Maximum empirical likelihood estimation of continuous-time models with conditional characteristic functions
Maximum empirical likelihood estimation of continuous-time models with conditional characteristic functions
复制标题
DOI：
10.1016/j.matcom.2008.01.007
复制
发表时间：
2008-07
期刊：
影响因子：
--
通讯作者：
Qingfeng Liu;Y. Nishiyama
中科院分区：
文献类型：
--
作者：
Qingfeng Liu;Y. Nishiyama
对于一些流行的金融连续时间模型，易于处理的似然函数的表达式是未知的。因此，最大似然估计方法是不可行的。幸运的是，这些模型中的一些的条件特征函数的封闭函数形式是已知的。我们构造了一个经验似然估计方法，使用易处理的条件特征函数来估计这样的模型。该方法解决了标准广义矩量法中协方差阵奇异的问题，充分利用了条件矩约束中的信息。它适用于许多流行的金融模型，如一些扩散模型，跳跃扩散模型和随机波动率模型。使用蒙特卡罗比较，我们表明，这种方法提供了上级性能相比，其他方法在某些情况下。
For some popular financial continuous-time models, tractable expressions of likelihood functions are unknown. For that reason, the maximum likelihood estimation method is infeasible. Fortunately, closed functional forms of conditional characteristic functions of some of these models are known. We construct an empirical likelihood estimation method using tractable conditional characteristic functions to estimate such a model. This method resolves the problem of covariance matrix singularity in the standard generalized method of moments and fully utilizes information in conditional moment restrictions. It is applicable to many popular financial models such as some diffusion models, jump diffusion models, and stochastic volatility models. Using a Monte Carlo comparison, we show that this method provides superior performance compared to other methods in some situations.