary episodes are preceded or followed

8 http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.

8
by a notable uncertainty shock. For example, the uncertainty peak in the index
corresponding to March 2000 appears one year before the economic contraction in
March 2001. Likewise, several recessions during the decades of the 40s, 50s and 60s do
not seem to be associated with episodes of high or even increasing uncertainty.
More importantly, uncertainty in the stock markets appears to correlate not only with
the volatility of fundamentals (i.e., recessions), but also with episodes of over-valuation
or bubbles in the market, as discussed for example in Yuhn et al. (2015), namely, those
of 1987 (Black Monday), 2000 (information technology boom) and 2007 (housing
market boom). Indeed, these episodes may well be the main drivers of uncertainty (even
more so than the recessions), at least in the last part of our sample. Many such episodes
have been identified in the recent literature and they constitute a particularly active area
of current research within the financial econometrics field (Phillips and Yu, 2011;
Phillips et al., 2011; Homm and Breitung, 2012; Phillips et al., 2015; Anderson and
Brooks, 2014) and even outside economics, especially in the application of statistical
mechanics tools to financial problems (see Zhou and Sornette (2003), Sornette and
Zhou (2004), Sornette et al. (2009), udinski-Petkovi et al. (2014) and references
therein).

9
Great Great
dot.com
Recession
8 Depression bubble
collapse

7 Recession
1937-38
Black
6 Monday
1987

1
Jan-27 Jan-37 Jan-47 Jan-57 Jan-67 Jan-77 Jan-87 Jan-97 Jan-07

Figure 1: Uncertainty Index: Jan-06-27 to Sept-30-14. The first 153 observations have been discarded
and the last 153 have been replaced by calculations using a (scaled) one-sided filter version of the GDFM
(Forni et al., 2005). The reason for doing this is that original GDFM are biased at the beginning and at the
end of the sample, because they make use of the estimation of the variance- covariance matrices of
order √ . Grey areas correspond to NBER recession dates (peak-to-trough), including the peaks and
troughs. The horizontal line corresponds to the 95 percentile of the empirical distribution of the index
from Jan-40 onwards. The original measure is rescaled by a factor of 100 in the plot.

The observation above can be rationalized under a framework of agents with

heterogeneous beliefs and bounded rationality as the one proposed by Hommes and
Wagener (2009). In their model, there is an endogenous switching mechanism,
governing the proportion of financial investors who follow a ‘perfect foresight’

9
forecasting rule (driven by market fundamentals), or alternative linear heuristics, such
as ‘biased beliefs’ and ‘past trends’. Instabilities may follow after an increasing in the
number of non-fundamentalist traders in the market and hence, produce the apparition
of persistent bubbles. Uncertainty, as measured by our index, is naturally related to this
possibility. That is, in high uncertainty regimes more agents may choose to switch to a
non-fundamentalist rule of prediction, driving the prices away from their fundamental
path.
In Table 1 we report descriptive statistics for a monthly (end-of-the-month) version of
the uncertainty index. We construct this monthly index to facilitate comparisons with
other macro-uncertainty proxies. The skewness, kurtosis, persistence and half-life of the
shocks for the full sample and for two sub-samples are presented (January 1927 to
March 1940 and April 1940 to September 2014). This break date was chosen after
testing for multiple breaks (Bai and Perron, 1998, 2003) in the autoregressive model of
the shocks persistence (AR(1) with drift)9.
Table 1. Summary statistics of the uncertainty index in two sub-samples
Sample period
Jan 1927-Sept Jan 1927-Mar Apr 1940-
Statistic
2014 1940 Sept 2014
Skewness 1.60 0.32 1.70
Kurtosis 4.74 1.97 6.62
Persistence, AR(1) 0.993 0.963 0.978
Half-life: months (years) 101 (8.42) 18.3 (1.53) 31.9 (2.65)

Table 1 shows that using the full sample to calculate persistence can lead to a spurious
estimation of the summary statistics. Indeed, the sample distribution of the uncertainty
index in the two sub-samples looks quite distinct. In the first part of the sample,
persistence is smaller and, therefore, the ‘shocks’ disappear in a shorter period of time
(1.53 years) than is the case in the second sub-sample (2.65 years). There are fewer
observations distant from the mean and, lastly, the distribution presents a slightly
asymmetric behavior (skewness equal to 0.32). In contrast, even when the second part
of the estimation presents shocks of a smaller magnitude (Figure 1), the distribution that
characterizes them tends to generate a higher number of ‘outliers’ (kurtosis equal to
6.92) and they are more likely to be above than below the mean (1.7 is the asymmetric
coefficient). This behavior may be interpreted as uncertainty showing some degree of
inconsistency across time, which is related to the knightian framework, in which
uncertainty is indeed understood as a non-predictable state.
Our estimations of persistence of macro-uncertainty are lower than those reported
elsewhere, for example, those provided by JLN. The latter estimate a persistence of
53.58 months, while in the second part of our sample our estimation is of 31.9 months
(41.2 months from Jan. 1960 to Sept. 2014). This could be interpreted as evidence that
financial-uncertainty shocks are not as persistent as macro-uncertainty shocks.
Nevertheless, it should be noted that JLN also report the persistence and half-lives of
frequently used proxies for uncertainty, including the VXO and the cross-sectional
standard deviation of the returns. They show that these uncertainty-related measures are
far less persistent than are macro-uncertainty shocks (with half-lives of 4.13 and 1.92
months). Thus, the half-life and persistence of our uncertainty measure are more similar

