By Matthew Roqnile
We investigate the properties of several nonparametric tests for jumps in financial markets. We derive a theoretical property of these tests not observed in any of the previous literature: when they are applied to finitely sampled data, they are generally biased toward finding too many jumps. This results from bias in finite-sample estimation of several important test components. The severity of the bias corresponds to the magnitude of change in volatility over the course of a day. We use data on
an intraday volatility pattern in several US equities, which results in quantitatively significant changes in the level of volatility during the day, to undertake Monte Carlo simulations of a price process without jumps. Applying several jump tests to the simulated data, we detect one-half to two-thirds as many jumps as in the observed data, suggesting that many jumps currently detected in empirical applications of these tests are spurious. We also present several possible modifications to jump tests that limit the effect of intraday patterns in volatility, all of which produce dramatically lower estimates of the frequency and importance of jumps.
Advisor: George Tauchen