用GARCH模型预测股票指数波动率

用GARCH模型预测股票指数波动率

目录

Abstract (2)

1.引言 (3)

2.数据 (6)

3.方法 (7)

3.1．模型的条件平均 (7)

3.2. 模型的条件方差 (8)

3.3 预测方法 (9)

3.4 业绩预测评价 (9)

4.实证结果和讨论 (12)

5.结论 (16)

References (18)

Abstract

This paper is designed to make a comparison between the daily conditional variance through seven GRACH models. Through this comparison, to test whether advanced GARCH models are outperforming the standard GARCH models in predicting the variance of stock index. The database of this paper is the statistics of 21 stock indices around the world from 1 January to 30 November 2013. By forecasting one –step-ahead conditional variance within different models, then compare the results within multiple statistical tests. Throughout the tests, it is found that the standard GARCH model outperforms the more advanced GARCH models, and recommends the best one-step-ahead method to forecast of the daily conditional variance. The results are to strengthen the performance evaluation criteria choices; differentiate the market condition and the data-snooping bias.

This study impact the data-snooping problem by using an extensive cross-sectional data establish and the advanced predictive ability test. Furthermore, it includes a 13 years’ period sample set, which is relatively long for the unpredictability forecasting studies. It is part of the earliest attempts to inspect the impact of the market condition on the forecasting performance of GARCH models. This study allows for a great choice of parameterization in the GARCH models, and it uses a broad range of performance evaluation criteria, including statistical loss function and the Mince-Zarnowitz regressions. Thus, the results are more robust and diffusely applicable as compared to the earliest studies.

KEY WORDS: GARCH models; volatility, conditional variance, forecast, stock indices.

1.引言

波动性预测可以运用到投资组合选择，期权定价，风险管理和以波动性为基础的交易策略。GARCH模型族被广泛的运用在模拟预测金融资产的波动性。另一个普遍运用的模式为简单的时间序列模型，例如指数加权移动平均（EWMA）模型和复杂随机波动性模型(Poon and Granger,2003)。对不同金融市场波动性的预测，Ederington在2005年发现GARCH模型通常的表现优异于EWMA模型。同样的，关于随机过程的波动率建模，有强有力的证据证明随机波动模型的样品性能堪比GARCH模型（Fleming and Kirby,2003）.

标准GARCH模型于1986年被Bollerslev提出后，为了规范条件方差，更多复杂的GRACH 模型参数被提出。这些先进的GARCH模型试图去更好的捕捉经验主义观察到条件方差的过程。例如，EGARC模型，GJR模型，TGARCH模型和NGARCH模型获得的负返回流的非对称性效应。更为广义的参数化，像APARCH模型和HGARCH模型，包含大量较为简单的GARCH模型（Zakoian, 1994）。尽管如此，用复杂的GARCH模型族来预测成绩并未让人留下深刻印象。Bali和Demirtas（2008）利用GARCH模型，EGARCH模型和TGARCH模型预测S&P500的未来指数。他们发现EGARCH模型最精准的预测了未来实际的波动性。Cao和Tsay在1992年提出EGARCH模型对小型股票提供了最好的长期预测，但是对于大型股票来说，其他时间序列模型会更为适合。Alberg（2008）发现EGARCH模型为Tel Aviv Stock Exchange(TASE)的股票指数提供了最好的方差预测。然而，Ederington 和Guan却指出在对大量资产种类波动性进行预测的过程中，GARCH模型和EGARCH模型是没有显著差别的。Lee在1991年提出，GARCH模型对样本外预测成绩取决于损失评估标准。2004年，Taylor比较了五种不同的GARCH 模型，发现GJR和IGARCH模型是最好的。利用均方根误差，平均绝对误差和平均绝对百分比误差的GJR模型被Brailsford认为是最好的（1996）。但是，Franses和Van在同年利用方差中值作为损失标准，发现QGARCH和GARCH模型在样本外预测上的表现优于GJR模型。预测汇率的波动性，Brooks和Bruke（1998）发现GARCH模型倾向于均方误差，但不建立在平均绝对误差的标准上。2004年，Balaban发现在预测汇率波动性上，EGARCH模型为最优，GJR模型为最差。但是，预测的优异取决于所选的损失标准。

因为严重参数化的模型更有利于获得多维度的波动性数据，因此一个好的实例在转变为样本外预测时可能并不重要。在样本外预测能力方面，简单的模型往往比复杂模型更有优势。通