9 See Perron (2006) for a survey of this literature.

10
to those of the macro-uncertainty shocks than to those derived from the volatility
measures.
5.2. Correlations with macro-uncertainty indexes
The closest measure of uncertainty to ours, methodologically speaking, is the
uncertainty index proposed by JLN, although their proposal might be interpreted more
directly as a ‘macro-uncertainty’ indicator, given its emphasis on economic variables as
opposed to purely financial ones. Given these circumstances, it seems to be a good
candidate with which to compare our index while seeking to identify any convergent
and divergent paths. In order to compare the indexes, we first reduce our sample to fit
theirs. Our resampled data start in January 1960 and end in May 201310. After so doing,
we recalculate our uncertainty index by aiming to use the same dates as those employed
by JLN. Second, we take the end-of-the-month value of our index, to resemble their
index frequency (monthly).
The results are reported in Figure 2. The shaded areas in the plot correspond to periods
of ‘high’ correlation. The Pearson’s correlation for the full sample between the indexes
is barely above 22%, which could be interpreted, at first glance, to indicate that different
forces lie behind the macro-uncertainty and the financial-uncertainty. However, this
correlation seems very volatile. We also calculate moving-window correlations of five
years during the sample and here our findings are more informative than the static
correlation. The correlation remains above 50% for most of the period (left panel).
Moreover, for the last part of the sample (from around February 2009 to May 2013),
this correlation remained above 90%, revealing practically no difference in the indexes’
dynamics. Even higher values were reached during the 70s and we observe correlations
between 40 and 80% in the period from May 1994 to February 2003 (right panel). There
are also two periods in which this correlation became negative, specifically from
January 1992 to August 1993 and December 2005 to September 2007. After these short
phases, the indexes started to move in the same direction once again, and in both cases
with a stronger impetus than before.
Finally, an analysis of the levels of the uncertainty indexes shows them to be
particularly different during the periods from March 1979 to May 1983 and July 1998 to
January 2003. Our intuition regarding the explanation for these divergent paths during
these periods is that while uncertainty in the financial markets is driven significantly by
bubble episodes, such episodes are not always the drivers of the recessions in the real
economy and, therefore, cannot be related on a one-to-one basis with macro-
uncertainty. Thus, the financial-uncertainty index highlights uncertainty associated with
bubble episodes (for instance, during the dot.com collapse) that did not materialize as
strong recessionary phases in the real economy and which, therefore, are not captured
by the JLN-uncertainty index. In the same vein, recessionary episodes not directly
related to the financial market (such as those from 1979 to 1983) are not especially
pronounced in our financial-uncertainty indicator.

10 The JLN-index is publicly available for this period on Sidney Ludvigson’s web page:
http://www.econ.nyu.edu/user/ludvigsons/

11
5 1 5
0,9
JLN JLN
4 4
0,8 0,7
U U
3 3
0,5
0,6
2 2
0,3

1 1
0,4 0,1

0 0 -0,1
0,2
-1 -1 -0,3

-2 0 -2 -0,5
Apr-65 Apr-74 Apr-83 Apr-92 Apr-01 Apr-10 Apr-65 Apr-74 Apr-83 Apr-92 Apr-01 Apr-10

Figure 2: Uncertainty Comparisons I. The solid line represents our Uncertainty Index (U), while the
dotted line represents the Jurado-Ludvigson-Ng’s Index (JLN) with forecast horizon , both from
Apr-65 to May-13. In the panel on the left, the shaded areas correspond to correlation periods above 0.5.
In the panel on the right, the shaded areas are the actual correlations. Correlations are measured on the
right axis of each panel. Correlations were calculated using rolling, moving windows of five years,
starting from January 1960.

We also compare our index with the VIX, another frequent proxy for macro- and
financial-uncertainty (Figure 3), but which is only available after January 1990. We
found a correlation of 65.2% using the full sample. The dynamics of the VIX and the
uncertainty index appear to be largely similar with a correlation above 70% for the last
ten years of the sample. However, these dynamics are considerably different
(considering the correlation levels) for the first ten years of the sample. Here again, the
results could be linked to the fact that volatility as a risk measure is inversely related to
the presence of over-valuation in the stock markets, whereas over-valuation appears to
be positively related to uncertainty.
6 1,0

5
0,8
4
VIX
0,6
3 U

2 0,4

1
0,2
0
0,0
-1

-2 -0,2
Jan-90 Jan-95 Jan-00 Jan-05 Jan-10
Figure 3: Uncertainty Comparisons II. The solid line represents our uncertainty index (U), while the
dotted line represents the VIX, both from Jan-90 to Sept-14. Shaded areas correspond to the five-year
rolling correlations and, therefore, start only after Jan-95. Correlations are measured along the right axis.

5.3. VAR dynamics: Uncertainty, economic activity and policy variables

In this section, we explore the dynamic relationship between our uncertainty index and
some macroeconomic and financial variables. To do so, we use the model proposed by

12
Christiano et al. (2005). This model has been widely studied in the literature and is,
therefore, useful for comparing our uncertainty estimates. The model is given in
reduced form by:
( ) , (9)

where, [ ] is a matrix ( ) containing the column-vectors of

the model. contains slow-moving variables which do not react contemporaneously
to a monetary policy shock: Production, Employment, Consumption, Inflation, New
Orders, Wages and Labor. refers to the Federal Funds Rate, understood as the
monetary policy instrument. refers to the fastest variables, which are assumed to
respond contemporaneously to the policy innovation, such as: the Stock Market Index
and M2. Finally, we place our Uncertainty Index U in last position (as do JLN and
Bloom, 2009)11. We estimate a VAR with 12 lags, as opposed to the four quarters used
in Christiano et al. (2005) to cover the same time-span. All the variables enter in log-
levels, with the exceptions of the Federal Funds Rate and Uncertainty, which enter in
original units, and M2, which enters in growth rates. We recover the structural
innovations by means of a Cholesky factorization of the variance-covariance matrix. As
is well known, the Cholesky decomposition implies a certain ordering of the set of
variables, depending on whether they react or not to other variables contemporaneously.
Following Christiano et al. (2005), the variables are sorted from more exogenous to
more endogenous as stated above. The impulse response functions are presented in
Figure 4.
The reactions of Production and Employment to uncertainty shocks have been studied
elsewhere, for example in JLN and Bloom (2009). The former report very similar
results to ours even when using their uncertainty index, which requires considerably
more information, processing time and modeling design than are required by our index
(see also section 5.4). Production reacts negatively to uncertainty increments and the
persistence of the shock extends beyond the two-year horizon. In the sixth months after
the innovation, 10.5% of the forecast error of the production series is explained by the
uncertainty shock, and up to 23.8% is explained 12 months on12.

