人工知能学会研究会資料 SIG-FPAI-B508-08 - - Predicting stock returns based on the time lag in information diffusion through supply chain networks 1 1 Yukinobu HAMURO 1 Katsuhiko OKADA 1 1 1 Kwansei Gakuin University Abstract: As Cohen and Frazzini demonstrates, stock prices do not promptly incorporate news about economically related firms, generating return predictability across assets. This paper demonstrates a unique methodology to exploit such phenomenon for alpha generation. Using a Fact Set Revere data, we construct a portfolio of stocks that has economic links with a suddenly rising firm, that is presumed to have met with good fundamental news. Calendar-time value-weighted portfolio generates large abnormal return after controlling for size and book-to-market ratio. We also found that the performance further improves when we create edge-betweenness weighted portfolio. 1 Cohen and Frazzini (supplier)- (customer) [1] (investor inattention) Cohen and Frazzini 2 2.1 t N t G t = (N t, E t ) E t 2 (supplier) (customer) G SC t = (N t, Et SC ) ( SC ) G CS t ( CS ) SC 2 a, b, ( (a, b) Et SC ) a b b a 662-8501 1-155 E-mail: hamuro@kwansei.ac.jp - 40 -
2.2 a t d r a t,d = (ra t d+1, ra t d+2,..., ra t ) rt a a t ra t = c a t /c a t 1 n t /n t 1 c a t t a n t t US S&P500 d 1 0 d u = (0, 0,..., 1) 2 c a t,d u sim(, ) ρ a t t t d + 1 t 1 t U t (1) U t = {a sim(r a t,d, u) ρ, δ L rt a 1 d 1 rt i a δ U } a Nt (1) d 1 i=1 d = 9, ρ = 0.8, δ L = 0.05, δ U = 0.1 d = 9 10 ( 2 ) δ U 2.3 t U t G t = (N t, E t ) t Q t (2) Q t = {b (a, b) E t, b U t } a Ut (2) a b Q t (SNS) (Node-Betweenness Centrality) (Edge-Betweenness Centrality) a C NB (a) (3) g ij i j g ij (a) i j a C NB (a) = i V j V ;j i g ij (a) g ij (3) a (a, b) C EB (a, b) (4) g ij (a, b) i j (a, b) C EB (a, b) = i V j V ;j i g ij (a, b) g ij (4) Apple Toyota - 41 -
t SC ( CS ) a Q t s(a) U t a wt NB (a) (5) ( ) wt EB (a) (6) 2012 5000 1 &'()*+,%-./!"#$%% 01%- 0,$- w NB t (a) = w EB t (a) = C NB (s(a)) i Q t C NB (s(i)) C EB (s(a), a) i Q t C EB (s(i), i) (5) (6) 1: 20% S&P500 3 FactSet Research Systems (Revere ) 2003 4 3 2017 5 31 3565 ( 15 ) 20% 3073 4127 1 () S&P500 20% 15 170%( 10%) S&P500 120%( 7.5%) Revere Factset (Form 10-K annual fillingsa) (investor presentations) Revere 2 2: ( ) ( ) ( ) 4 FactSet Revere t a Q t t h ( t + h) h = 1 G SC t G SC t 2 1) 2) wt NB 3) wt EB 3 6 3 3 Lyon Fama- French Three Factor Model[2] Three Factor Model - 42 -
9:;<=!,-&./&'01%*+!,-&456&'307*+,!-&456&'3%8*+!,-&2/&'3%(*+,!-&./&')7$*+,!-&2/&')%$*+!"#$%%&'()$*+ α (0.03%) 22.6% 31.4% 3 2008 3 max drawdown SC EB 3: 2008 SC: CS: org: NB: EB: : R p t R f t = α + β(r m t R f t ) + s(s t ) + h(h t ) + ϵ t (7) R p t t R f t 10 Rt m S&P500 S t H t α P P Three Factor Model Sharpe max drawdown Sharpe max drawdown 3 (CS SC ) α (α P SC:1.75%,CS:0.29%) CS NB, EB α SC NB, EB NB SC EB 5 5.1 h = 1 10 2 5 1 1: 1 5 α P (%) P <5% 1 2 3 4 5 EB 12.4 26.7 45.1 84.0 99.5 CS NB 26.7 87.4 79.3 83.3 76.9 org 1.75 29.0 29.8 78.7 62.5 EB 0.03 0.11 0.14 1.19 0.93 SC NB 1.35 4.54 2.14 7.56 15.1 org 0.29 0.65 0.64 2.62 2.44 CS SC BtoC iphone - 43 -
5.2 ρ ( δ L δ U ) ρ = 0.8 δ L = 0.05 δ U = 1.0 4 ρ = 0.7 u ρ = 0.8, 0.9 0.5 5.3 () u (-0.8 ) δ L = 0.1, δ U = 0.05 2 idiosyncratic (U t ) (Q t ) 2: Frama-French Three Factor α P (bp) (%) EB 3.56 13.9 CS NB 1.48 49.9 19085 org 3.52 9.64 EB 1.42 49.8 SC NB -0.65 74.0 15686 org -1.43 43.8 6 / 2 1.4 JST CREST (B) 25285127 27 28- -4101 [1] Cohen, L and Frazzini, A, Economic Links and predictable returns,journal of Finance, 63, 4, pp. 1977 2011 (2008) [2] Lyon, J., D., Brad M. Barber, Chih-Ling Tsai, Improved Methods for Tests of Long-Run Abnormal Stock Returns, The Journal of Finance, 54, 1, pp. 165 201 (1999) - 44 -
3: 6 Frama-French Three Factor 5% α bp 1/10000 α P SD Sharpe max draw-down (bp) (%) ( %) ( %) org 5.01 1.75 21.2 25.3 0.837-0.567 2007/04/24-2008/11/20 CS NB 2.68 26.79 85099 14.1 26.4 0.532-0.615 2007/04/25-2008/11/20 EB 3.87 12.49 18.1 28.3 0.642-0.967 2007/06/04-2008/11/20 org 6.32 0.29 22.6 23.8 0.952-0.505 2007/04/16-2008/03/17 SC NB 5.37 1.35 64819 19.8 23.7 0.837-0.473 2007/03/21-2009/03/17 EB 9.85 0.03 31.4 28.9 1.086-0.360 2007/04/24-2008/03/19 4: ρ, (δ L, δ U ) α P (%), P <5% ρ = 0.7 ρ = 0.8 ρ = 0.9 0-2.5% 2.5-5% 5-10% 0-2.5% 2.5-5% 5-10% 0-2.5% 2.5-5% 5-10% org 17.2 32.7 92.4 36.2 28.6 1.75 8.25 67.24 45.7 CS NB 28.4 52.8 73.7 47.8 20.3 26.7 8.63 55.33 58.3 EB 85.8 97.0 61.9 88.6 36.0 12.4 94.3 51.73 41.4 org 4.26 39.2 89.4 35.1 20.0 0.29 35.0 18.95 0.37 SC NB 42.8 10.1 95.6 49.6 14.9 1.35 20.5 17.24 0.99 EB 13.5 9.86 5.29 5.40 4.79 0.03 44.0 15.70 0.11-45 -