Measuring Uncertainty: Supplementary Material

Kyle Jurado Sydney C. Ludvigson Serena Ng

Columbia University NYU and NBER Columbia University
October 2, 2014

Abstract
This document contains supplementary material for the paper entitled “Measuring Uncer-
tainty”and has two parts. The …rst part provides the results of robustness exercises based on (i)
alternative weights used to aggregate individual uncertainty series; (ii) alternative estimates of
individual uncertainty; (iii) alternative conditioning information using recursive out-of-sample
forecasts to construct di¤usion index forecasts; (iv) dynamic responses in a VAR identical to
Bloom (2009) where HP …ltered data were used. The second part is a data appendix that
contains details on the construction of all data used in this study, including data sources.

Jurado: Department of Economics, Columbia University, 1019 International A¤airs Building, MC 3308, 420
West 118th Street, New York, NY 10027; Email: [email protected].
Ludvigson: Department of Economics, New York University, 19 W.4th Street, 6th Floor, New York, NY 10012;
Email: [email protected]; http://www.econ.nyu.edu/user/ludvigsons/.
Ng: Department of Economics, Columbia University, 1019 International A¤airs Building, MC 3308, 420 West
118th Street, New York, NY 10027; Email: [email protected]; http://www.columbia.edu/~sn2294/.
1 Robustness
y
Our baseline estimate of macro uncertainty U t (h) is constructed as the cross-sectional average
of the individual uncertainties Ujty (h), and each of these is based on evaluating (??) at the
posterior mean, over the full sample, of the state and parameters of the stochastic volatility
y 2 y y y
model (i.e.,flog( jt ) g, j , j ; and j ) and the OLS parameter estimates from the forecasting
y F W
model (i.e., j, j (L), and j (L)). This section assesses robustness of the results to these
assumptions.

1.1 Macro Uncertainty Factor

We …rst entertain the possibility that uncertainty has a factor structure. In such a case, macro
uncertainty at each t is a vector given by the common factor FtU (h) in

log Ujty (h) = cUj (h) + U0 U

hj Ft (h) + eUjt (h): (1)

Macro uncertainty is then summarized by FtU (h) while idiosyncratic uncertainty is eUhjt . Al-
though Ujty (h) is always positive, the principal components estimates do not constrain the (nor-
malized) estimated factors themselves to be positive. The log speci…cation is therefore used to
insure that both the domain and the range of the function (1) take on values on the entire real
line R. As a consequence of this log speci…cation, our PCA estimate of macro uncertainty Uty (h)
is the exponential of the PCA estimate FbU (h): Let Ubty (h) exp FbU (h) : To obtain such an
t t
estimate, we …rst need an estimate of the the common (log) uncertainty factor FtU (h). As many
Pt bU
uncertainty series appear non-stationary, this estimate is de…ned by FbtU (h) = f (h), k=2 k
y
where ftU
(h) is an rU 1 vector comprised of the rU principal components of log Ujt (h).1
As discussed in Bai and Ng (2004), this di¤erencing-recumulating approach ensures that the
factors are consistently estimated when the idiosyncratic errors are potentially non-stationary.
Because of the di¤erencing, the initial value in the sample of the common uncertainty factor,
F^1U (h); is not identi…ed. We initialize F^1U (h) to the average level of (log) uncertainty across all
P
N series; mathematically, N1 N y
j=1 log Uj1 (h).
The problem of determining rU , the number of common uncertainty factors f U (h) ; is non-
standard because the individual uncertainty measures are themselves estimated. Existing crite-
ria for determining the number of factors do not take the …rst step estimation error into account
and will likely overestimate the number of factors. However, there is strong evidence of a factor
structure as the largest eigenvalue of forecast error variance is distinctly large. In particular,
1 y
We observe log Ujt (h), a data matrix with T time-series observations and N cross-section observations. The
…rst di¤erenced data yield a (T 1) N vector of stationary variables. Let f U (h) f1U (h) ; f2U (h) ; :::; fTU (h)
0 U
and U = ( U 1 ; U 2 ; ::: U N ) : The principal component estimator of f (h) is the T 1 times the rU eigenvec-
y y 0
tors corresponding to the …rst rU largest eigenvalues of the (T 1) (T 1) matrix log Ujt (h) log Ujt (h) :

1
the …rst principal component of Ujty (h) explains 11% of the variance of the forecast errors for
h = 1, 14% for h = 3, and 22% for h = 12. We take rU to be one, which facilitates comparison
y
with the base-case estimate U t (h) that is based on simple averaging. We also calibrate the
uncertainty factor Ubty (h) to have the same mean and standard deviation as U t (h) over the
sample.
The right panel of Table 1 shows that the results using Ubty (h) are qualitatively and quan-
titatively similar to the base-case. The relative importance of the uncertainty factor and idio-
2
syncratic uncertainty is summarized in a Rjt (h) statistic analogous to (??). The main …nding
continues to be that variations in macro uncertainty constitute a larger fraction of variations
in individual uncertainty measures at longer horizons, and during recessions. Table 4 (second
column) also reports results for the eight variable VAR, but with U t (h) replaced by recursive
PCA estimates of uncertainty, Ubty (h). The uncertainty factor has very similar dynamic e¤ects
on production, employment, and hours as U t (h). If anything, the e¤ects due to the uncertainty
are somewhat larger than the base-case of equal weighting.

1.2 Alternative Estimates of Uncertainty

We next consider alternative estimates of individual uncertainty, and alternative ways of aggre-
gating these estimates to get macro uncertainty. The base-case implementation only requires
one evaluation of uncertainty for each series j since the posterior mean of each parameter is one
dimensional. Speci…cally, for h = 1; uncertainty in the variable j evaluated at the sth Monte
Carlo draw is q
: 2
Ujst (h) ( js ; xjst ) = exp js + js =2 + js xjst ;
y 2
where xjst ln( jst ) . When the function above is evaluated at the posterior mean (over all
s = 1; :::; S draws) of the parameters, we denote that Ujt (h) j ; xjt . In this notation, our base
case uncertainty estimate for the series j is Ujt (h) j ; xjt . But an uncertainty estimate can
also be obtained for each draw of the hyperparameters in the model for series j. Thus one can
also estimate Ujty (h) by the posterior mean of the draws of uncertainty for series j. In this case
P
we de…ne individual uncertainty as UjtS (h) = S1 Ss=1 Ujst (h) ( js ; xjst ), where the superscript S
denotes all S draws are used in the computation.2 Instead of the posterior mean, it is also
[s]
possible to consider other location statistics. Let Ujt (h) be the s-th percentile draw in the
sorted sequence of fUjst (h)gSs=1 . If [s] is 50, the median obtains. We use the 90th and the
10th percentiles of the posterior distribution of Ujst (h) ( js ; xjst ) to assess how extreme values
10 90
of individual uncertainty a¤ect aggregate uncertainty. These are denoted U t (h) and U t (h),
respectively.
2
To estimate the latter requires saving every posterior draw of Ujst (h) ( js ; xjst ) and is considerably
more computationally demanding than the base-case where uncertainty is evaluated once at the mean of the
parameters.

2
 Since we have three ways of estimating individual uncertainties two ways of aggregating
them, we have six measures of macro uncertainty summarized as follows:

Ut (h) Aggregator Ujt (h)