0,10 0,10
IP
E
0,00
0,00
-0,10

-0,10
-0,20

-0,30 -0,20
0 20 40 60 0 20 40 60
(months) (months)

11 See section 4 for a more detailed description of the data used in this section.
12 See Table 2 in the Appendix.

13
 1,00 0,20
NO 0,10 C
0,25
0,00

-0,50 -0,10
-0,20
-1,25
-0,30

-2,00 -0,40
0 20 40 60 0 20 40 60
(months) (months)

20,00 0,20
R SP
10,00 -0,05

0,00 -0,30

-10,00 -0,55

-20,00 -0,80

-30,00 -1,05
0 20 40 60 0 20 40 60
(months) (months)

Figure 4: Economic Dynamics under Uncertainty. We use a VAR (12) comprising 11 variables. The
axes are in percentages but the Federal Funds Rate is in basic points. The figure shows the reaction of the
variables to an unexpected increment of uncertainty. The estimation period runs from February 1959 to
September 2014. Confidence bands (86%) are calculated using bootstrapping techniques as explained in
Efron and Tibshirani (1993). The variables are defined as: IP: Industrial Production Index, E:
Employment, NO: New Orders, C: Consumption, R: Federal Funds Rate, SP: Standard and Poor’s 500.

Analogously, although at a smaller magnitude, employment decreases following a

positive uncertainty shock and the impact persists for two and a half years (that is, six
months more than in the case of production)13. Neither we nor JLN find any evidence
supporting the ‘rebound’ effect proposed by Bloom (2009) in the case of production.
However, the rebound effect is evident when analyzing the New Orders variable, which
is a better proxy for current investment. First, new orders decrease in the face of
uncertainty – a negative impact that lasts approximately eight months, but there is a
statistically significant rebound effect in months 16 to 19. The reason why a similar
effect is not detected in the production dynamics could be that following the original
uncertainty shocks, negative feedback is obtained from consumption and expected
demand.
Although there are theoretical claims explicitly linking uncertainty shocks and
consumption (see, for instance, Romer, 1990), little empirical evidence has been
presented to document this relationship. Here, we find that after an increment in
uncertainty, consumption is severely affected (indeed, more or less in the same

13 JLN report an impact of their uncertainty shock on production that persists for more than 60
months. We also find that the IRF tends to stabilize at a lower level following a shock, as can be seen
in Figure 4, although this is only true for the average level. Note that the bootstrapped confidence
intervals of our exercise prevent us from fixing the effects beyond three years as statistically
different from zero.

14
proportion as production, and more so than employment). However, the shock tends to
disappear more quickly (1.3 years before the upper confidence band reaches zero), but it
is also apparent that it causes the series to stabilize at a lower level relative to that of the
production series.
In line with the theory, financial prices, such as the stock market index, are significantly
affected by uncertainty in the financial markets. Indeed, the marked fall in the market
index in the face of uncertainty, and the stabilization of the sequence at a lower level, is
consistent with the theoretical discussion in Bansal and Yaron (2004). Basically, the
intuition is tied to the fact that markets do not like uncertainty and after an increase in
uncertainty, the discount of the expected cash flows is greater, causing the market to
reduce the price of the stock.
As can be seen from Table 2 in the Appendix, a variance decomposition of the forecast
errors of the series confirms the importance of uncertainty as a driver of the economy’s
dynamics. One year after the original structural innovation, it accounts for 23.8% of the
variance in production, 19.5% of new orders, 13.2% of employment and 15.9% of the
stock market prices. In all cases, it is the second or third largest source of variation. It
also affects other series, albeit to a lesser degree, including consumption (7.6%) and
Federal Funds (4.7%), being in these cases the fourth or fifth cause of variation among
the eleven variables considered.
Lastly, the Federal Funds Rate also seems to be sensitive to uncertainty. In the face of
an uncertainty shock the Federal Reserve tends to reduce the interest rate (thereby
confirming that the reduction in equity prices is due to uncertainty and not to possible
confounding interest movements). The reduction is particularly persistent during the
first year before it begins to disappear. Nevertheless, the uncertainty shock only
accounts for between 4 and 5% of the total variation in the Fed rate according to the
variance decomposition.
The Cholesky identification strategy allows us to distinguish the effect in the reverse
direction; in other words, it enables us to answer the question: Does an expansionary
monetary policy decrease uncertainty? As can be observed in Figure 5, a loosening
monetary policy does affect uncertainty. The effects are expected to occur with a lag of
one year, to last for a further year, and after this period, to disappear. This finding is in
line with similar effects documented by Bekaert et al. (2013), although they use non-
corrected uncertainty measures and an alternative strategy to differentiate it from risk.
Our results in this direction add to the research field by exploring the relationship
between policy intervention and uncertainty. However, the effects are small in
magnitude (see Table 2 in the Appendix), with between 2 and 6% being due to the
monetary policy innovations.

