Jumps and information asymmetry in the US Treasury market · Jumps and information asymmetry in the...

Jumps and information asymmetry in the USTreasury market

Ana-Maria H. DumitruGiovanni Urga

University of SurreyCass Business School, City University

Dumitru and Urga (2014) Jumps US Treasury 05/09/2014 1 / 27

Background and objectives


Jumps=big unexpected changes in the prices of financialsecuritiesrecent statistical tools to identify jumps

literature in high frequency econometrics




unanticipative nature: bring new info to the market

mark the incorporation of new info in prices

Hypothesis: no private info after the jump




Research objectives

MAIN OBJECTIVE: investigate the role of jumps in dissipatinginformational asymmetry in the US Treasury bond market

identify jumps in the US Treasury market

determinants of jumps: the liquidity puzzle



Outline

1 Background and objectives

2 Other research work in the area

3 Data and methodology

4 Jumps and determinants

5 Jumps and informativeness of the order flow

6 Conclusion and further developments


Other research work in the area

Other research work in the area

Jiang et al (2011) and Dungey et al (2009): US-Treasury bonds

Lahaye et al (2011): 8 different financial assets

Boudt and Petitjean (2014): Dow Jones Industrial Average indexconstituents

Gilder et al (2014): portfolio of US stocks

Common findings

identify jumps, co-jumps

most frequent jump determinants: public info, liquidity shocks


Data and methodology

Data

Brokertec trading platform

order book and trade data for the 2-, 5-, 10- and 30- year bondsand covering a period between January 2003 and March 2004

the 30-year- less liquid



Data

7:30 a.m. EST and 5:00 p.m. EST

sampling: every 5 and 15 minutes

mid-quotes for jump testing

trade data



Methodology- Jump identification

Dumitru and Urga (2012): thorough comparison between differentjump detection procedures:

evidence that combining different tests and different samplingfrequencies through reunions and intersections can lead to betterpower and size properties

final jumps: the ones identified on 15 minutes data by theLee-Mykland (2008) test corrected for periodicity in the volatilityfactor if they were also detected by the Barndorff-Nielsen andShephard (2006) procedure on either 5 or 15 minutes data



Working hypotheses: data generating process

semimartingale (Brownian semimartingale + jump)

µ drift; σ diffusion

usually µ, σ cadlag; µ predictable

dpt = µtdt +σtdWt +dJt ,

J(t) =Nt

∑j=1

c j,

where

Nt counting process, λt

c j size of the jump



Working hypotheses: what is to estimate?

quadratic variation of the process:

QVt =∫ t

0σ2

s ds+Nt

∑j=1

c2t j,

integrated volatility (variance)

IVt =∫ t

0σ2

s ds

interval [0, t] is split into n equal subintervals of length δj-th intraday return

r j = pt−1+ jδ − pt−1+( j−1)δ



Barndorff-Nielsen and Shephard (2006) test for jumps

H0: no jumps in the sample path in the interval [0, t]

RVt =n

∑j=1

r2j

p→ QVt ,

BVt =π2

n

∑j=1

|r j||r j+1|p→ IVt

1− BVtRVt√

0.61δ max(

1, T Qt

BV 2t

)

where T Qt = n1.74(

nn−2

)∑n

j=3 |r j−2|4/3|r j−1|

4/3|r j|4/3 consistent for

integrated quarticity



Lee-Mykland (2008) test for jumps

H0: no jumps in the sample path at time t j

z j = |r j|/√

V̂j,

where V̂j = BVt j/(K −2), with K the window size on which BVt j iscalculated

(max(z j)−Cn)/SnL→ ξ , P(ξ ) = exp(−e−x)

Nonparametric correction for periodicity based on Boudt et al. (2011).


Jumps and determinants

Detected jumps

the 2-year bonds jump in 14.5% of the days, the 5-year in 10.6%,the 10-year in 9.6% and finally, the 30-year in 17.91% of the days

jump size: increasing with the maturity

Average Y2 Y5 Y10 Y30size 0.081% 0.24% 0.4% 0.77%



Jumps and macroeconomic announcements

Y2 Y5 Y10 Y30Match 94 92.16% 79 100.00% 60 93.75% 72 83.72%

No match 8 7.84% 4 6.25% 14 16.28%Total 102 100.00% 79 100.00% 64 100.00% 86 100.00%