Baseline CSA: U t (h) CSA Ujt (h) j ; xjt

Baseline PCA: Ubt (h) PCA Ujt (h) j ; xjt
S 1
PS
Posterior Mean CSA: U t (h) CSA S Ps=1 jst
U (h) ( js ; xjst )
Posterior Mean PCA: UbtS (h) PCA 1
S
S
s=1 Ujst (h) ( js ; xjst )
[s] [s]
Posterior s-Percentile CSA: U t (h) CSA Ujt (h)
Posterior s-Percentile PCA: Ubt (h)
[s] [s]
PCA Ujt (h)

where CSA stands for simple averaging over Ny series, and PCA stands for for the principal
component of the Ny individual uncertainties constructed using the methodology as discussed
above.
Figure (1) shows the baseline and posterior mean estimates of aggregate uncertainty when
h = 1. Each of these measures are highly correlated with one another. Indeed, the estimates
based on the average across draws of the parameters versus the posterior mean of the uncertainty
draws are virtually indistinguishable. The estimates based on cross-section averaging are also
very highly correlated with those based on the principal component estimates. Given the
similarity between the CSA and PCA estimates, Figure (2) shows our base-case estimate of
S 10 90
uncertainty U t (h), the CSA variant of U t (h), along with the CSA variant of U t (h) and U t (h).
As for the above variations, di¤erent percentiles of the distribution have the e¤ect of shifting our
estimate of uncertainty by a constant amount only but do not much a¤ect the dynamics of our
uncertainty estimates. The 90th and 10th percentiles of the distribution have a correlation with
our baseline estimate each in excess of 0.998. We conclude that results regarding the number of
large uncertainty episodes, their timing, or their dynamic relation with economic activity are
robust to using more extreme estimates of individual uncertainty. Overall, the results suggest
that the …ndings reported above are not sensitive to using these alterative estimates of aggregate
uncertainty.
Finally, we consider using GARCH or EGARCH to estimate the volatility of individual
series. Figure (3) shows that, when we aggregate in exactly the same way, our estimates
of aggregate uncertainty over time are very similar to the baseline stochastic volatility case.
Results based on the GARCH/EGARCH estimates indicate the number and timing of big
uncertainty episodes, as well as the persistence of uncertainty, is very similar to that reported
here using our base-case measure of macrouncertainty. What is is di¤erent is the real e¤ect of
uncertainty innovations from a VAR, once orthogonalized shocks are analyzed. This is to be
expected because GARCH type models (unlike stochastic volatility) have a shock to the second
moment that is not independent of the …rst moment. This is inconsistent with the assumptions

3
of an independent uncertainty shock presumed in the uncertainty literature. Using a GARCH-
based uncertainty index thus creates additional identi…cation problems that are beyond the
scope of this paper.

1.3 Recursive Out-of-Sample Estimation

y F W
We next consider the sensitivity of the forecasting parameters j, j (L), and j (L) to the
estimation sample. Instead of full sample estimation (and hence in-sample forecasts), we also
form out-of-sample forecasts for the monthly macro dataset.3 This procedure involves fully
recursive factor estimation and parameter estimation using data only through time t for fore-
casting at time t + 1. Notice that, since the forecasting parameters evolve over time as new data
becomes available, such recursive forecasts are informative about the extent to which parame-
ter instability in the conditional mean forecasting relation in‡uences the uncertainty estimates.
We use the …rst 10 years of data (t = 1,2,...,120, 1959:01-1969:01) as an initial estimation
period to estimate both the factors and the parameters of the conditional mean (forecasting)
regression, and to perform model selection. Next, the forecasting regressions are run over the
period t =1959:01,...,1969:01, and the values of the regressors at t =1969:01 are used to fore-
cast yj1969:02 . All parameters, factors and model selection criteria are then re-estimated from
1959:01 through 1969:02, and forecasts are recomputed for yj1969:03 , and so on, until the …nal
out-of-sample forecast is made for yj2011:12 : Since our dataset has 622 months total, this leaves
y
502=622-120 forecast errors. The forecast error variances are used to compute U jt (h), and
averaging over j gives macro uncertainty. The resulting uncertainty estimate is plotted in Fig-
ure 4 along with the original estimate. The measure is extremely highly correlated with that
based on in-sample forecasts.4 Although use of the full sample slightly under-states the level of
uncertainty, it does an excellent job of capturing its time-series variation, only in‡uencing the
estimates by a constant amount. We can con…rm that our VAR analysis is little e¤ected by
whether we use out-of-sample or in-sample forecasts, having virtually no bearing on the num-
ber of uncertainty episodes, their timing, or their dynamic relationship with economic activity.
These …ndings are consistent with evidence that dynamic factor analysis provides robustness
against the temporal parameter instability that often plagues low-dimensional forecasting re-
gressions (Stock and Watson (2002a)). The reason is that such instabilities can “average out”in
the construction of common factors if the instability is su¢ ciently dissimilar from one series to
the next. In the recursive VAR estimation the parameters of the forecasting relation change
every period, so this speaks directly to the question of the role played by parameter stability
3
This procedure closely follows the real-time simulation procedure of Stock and Watson (2002b).
4
Note that this measure is feasible to compute only for h = 1. The multi-step ahead forecasts that are
needed for uncertainty with h > 1 are computed once by rolling forward one-step ahead forecasts from the
VAR. Recomputing the VAR in every time period would require recomputing uncertainty in every time period,
which is not possible in reasonable time.

4
in our estimates.
The recursive out-of-sample approach is only feasible in the h = 1 case. This is because we
obtain our estimates of uncertainty for h > 1 by are computed once by rolling ahead one-step-
ahead forecasts from the VAR stacked in companion form. By design, this approach relies on
the parameters of the VAR being …xed over the sample. Nevertheless we …nd the robustness
of the results in the h = 1 case along this dimension to be comforting and suggestive of what
would be likely for the other cases.

1.4 Bloom (2009) VAR

The Bloom VAR results thus far have used an ordering that puts uncertainty second in a list of
eight variables, following Bloom (2009). Table 5 reports VAR variance decomposition results
with uncertainty ordered last to allow uncertainty to respond contemporaneously to the …ve
variables ordered after it. Figure 6 reports the impulse responses to orthogonal shocks created
from a Cholesky decomposition of the VAR with this alternative ordering. Some variations
previously attributed to uncertainty are now allocated to the orthogonalized innovations in the
fed funds rate, wages, CPI, hours, employment, and industrial production. This is not surprising
because our measure of uncertainty is contemporaneously correlated with these measures of
economic activity, thus once we remove the variation in uncertainty that is attributable to
these correlations, the e¤ect is smaller. We again caution, however, that these results as well as
the previous ones tell us only about dynamic correlations (not true causality) and di¤er only
because of a change in the assumption about the timing of shocks. For the sake of comparison,
the last column of Table 5 reports results with VXO ordered last. As documented earlier, stock
market volatility and uncertainty are correlated but have signi…cant independent variations. As
expected, because our measures of uncertainty are more highly contemporaneously correlated
with real activity than is VXO, the e¤ect on production, employment, and hours attributed to
uncertainty shocks is smaller compared to the results in Table 2 when uncertainty is ordered
second. By contrast, the decomposition of forecast error variances to VXO shocks is not greatly
a¤ected by the ordering of VXO in the VAR, implying that VXO shocks are not as strongly
contemporaneously correlated with the …ve real activity variables in the system as are our
uncertainty estimates. These results reinforce the conclusion that the stock market can move
signi…cantly in the absence changes in fundamentals in the economy. It is thus not a good proxy
for macroeconomic uncertainty, which we have found does move with these fundamentals.

5
 Variance Decompositions from VAR(12)
Uncertainty Ordered Last
k U(1) U(3) U(12) VXO
Production
1 0.00 0.00 0.00 0.00
3 1.16 1.31 1.03 1.04
12 6.18 8.95 6.11 5.84
1 5.51 7.26 6.33 4.14
max 6.78 9.45 6.62 7.19
max k 10 10 10 8
Employment
1 0.00 0.00 0.00 0.00
3 0.60 0.59 0.43 1.11
12 5.97 9.20 6.58 8.88
1 4.99 7.03 6.18 5.18
max 6.05 9.20 6.58 9.61
max k 11 12 12 9
Hours
1 0.00 0.00 0.00 0.00
3 1.42 1.57 0.89 1.70
12 5.82 8.00 5.56 7.12
1 5.94 7.97 6.81 5.98
max 6.21 8.40 6.81 7.86
max k 8 10 38 8

y
Table 5: Eight-variable VAR(12) using the VXO Index or U t (h) for h = 1; 3; 12 as a
measure of uncertainty, estimated from the monthly macro dataset. Each VAR(12) contains,
in the following order: log(S&P 500 Index), federal funds rate, log(wages), log(CPI), hours,
log(employment), log(industrial production), and uncertainty. As in Bloom (2009), all variables
are HP …ltered, except for the uncertainty measures, which enter in raw levels. The data are
monthly and span the period 1960:07-2011:12.