15
 20,00 0,10
R U
10,00
0,05
0,00
0,00
-10,00
-0,05
-20,00

-30,00 -0,10
0 20 40 60 0 20 40 60
(months) (months)

Figure 5: Policy intervention and uncertainty. We use a VAR (12) comprising 11 variables. The axes
are in basic points and units, respectively. We replicate the left panel from Figure 5 and we multiply by
minus one the response to an increase in the Federal Funds Rate, to be consistent with the text. The
estimation period runs from February 1959 to September 2014. Confidence bands (86%) are calculated
using bootstrapping techniques as explained in Efron and Tibshirani (1993).

Finally, in Figure 6, using our proposed index and JLN’s index, we compare the
responses of the variables facing uncertainty. However, the qualitative and quantitative
results reported above do not vary significantly depending on the uncertainty measure
used.

0,10 0,10
E
IP
0,00
0,00
-0,10
-0,10
-0,20

-0,30 -0,20
0 20 40 60 0 20 40 60
(months) (months)

1,00 0,20
NO C
0,10
0,25
0,00
-0,50 -0,10
-0,20
-1,25
-0,30
-2,00 -0,40
0 20 40 60 0 20 40 60
(months) (months)

16
 20,00 0,20
SP
R
10,00 -0,05

0,00 -0,30

-10,00 -0,55

-20,00 -0,80

-30,00 -1,05
0 20 40 60 0 20 40
(months) (months)

Figure 6: Economic Dynamics under Uncertainty. Comparison of the JLN and U Indexes. We use a
VAR (12) comprising 11 variables. The figure displays the reaction of the variables to an unexpected
increment in two standardized uncertainty measures, the U index (solid line) and the JLN index (dotted
line). The estimation period for the U index runs from February 1959 to September 2014 whereas the JLN
index is only publicly available from July 1960 to May 2013 on one of its author’s web pages; therefore,
we use this latter period to estimate the IRFs in this case. The variables are defined as: IP: Industrial
Production Index, E: Employment, NO: New Orders, C: Consumption, R: Federal Funds Rate, SP:
Standard and Poor’s 500.

5.4. Robustness
We perform several robustness exercises varying the econometric methodology
employed to extract the idiosyncratic component. We estimate the uncertainty index
using DFM instead of GDFM; we also use the ‘one-sided’ filter version of the GDFM
proposed by Forni et al. (2005) as opposed to the two-sided original GDFM, for the full
sample; we estimate the index as the stochastic volatility without using any factor
model to extract the idiosyncratic component and, finally, we estimate the idiosyncratic
component in a recursive fashion, recalculating each model with rolling windows of 80
days (approx. one quarter). The latter approach speaks directly about parameter
stability. The main results are summarized in Figure 7.

10 10

8 8
UOneSide

6 6
UDFM
4 4

2 2

0 0

-2 UGDFM UGDFM
-2

-4 -4
Apr-60 Apr-70 Apr-80 Apr-90 Apr-00 Apr-10 Apr-60 Apr-70 Apr-80 Apr-90 Apr-00 Apr-10

17
 10 10

8 8
VOL
6 6
UITER
4 4

2 2

0 0

UGDFM UGDFM
-2 -2

-4 -4
Apr-60 Apr-70 Apr-80 Apr-90 Apr-00 Apr-10 Apr-60 Apr-70 Apr-80 Apr-90 Apr-00 Apr-10

Figure 7: Robustness exercises. The uncertainty index using GDFM (solid line) is compared with
different alternatives: a DFM (top left), a one-sided filter version of the GDFM (top right), a recursive
algorithm (bottom left) and a conditional volatility measure of the original series (bottom right). All the
indexes have been standardized to make proper comparisons.

In general the uncertainty index behaves in a very similar fashion, regardless of the
factor methodology used to extract the idiosyncratic components of the series. Nor does
it change when we use recursive estimations. Nevertheless, its behavior is considerably
different to that of the stochastic volatility of the original series. This, however, is not
surprising and is indeed in-line with previous findings in the literature. Volatility
measures tend to overestimate the uncertainty of the economy because they confuse
uncertainty with risk or risk aversion.
6. Conclusions
We propose an index of time-varying financial uncertainty. The construction of this
index is relatively simple as it does not rely on excessive data mining devices nor does
it have to satisfy demanding information requirements. We construct the index on a
daily basis, for the United States’ economy between 1927 and 2014. As such, the index
can be used to perform event studies, that is, to evaluate the impact of policy treatments
on economic uncertainty, thanks to the higher frequency it offers compared to other
proposals.
Our estimations allow us to identify several periods of uncertainty, some of which
coincide with well-documented episodes, including major recessions, wars, and political
upheavals. Others, especially those occurring in more recent decades, are more closely
associated with bubble regimes in the stock market. We also document a change in the
persistence of uncertainty between 1940 and 2014 compared to that recorded between
1927 and 1940. Current uncertainty is more persistent and is plagued with more extreme
observations, although current periods tend to be smaller in magnitude than earlier
periods.
We discuss the circumstances under which our index is a better measure of financial
uncertainty and when it is in agreement with measures available elsewhere. We
conclude that significant departures between macro-uncertainty and financial
uncertainty can be expected during bubble episodes and we present evidence of this.

18
However, the economic dynamics that we document here (using a VAR model) are
consistent with theoretical expectations and previous empirical studies (when available).
For example, we find that after an uncertainty shock, production and employment react
negatively and the effects of the shock tend to disappear slowly. We also present novel
empirical evidence regarding the negative effect of uncertainty on consumption,
inventory investment (including overshooting) and stock market prices.
Finally, we explore the relationship between uncertainty and policy variables. We find
that there is indeed a relation between the reference interest rate in the economy and
uncertainty. The interest rate tends to decrease in the face of an uncertainty shock, while
the uncertainty shock decreases following a loosening of the monetary policy position,
with a lag of one-year. However, this latter effect is very small in terms of accounting
for the total variation of the forecast errors of the uncertainty variable. This result raises
questions regarding the capability of the central banks to combat uncertainty by means
of traditional monetary policy.

