GOODNESS-OF-FIT TEST FOR THE SVM BASED ON NOISY OBSERVATIONS

In financial high frequency data analysis, the efficient price of an asset is commonly assumed to follow a continuous-time stochastic volatility model, contaminated with a microstructure noise. In this study, we consider a goodness-of-fit test problem for the efficient price models based on discretely observed samples and employ a goodness-of-fit test based on the empirical characteristic function. We show that the proposed test is asymptotically a weighted sum of products of centered normal random variables. To evaluate the proposed test, we conducted a simulation study using a bootstrap method. A data analysis is provided for illustration.