demand for information, uncertainty and the response of u
TRANSCRIPT
Demand for Information, Uncertainty and the Responseof U.S. Treasury Securities to News
Hedi Benamar 1 Thierry Foucault2 Clara Vega1
1Federal Reserve Board of Governors
2HEC Paris
Disclaimer: The opinions expressed here are our own, and do not reflect the views of theBoard of Governors or its staff.
NBER, March 2019
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 1 / 31
Summary
What is this paper about?
I We argue that information demand is a good proxy for uncertainty
I Theory:1. Information demand and uncertainty are positively correlated
I Empirical Analysis:1. Information demand is easy to measure using newly available (big) data on
who reads what and when. Bitly data.2. Information demand related to employment (nonfarm payroll) is positively
correlated with measures of monetary policy, macroeconomic and interest rateuncertainty
I However, my advisor told me many times, there are many coincidentalindicators, but very few leading indicators
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 2 / 31
Summary
What is this paper about?
I To fully convince you that information demand is a good proxy foruncertainty
I Theory:1. High uncertainty and information demand predicts future news have a bigger
impact on prices
I Empirical Analysis:1. High information demand prior to release of employment figures predicts
employment news have a bigger impact on U.S. Treasury yields2. In our sample period, information demand is one of the few variables that
predicts the impact of news on U.S. Treasury yields3. Why do other measures of uncertainty do not do so well? All the measures are
imperfect in some way, e.g., unconditional variance rather than conditional,risk premia, staleness
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 3 / 31
Importance
Why is this paper important?
I Uncertainty plays a central role in finance and economics:1. Uncertainty affects investment decisions2. Uncertainty affects the price discovery process (predicts when news have a big
impact on prices)
I But it is difficult to measure uncertainty1. Uncertainty regarding asset returns: variance of this return conditional on
investors’ information
I Previous literature:1. Realized volatility (unconditional variance)2. Implied volatility (conditional variance, but contains risk premia)3. Dispersion of analysts’ forecasts4. News-based measures of policy and macroeconomic uncertainty
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 4 / 31
Theory
Theory
I Model with four dates
I Risky asset with payoff F realized on the last date
I Public and private information about F
I Three types of agents1. Noise traders: trade for exogenous reasons2. Informed traders: pay to acquire information prior to a public announcement3. Market maker: sets prices conditional on order flow and public information
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 5 / 31
Theory
Timeline
Time 0: Information acquisition Time 1: Trading ahead of news Time 2: News arrives Time 3: Uncertainty resolved 1‐ Investors pay c(τi) for information about the macroeconomic fundamental and obtain a signal: si = F + ηi
with precision τi 2‐ Demand for information is realized
3‐ Investors submit their demand for the asset 4‐ Price is set and equal to the expected value of the asset conditional on aggregate demand p1 = E[F|D(p1)]
5‐ New arrives: NFP = F + ε 6‐ Price is updated p2 = E[F|D(p1), NFP]
6‐ Asset payoff is realized F
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 6 / 31
Theory
Equilibrium Price at Date 1
I Equilibrium price at date 1:
p1 = E(F |D(p1)) = E(F |z1) = λz1, (1)
where z1 = F + χD
I z1 is the signal conveyed by the order flow D(p1) at date 1
I χD = γτ−1η u is noise in this signal
I λ is the weight a Bayesian learning dealer puts on the signal Cov(F ,z1)Var(z1)
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 7 / 31
Theory
Uncertainty Prior to the NewsI After trading, the variance of the asset payoff conditional on public
information (z1) is
Uncertainty = Var(F |z1) =
Exogenous︷ ︸︸ ︷Var(F )
Endogenous︷ ︸︸ ︷Var(χD)
Var(F ) + Var(χD).
I The informativeness of the order flow (1/Var(χD) is higher if speculators’information demand is stronger (Var(χD) decreases with τ̄η)
I ⇒ Other things equal, uncertainty about F is reduced before theannouncement at date 2 if investors acquire more information before theannouncement
I But information demand is endogenous (not a parameter)
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 8 / 31
Theory
Equilibrium Price at Date 2
I Price at date 2 (holding information demand constant):
p2 = p1 + β (NFP − E(NFP |z1))︸ ︷︷ ︸Surprise
, (2)
withβ =
Cov(F ,NFP|z1)
Var(NFP|z1)=
UncertaintyUncertainty + Var(ε)
(3)
I Uncertainty = Var(F |z1) and NFP = F + ε
I β measures the sensitivity of the price reaction to the surprise in thenews. It increases in1. The precision of the news (1/Var(ε)).2. The level of uncertainty prior to the news (Var(F | z1)). This is endogenous
(depends on information demand).