Acknowledgements

This work was supported by the Spanish Ministry of Economy [grant numbers
ECO2013-48326-C2-1-P, ECO2015-66314-R, ECO2016-76203-C2-2-P]; and ICREA
Academia. The authors are very grateful to Angel Pardo for his valuable comments to a
preliminary draft.

References
Abel, A.B., & Eberly, J.C. (1996). Optimal investment with costly reversibility. The Review of Economic
Studies, 63(4), 581-593.
Aghion, P., Angeletos, G.M., Banerjee, A., & Manova, K. (2010). Volatility and growth: credit
constraints and the composition of investment. Journal of Monetary Economics, 57(3), 246-265.
Alexopoulos, M., & Cohen, J. (2015). The power of print: Uncertainty shocks, markets, and the economy.
International Review of Economics & Finance, 40, 8-28.
Anderson, K., & Brooks, C. (2014). Speculative bubbles and the cross-sectional variation in stock returns.
International Financial Analysis, 35, 20-31.
Arellano, C., Bai, Y., & Kehoe, P. J. (2012). Financial markets and fluctuations in volatility. Staff Report
No. 466, Federal Reserve Bank of Minneapolis,
https://www.minneapolisfed.org/research/sr/sr466.pdf
Bachmann, R., & Bayer, C. (2013). ‘Wait-and-see’ business cycles? Journal of Monetary Economics,
60(6), 704 -719.
Bachmann, R., Elstner, S. & Sims, E. (2013). Uncertainty and economic activity: evidence from business
survey data. American Economic Journal: Macroeconomics, 5(2), 217-249.
Bai, J., & Perron, P. (1998). Estimating and testing linear models with multiple structural changes
.Econometrica, 66(1), 47–78.
Bai, J., & Perron, P. (2003). Computation and analysis of multiple structural change models. Journal of
Applied Econometrics, 18(1), 1–22.
Bai, J., & Ng, S. (2002). Determining the number of factors in approximate factor models. Econometrica,
70(1), 191-221.
Bai, J., & Ng, S. (2007). Determining the number of primitive shocks in factor models. Journal of
Business and Economic Statistics, 25(1), 52-60.
Bai, J., & Ng, S. (2008). Large dimensional factor analysis. Foundations and Trends in Econometrics,
3(2), 89–163.
Bai, J., & Wang, P. (2012). Identification and estimation of dynamic factor models. Columbia University
Academic Commons, http://hdl.handle.net/10022/AC:P:13083.
Baker, S. R., Bloom, N., & Davis, S. J. (2016). Measuring Economic Policy Uncertainty, The Quarterly

19
 Journal of Economics, doi: 10.1093/qje/qjw024.
Bansal, R., & Yaron, A. (2004). Risks for the long run: a potential resolution of asset pricing puzzles.
Journal of Finance, 59(4), 1481–1509.
Baum, C., Caglayan, M., & Ozkan, N. (2013). The role of uncertainty in the transmission of monetary
policy effects on bank lending. Manchester School, 81(2), 202-225.
Bekaert, G., Engstrom, E., & Xing, Y. (2009). Risk, uncertainty, and asset prices. Journal of Financial
Economics, 91(1), 59–82.
Bekaert, G., Hoerova, M., & Lo Duca, M. (2013). Risk, uncertainty and monetary policy. Journal of
Monetary Economics, 60(7), 771-788.
Bernanke, B. (1983). Irreversibility, uncertainty and cyclical investment. Quarterly Journal of
Economics, 98(1), 85-106.
Bernstein, P. (1998). Against The Gods: The Remarkable Story of Risk. New York: John Wiley & Sons.
Bertola, G., & Caballero, R. (1994). Irreversibility and aggregate investment. The Review of Economic
Studies, 61(2), 223-246.
Bjerkholt, O. (2014). Econometric Society 1930: How it Got Founded, Department of Economics
University of Oslo, working paper, http://dx.doi.org/10.2139/ssrn.2532106.
Bloom, N. (2009). The impact of uncertainty shocks. Econometrica, 77(3), 623-685.
Bloom, N., Bond, S., & Van Reenen J. (2007). Uncertainty and investment dynamics. The Review of
Economic Studies, 74(2), 391-415.
Bloom, N., Floetotto, M., Jaimovich, N., Saporta-Eksen, I., & Terry, S. J. (2013). Really Uncertain
Business Cycles, CEP discussion papers, http://cep.lse.ac.uk/pubs/download/dp1195.pdf
Bomfim, A.N. (2003). Pre-announcement effects, news, and volatility: monetary policy and the stock
market. Journal of Banking and Finance, 27(1), 133–151.
Brainard, W. (1967). Uncertainty and the effectiveness of policy. The American Economic Review, 57(2),
411–425.
udinski-Petkovi , L.J., Lon revi , I., Jak i , .M., rhovac, S. . (2014). Fractal properties of
financial markets. Physica A: Statistical Mechanics and its Applications, 410, 43-53.
Caballero, R. J., & Pindyck, R.S. (1996). Uncertainty, investment, and industry evolution. International
Economic Review,37(3), 641-662.
Carriero, A., Clark, T. E., & Marcellino, M. (2015). Common drifting volatility in large bayesian vars,
Journal of Business and Economic Statistics, doi: 10.1080/07350015.2015.1040116
Christiano, L. J., Motto, R., & Rostagno, M. (2014). Risk shocks. American Economic Review, 104(1),
27–65.
Christiano, L.J., Eichenbaum, M., & Evans, C. (2005). Nominal rigidities and the dynamic effects of a
shock to monetary policy. Journal of Political Economy, 113(1),1-45.
Chuliá, H., Martens, M., & Van Dijk, D. (2010). Asymmetric effects of federal funds target rate changes
on s&p stock returns, volatilities and correlations. Journal of Banking and Finance, 34, 834-839.
Cochrane, J. (2005). Financial markets and the real economy. Foundations and Trends in Finance, 1(1),
1-101.
Davis, J., Fama, E., & French, K. (2000). Characteristics, covariances, and average returns: 1929–1997.
Journal of Finance, 55(1), 389–406.
Dick, C.D., Schmeling, M., & Schrimpf, A. (2013). Macro Expectations, aggregate uncertainty, and
expected term premia. European Economic Review, 58(C), 58-80.
Diebold, F., Doherty, N., & Herring R. (2010). The Known, the Unknown, and the Unknowable in
Financial Risk Management: Measurement and Theory Advancing Practice. New York:
Princeton University Press.
Diether, K. B., Malloy, C., & Scherbina, A. (2002). Differences of opinion and the cross section of stock
returns. The Journal of Finance, 57(5), 2113-2141.
Efron, B., & Tibshirani, R. J. (1993). An Introduction to the Bootstrap. New York: Chapman & Hall.
Fair, R.C. (2002). Events that shook the market. Journal of Business, 75, 713–731.
Fendoǧlu, S. (2014). Optimal Monetary policy rules, financial amplification, and uncertain business
cycles. Journal of Economics Dynamics and Control, 46, 271-305.
Forni, M., & Reichlin, L. (1998). Let’s get real: factor to analytical cycle approach dynamics
disaggregated business. The Review of Economic Studies, 65(3), 453–473.
Forni, M., Hallin, M., Lippi, M., & Reichlin, L. (2000). The generalized dynamic-factor model:
identification and estimation. The Review of Economics and Statistics, 82(4), 540–554.
Forni, M., Hallin, M., Lippi, M., & Reichlin, L. (2005). The generalized dynamic factor model: one-sided
estimation and forecasting. Journal of the American Statistical Association, 100(471), 830–840.
Ganegoda, A., & Evans, J. (2014). A framework to manage the measurable, immeasurable and the
unidentifiable financial risk. Australian Journal of Management, 39(1), 5-34.

