资产配置模型在中国资本市场真的有效吗?

The Performance of Asset Allocation Models in Chinese Capital Market

本文在标准的均值—方差模型框架下,以改善估计误差为主线,选取经典的样本外MV模型和其他15种具有代表性的资产配置模型,运用六组中国资本市场数据进行对比研究。结果表明,增加估计窗口会减少估计误差,11种非卖空限制策略表现优于样本外MV模型,卖空限制模型优于非卖空限制模型。同时,本文选取了简单的等权重(EW)策略作为基准策略,发现估计误差对资产配置产生了严重影响,导致几乎没有模型的确定性等价收益(Certainty Equivalent return,CEQ)一致优于EW策略。为进一步了解估计窗口长度和卖空限制对估计误差的影响,本文以证监会分类的13个行业数据为例进行模拟,结果表明,当估计窗口从60个月增加到600个月时,非卖空限制模型的CEQ大幅提高(除EW策略外),但是,当估计窗口达到600个月后,增加估计窗口对估计误差的改善效果在减少;相对于非卖空策略,卖空限制策略在估计窗口较短时对估计误差改善程度较大,当估计窗口达到480个月后,CEQ大致保持不变,说明当估计窗口达到一定临界值后,卖空限制会抑制投资组合模型的配置效果。最后,本文根据这些实证结果提出了相应的完善资本市场的政策建议。

The performance of asset allocation has been widely concerned since Markowtiz （1952） developed the mean-variance framework. Due to the estimation errors in performance expected returns and the covariance, the out-of sample for Markowtiz＇s approach is not good as expected. Thus, efforts for improving Markowtiz＇s approach have been made for decades, and many improved asset allocation models have been proposed, such as： equM1 risk contribution portfolio,most diversified portfolio （MDP）. Comparison between each of those recently y weighted developed models and Markowtiz＇s approach has been made, yet comparison among those diversification models is more needed to be done. DeMieguel et al. （2009） compare the out-of-sample performance of 14 asset allocation models using da- ta from developed stock markets for the first time, and find no asset allocation model outperforms a very naive strate- gy：equally andalso its weighted portfolio. As one of the fastest growth developing economy, China plays more and more roles capital market attracts more and more international investors. The performance of existing asset alloca- tion models in Chinese market becomes one important issue which has not been investigated yet. This paper evaluates the out-of-sample performance of 16 asset allocation models by using six datasets from Ju- ly, 1997 to June ,2013 in Chinese capital market and includes more recent asset allocation models relative to the em- pirical study by DeMieguel et al. （2009）. And also the performance of those models with different length of rolling estimation windows is investigated. Empirical study in this paper shows that no strategy outperforms the benchmark： equ^Uy weighted portfolio in general. In addition, for the six datasets in Chinese capital market, we find that increas- ing the length of estimation window can reduce the estimation error. The 11 unconstrained models perform better than the mean-variance model, the 4 constrained models better than the other unconstrained models. And none of the models is consistently better than the benchmark in terms of certainty-equivalent return （ CEQ）, which indicates that the gain from optimal diversification is largely offset by estimation error. Based on the 13 sector portfolios in Chinese capital market, our simulation results show that window length and short-sale constraints have significant effect on the estimation error. Unconstrained models, except for equally weighted strategy, improves performance greatly when window length increases from 60 months to 600 months. While the constrained models performs better when window length is relatively small, but their CEQs remains nearly unchanged after 480 months, which indicates that short-sale constrains will have depressing effect on asset allocation after s certain window length. This paper implicates that even in Chinese capital market which is supposed to be less efficient relative to U. S. market,none of those complicated asset allocation models outperforms the naive equally weighted strategy on the basis of out-of-sample tests. And lengthening the estimation window indeed helps to reduce the estimation errors in e the xpected returns and covariance ,yet presents marginal length improvement on the performance in terms of CEQ, when of estimation window is greater than 20 years. Our results suggest that estimation errors has less impact on constrained models relative to the unconstrained models which are considered in this paper,which indicates that un- constrained models should be carefully treated when they are put into practice. Our results are robust to the length of estimation window, number of individual assets in the portfolio and the coefficient of risk aversion. Our findings call into question the efficiency of asset allocation models when the length of estimation window is small and naive diversification strategy is suggested in practice.