The present dissertation includes my research on modeling financial market returns and volatility on high frequency data. It consists of three chapters. The first chapter introduces a novel high-frequency volatility estimator based on large price moves, which constitutes a generalization of the standard range. An asymptotic theory is developed in a jump-diffusion setting, originating a family of strongly consistent diffusive volatility estimators that are robust to jumps. This theory tackles maximization on a time grid not previously studied in the volatility literature and uncovers valuable distributional properties of Brownian peaks and troughs, formalized as h-extrema. Supplementary extensive simulation evidence documents that the generalized range behaves in accordance with the derived theory and compares favorably to other known estimators of diffusive volatility that are robust to jumps. The second chapter lays down basic applications and extensions of the generalized range. The main empirical finding is that the generalized range proves valuable for intraday jump detection and short-term forecasting of stock return volatility. It is also largely robust to microstructure noise when calculated on bid-ask quotes. In a model-free environment, the capability of the generalized range to identify large zig-zag price moves appears to be directly applicable to relative value arbitrage strategies. By illustrating two different ways of using the generalized range on real data, this chapter provides the foundations for developing further applications to risk management, portfolio choice, and market microstructure. The third chapter develops novel conditional moment tests and related sequential procedure for testing no-arbitrage semi-martingale restrictions on return distributions. In particular, by relying on nonparametric volatility measures on intraday data and explicitly allowing for jumps, leverage effects, and microstructure noise, the developed methodology allows for testing the general adequacy of jump-diffusion models for return distributions. To this end, appropriate conditional moment tests are applied along with a new jump detection technique and a novel robust-to-jumps approach alleviating microstructure frictions for realized volatility estimation. Size and power of the procedure are explored through Monte Carlo methods. The obtained empirical findings provide additional justification for the jump-diffusion setting adopted in the preceding chapters in the context of the generalized range theory.

Last modified

• 05/28/2018

Creator

• Dobrislav Dobrev

DOI

• https://doi.org/10.21985/N2PH65

Subject

• Finance

Keyword

• Normality Tests
• Jump Detection
• Generalized Range
• Realized Volatility
• Jump-Diffusion Processes
• High-Frequency Data

Date created

• 2007-05-11

Resource type

• Dissertation

Rights statement

• In Copyright