Nonparametric Density Estimation and Tests of Continuous Time Interest Rate Models
研究了短期利率非参数密度估计的有限样本性质，发现渐近分布低估了偏差和方差，且检验过度拒绝，调整后对Vasicek和CIR模型的区分力较低。
Nonparametric kernel density estimation has recently been used to estimate and test short-term interest rate models, but inference has been based on asymptotics. We derive finite sample properties of kernel density estimates of the ergodic distribution of the short-rate when it follows a continuous time AR(1) as in Vasicek. We find that the asymptotic distribution substantially understates finite sample bias, variance, and correlation. Also, estimator quality and bandwidth choice depend strongly on the persistence of the interest rate process and on the span of the data, but not on sampling frequency. We also examine the size and power of one of Ait-Sahalia's nonparametric tests of continuous time interest rate models. The test rejects too often. This is probably because the quality of the nonparametric density estimate depends on persistence, but the asymptotic distribution of the test does not. After critical values are adjusted for size, the test has low power in distinguishing between the Vasicek and Cox-Ingersoll-Ross models relative to a conditional moment-based specification test.