6
2 Data Appendix
The …rst dataset, denoted X m , is an updated version of the of the 132 mostly macroeconomic
series used in Ludvigson and Ng (2010). The 132 macro series in X m are selected to represent
broad categories of macroeconomic time series: real output and income, employment and hours,
real retail, manufacturing and trade sales, consumer spending, housing starts, inventories and
inventory sales ratios, orders and un…lled orders, compensation and labor costs, capacity utiliza-
tion measures, price indexes, bond and stock market indexes, and foreign exchange measures.
The 147 …nancial series in X f consists of a number of indicators measuring the behavior
of a broad cross-section of asset returns, as well as some aggregate …nancial indicators not
included in the macro dataset. These data include valuation ratios such as the dividend-price
ratio and earnings-price ratio, growth rates of aggregate dividends and prices, default and term
spreads, yields on corporate bonds of di¤erent ratings grades, yields on Treasuries and yield
spreads, and a broad cross-section of industry equity returns. Following Fama and French
(1992), returns on 100 portfolios of equities sorted into 10 size and 10 book-market categories.
The dataset X f also includes a group of variables we call “risk-factors,” since they have been
used in cross-sectional or time-series studies to uncover variation in the market risk-premium.
These risk-factors include the three Fama and French (1993) risk factors, namely the excess
return on the market M KTt , the “small-minus-big” (SM Bt ) and “high-minus-low” (HM Lt )
portfolio returns, the momentum factor U M Dt , the bond risk premia factor of Cochrane and
Piazzesi (2005), and the small stock value spread R15 R11.
The raw data used to form factors are always transformed to achieve stationarity. In addi-
tion, when forming forecasting factors from the large macro and …nancial datasets, the raw data
(which are in di¤erent units) are standardized before performing PCA. When forming common
uncertainty from estimates of individual uncertainty, the raw data (which are in this case in
the same units) are demeaned, but we do not divide by the observation’s standard deviation
before performing PCA.
Throughout, the factors are estimated by the method of static principal components (PCA).
p
Speci…cally, the T rF matrix F^t is T times the rF eigenvectors corresponding to the rF
largest eigenvalues of the T T matrix xx0 =(T N ) in decreasing order. In large samples (when
p
T =N ! 1), Bai and Ng (2006) show that the estimates F^t can be treated as though they
were observed in the subsequent forecasting regression. There is no need to correct standard
errors for uncertainty in this estimate, unlike the generated regressor case analyzed in Pagan
(1984) when N is …xed. This asymptotic result allows for time variation in the volatility of the
forecast error.

7
2.1 Macro Dataset
This appendix lists the short name of each series in the macro dataset, its code in the source
database, the transformation applied to the series, and a brief data description. All series are
from the IHS Global Insights database, unless the source is listed (in parentheses) as FRED
(St. Louis Federal Reserve Economic Data), BLS (Bureau of Labor Statistics), S (R. J. Shiller
website), BEA (Bureau of Economic Analysis), IMF (IMF International Financial Statistics
database), B (R Barnichon website), UM (Thomson Reuters/University of Michigan Surveys
of Consumers) or AC (author’s calculation). The data are available from 1959:01-2011:12.
Let Xit denote variable i observed at time t after e.g., logarithm and di¤erencing transfor-
mation, and let XitA be the actual (untransformed) series. Let = (1 L) with LXit = Xit 1 .
There are six possible transformations with the following codes:

1 Code lv: Xit = XitA .

2 Code lv: Xit = XitA XitA 1 .

2 2
3 Code lv: Xit = XitA .

4 Code ln: Xit = ln(XitA ).

5 Code ln: Xit = ln(XitA ) ln(XitA 1 ).

2 2
6 Code ln: Xit = lnXitA .

Group 1: Output and Income

No. Gp Short Name Code Tran Descripton
1 1 PI M_14386177 ln Personal Income
6 1 IP: total M_116460980 ln Industrial Production Index - Total Index
7 1 IP: products M_116460981 ln Industrial Production Index - Products, Total
8 1 IP: …nal prod M_116461268 ln Industrial Production Index - Final Products
9 1 IP: cons gds M_116460982 ln Industrial Production Index - Consumer Goods
10 1 IP: cons dble M_116460983 ln Industrial Production Index - Durable Consumer Goods
11 1 IP: cons nondble M_116460988 ln Industrial Production Index - Nondurable Consumer Goods
12 1 IP: bus eqpt M_116460995 ln Industrial Production Index - Business Equipment
13 1 IP: matls M_116461002 ln Industrial Production Index - Materials
14 1 IP: dble matls M_116461004 ln Industrial Production Index - Durable Goods Materials
15 1 IP: nondble matls M_116461008 ln Industrial Production Index - Nondurable Goods Materials
16 1 IP: mfg M_116461013 ln Industrial Production Index - Manufacturing
17 1 IP: res util M_116461276 ln Industrial Production Index - Residential Utilities
18 1 IP: fuels M_116461275 ln Industrial Production Index - Fuels
19 1 NAPM prodn M_110157212 lv Napm Production Index
20 1 Cap util M_116461602 lv Capacity Utilization

8
 Group 2: Labor Market
No. Gp Short Name Code Tran Descripton
21 2 Help wanted indx - lv Index Of Help-Wanted Advertising (B)
22 2 Help wanted/unemp M_110156531 lv Ratio of Help-Wanted Ads/No. Unemployed (AC)
23 2 Emp CPS total M_110156467 ln Civilian Labor Force: Employed, Total
24 2 Emp CPS nonag M_110156498 ln Civilian Labor Force: Employed, Nonagric.Industries
25 2 U: all M_110156541 lv Unemployment Rate: All Workers, 16 Years & Over
26 2 U: mean duration M_110156528 lv Unemp By Duration: Average Duration In Weeks
27 2 U < 5 wks M_110156527 ln Unemploy By Duration: Persons Unempl Less Than 5 Wks
28 2 U 5-14 wks M_110156523 ln Unemploy By Duration: Persons Unempl 5 To 14 Wks
29 2 U 15+ wks M_110156524 ln Unemploy By Duration: Persons Unempl 15 Wks +
30 2 U 15-26 wks M_110156525 ln Unemploy By Duration: Persons Unempl 15 To 26 Wks
31 2 U 27+ wks M_110156526 ln Unemploy By Duration: Persons Unempl 27 Wks +
32 2 UI claims M_15186204 ln Initial Claims for Unemployement Insurance
33 2 Emp: total M_123109146 ln Employees On Nonfarm Payrolls: Total Private
34 2 Emp: gds prod M_123109172 ln Employees On Nonfarm Payrolls - Goods-Producing
35 2 Emp: mining M_123109244 ln Employees On Nonfarm Payrolls - Mining
36 2 Emp: const M_123109331 ln Employees On Nonfarm Payrolls - Construction
37 2 Emp: mfg M_123109542 ln Employees On Nonfarm Payrolls - Manufacturing
38 2 Emp: dble gds M_123109573 ln Employees On Nonfarm Payrolls - Durable Goods
39 2 Emp: nondbles M_123110741 ln Employees On Nonfarm Payrolls - Nondurable Goods
40 2 Emp: services M_123109193 ln Employees On Nonfarm Payrolls - Service-Providing
41 2 Emp: TTU M_123111543 ln Employees On Nonfarm Payrolls - Trade, Transportation, And Utilities
42 2 Emp: wholesale M_123111563 ln Employees On Nonfarm Payrolls - Wholesale Trade.
43 2 Emp: retail M_123111867 ln Employees On Nonfarm Payrolls - Retail Trade
44 2 Emp: FIRE M_123112777 ln Employees On Nonfarm Payrolls - Financial Activities
45 2 Emp: Govt M_123114411 ln Employees On Nonfarm Payrolls - Government
*46 2 Agg wkly hours - lv Index of Aggregate Weekly Hours (BLS)
*47 2 Avg hrs M_140687274 lv Avg Weekly Hrs of Prod or Nonsup Workers Private Nonfarm - Goods-Producing
*48 2 Overtime: mfg M_123109554 lv Avg Weekly Hrs of Prod or Nonsup Workers Private Nonfarm - Mfg Overtime
*49 2 Avg hrs: mfg M_14386098 lv Average Weekly Hours, Mfg.
50 2 NAPM empl M_110157206 lv NAPM Employment Index
2
129 2 AHE: goods M_123109182 ln Avg Hourly Earnings of Prod or Nonsup Workers Private Nonfarm - Goods-Produ
2
130 2 AHE: const M_123109341 ln Avg Hourly Earnings of Prod or Nonsup Workers Private Nonfarm - Construction
2
131 2 AHE: mfg M_123109552 ln Avg Hourly Earnings of Prod or Nonsup Workers Private Nonfarm - Manufacturin

9
 Group 3: Housing
No. Gp Short Name Code Tran Descripton
*51 3 Starts: nonfarm M_110155536 ln Housing Starts:Nonfarm(1947-58);Total Farm&Nonfarm(1959-)
*52 3 Starts: NE M_110155538 ln Housing Starts:Northeast
*53 3 Starts: MW M_110155537 ln Housing Starts:Midwest
*54 3 Starts: South M_110155543 ln Housing Starts:South
*55 3 Starts: West M_110155544 ln Housing Starts:West
*56 3 BP: total M_110155532 ln Housing Authorized: Total New Priv Housing Units
*57 3 BP: NE M_110155531 ln Houses Authorized By Build. Permits:Northeast
*58 3 BP: MW M_110155530 ln Houses Authorized By Build. Permits:Midwest
*59 3 BP: South M_110155533 ln Houses Authorized By Build. Permits:South
*60 3 BP: West M_110155534 ln Houses Authorized By Build. Permits:West