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 9 / 31
Theory
Equilibrium Demand for Information
I The certainty equivalent of speculator i ’s expected utility at date 0when she acquires a signal of precision τηi at date 0 is:
Π(τηi , τη) =12γ (ln(1 + τηi × Var(F | z1))− c(τηi ). (4)
I Equilibrium:1. An equilibrium at date 0 is a vector (τ∗ηi )i∈{1,...,n} for each investor such that
τ∗ηi maximizes Π(τ∗ηi , τ∗η) with τ∗η =
∫τ∗ηi di .
2. As all speculators are identical, consider symmetric equilibria in which they allchoose the same precision: τ∗ηi = τ∗η, ∀i .
3. Equilibrium Condition: marginal profit = marginal cost
∂Π(τ∗η, τ∗η)
∂τ∗ηi= 0.
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 10 / 31
Theory
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 11 / 31
Theory
Implication
I An increase in the variance of the asset payoff has two effects:1. Holding information demand constant, it raises uncertainty (Var(F |z1) ↑
when Var(F ) ↑)2. It raises information demand (τ̄η ↑ when Var(F ) ↑).
I The second effect dampens the first but never offsets it completely
I Corollary: An increase in the variance of the asset payoff results in a jointincrease in (i) Information demand and (ii) The reaction of prices to news(β).
I ⇒ Prediction: An increase in information demand should be predictive of astronger reaction of prices to news (stronger β).
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 12 / 31
Empirical Analysis
Our Measure of Information Demand
I We use Bitly data to measure information demand (who reads what when)
I Bitly is a provider of short-URL-links and tracks readership statistics
I Original address: http//blogs.wsj.com/economics/2016/why-december-private-payrolls-arent-a-great-predictor-of-the-jobs-report/
I Short-URL-link: : http//on.wsj.com/2kMHdvJ
I Why do journalists and investors use short-URL-links?
I Bitly tracks readership statistics for journalists
I News-readers share short-URL-links more widely
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 13 / 31
Empirical Analysis
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 14 / 31
Empirical Analysis
Bitly Data
I Data from January 2012 through July 2018
I Clicks on news articles provided by 59 major news providers.I Financial news providers are most useful: WSJ and Bloomberg
I Big data: 70 million links with 10 billion clicks.
I The data contains:
I original URL (identify payroll news and source)I login name of link creatorI time stamp of when the article was sharedI time stamp of when the article was read (clicked on)I clicker’s geographical locationI access mode (direct, Tweeter, Facebook, etc.)
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 15 / 31
Empirical Analysis
NFP Clicks
I We identify articles related to nonfarm payroll news using keywords: “payroll”or “jobs report” or “unemployment”
I These words are the most frequent words used in a headline during the twohours before and after a nonfarm payroll announcement
I Limitation: our method to identify news related to employment works betterthe closer you are to the announcement
I NFP clicks: 21, 283 clicks two before and 148, 971 two hours after across 79announcements
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 16 / 31
Empirical Analysis
Our measure of information demand
I Bitly Counts: Sum number of clicks on NFP articles over a given interval(over a month or over the 2 hours preceding the announcement) standardizedby its standard deviation
I HighBitly : A dummy variable equal to 1 when this number is higher than itsmedian value
I Assumption: Increase in NFP clicks is positively correlated with investors’information demand about NFP/future level of interest rates. We do notclaim that all NFP clicks originate from investors themselves, nor do we claimthat this is the only source of information for investors.
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 17 / 31
Empirical Analysis
What are the characteristics of our information demandmeasure?
