【AP】Multivariate GARCH models and the BL approach for tracking error constrained portfolios

最新研究 同时被 3 个专栏收录
30 篇文章 1 订阅
6 篇文章 0 订阅

Paper Link

Multivariate GARCH models and the BL approach for tracking error constrained portfolios

Pub Date

2008 vol 10. No. 4 Global Business and Economics Review

Abstract

In a typical tatical asset allocation setup, managers generally make their choices with the aim of beating a benchmark portfolio. In this context, the pure markowitz (1959) strategy does not take two aspects into account: asset returns often show changes in volatility and manager's decisions depend on private information.(指出经典MV模型存在的两个缺陷: 设置了资产收益率的波动性为常数以及模型为静态决策)
This paper provides an empirical model for large-scale tactical asset allocation with multivariate GARCH estimates, given a tracking error constraint. Moreover, the Black and Litterman (1991) approach makes it possible to tactically manage the selected portfolio by combining information taken from the time-varying volatility model with some personal views about asset returns.

Introduction

基于追求主动收益的资产配置策略(Tactial Asset Allocation, TAA) ,也被称为alpha策略,要求在控制组合波动性的前提下最大化主动收益. 该策略的基本假设是:最优投资组合由三部分组成,最小方差(minimum variance), 投资策略(strategic)战术配置(tactical).

The aim of this work is to show how it is possible to make a portfolio optimization in the presence of a large number of assets by combining two different types of information:

本文的主要目的是结合两种信息的情况下完成资产优化配置,第一种信息是估计出时变的波动率模型,第二种信息是投资经理人是否可以押注时变的资产超额收益.(private information that derives from the manager's bets about the evolution in time of asset excess returns).

Portfolio frontiers

给定 n n n维向量 R R R表示表示资产的超额收益率, Ω \Omega Ω表示资产的方差-协方差矩阵, P P P表示投资经理选择的投资组合,组合的期望超额收益和方差分别为 R p R_p Rp σ p 2 \sigma_p^2 σp2,向量 ω \omega ω表示组合权重,相应的,符号 R B R_B RB σ B 2 \sigma_B^2 σB2 ω B \omega_B ωB表示基准组合 B B B的对应参数. 可以在mean-variance框架下求解出有效前沿.

Constant TE frontier

主动资产配置问题的优化模型为满足约束的条件下,策略组合 q q q和有效前沿权重 ω \omega ω的组合收益差距最大
max ⁡ ( ω − q ) ′ R s . t . { ( ω − q ) ′ 1 = 0 ( ω − q ) ′ Ω ( ω − q ) = T E ω ′ Ω ω = σ p 2 \max (\omega-q)'R\\ s.t. \begin{cases} (\omega-q)'\mathbf{1}=0\\ (\omega-q)'\Omega(\omega-q)=TE\\ \omega'\Omega\omega=\sigma_p^2 \end{cases} max(ωq)Rs.t.(ωq)1=0(ωq)Ω(ωq)=TEωΩω=σp2

The BL model

BL模型下得到的后验超额均值向量及观点不确定矩阵如下
vector

The aim of BL model is to insert uncertain personal views into the equilibrium returns to modify the portfolio weights in the direction of manager’s hypothesised scenarios.

但是BL模型存在着两个问题:
第一,在先验信息和投资者观点中的联合正态分布假设(joint normality assumption)与实际情况不相符合,在Fabozzi(2007)1的论文中放松了资产收益率符合Gaussian distribution的假设,而选择使用一些厚尾分布,如 α − \alpha- αstable或者student-t分布.

they find that information depends on how the different distributions impact the optimal portfolio. This is true for the marginal distributions of expected returns.

第二,模型的本质是Bayesian,后验分布由经理人的投资观点而不是市场实际数据产生.

Meucci(2005) solves this problem by using a copula and opinion-pooling methodology to determine the posterior market distribution. Moreover, he claims that his extension to the BL model be applied to any market distirbution and non-normal views.

The model

data

DJ Euro Stoxx 50作为基准投资组合

the time-varying volatility model

TAA策略中需要使用时变的波动率模型估计和预测returnscovariance. 关于在TAA中使用波动率模型的理由论述如下

The use of this approach yields two benefits: first modelling heteroskedasticity explicitly leads to the increased efficiency in the estimation of the parameters of the conditional mean. Moreover, forecasting the covariance matrix for different time horizons would be useful for the TAA, especially because the forecast of the conditional covariances is likely to be the main object interest.

本文使用了Billio等人开发的Flexible Dynamic Conditional Correlations, FDCC模型,给定一个 n n n维向量 y t y_t yt,标准的FDCC模型表示如下
{ y t = μ + Π y t − 1 + ε t E ( ε t ∣ F t − 1 ) = 0 E ( ε t ε t ′ ∣ F t ) = Ω t Ω t = D t − 1 / 2 R t D t − 1 / 2 (8) \begin{cases} y_t=\mu+\Pi y_{t-1}+\varepsilon_t\\ E(\varepsilon_t\mid\mathcal{F}_{t-1})=0\\ E(\varepsilon_t\varepsilon_t'\mid \mathcal{F}_t)=\Omega_t\\ \Omega_t=D_t^{-1/2}R_tD_t^{-1/2} \end{cases}\tag{8} yt=μ+Πyt1+εtE(εtFt1)=0E(εtεtFt)=ΩtΩt=Dt1/2RtDt1/2(8)
FDCC模型的推导分为两个步骤:

  1. 估计出时间序列的条件均值和条件方差
  2. 推断动态相关性矩阵(dynamic correlations matrix)

参数估计由以下方程得到
y i t = μ i + π i y t − 1 + ε i , t y_{it}=\mu_i+\pi_iy_{t-1}+\varepsilon_{i,t} yit=μi+πiyt1+εi,t

h i , t = ω + α ε i , t − 1 2 + β h i , t − 1 h_{i,t}=\omega+\alpha\varepsilon_{i, t-1}^2+\beta h_{i, t-1} hi,t=ω+αεi,t12+βhi,t1
使用FDCC模型的目的如下
FDCC

Blend Views

本文在BL模型中设置观点不确定矩阵原则如下
views

Concluding remarks and further research

conclusions


  1. stable distributions in BL approach to asset allocation ↩︎

  • 1
    点赞
  • 0
    评论
  • 1
    收藏
  • 一键三连
    一键三连
  • 扫一扫,分享海报

©️2021 CSDN 皮肤主题: 点我我会动 设计师:白松林 返回首页
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、C币套餐、付费专栏及课程。

余额充值