Group 4: Consumption, Orders, and Inventories

No. Gp Short Name Code Tran Descripton
61 4 PMI M_110157208 lv Purchasing Managers’Index
62 4 NAPM new ordrs M_110157210 lv Napm New Orders Index
63 4 NAPM vendor del M_110157205 lv Napm Vendor Deliveries Index
64 4 NAPM Invent M_110157211 lv Napm Inventories Index
65 4 Orders: cons gds M_14385863 ln Mfrs’New Orders, Consumer Goods And Materials
66 4 Orders: dble gds M_14386110 ln Mfrs’New Orders, Durable Goods Industries
67 4 Orders: cap gds M_178554409 ln Mfrs’New Orders, Nondefense Capital Goods
68 4 Unf orders: dble M_14385946 ln Mfrs’Un…lled Orders, Durable Goods Indus.
69 4 M&T invent M_15192014 ln Manufacturing And Trade Inventories
70 4 M&T invent/sales M_15191529 lv Ratio, Mfg. And Trade Inventories To Sales
3 4 Consumption M_123008274 ln Real Personal Consumption Expenditures (AC)
4 4 M&T sales M_110156998 ln Manufacturing And Trade Sales
5 4 Retail sales M_130439509 ln Sales Of Retail Stores
132 4 Consumer expect hhsntn lv U. Of Mich. Index Of Consumer Expectations (UM)

Group 5: Money and Credit

No. Gp Short Name Code Tran Descripton
2
71 5 M1 M_110154984 ln Money Stock: M1
2
72 5 M2 M_110154985 ln Money Stock: M2
2
73 5 Currency M_110155013 ln Money Stock: Currency held by the public
74 5 M2 (real) M_110154985 ln Money Supply: Real M2 (AC)
2
75 5 MB M_110154995 ln Monetary Base, Adj For Reserve Requirement Changes
2
76 5 Reserves tot M_110155011 ln Depository Inst Reserves:Total, Adj For Reserve Req Chgs
2
77 5 Reserves nonbor M_110155009 ln Depository Inst Reserves:Nonborrowed,Adj Res Req Chgs
2
78 5 C&I loans BUSLOANS ln Commercial and Industrial Loans at All Commercial Banks (FRED)
79 5 C&I loans BUSLOANS lv Change in Commercial and Industrial Loans at All Commercial Banks (FRED)
2
80 5 Cons credit M_110155009 ln Consumer Credit Outstanding - Nonrevolving
81 5 Inst cred/PI M_110154569 lv Ratio, Consumer Installment Credit To Personal Income

10
 Group 6: Bond and Exchange Rates
No. Gp Short Name Code Tran Descripton
86 6 Fed Funds M_110155157 lv Interest Rate: Federal Funds
87 6 Comm paper CPF3M lv 3-Month AA Financial Commercial Paper Rate (FRED)
88 6 3 mo T-bill M_110155165 lv Interest Rate: U.S.Treasury Bills,Sec Mkt,3-Mo.
89 6 6 mo T-bill M_110155165 lv Interest Rate: U.S.Treasury Bills,Sec Mkt,6-Mo.
90 6 1 yr T-bond M_110155165 lv Interest Rate: U.S.Treasury Const Maturities,1-Yr.
91 6 5 yr T-bond M_110155174 lv Interest Rate: U.S.Treasury Const Maturities,5-Yr.
92 6 10 yr T-bond M_110155169 lv Interest Rate: U.S.Treasury Const Maturities,10-Yr.
93 6 Aaa bond M_14386682 lv Bond Yield: Moody’s Aaa Corporate
94 6 Baa bond M_14386683 lv Bond Yield: Moody’s Baa Corporate
95 6 CP-FF spread - lv CP-FF spread (AC)
96 6 3 mo-FF spread - lv 3 mo-FF spread (AC)
97 6 6 mo-FF spread - lv 6 mo-FF spread (AC)
98 6 1 yr-FF spread - lv 1 yr-FF spread (AC)
99 6 5 yr-FF spread - lv 5 yr-FF spread (AC)
100 6 10 yr-FF spread - lv 10 yr-FF spread (AC)
101 6 Aaa-FF spread - lv Aaa-FF spread (AC)
102 6 Baa-FF spread - lv Baa-FF spread (AC)
103 6 Ex rate: avg - ln Nominal E¤ective Exchange Rate, Unit Labor Costs (IMF)
104 6 Ex rate: Switz M_110154768 ln Foreign Exchange Rate: Switzerland - Swiss Franc Per U.S.$
105 6 Ex rate: Japan M_110154768 ln Foreign Exchange Rate: Japan - Yen Per U.S.$
106 6 Ex rate: UK M_110154772 ln Foreign Exchange Rate: United Kingdom - Cents Per Pound
107 6 EX rate: Canada M_110154744 ln Foreign Exchange Rate: Canada - Canadian $ Per U.S.$

11
 Group 7: Prices
No. Gp Short Name Code Tran Descripton
2
108 7 PPI: …n gds M_110157517 ln Producer Price Index: Finished Goods
2
109 7 PPI: cons gds M_110157508 ln Producer Price Index: Finished Consumer Goods
2
110 7 PPI: int materials M_110157527 ln Producer Price Index:I ntermed Mat.Supplies & Components
2
111 7 PPI: crude matâeT M ls M_110157500 ln Producer Price Index: Crude Materials
2
112 7 Spot market price M_110157273 ln Spot market price index: bls & crb: all commodities
2
113 7 PPI: nonferrous materials M_110157335 ln Producer Price Index: Nonferrous Materials
114 7 NAPM com price M_110157204 lv Napm Commodity Prices Index
2
115 7 CPI-U: all M_110157323 ln Cpi-U: All Items
2
116 7 CPI-U: apparel M_110157299 ln Cpi-U: Apparel & Upkeep
2
117 7 CPI-U: transp M_110157302 ln Cpi-U: Transportation
2
118 7 CPI-U: medical M_110157304 ln Cpi-U: Medical Care
2
119 7 CPI-U: comm. M_110157314 ln Cpi-U: Commodities
2
120 7 CPI-U: dbles M_110157315 ln Cpi-U: Durables
2
121 7 CPI-U: services M_110157325 ln Cpi-U: Services
2
122 7 CPI-U: ex food M_110157328 ln Cpi-U: All Items Less Food
2
123 7 CPI-U: ex shelter M_110157329 ln Cpi-U: All Items Less Shelter
2
124 7 CPI-U: ex med M_110157330 ln Cpi-U: All Items Less Midical Care
2
125 7 PCE de‡ gmdc ln Pce, Impl Pr De‡:Pce (BEA)
2
126 7 PCE de‡: dlbes gmdcd ln Pce, Impl Pr De‡:Pce; Durables (BEA)
2
127 7 PCE de‡: nondble gmdcn ln Pce, Impl Pr De‡:Pce; Nondurables (BEA)
2
128 7 PCE de‡: service gmdcs ln Pce, Impl Pr De‡:Pce; Services (BEA)

Group 8: Stock Market

No. Gp Short Name Code Tran Descripton
82 8 S&P 500 M_110155044 ln S&P’s Common Stock Price Index: Composite
83 8 S&P: indust M_110155047 ln S&P’s Common Stock Price Index: & Industrials
84 8 S&P div yield - lv S&P’s Composite Common Stock: Dividend Yield Real (S)
85 8 S&P PE ratio - ln S&P’s Composite Common Stock: Price-Earnings Ratio Real (S)

12
 Notes:
1. Series # 87, 104 and 105 were spliced with the data available on the previous data
set.
2. Series # 3 and 74 were calculated dividing the series by # 125.
3. Series # 21 is a vacancy posting index built by R. Barnichon by combining the print
help-wanted index and the on-line help-wanted index. See Barnichon, R. , Building a composite
Help-Wanted Index, Economic Letters Dec 2010, for more details.
4. Series # 22 was computed dividing series # 21 by series M_110156531 of the IHS GI
database.
5. Series # 84 was computed as Dt =Pt . Both Price and Dividends are real.
6. Series # 85 was computed as Pt =AV ERAGE (Et 1 ; :::Et 12 ). Both Price and Earn-
ings are real.
7. Series 125-128 (implicit price de‡ators) were calculated as (Nominal Cons / Real
Cons) * 100. Real consumption is computed as: RealConst = RealConsbase * Qindext /Qindexbase .
The quantity indices are from table 2.8.3. The Base is Jan 2005, Real Consumption for the
base comes from table 2.8.6. The Nominal consumption comes from table 2.8.5.

