Detecting spurious jumps in high frequency data

We propose a technique to avoid spurious detection of jumps in high frequency data via an explicit thresholding on available test statistics. We prove that it eliminates asymptotically all spurious jumps. Monte Carlo results show that it performs also well in finite samples. Our empirical investigation of Dow Jones stocks reveals that the spurious detections represent up to 50% of the jumps detected initially. After eliminating the spurious detections with our method, the average number of jumps amounts to around 40 a year. For the majority of Dow Jones stocks, we do not detect clustering in time of jumps occurrences. During the three years of our study, we find no single cojump affecting all Dow Jones constituents. However, if we consider industry sectors separately, we observe a number of cojumps significantly larger than if the stocks were independent. Finally, we relate detected jumps to news releases. JEL classification: C10, C22, G10.