Econ201FS- New Jump Test Haolan Cai
Lee and Mykland (2007) Introduces a new Jump Test with New test statistic (ζ) Describes Asymptotic Behavior of ζ Describes Misclassifications of ζ Becomes negligible at high frequencies Tests jump detection using MCMC Simulation Claims it detects jumps better than Barndoff-Nielsen and Shephard (2006) and Jiang and Oomen (2005)
My Contribution Investigation of Jump Test Detection in real data (GE and Intel) Comparison to existing BN-S tests at varying time intervals
What is ζ? So basically, ζ is the proportion of logged returns to the realized bipower variation over a given window size, K. Claim that using bipower variation for estimating instantaneous volatility is robust.
How is ζ distributed? Ui is a standard normal variable which means its distributed with a mean of zero and variance of 1. C is a constant approximately equal to .7979.
What does K mean? Lee & Mykland (2007)
Implementation Data: Test Parameters: 1-minute prices from General Electric and Intel Corporation GE: April 9, 2007 9:35 am to Jan 24, 2008 3:59 pm INTC: April 16, 2007 9:35 am to Jan 24, 2008 3:59 pm Test Parameters: K = 10 (for Lee & Mykland test) Use Quad-power Quarticity (for BN-S test) Time Intervals: 1-10 min. Significance Level: .01% (z-score = 3.09)
Results- INTC (in days) Barndoff-Nielsen & Shephard Lee & Mykland
Results- INTC (in percentages) Lee & Mykland Barndoff-Nielsen & Shephard
Similar Results for GE Barndoff-Nielsen & Shephard Lee & Mykland
Remarks Lee and Mykland test does seem to be less affected by increase in time interval Consistently finds the same number/percentage of jump days Are these jumps ‘real jumps’ or due to microstructure noise? Are the jumps being detected by the L & M test the same as the BN-S test?