20
Gilchrist, S., Sim, J. W., & Zakrajsek, E. (2014). Uncertainty, Financial Frictions, and Investment
Dynamics. NBER working papers 2003, http://www.nber.org/papers/w20038.
Gomery, R. (1995) .The known, the unknown and the unknowable. Scientific American, June.
Hendry, D. (1980). Econometrics: Alchemy or Science?. Economica, 47(188), 387-406.
Homm, U. & Breitung J. (2012). Testing for speculative bubbles in stock markets: a comparison of
alternative methods. Journal of Financial Econometrics, 10(1), 198-231.
Hommes, C., & Wagener, F. (2009). Complex Evolutionary Systems in Behavioral Finance. In T. Hens,
& K.R., Schenk-Hoppé(Eds.), Handbook of Financial Markets Dynamics and Evolution, (pp.
217-276). Amsterdam: North-Holland.
Jurado, K., Ludvigson, S.C. & Ng, S. (2013). Measuring Uncertainty: Supplementary Material, NBER
working paper 19456, http://www.nber.org/data-appendix/w19456/jln_supp.pdf
Jurado, K., Ludvigson, S.C., & Ng, S. (2015). Measuring uncertainty. American Economic Review. 105
(3),1177-1216.
Kastner, G. (2016). Dealing with stochastic volatility in time series using the r package stochvol. Journal
of Statistical Software, 69(5), doi 10.18637/jss.v069.i05
Kastner, G., & Frühwirth-Schnatter, S. (2014). Ancillarity-sufficiency interweaving strategy (asis) for
boosting mcmc estimation of stochastic volatility models. Computational Statistics and Data
Analysis, 76, 408-423.
Keuzenkamp, H. (2004) Probability, Econometrics And True: The Methodology Of Econometrics,
Cambridge: Cambridge University Press.
Keynes, J.M. (1921). A Treatise on Probability, New York: St Martin's Press.
Keynes, J.M. (1939). Professor Tinbergen's method. Economic Journal ,49, 558-68.
Knight, F. (1921). Risk, uncertainty, and profit. Boston: Hart, Schaffner & Marx; Houghton Mifflin Co.
Lahiri, K., & Sheng, X. (2010). Measuring forecast uncertainty by disagreement: the missing link.
Journal of Applied Econometrics, 25(4), 514-538.
Leahy, J., & Whited, T. M. (1996). The effect of uncertainty on investment: some stylized facts. Journal
of Money, Credit and Banking, 28(1), 64-83.
Nelson, S., & Katzenstein P. (2014). Uncertainty, risk, and the financial crisis of 2008. International
Organization, 68(2), 361-392.
Panousi, V., & Papanikolaou, D. (2012). Investment, idiosyncratic risk, and ownership. The Journal of
Finance, 67(3),1113-1148.
Perron, P. (2006). Dealing with Structural Breaks, in K. Patterson, & T. C. Mills (Eds.), Palgrave
Handbook of Econometrics (1), London: Palgrave MacMillan.
Petersen, K. (1996). Ergodic theorems and the basis of science. Synthese, 108(2), 171-183.
Phillips, P.C.B, & Yu, J. (2011). Dating the timeline of financial bubbles during the subprime crisis.
Quantitative Economics, 2(3), 455-491.
Phillips, P.C.B., Shi, S.P., & Yu, J. (2015). Testing for multiple bubbles: historical episodes of
exuberance and collapse in the S&P 500, International Economic Review, 56(4), 1043-1078.
Phillips, P.C.B., Whu, Y., & Yu, J. (2011). Explosive behavior in the 1990s nasdaq: when did
exuberance escalate asset values? International Economic Review.52(1), 201-226.
Ramey, G., & Ramey, V.A. (1995). Cross-country evidence on the link between volatility and growth.
American Economic Review, 85(5), 1138-1151.
Romer, C.D. (1990). The great crash and the onset of the great depression. Quarterly Journal of
Economics, 105(3), 597-624.
Scotti, C. (2016). Surprise and uncertainty indexes: real-time aggregation on real-activity macro surprises,
Journal of Monetary Economics, 82, 1-19.
Sentana, E. (2004). Factor representing portfolios in large asset markets. Journal of Econometrics,
119(2), 257-289.
Shi, Y., Ho, K-Y., & Liu, W-M. (2016). Public information arrival and stock return volatility: Evidence
from news sentiment and Markov Regime-Switching Approach. International Review of
Economics and Finance, 42, 291-312.
Sornette, D., & Zhou, W.X. (2004). Evidence of fueling of the 2000 new economy bubble by foreign
capital inflow: implications for the future of the us economy and its stock market. Physica A:
Statistical Mechanics and its Applications, 332, 412-440.
Sornette, D., Woodard, R., & Zhou, W.X. (2009). The 2006-2008 oil bubble: evidence of speculation and
prediction. Physica A: Statistical Mechanics and its Applications, 388(8), 1571-1576.
Stock, J. H., & Watson, M. W. (2002). Forecasting using principal components from a large number of
predictors. Journal of the American Statistical Association, 97(460), 1167–1179.
Stock, J. H., & Watson, M. W. (2011). Dynamic factor models. In M.P. Clements & D.F. Hendry (Eds.),
The Oxford Handbook of Economic Forecasting, (pp. 35-59). Oxford: Oxford University Press.