2.2 Financial Dataset

The data set is at monthly frequency, with 147 observations spanning the period 1960:01-
2013:01. All returns and spreads are expressed in logs (i.e. the log of the gross return or
spread), are displayed in percent (i.e. multiplied by 100), and are annualized by multiplying by
12, i.e., if x is the original return or spread, we transform to 1200 ln (1 + x=100). Federal Reserve
data are annualized by default and are therefore not “re-annualized.”Note: this annualization
means that the annualized standard deviation (volatility) is equal to the data standard deviation
p
divided by 12. The data series used in this dataset are listed below by data source. Additional
details on data transformations are given below the table.

13
No. Short Name Source Tran Description
1 D_log(DIV) CRSP ln log Dt see additional details below
2 D_log(P) CRSP ln log Pt see additional details below
3 D_DIVreinvest CRSP ln log Dtre; see additional details below
4 D_Preinvest CRSP ln log Ptre; see additional details below
5 d-p CRSP ln log(Dt ) log Pt see additional details below
6 R15-R11 Kenneth French lv (Small, High) minus (Small, Low) sorted on (size, book-to-market)
7 CP Monika Piazzesi lv Cochrane-Piazzesi factor (Cochrane and Piazzesi (2005))
8 Mkt-RF Kenneth French lv Market excess return
9 SMB Kenneth French lv Small Minus Big, sorted on size
10 HML Kenneth French lv High Minus Low, sorted on book-to-market
11 UMD Kenneth French lv Up Minus Down, sorted on momentum
12 Agric Kenneth French lv Agric industry portfolio
13 Food Kenneth French lv Food industry portfolio
14 Beer Kenneth French lv Beer industry portfolio
15 Smoke Kenneth French lv Smoke industry portfolio
16 Toys Kenneth French lv Toys industry portfolio
17 Fun Kenneth French lv Fun industry portfolio
18 Books Kenneth French lv Books industry portfolio
19 Hshld Kenneth French lv Hshld industry portfolio
20 Clths Kenneth French lv Clths industry portfolio
21 MedEq Kenneth French lv MedEq industry portfolio
22 Drugs Kenneth French lv Drugs industry portfolio
23 Chems Kenneth French lv Chems industry portfolio
24 Rubbr Kenneth French lv Rubbr industry portfolio
25 Txtls Kenneth French lv Txtls industry portfolio
26 BldMt Kenneth French lv BldMt industry portfolio
27 Cnstr Kenneth French lv Cnstr industry portfolio
28 Steel Kenneth French lv Steel industry portfolio
39 Mach Kenneth French lv Mach industry portfolio
30 ElcEq Kenneth French lv ElcEq industry portfolio
31 Autos Kenneth French lv Autos industry portfolio
32 Aero Kenneth French lv Aero industry portfolio
33 Ships Kenneth French lv Ships industry portfolio
34 Mines Kenneth French lv Mines industry portfolio
35 Coal Kenneth French lv Coal industry portfolio
36 Oil Kenneth French lv Oil industry portfolio
37 Util Kenneth French lv Util industry portfolio
38 Telcm Kenneth French lv Telcm industry portfolio
39 PerSv Kenneth French lv PerSv industry portfolio
40 BusSv Kenneth French lv BusSv industry portfolio
41 Hardw Kenneth French lv Hardw industry portfolio
42 Chips Kenneth French lv Chips industry portfolio
43 LabEq Kenneth French lv LabEq industry portfolio
44 Paper Kenneth French lv Paper industry portfolio
45 Boxes Kenneth French lv Boxes industry portfolio
46 Trans Kenneth French lv Trans industry portfolio
47 Whlsl Kenneth French lv Whlsl industry portfolio
48 Rtail Kenneth French lv Rtail industry portfolio
49 Meals Kenneth French lv Meals industry portfolio
50 Banks Kenneth French lv Banks industry portfolio
51 Insur Kenneth French lv Insur industry portfolio
52 RlEst Kenneth French lv RlEst industry portfolio
53 Fin Kenneth French lv Fin industry portfolio
54 Other Kenneth French lv Other industry portfolio

14
List of Financial Dataset Variables (Cont’d)

No. Short Name Source Tran Description

55 1_2 Kenneth French lv (1, 2) portfolio sorted on (size, book-to-market)
56 1_4 Kenneth French lv (1, 4) portfolio sorted on (size, book-to-market)
57 1_5 Kenneth French lv (1, 5) portfolio sorted on (size, book-to-market)
58 1_6 Kenneth French lv (1, 6) portfolio sorted on (size, book-to-market)
59 1_7 Kenneth French lv (1, 7) portfolio sorted on (size, book-to-market)
60 1_8 Kenneth French lv (1, 8) portfolio sorted on (size, book-to-market)
61 1_9 Kenneth French lv (1, 9) portfolio sorted on (size, book-to-market)
62 1_high Kenneth French lv (1, high) portfolio sorted on (size, book-to-market)
63 2_low Kenneth French lv (2, low) portfolio sorted on (size, book-to-market)
64 2_2 Kenneth French lv (2, 2) portfolio sorted on (size, book-to-market)
65 2_3 Kenneth French lv (2, 3) portfolio sorted on (size, book-to-market)
66 2_4 Kenneth French lv (2, 4) portfolio sorted on (size, book-to-market)
67 2_5 Kenneth French lv (2, 5) portfolio sorted on (size, book-to-market)
68 2_6 Kenneth French lv (2, 6) portfolio sorted on (size, book-to-market)
69 2_7 Kenneth French lv (2, 7) portfolio sorted on (size, book-to-market)
70 2_8 Kenneth French lv (2, 8) portfolio sorted on (size, book-to-market)
71 2_9 Kenneth French lv (2, 9) portfolio sorted on (size, book-to-market)
72 2_high Kenneth French lv (2, high) portfolio sorted on (size, book-to-market)
73 3_low Kenneth French lv (3, low) portfolio sorted on (size, book-to-market)
74 3_2 Kenneth French lv (3, 2) portfolio sorted on (size, book-to-market)
75 3_3 Kenneth French lv (3, 3) portfolio sorted on (size, book-to-market)
76 3_4 Kenneth French lv (3, 4) portfolio sorted on (size, book-to-market)
77 3_5 Kenneth French lv (3, 5) portfolio sorted on (size, book-to-market)
78 3_6 Kenneth French lv (3, 6) portfolio sorted on (size, book-to-market)
79 3_7 Kenneth French lv (3, 7) portfolio sorted on (size, book-to-market)
80 3_8 Kenneth French lv (3, 8) portfolio sorted on (size, book-to-market)
81 3_9 Kenneth French lv (3, 9) portfolio sorted on (size, book-to-market)
82 3_high Kenneth French lv (3, high) portfolio sorted on (size, book-to-market)
83 4_low Kenneth French lv (4, low) portfolio sorted on (size, book-to-market)
84 4_2 Kenneth French lv (4, 2) portfolio sorted on (size, book-to-market)
85 4_3 Kenneth French lv (4, 3) portfolio sorted on (size, book-to-market)
86 4_4 Kenneth French lv (4, 4) portfolio sorted on (size, book-to-market)
87 4_5 Kenneth French lv (4, 5) portfolio sorted on (size, book-to-market)
88 4_6 Kenneth French lv (4, 6) portfolio sorted on (size, book-to-market)
89 4_7 Kenneth French lv (4, 7) portfolio sorted on (size, book-to-market)
90 4_8 Kenneth French lv (4, 8) portfolio sorted on (size, book-to-market)
91 4_9 Kenneth French lv (4, 9) portfolio sorted on (size, book-to-market)
92 4_high Kenneth French lv (4, high) portfolio sorted on (size, book-to-market)
93 5_low Kenneth French lv (5, low) portfolio sorted on (size, book-to-market)
94 5_2 Kenneth French lv (5, 2) portfolio sorted on (size, book-to-market)
95 5_3 Kenneth French lv (5, 3) portfolio sorted on (size, book-to-market)
96 5_4 Kenneth French lv (5, 4) portfolio sorted on (size, book-to-market)
97 5_5 Kenneth French lv (5, 5) portfolio sorted on (size, book-to-market)
98 5_6 Kenneth French lv (5, 6) portfolio sorted on (size, book-to-market)
99 5_7 Kenneth French lv (5, 7) portfolio sorted on (size, book-to-market)
100 5_8 Kenneth French lv (5, 8) portfolio sorted on (size, book-to-market)
101 5_9 Kenneth French lv (5, 9) portfolio sorted on (size, book-to-market)
102 5_high Kenneth French lv (5, high) portfolio sorted on (size, book-to-market)

