jpxwp_032917r3_upload

Size: px

Start display at page:

Download "jpxwp_032917r3_upload"

いつやごみぶち
5 years ago
Views:

1 : ( ),, CRESTJST ( ) /21 (lead-lag).., Hoffman, Rosenbaum and Yoshida (2013) Dobrev and Schaumburg (2015) H. and Koike (2016) (Cross-marke, single-asse analysis) 3 (, Japannex,Chi-XJapan) (2015, 2016, 2017) ( ) /21

2 Hoffmann-Rosenbaum-Yoshida s mehod Hoffmann, Rosenbaum and Yoshida(2013) Formulaed a lead/lag esimaion problem and proposed an esimaor based on H. and Yoshida (2005) s covariance esimaor ( HY esimaor ) no synchronizaion needed Two coninuous semimaringales: X, Y Shif J i (inerval for Y )ohelef(pas)byθ: J i θ =(T i 1 θ,t i θ] θ-shifed HY esimaor: where K i,j θ U n (θ) T := i,j=1 S i T =1 {I i j θ shifed HY facor J θ }: ( i X)( j Y )K i,j θ, (1) ( ) /21 HRY s mehod Hoffmann, Rosenbaum and Yoshida(2013) Assumpion: There exiss a laen process Ỹ s.. Y = Ỹ θ. θ is an unknown consan. Ỹ is no observed in real ime, bu is observed wih (consan) delay θ θ > 0: X leads Y θ < 0: Y leads X ( ) /21

3 HRY s mehod Hoffmann, Rosenbaum and Yoshida(2013) HRY(2013) s proposal: HRY esimaor for he unknown rue lag parameer θ θ HRY := θ n := arg max θ G n U n (θ) T (2) where G n is a finie grid saisfying cerain regulariy condiions. In he high-frequency seing, θ HRY is consisen for θ (Theorem 1 of Hoffman, e al(13)). ( ) /21 Esimaed HRY measures: micro price, TSE vs JNX ime series plo of ˆθ s8306, j, y s8316, j, y s8411, j, y s8604, j, y s8766, j, y s8801, j, y Figure 1 : Daily HRY measures, Misubishi UFJ FG(8306) Misui Real Esae (8801): TSE JNX. ( ) /21

4 Alernaive approach: Dobrev-Schaumburg (2015) s mehod Esimaed HRY measures: Unsable Affeced by microsrucure noise Sensiive o jumps An alernaive approach by Dobrev and Schaumburg: High-Frequency Cross-Marke Trading: Model Free Measuremen and Applicaions. Working paper, Uilizes no price changes bu imesamps of he rading aciviies Is no (direcly) influenced by microsrucure noise peraining o he behavior of he observed, inefficien prices ( ) /21 DS s mehod I i =1if raded a ; 0 oherwise. =0, 1,...,N, j =1, 2. regular spacing wih = T N. Objecive funcion o be maximized: A(h) := 1 A N h =1 I 1 I 2 +h where A := min ( I1, I2 ). Modified objecive funcion: Ã(h) :=A(h) Ā where Ā := 1 2H+1 H h= H A(h). ( ) /21

5 A(h) :,.,,,, 2. Figure 2 : A(h) ( FG (8411), ). 2 ( ) 6 ( ). ( ) / 21

6 Esimaed DS measures: micro price, TSE vs JNX s8306, se_jnx, y s8316, se_jnx, y s8411, se_jnx, y s8604, se_jnx, y s8766, se_jnx, y s8801, se_jnx, y Figure 3 : DS (, ): UFJ (8306) (8801): TSE JNX. ( ) / 21 Summary saisics of esimaed DS measures summary over Period 7/22/14 12/30/14: Micro price, TSE vs JNX. Period 1 Period 2 Period code mean sdev mean sdev mean sdev s s s s s s s s s s s s Table 1 : DS ( : ): TSE JNX,. ( ) / 21

7 -1 (TSE ChiX, JNX ChX,.) 1., TSE ( 1 2 ), : TSE JNX, TSE ChiX: TSE PTS, 4. 2, TSE JNX, TSE. JNX ChiX: 1 ( ). 2, JNX ChiX, JNX ChiX. ( ) / 21 DS ( ) TSE vs JNX TSE vs ChiX JNX vs ChiX Period 1 Buy Sell Buy Sell Buy Sell mean sdev Period 2 Buy Sell Buy Sell Buy Sell mean sdev Period 3 Buy Sell Buy Sell Buy Sell mean sdev Table 2 : DS (TOPIX100, : ). ( ) / 21