Table: Number and percentages of jumps matched with macroeconomicannouncements



The liquidity around jump times puzzle

−25 −20 −15 −10 −5 0 5 10 15 20 250

20

40

60

80

100

120

140Depth on the ask side

minutes−25 −20 −15 −10 −5 0 5 10 15 20 25

0

20

40

60

80

100

120

140

160Depth on the bid side

minutes

−25 −20 −15 −10 −5 0 5 10 15 20 250

5

10

15Spread

minutes−25 −20 −15 −10 −5 0 5 10 15 20 25

0

2

4

6

8

10

12x 10

6 Trading volume

minutes

Figure: Various liquidity measures around the time of jump for the 2-YearbondDumitru and Urga (2014) Jumps US Treasury 05/09/2014 16 / 27


The liquidity around jump times puzzle

most jumps occur as a result of macroeconomic announcements

before jumps, a severe liquidity withdrawal is observed

before announcements- market awaits

simultaneity of jumps and liquidity withdrawal= both caused byannouncements

endogeneity issue



Regression explaining jump occurrence

Pr{I jump = 1|X}= f (Surprise,BVt) ,

where f is extreme value



Determinants of jump occurrence

Coefficient p-value Goodness of fitY2 C -2.53 0.0000 H-L Statistic 5.74

Surprise 0.29 0.0021 Prob. Chi-Sq(8) 0.68Volatility (BV) 1881.19 0.0000

Y5 C -2.37 0.0000 H-L Statistic 4.36Surprise 0.40 0.0002 Prob. Chi-Sq(8) 0.82

Volatility (BV) 432.54 0.0002






Jumps and informativeness of the order flow

Analysis of the informativeness of the order flow in theproximity of jumps

starting point: Madhavan, Richardson and Roomans (1997)’model of price formation:

pti − pti−1 = (φ +θ)xti − (φ +ρθ)xti−1 + eti

Green(2004) uses this model to explore the informativeness of theorderflow when news impact the market



Analysis of the informativeness of the order flow in the proximity of jumps

Estimated models:Model 1

pti − pti−1 =(φJ +θJ)IJ,ti xt − (φJ +ρθJ)IJ,ti xti−1 +(φNJ +θNJ)INJ,ti xt−

(φNJ +ρθNJ)INJ,ti xti−1 + eti ,

Model 2

pti − pti−1 =(φJ0+θJ0)IJ,ti IJ0,ti xti − (φJ0+ρθJ0)IJ,ti IJ0,ti xti−1 +(φB +θB)IJ,ti IB,ti xti−

(φB +ρθB)IJ,ti IB,ti xti−1 +(φA +θA)IJ,ti IA,ti xti − (φA +ρθA)IJ,ti IA,ti xti−1+

(φNJ +θNJ)INJ,ti xti − (φNJ +ρθNJ)INJ,ti xti−1+ eti ,

Model 3

pti − pti−1 =(φJ0+θJ0)IJ,ti IJ0,ti xti − (φJ0+ρθJ0)IJ,ti IJ0,ti xti−1+

(φB5+θB5)IJ,ti IB5,ti xti − (φB5+ρθB5)IJ,ti IB5,ti xti−1+

(φA5+θA5)IJ,ti IA5,ti xti − (φA5+ρθA5)IJ,ti IA5,ti xti−1+

(φother +θother)Iother,ti xti − (φother +ρθother)Iother,ti xti−1 + eti



Analysis of the informativeness of the order flow in the proximity of jumps

Estimated models:Model 4




(φother +θother)Iother,ti xti − (φother +ρθother)Iother,ti xti−1 + eti ,

Model 5




(φother +θother)Iother,ti xti − (φother +ρθother)Iother,ti xti−1 + eti ,



Analysis of the informativeness of the order flow in theproximity of jumps

estimation method:GMM

for Model 1,

E

xtixti−1 − x2tiρ

eti −α(eti −α)IJ,tixti(eti −α)IJ,tixti−1

(eti −α)INJ,tixti(eti −α)INJ,tixti−1

= 0



Analysis of the informativeness of the order flow in the proximity of jumps.

Results

Model 1 Model 2 Model 3Coef Prob. Coef Prob. Coef Prob.