15
List of Financial Dataset Variables (Continued)

No. Short Name Source Tran Description

103 6_low Kenneth French lv (6, low) portfolio sorted on (size, book-to-market)
104 6_2 Kenneth French lv (6, 2) portfolio sorted on (size, book-to-market)
105 6_3 Kenneth French lv (6, 3) portfolio sorted on (size, book-to-market)
106 6_4 Kenneth French lv (6, 4) portfolio sorted on (size, book-to-market)
107 6_5 Kenneth French lv (6, 5) portfolio sorted on (size, book-to-market)
108 6_6 Kenneth French lv (6, 6) portfolio sorted on (size, book-to-market)
109 6_7 Kenneth French lv (6, 7) portfolio sorted on (size, book-to-market)
110 6_8 Kenneth French lv (6, 8) portfolio sorted on (size, book-to-market)
111 6_9 Kenneth French lv (6, 9) portfolio sorted on (size, book-to-market)
112 6_high Kenneth French lv (6, high) portfolio sorted on (size, book-to-market)
113 7_low Kenneth French lv (7, low) portfolio sorted on (size, book-to-market)
114 7_2 Kenneth French lv (7, 2) portfolio sorted on (size, book-to-market)
115 7_3 Kenneth French lv (7, 3) portfolio sorted on (size, book-to-market)
116 7_4 Kenneth French lv (7, 4) portfolio sorted on (size, book-to-market)
117 7_5 Kenneth French lv (7, 5) portfolio sorted on (size, book-to-market)
118 7_6 Kenneth French lv (7, 6) portfolio sorted on (size, book-to-market)
119 7_7 Kenneth French lv (7, 7) portfolio sorted on (size, book-to-market)
120 7_8 Kenneth French lv (7, 8) portfolio sorted on (size, book-to-market)
121 7_9 Kenneth French lv (7, 9) portfolio sorted on (size, book-to-market)
122 8_low Kenneth French lv (8, low) portfolio sorted on (size, book-to-market)
123 8_2 Kenneth French lv (8, 2) portfolio sorted on (size, book-to-market)
124 8_3 Kenneth French lv (8, 3) portfolio sorted on (size, book-to-market)
125 8_4 Kenneth French lv (8, 4) portfolio sorted on (size, book-to-market)
126 8_5 Kenneth French lv (8, 5) portfolio sorted on (size, book-to-market)
127 8_6 Kenneth French lv (8, 6) portfolio sorted on (size, book-to-market)
128 8_7 Kenneth French lv (8, 7) portfolio sorted on (size, book-to-market)
129 8_8 Kenneth French lv (8, 8) portfolio sorted on (size, book-to-market)
130 8_9 Kenneth French lv (8, 9) portfolio sorted on (size, book-to-market)
131 8_high Kenneth French lv (8, high) portfolio sorted on (size, book-to-market)
132 9_low Kenneth French lv (9, low) portfolio sorted on (size, book-to-market)
133 9_2 Kenneth French lv (9, 2) portfolio sorted on (size, book-to-market)
134 9_3 Kenneth French lv (9, 3) portfolio sorted on (size, book-to-market)
135 9_4 Kenneth French lv (9, 4) portfolio sorted on (size, book-to-market)
136 9_5 Kenneth French lv (9, 5) portfolio sorted on (size, book-to-market)
137 9_6 Kenneth French lv (9, 6) portfolio sorted on (size, book-to-market)
138 9_7 Kenneth French lv (9, 7) portfolio sorted on (size, book-to-market)
139 9_8 Kenneth French lv (9, 8) portfolio sorted on (size, book-to-market)
140 9_high Kenneth French lv (9, high) portfolio sorted on (size, book-to-market)
141 10_low Kenneth French lv (10, low) portfolio sorted on (size, book-to-market)
142 10_2 Kenneth French lv (10, 2) portfolio sorted on (size, book-to-market)
143 10_3 Kenneth French lv (10, 3) portfolio sorted on (size, book-to-market)
144 10_4 Kenneth French lv (10, 4) portfolio sorted on (size, book-to-market)
145 10_5 Kenneth French lv (10, 5) portfolio sorted on (size, book-to-market)
146 10_6 Kenneth French lv (10, 6) portfolio sorted on (size, book-to-market)
147 10_7 Kenneth French lv (10, 7) portfolio sorted on (size, book-to-market)

2.2.1 CRSP Data Details

Value-weighted price and dividend data were obtained from the Center for Research in Security
Prices (CRSP). From the Annual Update data, we obtain monthly value-weighted returns series
vwretd (with dividends) and vwretx (excluding dividends). These series have the interpretation

Pt+1 + Dt+1
V W RET Dt =
Pt
Pt+1
V W RET Xt =
Pt

16
From these series, a normalized price series P , can be constructed using the recursion

P0 = 1
Pt = Pt 1 V W RET Xt :

A dividend series can then be constructed using

Dt = Pt 1 (V W RET Dt V W RET Xt ):

We de…ne the series

Dt = (Dt + Dt 1 + Dt 2 + Dt 3 ):

For the price and dividend series under “reinvestment,” we calculate the price under reinvest-
ment, Ptre , as the normalized value of the market portfolio under reinvestment of dividends,
using the recursion

P0re = 1
Ptre = Pt 1 V W RET Dt

Similarly, we can de…ne dividends under reinvestment, Dtre , as the total dividend payments on
this portfolio (the number of “shares”of which have increased over time) using

Dtre = Ptre1 (V W RET Dt V W RET Xt ):

As before, we de…ne the series

Dtre; = (Dtre + Dtre 1 + Dtre 2 + Dtre 3 )

Five data series are constructed from the CRSP data as follows:

D_log(DIV): log Dt .

D_log(P): log Pt .

D_DIVreinvest: log Dtre;

D_Preinvest: log Ptre;

d-p: log(Dt ) log(Pt )

17
2.2.2 Kenneth French Data Details

The following data are obtained from the data library of Kenneth French’s Dartmouth website
(http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html):

Fama/French Factors: From this dataset we obtain the data series RF, Mkt-RF, SMB,
HML.

25 Portolios formed on Size and Book-to-Market (5 x 5): From this dataset we obtain the
series R15-R11, which is the spread between the (small, high book-to-market) and (small,
low book-to-market) portfolios.

Momentum Factor (Mom): From this dataset we obtain the series UMD, which is equal
to the momentum factor.

49 Industry Porfolios: From this dataset we use all value-weighted series, excluding any
series that have missing observations from Jan. 1960 on, from which we obtain the series
Agric through Other. The omitted series are: Soda, Hlth, FabPr, Guns, Gold, Softw.

100 Portfolios formed in Size and Book-to-Market: From this dataset we use all value-
weighted series, excluding any series that have missing observations from Jan. 1960 on.
This yields variables with the name X_Y where X stands for the index of the size variable
(1, 2, ..., 10) and Y stands for the index of the book-to-market variable (Low, 2, 3, ..., 8,
9, High). The omitted series are 1_low, 1_3, 7_high, 9_9, 10_8, 10_9, 10_high.

