Higher Moments and Efficiency Gains in Recursive Structural Vector Autoregressions
研究了递归SVAR模型中利用高阶矩条件（如协偏度和协峰度）能否提升估计效率，发现仅用协方差条件会损失效率，但部分高阶矩条件总是冗余，排除它们才能在小样本中获益。
ABSTRACT Recursive SVAR models are identified by covariance conditions derived from the assumption of uncorrelated shocks. Recent literature has advocated using additional higher‐order moment conditions implied by independent shocks. This paper characterizes the redundancy properties of these higher‐order coskewness and cokurtosis conditions by showing that recursive SVAR estimators that rely exclusively on covariance conditions, neglecting the additional identifying information in higher‐order moments, are asymptotically inefficient. Moreover, we prove that some higher‐order moment conditions are always redundant and provide no improvement in asymptotic efficiency. A simulation demonstrates that excluding redundant conditions is essential to achieve performance gains in small samples.