8 DS ( ): Buy: se_jnx, opix100 Buy: se_chx, opix100 Buy: jnx_chx, opix Sell: se_jnx, opix100 Sell: se_chx, opix100 Sell: jnx_chx, opix Figure 4 : DS, ( ) ( ) (TOPIX100 ): TSE JNX ( ), TSE ChiX ( ), JNX ChiX ( ). ( ), ( ). ( ) / 21-2 : 1,.,. 2,., TSE, TSE JNX, TSE,. TSE ChiX, TSE,. JNX ChiX, ChiX, ChiX. 1 : U. 1., 1, U ( J ). ( ) / 21

9 Panel regression analysis Wan o explain how he lead/lag indices are generaed in erms of observable characerisics of he socks. i-h sock, j-h, -h ime bin Covariaes (group 1): Raios beween marke pairs RV r ij =log(rvij/rv X ij): Y Realizedvolailiyraio NQr ij =log(nq X ij/n Q Y ij): #quoerevisionsraio LT Sr ij =log(lt S X ij/lt S Y ij): averagelosizeraio Spr ij =log(spr X ij/spr Y ij): Spreadraio(omidprice)raio QAr ij =log(qa X ij/qa Y ij): Deph(onBesAsk)raio. QBr ij =log(qb X ij/qb Y ij): Deph(onBesBid)raio Covariaes (group 2): Aggregae values of he hree markes RV ag ij :RVofheaggregaemarke RET s ij :LogreurnonheTSE TVag ij :Toalsharevolumeofheaggregaemarke All he variables are sandardized. ( ) / 21 Panel regression analysis Fi a linear mixed effecs model o he panel daase. y ij = ˆθ ij : i-h sock, j-h, -h ime bin T : ime-of-he- effec ( = 1,...,10), 6-level facor Spli he rading hours (300 min.) ino en 30 min. inervals. Bin 1: 9-9:30, Bin 2: 9:30-10:00,..., Bin 10: 14:30-15:00 Use only Bins 2, 3, 4, 7, 8, 9. Dir: Buy direcion/ Sell direcion, 2-level facor y (1 Code)+(1 Ymd) + }{{} }{{} T random effecs 6 level facor + Dir }{{} 2 level facor RV r + NQr + LT Sr + SPRr + QBr + QAr }{{} covariaes (group1) +RV ag + RET s + TVag }{{} covariaes (group2) ( ) / 21

10 Panel regression 3,.. : :,,. :.,. (2015, 2016).. ( ) / 21 Acknowledgemen Gran-in-Aid for Scienific Research (C), Japan Sociey for he Promoion of Science (JSPS), no.16k CREST JST Projec, Graduae School of Mahemaical Sciences, Universiy of Tokyo. Japan Exchange Group, Inc. SBI Japannex Co., Ld Chi-X Japan Ld ( ) / 21

11 References (2015) , 53-4, (2016). 3., Dobrev and Schaumburg. (2015). High-Frequency Cross-Marke Trading: Model Free Measure- men and Applicaions. Unpublished manuscrip. H. and Koike (2016). Wavele-based mehods for high-frequency lead-lag analysis. arxiv: Hoffmann, Rosenbaum, and Yoshida (2013). Esimaion of he lead-lag parameer from non-synchronous daa, Bernoulli, 19 2, ( ) / 21

IPSJ SIG Technical Report Vol.2010-CVIM-172 No /5/ Object Tracking Based on Generative Appearance Model 1. ( 1 ) ( 2 ) ( 3 ) 1 3) T

IPSJ SIG Technical Report Vol.2010-CVIM-172 No /5/ Object Tracking Based on Generative Appearance Model 1. ( 1 ) ( 2 ) ( 3 ) 1 3) T 1 2 2 3 1 Objec Tracking Based on Generaive Appearance Model 1. ( 1 ) ( 2 ) ( 3 ) 1 3) Tasuya YONEKAWA, 1 Kazuhiko KAWAMOTO, 2 Asushi IMIYA 2 and Akihiro SUGIMOTO 3 We propose a mehod for racking objecs