2.3 Firm-level Dataset

Firm level observations are from COMPUSTAT Fundamentals Quarterly dataset. The unit
of observation is the change in …rm pre-tax pro…ts Pi;t , normalized by a two-period moving
average of sales, Si;t ; following Bloom (2009). Bloom constructs

dpretaxi;t = (Pit Pit 1 ) = (0:5 Sit + 0:5 Sit 1 ) ; (2)

for each …rm i in quarter t. This is the same measure reported on in Bloom (2009), Table 1,
and discussed in footnote c. We …nd, however, that (2) exhibits clear seasonality patterns, thus
we instead use year-over-year changes for the variable (2), normalized by average sales:

Yi;t = dpretaxyi;t = (Pit Pit 4 ) = (0:5 Sit + 0:5 St 4 ) ; (3)

We follow the trimming procedures used by Bloom, which includes considering any observation
with sales S = 0 a missing value, and windsorizing observations at the top and bottom 0.05%
values (replacing values in the top and bottom 0.05% with the values at the 0.05th and 99.95th

18
percentile values).5 After converting to a balanced panel, we are left with 155 …rms from
1970:Q1-2011:Q2 without missing values.
These variables are constructed from COMPUSTAT Fundamentals Quarterly dataset. It
contains 155 …rms observed from 1970Q1 to 2011Q2 that have non-missing observations for Pi;t
(Compustat identi…er piq) and Si;t (Compustat identi…er for net sales saleq) across the entire
time period.6

gvkey: …rm identi…er

date: period (1 to 166)

dpretax: quarterly change in pretax pro…ts scaled by average sales in current and past
quarter:
piqi;t piqi;t 1
dpretaxi;t = :
0:5 (saleqi;t + saleqi;t 1 )

dpretaxy: year-over-year change in quarterly pretax pro…ts scaled by average sales:

piqi;t piqi;t 4
dpretaxyi;t =
0:5 (saleqi;t + saleqi;t 4 )

2.4 Data for VAR Analysis

2.4.1 Monthly VAR Data

REX 3M: Log Excess Equity return, NSA (CRSP and Board of Governors)
The log equity return is the VWRETD series obtained from CRSP. For each month, we
create the quarterly return by adding over the log return for that month and the following two
months.
To obtain the quarterly excess return, we subtract the 3-month log t-bill return (secondary
market), obtained from the Board of Governors via FRED (series name: TB3MS).
For example, the January excess return is de…ned as the sum of the January, February, and
March log equity returns, minus the log 3-month t-bill return for January.
Log returns are multiplied by 100 to express in percent.
REX 1Y: Log 1-year excess return.
Equity return is obtained by compounding the log of the CRSP series VWRETD over 12
consecutive months and subtracting o¤ the 1-year log T-Bill return.
5
A detailed description of these procedures are given in the code to Bloom (2009)
http://www.stanford.edu/~nbloom/Uncertainty_shocks_code.zip.
6
This item represents operating and nonoperating income before provisions for income taxes and minority
interest. Earnings (COMPUSTAT code ibq) are measured as the income of a company after all expenses,
including special items, income taxes, and minority interest, but before provisions for common and/or preferred
dividends. Formally: ibq = piq txt (income taxes) mii (minority interest).

19
 For example, a January observation is given by the sum of January through December equity
returns, minus the January T-Bill return.
The 1-year T-Bill series is the constant maturity series, obtained from the Board of Gover-
nors, via FRED (series name: GS1).
REX 5Y: Log 5-year excess return.
Equity return is obtained by compounding the log of the CRSP series VWRETD over 60
consecutive months and subtracting o¤ the 1-year log T-Bill return.
For example, a January observation is given by the sum of January through December …ve
years hence equity returns, minus the January T-Bill return of the initial year.
The 5-year T-Bill series is the constant maturity series, obtained from the Board of Gover-
nors, via FRED (series name: GS5).
FEDFUNDS: Log E¤ective Federal Funds Rate, NSA (Board of Governors)
Obtained via FRED (series name: FEDFUNDS).
Log returns are multiplied by 100 to express in percent.
EARN_ALL: Average Hourly Earnings of Production and Nonsupervisory Employees: To-
tal Private, SA (BLS)
Obtained via FRED (series name: AHETPI).
EARN_MAN: Average Hourly Earnings Of Production And Nonsupervisory Employees:
Manufacturing, SA (BLS)
Obtained via FRED (series name: AHEMAN).
CPI: Consumer Price Index for All Urban Consumers: All Items (BLS)
Obtained via FRED (series name: CPIAUSCL).
HOURS_ALL: Average Weekly Hours Of Production And Nonsupervisory Employees: To-
tal Private, SA (BLS)
Obtained via FRED (series name: AWHNONAG).
HOURS_MAN: Average Weekly Hours of Production and Nonsupervisory Employees: Man-
ufacturing, SA (BLS)
Obtainted via FRED (series name: AWHMAN).
EMP_ALL: All Employees: Total Private Industries, SA (BLS)
Obtained via FRED (series name: USPRIV).
EMP_MAN: All Employees: Manufacturing, SA (BLS)
Obtained via FRED (series name: MANEMP)
IP_ALL: Industrial Production Index, SA (Board of Governors)
Obtained via FRED (series name: INDPRO).
IP_MAN: Industrial Production: Manufacturing (NAICS) (Board of Governors)
Obtained via FRED (series name: IPMAN)

20
 Quarterly dates are expressed as the month in the BEGINNING of the quarter (i.e. Jan for
Q1).
Variables in QDATA.xls:
REX 3M: Log Excess Equity return, NSA (CRSP and Board of Governors)
The log equity return is the quarterly VWRETD series obtained from CRSP. For each
month, we create the quarterly return by adding over the log return for that month and the
following two months.

Monthly Macro VAR Endogenous variables, in order:

(1) log(IP)
(2) log(Employment)
(3) log(Real Consumption)
(4) log(Price Level)
(5) log(Real Value of New Orders)
(6) log(Real Wage)
(7) log(Hours)
(8) Federal Funds Rate
(9) log(S&P 500)
(10) growth rate of M2
(11) uncertainty (various meausres)
IP = Industrial Production Index: total; jlndata series 6.
Employment = All employees, total nofarm; FRED series PAYEMS.
Real Consumption = jlndata series 3.
Price Level = PCE Implicit Price De‡ator; jlndata series 125.
New Orders = Value of Manufacturers New Order: consumer goods and materials +
Value of Manufacturers’ New Orders: nondefense capital goods; jlndata series 65 + 67.
Real Value of New Orders = New Orders/Price Level.
Wage = Average Hourly Earnings of Production and Nonsupervisory Workers: Manufac-
turing; jlndata series 131.
Real Wage = Wage/Price Level.
Hours = Average Weekly Hours of Production and Nonsupervisory Workers: manufacturing;
jlndata series 49.
Federal Funds Rate = E¤ective Federal Funds Rate; jlndata series 86.
S&P 500 = jlndata series 82.
M2 = jlndata series 72.

21
Monthly Bloom (2009) VAR Endogenous variables, in order:
(1) log(S&P 500)
(2) uncertainty (various measures)
(3) Federal Funds Rate
(4) log(Nominal Wage)
(5) log(Price Level)
(6) Hours
(7) log(Employment)
(8) log(Industrial Production)
S&P 500 = jlndata series 82.
Federal Funds Rate = e¤ective federal funds rate; jlndata series 86.
Nominal Wage = average hourly earnings in manufcaturing; jlndata series 131.
Price Level = CPI-U: all items; jlndata series 115.
Hours = Average Weekly Hours of Production and Nonsupervisory Workers: manufacturing;
jlndata series 49.
Employment = Employees on Nonfarm Payrolls: manufacturing; jlndata series 37.
Industrial Production = Industrial Production Index: manufacturing; jlndata series 16.

2.4.2 Quarterly VAR Data

To obtain the quarterly excess return, we subtract the 3-month log t-bill return (secondary
market), obtained from the Board of Governors via FRED (series name: TB3MS).
For example, the Q1 log excess return is the annualized Q1 quarterly log equity return,
minus the log 3-month t-bill return for January of that year.
Log returns are multiplied by 100 to express in percent.
REX 1Y: Log 1-year excess return.
Equity return is obtained by compounding the log of the quarterly CRSP series VWRETD
over 12 consecutive months and subtracting o¤ the 1-year log T-Bill return.
For example, a January observation is given by the sum of January through December equity
returns, minus the January T-Bill return.
The 1-year T-Bill series is the constant maturity series, obtained from the Board of Gover-
nors, via FRED (series name: GS1).
For example, the Q1 log excess return is the compounded Q1-Q4 quarterly log equity return,
minus the log 1 year t-bill return for January of that year.
Log returns are multiplied by 100 to express in percent.
REX 5Y: Log 5-year excess return.
Equity return is obtained by compounding the log of the quarterly CRSP series VWRETD
over 60 consecutive months and subtracting o¤ the 1-year log T-Bill return.