I Serial correlation:1. High Bitly count before announcement predicts high Bitly count after
announcement (coefficient 1.74)2. Monthly serial correlation coefficient is 0.74
I Positive correlation with measures of uncertainty, other measures ofinformation demand (Google Trends), and measures of information supply(Ravenpack news)
I Spikes up at 8:30 am after the release of the announcement
I Spikes up during announcement days
I Spikes up after large “U.S.” and “global” shocks
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 18 / 31
Empirical Analysis
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 19 / 31
Empirical Analysis
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 20 / 31
Empirical Analysis
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 21 / 31
Empirical Analysis
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 22 / 31
Empirical Analysis
Table 1: Popular News Sources of Articles Shared using Bitly
News Source Number of Clicks Percent of Total Number Cumulative Percent
Panel A: Prior to nonfarm payroll release, from 6:29 am ET to 8:29 am ETWall Street Journal 7,137 34% 34%Bloomberg 5,538 26% 60%CNN 4,365 21% 80%New York Times 1,219 6% 86%USA Today 1,144 5% 91%
Panel B: During and after nonfarm payroll release, from 8:30 am ET to 10:30 am ETWall Street Journal 26,200 18% 18%CNN 24,291 16% 34%Bloomberg 21,853 15% 49%New York Times 21,092 14% 63%USA Today 14,123 9% 72%
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 23 / 31
Yield Reaction to News
Yield Reaction to News
I To measure the response of treasury prices to nonfarm payrollannouncements, we estimate:
∆yt = α + βSSurpriset + εt , (5)
where1. ∆yt is the 30-minute difference in yields of the two-, five- and ten-year U.S.
Treasury note during the announcement period (79 observations)
2. Surpriset is the difference between the actual release of the nonfarm payrollfigure on day t and the median forecast of this figure submitted to Bloombergby professional forecasters (standardized by its standard deviation).
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 24 / 31
Yield Reaction to News
Table 2: Yield Response to News
Jan. 2004 - Jul. 2018 Jan. 2012 - Jul. 2018(1) (2) (3)
Panel A: Response of the Two-Year U.S. Treasury Note FuturesNonfarm Payroll Surprise 4.953*** 3.188*** 3.584***
(0.606) (0.701) (0.819)NFP Surprise × SW ZLB Period -2.373**
(0.920)SW ZLB Period -1.076
(0.646)Constant 0.632 0.0648 0.249
(0.441) (0.452) (0.526)Number of Observations 175 79 79Adjusted R-squared 0.335 0.233 0.259
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 25 / 31
Yield Reaction to News
Table 3: Yield Response to News Interacted with Information Demand
(1) (2) (3) (4) (5) (6)Two-Year Five-Year Ten-Year
NFP Surprise 0.362 0.598 3.076** 3.027** 4.577*** 4.406***(0.677) (0.620) (1.208) (1.301) (1.415) (1.566)
NFP × Bitly Count 2.873*** 3.418*** 2.660**(0.819) (1.125) (1.191)
Bitly Counts 1.010* 1.200 1.093(0.574) (0.775) (0.746)
NFP × I(Bitly Count) 4.446*** 5.774*** 4.657**(1.159) (1.829) (2.068)
High Bitly Count 0.194 0.538 0.788(0.820) (1.273) (1.358)
Constant -1.004** -0.382 -1.414* -0.879 -1.309 -0.964(0.440) (0.378) (0.749) (0.721) (0.834) (0.840)
Number of Obs. 79 79 79 79 79 79Adjusted R-squared 0.442 0.338 0.450 0.408 0.434 0.412
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 26 / 31
Two-Year Yield Reaction to News
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 27 / 31
Five-Year Yield Reaction to News
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 28 / 31
Order Flow Impact
Order Flow Impact
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 29 / 31
Informed or Noise Trading?
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 30 / 31
Conclusion
Conclusion
I Data on news consumption by investors (e.g., search for news on Bloombergterminals, Bitly clicks etc.) offer opportunities to easily measure informationdemand
I Information demand plays a central role in cornerstone theories of priceformation in securities markets (e.g., Grossman and Stiglitz (1980))
I We argue that information demand is a good proxy for uncertainty
I We show that information demand is one of the few predictors of a largeresponse to news
I The study may help policy makers in forecasting when asset prices maybemore sensitive to communications
Benamar, Foucault, Vega (Fed and HEC) Demand for Information and Uncertainty NBER 2019 31 / 31