21
Taleb, N.N. (2007).The black swan: the impact of the highly improbable. New York: Random House.
Williams, J.C. (2013). A defense of moderation in monetary policy. Journal of Macroeconomics, 38
(PB), 137-150.
Williams, N. (2012). Monetary policy under financial uncertainty. Journal of Monetary Economics,
59(5), 449-465.
Yu, Y., & Meng, X.L. (2011). To center or not to center: that is not the question—an ancillarity-suffiency
Interweaving Strategy (ASIS) for Boosting MCMC Efficiency. Journal of Computational and
Graphical Statistics,20(3), 531–570.
Yuhn, K.H., Kim, S.B., & Nam, J. H. (2015). Bubbles and the Weibull distribution: was there an
explosive bubble in us stock prices before the global economic crisis? Applied Economics, 47(3),
255-271.
Zhou, W. X., & Sornette, D. (2003). 2000-2003 real state bubble in the UK but not in the USA. Physica
A: Statistical Mechanics and its Applications, 329(1-2), 249-263.

22
 APPENDIX

In the estimations we make use of some routines from the web page of Serena Ng
(http://www.columbia.edu/~sn2294/) to estimate the DFM, and to select the optimal
number of static and dynamic factors. To estimate the GDFM, both, one-side and two-
sides filters, we use codes from the web page of Mario Forni.
(http://morgana.unimore.it/forni_mario/matlab.htm). To estimate stochastic volatilities
we use the r-package ‘stochvol’ (Kastner, 2016), to estimate structural breaks in the
index we employ the r-package ‘strucchange’ and to estimate the VAR model the r-
package ‘vars’ was used.

Table 2: Variance Decomposition of the Forecast Errors

Industrial Production

Period 1 6 12 24 48 Max

Ind. Production 95.2% 68.2% 41.8% 23.7% 16.8% 95.2%

Employment 0.7% 3.6% 3.2% 2.1% 5.3% 7.1%
Consumption 0.1% 0.2% 1.0% 0.8% 1.7% 2.2%
Inflation 0.3% 0.2% 2.4% 15.4% 17.0% 18.7%
New Orders 2.5% 8.1% 4.6% 4.9% 3.6% 8.2%
Wage 0.0% 0.1% 0.2% 0.5% 1.0% 1.1%
Hours 0.8% 0.6% 0.4% 0.7% 0.4% 0.9%
R 0.0% 1.6% 4.5% 12.8% 26.0% 26.3%
S&P500 0.0% 5.0% 11.8% 9.8% 6.8% 13.7%
M2 0.0% 1.8% 6.3% 7.7% 7.7% 7.9%
Uncertainty 0.3% 10.5% 23.8% 21.7% 13.7% 25.3%

New Orders

Period 1 6 12 24 48 Max

Ind. Production 10.9% 7.5% 8.4% 7.7% 7.3% 10.9%

Employment 3.1% 5.3% 5.9% 5.4% 5.0% 6.1%
Consumption 2.9% 1.9% 1.8% 1.5% 1.4% 3.1%
Inflation 1.9% 2.7% 9.2% 12.6% 12.6% 12.8%
New Orders 78.7% 48.2% 39.9% 33.8% 31.5% 78.7%
Wage 0.0% 0.3% 0.4% 0.5% 0.5% 0.5%
Hours 0.5% 0.8% 1.7% 1.5% 1.5% 1.7%
R 0.0% 5.7% 7.2% 8.8% 9.7% 10.5%
S&P500 1.6% 4.9% 4.5% 10.5% 12.7% 13.3%
M2 0.2% 1.2% 1.5% 1.4% 1.4% 1.6%
Uncertainty 0.1% 21.5% 19.5% 16.4% 16.4% 22.6%

