随机条件久期模型的贴近度新息构建方法

针对随机条件久期(SCD)模型伪似然估计方法的非有效性问题,分析了新息对数正态分布的可能,证明了经修正的对数正态分布不仅满足SCD模型新息构建要求,其在期望值设定和随机取值方面优于原有新息分布,并进一步在此分布基础上提出了贴近度新息构建方法,该方法不同于期望新息构建方法,目的在于产生更少极端值使拟合久期接近实际久期,进而提高模型拟合和预报性能.同时也证明了新方法能包含原新息构建方法,具有一般性.当然,新息的对数正态分布也保证了SCD模型卡尔曼滤波解具备有效性,属于非伪似然估计.最后,通过一组模拟数据证实了新息对数正态分布的可行性,以及所提贴近度新息构建方法的优越性.

Construction of the closeness innovation in the stochastic conditional duration model

To solve the non-effectiveness of pseudo-likelihood estimation approach in SCD(stochastic conditional durations) model,this paper analyzes lognormal distribution and amends it to suit SCD model.Further, this paper gives a closeness innovation on the basis of lognormal distribution to produce less extreme values and make the fitting durations close to the practical durations,which is different from the old expectation innovation. Then,this paper proves that the new innovation is a general innovation and can produce the original expectation innovation.Of course,the log-normal distribution of the new innovation also ensures the effectiveness of the Kalman filter solution in solving the SCD model,and is a non-pseudo-likelihood estimation. Finally,by a set of simulated data,the feasibility and the superiority of this new method about the innovation based on the lognormal distribution in SCD model are demonstrated.