22
 For example, a January observation is given by the sum of January through December …ve
years hence equity returns, minus the January T-Bill return of the initial year.
The 5-year T-Bill series is the constant maturity series, obtained from the Board of Gover-
nors, via FRED (series name: GS5).
For example, the Q1 log excess return is the compounded quarterly log equity return over
5 years annualized, minus the annualized log 5 year t-bill return for January of that year.
Log returns are multiplied by 100 to express in percent.
FEDFUNDS: Log E¤ective Federal Funds Rate, Not Seasonally Adjusted (Board of Gover-
nors)
Obtained via FRED (series name: FEDFUNDS).
Quarterly log returns are obtained by averaging monthly log returns over the quarter.
Log returns are multiplied by 100 to express in percent.
EARN_ALL: Average Hourly Earnings of Production and Nonsupervisory Employees: To-
tal Private, Seasonally Adjusted (BLS)
Obtained via FRED (series name: AHETPI).
Quarterly series is obtained by averaging over the quarter.
EARN_MAN: Average Hourly Earnings Of Production And Nonsupervisory Employees:
Manufacturing, SA (BLS)
Obtained via FRED (series name: AHEMAN).
Quarterly series is obtained by averaging over the quarter.
CPI: Consumer Price Index for All Urban Consumers: All Items (BLS)
Obtained via FRED (series name: CPIAUSCL).
Quarterly series is obtained by averaging over the quarter.
HOURS_ALL: Average Weekly Hours Of Production And Nonsupervisory Employees: To-
tal Private, Seasonally Adjusted (BLS)
Obtained via FRED (series name: AWHNONAG).
Quarterly series is obtained by averaging over the quarter.
HOURS_MAN: Average Weekly Hours of Production and Nonsupervisory Employees: Man-
ufacturing, SA (BLS)
Obtainted via FRED (series name: AWHMAN).
Quarterly series is obtained by averaging over the quarter.
EMP_ALL: All Employees: Total Private Industries, Seasonally Adjusted (BLS)
Obtained via FRED (series name: USPRIV).
Quarterly series is obtained by averaging over the quarter.
EMP_MAN: All Employees: Manufacturing, SA (BLS)
Obtained via FRED (series name: MANEMP)
Quarterly series is obtained by averaging over the quarter.

23
GDP: Real Gross Domestic Product, 1 Decimal, Seasonally Adjusted Annual Rate (BEA)
Obtained via FRED (series name: GDPC1).

24
References
Bai, J., and S. Ng (2004): “A PANIC Attack on Unit Roots and Cointegration,”
Econometrica, 72(4), 1127–1177.
(2006): “Con…dence Intervals for Di¤usion Index Forecasts and Inference for
Factor-Augmented Regressions,”Econometrica, 74(4), 1133–1150.
Bloom, N. (2009): “The Impact of Uncertainty Shocks,”Econometrica, 77, 623–685.
Cochrane, J. H., and M. Piazzesi (2005): “Bond Risk Premia,”American Economic
Review, 95(1), 138–160.
Fama, E. F., and K. R. French (1992): “The Cross-Section of Expected Returns,”
Journal of Finance, 47, 427–465.
(1993): “Common Risk Factors in the Returns on Stocks and Bonds,” Journal
of Financial Economics, 33, 3–56.
Ludvigson, S. C., and S. Ng (2010): “A Factor Analysis of Bond Risk Premia,” in
Handbook of Empirical Economics and Finance, ed. by A. Ulah, and D. E. A. Giles,
vol. 1, pp. 313–372. Chapman and Hall, Boca Raton, FL.
Pagan, A. R. (1984): “Econometric Issues in the Analysis of Regressions with Generated
Regressors,”International Economic Review, 25(1), 221–247.
Stock, J. H., and M. W. Watson (2002a): “Forecasting Using Principal Components
From a Large Number of Predictors,”Journal of the American Statistical Association,
97(460), 1167–1179.
(2002b): “Macroeconomic Forecasting Using Di¤usion Indexes,”Journal of Busi-
ness and Economic Statistics, 20(2), 147–162.

25
 1.5
Baseline CSA
1.4 Baseline PCA, corr with baseline CSA = 0.8513
Posterior mean CSA, corr with baseline CSA = 0.9998
Posterior mean PCA, corr with baseline CSA = 0.8876
1.3

1.2

1.1

0.9

0.8

0.7

0.6

0.5
1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

Figure 1: Di¤erent estimates of macro uncertainty when h = 1. Baseline CSA is U t (1) =

1
PNy
Ny j=1 Ujt (1) j ; xjt . Baseline PCA shows the principal component based on Ujt (1) j ; xjt .
P S
Posterior mean CSA is the cross-section average of S1 s=1 Ujst (1) ( js ; xjst ). Posterior mean
P
PCA shows the …rst principal component based on S1 Ss=1 Ujst (1) ( js ; xjst ) . The full sample
spans the period 1960:01-2011:12.

26
 1.6
Baseline, corr with IP = −0.62
Posterior mean, corr with Baseline = 0.9998
90−10 percentiles, corr with Baseline = (0.9987,0.9988)
1.4

1.2

0.8

0.6

0.4

1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

Figure 2: Percentile-based estimates of aggregate uncertainty when h P = 1. Baseline de-

1 Ny
notes our base-case CSA estimate of macro uncertainty: U t (1) = Ny j=1 Ujt (1) j ; xjt
and j and xjt are posterior means over S draws. Posterior mean CSA is U t (1) =
1
PNy 1 PS 1
PNy [s]
Ny j=1 S s=1 Ujst (1) ( js ; xjst ). The posterior percentile-s CSA is U t (1) = Ny j=1 Ujt (1)
[s]
where Ujt (1) is the s-th percentile draw in the ordered sequence of Ujst (1)( js ; xjst ), for
s = 1; : : : ; S. The sample spans the period 1960:01-2011:12.

27
 1.5
correlation = 0.95
1.4

1.3

1.2

1.1

1 EGARCH

0.9

0.8

0.7

0.6 y
U t (1)

0.5
1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

y
Figure 3: EGARCH Aggregate Uncertainty: U t (1) computed using baseline stochastic volatility
estimates, and EGARCH(1,1) estimates with t-distributed errors. Aggregate uncertainty is
calculated as before, using a simple cross-sectional average. The data are monthly and span
the period 1960:07-2011:12.

28
 5
Baseline, corr with IP = −0.62
Real−time, corr with IP = −0.60
4

y
0 U t (1)

y
−1 U t (1)

Correlation = 0.98
−2
1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

y
Figure 4: Uncertainty factor based on recursive forecasts. This plot displays U t (h) based
on forecasts which use information from the full sample (“Baseline”), and based on recursively
computed out-of-sample forecasts (“Real-time”), expressed in standardized units. The recursive
forecasting procedure involves estimating model parameters and predictor variables only using
information available up to time t. A training sample of 10 years (120 observations) is used
to compute the …rst out-of-sample forecast, for 1970:01. The full sample spans the period
1960:01-2011:12.

29
 Production Employment
4 4
2 2
h=1

0 0
−2 −2
−4 −4
0 20 40 60 0 20 40 60
4 4
2 2
h=3

0 0
−2 −2
−4 −4
0 20 40 60 0 20 40 60
4 4
2 2
h = 12

0 0
−2 −2
−4 −4
0 20 40 60 0 20 40 60
4 4
VXO Index

2 2
0 0
−2 −2
−4 −4
0 20 40 60 0 20 40 60

y
Figure 5: Eight-variable VAR(12) using the VXO Index or U t (h) for h = 1; 3; 12 as a measure of
uncertainty. Each VAR(12) contains, in the following order: log(S&P 500 Index), uncertainty,
federal funds rate, log(wages), log(CPI), hours, log(employment), and log(industrial produc-
tion). All shocks are a 4 standard deviation impulse, which is the same magnitude considered
in Bloom (2009) Figure A.1. As in Bloom (2009), all variables are HP …ltered, except for the
uncertainty measures, which enter in raw levels. The data are monthly and span the period
1960:07-2011:12.

30
 Production Employment
4 4
2 2
h=1

0 0
−2 −2
−4 −4
0 20 40 60 0 20 40 60
4 4
2 2
h=3

0 0
−2 −2
−4 −4
0 20 40 60 0 20 40 60
4 4
2 2
h = 12

0 0
−2 −2
−4 −4
0 20 40 60 0 20 40 60
4 4
VXO Index

2 2
0 0
−2 −2
−4 −4
0 20 40 60 0 20 40 60

Figure 6: Eight-variable VAR(12) with uncertainty ordered last. Uncertainty is measured using
y
the VXO Index or U t (h) for h = 1; 3; 12 as a measure of uncertainty. Each VAR(12) contains,
in the following order: log(S&P 500 Index), federal funds rate, log(wages), log(CPI), hours,
log(employment), log(industrial production), and uncertainty. All shocks are a 4 standard
deviation impulse, which is the same magnitude considered in Bloom (2009) Figure A.1. As in
Bloom (2009), all variables are HP …ltered, except for the uncertainty measures, which enter in
raw levels. The data are monthly and span the period 1960:07-2011:12.

31