23
 Consumption

Period 1 6 12 24 48 Max

Ind. Production 2.9% 5.3% 3.9% 2.1% 1.7% 6.7%

Employment 0.7% 4.8% 3.5% 1.8% 3.4% 5.3%
Consumption 93.8% 62.7% 45.0% 31.9% 25.4% 93.8%
Inflation 0.6% 6.4% 14.4% 24.3% 25.4% 26.1%
New Orders 0.3% 0.8% 2.1% 5.0% 4.8% 5.2%
Wage 0.0% 0.3% 0.4% 0.3% 0.4% 0.4%
Hours 0.0% 0.8% 1.0% 0.9% 0.7% 1.1%
R 0.5% 7.4% 12.1% 19.0% 23.6% 23.8%
S&P500 0.7% 3.9% 4.8% 3.3% 2.1% 5.0%
M2 0.2% 2.3% 5.1% 6.6% 9.5% 10.8%
Uncertainty 0.3% 5.3% 7.6% 4.7% 3.1% 7.8%

Employment

Period 1 6 12 24 48 Max

Ind. Production 32.8% 29.5% 19.1% 11.8% 8.8% 35.1%

Employment 66.1% 53.2% 42.5% 26.3% 11.5% 66.1%
Consumption 0.1% 0.6% 0.4% 0.5% 0.3% 0.8%
Inflation 0.0% 0.1% 0.8% 9.0% 13.3% 14.1%
New Orders 0.7% 4.4% 2.3% 1.9% 2.0% 4.5%
Wage 0.1% 0.1% 0.3% 0.8% 1.4% 1.4%
Hours 0.1% 0.1% 0.4% 1.8% 2.2% 2.3%
R 0.0% 2.5% 7.4% 19.6% 41.4% 44.5%
S&P500 0.1% 3.9% 10.4% 9.2% 7.5% 12.5%
M2 0.0% 0.9% 3.3% 4.2% 3.4% 4.2%
Uncertainty 0.0% 4.6% 13.2% 14.7% 8.2% 15.5%

Standard & Poor's 500

Period 1 6 12 24 48 Max

Ind. Production 0.3% 0.5% 0.4% 0.5% 1.2% 1.2%

Employment 0.1% 1.3% 2.7% 4.1% 5.6% 6.2%
Consumption 0.3% 0.9% 1.8% 1.8% 1.6% 1.8%
Inflation 0.5% 0.4% 4.0% 6.9% 5.8% 6.9%
New Orders 0.3% 1.3% 3.8% 5.7% 4.6% 5.7%
Wage 0.0% 0.2% 2.2% 4.3% 8.0% 9.0%
Hours 0.6% 1.0% 0.8% 0.9% 1.0% 1.0%
R 1.0% 1.5% 1.1% 1.2% 2.1% 2.1%
S&P500 94.5% 73.6% 63.6% 54.1% 44.7% 94.5%
M2 0.2% 3.4% 3.6% 3.9% 3.7% 4.0%
Uncertainty 2.2% 15.9% 15.9% 16.7% 21.7% 23.4%

24
 Federal Funds -R

Period 1 6 12 24 48 Max

Ind. Production 0.0% 6.4% 5.4% 4.9% 6.2% 6.5%

Employment 0.0% 1.7% 6.5% 8.6% 8.2% 9.1%
Consumption 0.0% 0.5% 2.5% 3.3% 8.5% 11.0%
Inflation 0.0% 2.2% 3.7% 3.5% 4.0% 4.0%
New Orders 0.0% 10.6% 11.2% 9.2% 7.6% 11.2%
Wage 0.0% 0.8% 0.7% 0.8% 0.8% 0.9%
Hours 0.0% 1.0% 1.1% 1.1% 1.3% 1.3%
R 0.0% 72.8% 55.9% 47.8% 42.2% 91.7%
S&P500 0.0% 1.7% 6.8% 13.3% 14.4% 16.9%
M2 0.0% 0.5% 1.6% 1.7% 1.5% 1.7%
Uncertainty 0.0% 1.9% 4.7% 5.9% 5.4% 6.1%

Uncertainty

Period 1 6 12 24 48 Max

Ind. Production 0.5% 1.9% 2.4% 2.0% 2.2% 2.4%

Employment 0.1% 0.8% 1.0% 1.4% 1.2% 1.5%
Consumption 0.0% 0.5% 1.6% 1.3% 1.1% 1.6%
Inflation 0.4% 2.6% 5.9% 4.8% 5.6% 6.0%
New Orders 0.1% 0.3% 0.4% 1.0% 2.0% 2.1%
Wage 0.0% 0.7% 3.7% 3.5% 3.3% 4.3%
Hours 0.0% 0.7% 1.4% 1.9% 2.2% 2.2%
R 0.0% 0.1% 0.2% 4.0% 4.8% 5.0%
S&P500 1.3% 3.8% 7.1% 22.6% 28.2% 28.6%
M2 1.8% 3.1% 3.1% 2.6% 3.3% 3.4%
Uncertainty 95.7% 85.6% 73.2% 54.9% 46.1% 95.7%

NOTE: We use a VAR (12) comprising 11 variables, in the following Cholesky-order from
contemporaneously exogenous to contemporaneously endogenous: Production, Employment,
Consumption, Inflation, New Orders, Wages, Labor, (Federal Funds Rate), Stock Market Index, M2
and the Uncertainty Index. All the variables are in logs except the Fed rate in percentage, the uncertainty
index in units and M2 in growth rates.

Highlights
 A daily index of time-varying stock market uncertainty is proposed.
 Stock market uncertainty reacts to economic recessions, but also to bubble
episodes.
 Uncertainty impacts negatively investment, consumption, production and
employment.
 The impact of uncertainty on stock prices is negative and persistent.

25