文摘
Since the pioneering work of Mandelbrot 1963) and Fama 1963),researchers have extensively studied the regularities of financial series. A common statistical modeling assumption is the errors are from the Gaussian distribution. With this assumption comes the maladies that affect the Gaussian estimator: Lack of robustness,inefficiency under non-Gaussian data-generating processes,and the inability to capture the observed skewness and excess kurtosis of most financial series. In this paper we develop an adaptive methodology by using a local basis based on polynomial splines for modeling conditional location and a time-varying scale. The time-varying scale feature of the model is an adaptation of Bollerslevs 1986) LARCH model. By modeling via a local basis we can reduce the influence of highly discrepant observations and effectively model the excess kurtosis and skewness of many series. Simulation evidence suggests the loss of efficiency with respect to the Gaussian estimator is mitigated for leptokurtic and asymmetric distributions. An example using the $/£; exchange rate illustrates how our model can effectively capture the heterogeneity and nonlinearities of many time series. With the proliferation of computing power and storage capabilities,financial data can be collected at shorter intervals than just a few years ago; for example,each transaction from an exchange can be recorded. Unlike previous studies that model the time between transactions completely parametrically,in this paper we use the semiparametric survival model of Kooperberg,Stone,and Troung 1995). The primary objective of this paper is to examine how important trade characteristics are on the prices spacings and a measure of instantaneous volatility using the semiparametric survival model. Graphical methods and specification tests indicate the significant dependence between arrival times can be sufficiently modeled in the semiparametric framework. When the semiparametric model is compared to a theoretical model of geometric Brownian motion,diagnostics reveal the semiparametric model outperforms the hypothetical model. The empirical findings are that information flow variables,such as volume,spreads and trading imbalances,predict more rapid price revisions. Tests of different market microstructure models lends credence to the theoretical assertions that movement in prices are due to informed traders and not liquidity traders.