• deep learning statistical arbitrage


    deep learning statistical arbitrage

    • empirial
    • stanford
    • Jorge guijarro
    • markus

    Motivation

    • Pair trading
    • GM and Ford
    • Assumption
      • prices are on average similar
    • Exploit temporal price different between similiar seests

    Three components of statisical arbitarge

    • contrict protolio
    • trading signal

    Foundational problem

    Research question

    • arbitrage portolios
    • arbitarge signals

    Contributions

    • Novel conceptual framework
    • Unified framework
    • To compare different statistical arbitrage methods
      • Portolio generation
      • signal extraction
      • allocation decision
    • Study each component and compare with conventional models

    Novel methods

    • statistical factor
    • Convolution neural network

    Empirical

    • substantially outperforms
    • sharpe ratios

    Parametric models

    • PCA
    • cOINTEGRATION
    • STOCHASTIC CONTROL
    • SIMPLE PAIRS TRADING
    • INTRACTABLE PARAMETRIC MODELS WITH ml

    Model

    R n , t = β n , t − 1 T F t + ε R_{n,t}=\beta^T_{n,t-1}F_t+\varepsilon Rn,t=βn,t1TFt+ε

    x : = ε t L : = ( ε n , t − L ) x:=\varepsilon_t^L:=(\varepsilon_{n,t-L}) x:=εtL:=(εn,tL)

    w t − 1 ε = w ε ( θ ( ε t − 1 L ) ) w_{t-1}^\varepsilon=w^\varepsilon(\theta(\varepsilon_{t-1}^L)) wt1ε=wε(θ(εt1L))

    w t − 1 R = w_{t-1}^R=\frac{}{} wt1R=

    d X t = κ ( μ − X t ) dX_t = \kappa(\mu-X_t) dXt=κ(μXt)

    θ i = ∑ j = 1 L W j f i l t e r X j \theta_i=\sum_{j=1}^{L}W_j^{filter}X_j θi=j=1LWjfilterXj

    W W^{} W

    θ C N N + T r a n s ( X ) \theta^{CNN+Trans}(X) θCNN+Trans(X)

    y I ( 0 ) = ∑ m = 1 D s i z e W m l o c a l X y_I^{(0)}=\sum_{m=1}^{D_{size}}W_m^{local}X yI(0)=m=1DsizeWmlocalX

    h i = ∑ I = 1 L α i , I x I ~ h_i=\sum_{I=1}^{L}\alpha_i,I\widetilde{x_I} hi=I=1Lαi,IxI

    F a m a − F r e n c h F a c t o r Fama-French Factor FamaFrenchFactor

    C N N + T r a n s f o r m CNN+Transform CNN+Transform

    α , t α , R 2 \alpha, t_\alpha,R^2 α,tα,R2

    t μ t_\mu tμ

    w t − 1 = w t − 1 w_{t-1}=\frac{w_{t-1}^{}}{} wt1=wt1

    L = 60 L=60 L=60

    F F N FFN FFN

    < 1 % <1\% <1%

    T t r a i n = 4 T_{train}=4 Ttrain=4

    f a s t − r e v e r s a l fast-reversal fastreversal

    • fast reversal
    • early momemtum
    • low frequency downturn
    • low frequency momentum

    • smooth trends or local curvature
    • most recent 14 days get more attention for trading decision

    • more complex than simple reversal patterns

    c o s t ( w t − 1 R , w t − 2 R ) = 0.0005 ∣ ∣ w t − 1 cost(w_{t-1}^R, w_{t-2}^R)=0.0005||w_{t-1} cost(wt1R,wt2R)=0.0005wt1

    B = 7 B=7 B=7

    S R = 1 SR=1 SR=1

    a r b i t r a g e arbitrage arbitrage

    m e a n mean mean

    Δ P = P 2 − P 1 \Delta P=P_2-P_1 ΔP=P2P1

    V = ∑ V=\sum V=

    V = ∣ β 0 + β 1 Δ P ∣ V=|\beta_0+\beta_1\Delta P| V=β0+β1ΔP

    β 0 = c ( μ A − μ B ) \beta_0=c(\mu_A-\mu_B) β0=c(μAμB)

    β 1 = f ( r i s k ) \beta_1=f(risk) β1=f(risk)

    E ( V ) = E [ ∣ β 0 + β 1 σ P Z ∣ ] E(V)=E[|\beta_0+\beta_{1\sigma P}Z|] E(V)=E[β0+β1σPZ]

    Z Z Z

    N ( 0 , 1 ) N(0,1) N(0,1)

    E ( V ) = c o n s t a n t E(V)=constant E(V)=constant

    1 1 + ϕ ( h ∣ β 0 ∣ β 1 ) \frac{1}{1+\phi (\frac{h|\beta_0|}{\beta_1})} 1+ϕ(β1hβ0)1

    K > S T K>S_T K>ST

    K ≤ S τ K \le S_\tau KSτ

    K − S τ K-S_\tau KSτ

    K > S 0 , k = S 0 K>S_0, k=S_0 K>S0,k=S0

    m o n e y n e s s = l o g ( K S 0 ) σ τ moneyness=\frac{log(\frac{K}{S_0})}{\sigma \sqrt{\tau}} moneyness=στ log(S0K)

    l o n g d a t e d = l a r g e τ long dated = large \tau longdated=largeτ

    M o n e y n e s s = l o g ( K S 0 ) σ τ Moneyness = \frac{log(\frac{K}{S_0})}{\sigma\sqrt{\tau}} Moneyness=στ log(S0K)

    S P X SPX SPX

    3 b i l l i o n 3 billion 3billion

    R V t o p t i o n = ∑ i ( r i , t o p t i o n ) 2 RV_t^{option}=\sum_i (r_{i,t}^{option})^2 RVtoption=i(ri,toption)2

    realized variance
    R V t o p t i o n = ∑ i ( r i , t o p t i o n ) 2 RV_t^{option}=\sum_i(r_{i,t}^{option})^2 RVtoption=i(ri,toption)2

  • 相关阅读:
    MySQL函数运用
    接口基础知识1:认识接口
    【Vue】单项数据流的理解
    什么是JVM常用调优策略?分别有哪些?
    小学生python游戏编程arcade----游戏界面按钮实现事件实现的三种方法
    【Java分享客栈】一个包装过简历的新同事写完微信支付引起事故后果断离职了
    Scala 练习一 将Mysql表数据导入HBase
    《一》基础
    四、nginx负载均衡[轮询]
    A-level物理真题得这么刷
  • 原文地址:https://blog.csdn.net/Dequn_Teng_CSDN/article/details/125476020