α 0.003 0.003 α 0.003 0.004 α 0.003 0.004φJ 0.031 0.028 φJ0 -0.852 0.114 φJ0 -0.842 0.118θJ 0.376 0.000 θJ0 1.378 0.001 θJ0 1.527 0.000

φNJ 0.046 0.000 φB 0.065 0.008 φB5 0.191 0.172θNJ 0.326 0.000 θB 0.286 0.000 θB5 0.345 0.000

φA 0.032 0.011 φA5 -0.135 0.286θA 0.360 0.000 θA5 0.667 0.000

φNJ 0.046 0.000 φother 0.046 0.000θNJ 0.326 0.000 θother 0.329 0.000

Model 4 Model 5Coef Prob. Coef Prob.

α 0.003 0.004 α 0.003 0.004φJ0 -0.842 0.118 φJ0 -0.842 0.119θJ0 1.528 0.000 θJ0 1.419 0.001

φB10 0.154 0.054 φB20 0.124 0.046θB10 0.246 0.000 θB20 0.266 0.000φA10 -0.083 0.279 φA20 -0.056 0.236θA10 0.573 0.000 θA20 0.506 0.000

φother 0.047 0.000 φother 0.047 0.000θother 0.328 0.000 θother 0.326 0.000

Estimated coefficients and p-values for Models 1-5 for the 2-Year bond. Throughout all the models, we use a unique correlation

coefficient for the order flow: ρ̂ = 0.661




Results


α 0.004 0.010 α 0.004 0.011 α 0.004 0.011φJ -0.171 0.000 φJ0 -1.618 0.076 φJ0 -1.671 0.067θJ 0.858 0.000 θJ0 2.958 0.001 θJ0 3.277 0.000

φNJ -0.177 0.000 φB -0.175 0.000 φB5 -0.216 0.144θNJ 0.845 0.000 θB 0.761 0.000 θB5 0.773 0.000

φA -0.130 0.000 φA5 -0.017 0.953θA 0.788 0.000 θA5 1.144 0.000

φNJ -0.177 0.000 φother -0.173 0.000θNJ 0.845 0.000 θother 0.841 0.000


α 0.004 0.011 α 0.004 0.011φJ0 -1.670 0.068 φJ0 -1.607 0.078θJ0 3.277 0.000 θJ0 3.032 0.001

φB10 -0.263 0.012 φB20 -0.315 0.036θB10 0.729 0.000 θB20 0.805 0.000φA10 -0.181 0.281 φA20 -0.120 0.237θA10 0.944 0.000 θA20 0.841 0.000

φother -0.172 0.000 φother -0.172 0.000θother 0.840 0.000 θother 0.840 0.000






Results


α 0.003 0.310 α 0.003 0.299 α 0.003 0.306φJ -0.104 0.016 φJ0 -0.836 0.637 φJ0 -0.910 0.608θJ 1.322 0.000 θJ0 0.000 1.000 θJ0 0.517 0.739

φNJ -0.202 0.000 φB -0.030 0.707 φB5 0.516 0.677θNJ 1.365 0.000 θB 1.037 0.000 θB5 2.312 0.017

φA -0.088 0.035 φA5 0.153 0.606θA 1.263 0.000 θA5 1.715 0.000

φNJ -0.202 0.000 φother -0.190 0.000θNJ 1.365 0.000 θother 1.357 0.000


α 0.003 0.305 α 0.004 0.286φJ0 -0.910 0.608 φJ0 -0.873 0.624θJ0 0.520 0.737 θJ0 -0.053 0.973

φB10 0.279 0.574 φB20 -0.308 0.448θB10 1.567 0.018 θB20 1.932 0.000φA10 -0.020 0.943 φA20 0.154 0.350θA10 1.509 0.000 θA20 1.363 0.000

φother -0.191 0.000 φother -0.195 0.000θother 1.356 0.000 θother 1.359 0.000




Conclusion

Conclusions

the paper analyzes the occurrence of jumps and role in reducingthe informational asymmetry in the US Treasury market.

we detect and estimate jumps in the US Treasury 2-,5-,10- and30-year bondsthe release of macroeconomic news is found to be the major causeof jumps in the bond prices

2- and 5- year maturities: the level of information asymmetryincreases immediately after jumps occur and remains high up to20 minutes after the jump

before a jump: low degree of informational asymmetry, consistentwith a low extent of information leakage

10- year maturity behaves differently (different explanations)


Jumps and information asymmetry in the US Treasury market · Jumps and information asymmetry in the...

Documents

Transcript of Jumps and information asymmetry in the US Treasury market · Jumps and information asymmetry in the...