Econometrics
Financial economics
Empirical finance

Econometric Theory
Applied Econometrics
Financial Markets
Derivatives and Option Pricing

October 1997 (Oral) Econometrics, Financial Economics
May 1996 (Written) Microeconomic Theory (with distinction), Macroeconomic Theory

Semiparametric Modeling of Competing Risks in a Limit Order Market

Professor Peter C.B. Phillips
Professor Donald W. Andrews
Professor Matthew Spiegel
Professor Robert Shiller

May 2001

M.Phil., Yale University, 1999
M.A. (with distinction), New Economic School (Moscow), 1995
Diploma cum laude, Moscow State University, 1991

Young Economist Best Paper Award, Central Bank of Turkey, 2000
Dissertation Fellowship, Yale University, 2000
Carl A. Anderson Fellowship, Cowles Foundation, 1999
Graduate School Fellowship, Yale University, 1997–1999
Falk Fellowship, Yale University, 1995–1997

Teaching Assistant, Introductory Macroeconomics (undergraduate course), Professor George J. Hall, Spring 2001
Teaching Assistant, Financial Theory (undergraduate course), Professor John Geanakoplos, Fall 2000
Teaching Assistant, Time Series Econometrics: Unit Roots and Cointegration, Professor Peter C.B. Phillips, Spring 2000
Head Teaching Assistant, Finance: Theory and Applications (undergraduate course), Professor Robert J. Shiller, Spring 1999
Teaching Assistant, Econometrics (graduate course), Professor Oliver Linton, Fall 1998
Teaching Assistant, Microeconomic Theory (graduate course), Professors Herbert E. Scarf and John Geanakoplos, Spring 1998
Teaching Assistant, Probability and Statistics (undergraduate course), Professor Donald W. Andrews, Fall 1997

Research Assistant, Professor Donald W. Andrews, Spring 2000
Research Assistant, Professor Peter C.B. Phillips, Summer 1999

"Semiparametric Estimation of Competing Risks and Order Flow Dynamics in FOREX Brokerage", 2000, accepted for publication in Istanbul Stock Exchange Review.

"The Occupation Density of Fractional Brownian Motion and Some of Its Applications" (with Peter C.B. Phillips), mimeo, Yale University, 1999.

"Large Sample Properties of Adaptive Hazard Functions Estimators in the Competing Risks Environment", in progress.

"Forecasting the High-Frequency Momentum in Tick-by-Tick FOREX Quotes Using Information on the Limit Order Book", in progress.

"Empirical Tests of the Arbitrage Pricing Theory on Japanese Stock Market Data", mimeo, Yale University, 1998.

Conference Presentations:
"Semiparametric Estimation of Competing Risks and Order Flow Dynamics in FOREX Brokerage", presented at the ERC/METU International Conference in Economics, Ankara, Turkey, September 13–16, 2000.

"The Occupation Density of Fractional Brownian Motion and Some of Its Applications", presented at the Cowles Foundation Econometrics Conference, Yale University, October 23–24, 1999.

Other papers:
"The Occupation Density of Fractional Brownian Motion and Some of Its Applications", which is joint work with Peter C.B. Phillips, develops a theory of chronological local time for fractionally integrated processes. The new approach employed in the paper is based on the approximation of the chronological local time of a fractional Brownian motion by local times of continuous diffusion mixtures and can be applied to generalize the asymptotic theory of local times of Wiener processes to the case of a fractional Brownian motion. An almost sure approximation of the spatial density based on a discrete sample of observations is obtained for a wide class of long-memory processes. The methods of the paper can be used to generate empirical estimates of spatial densities of fractionally integrated time series in economics and finance and to conduct inference with nonlinear

This dissertation studies various aspects of the microstructural dynamics of an order-driven electronic financial market in the statistical framework of competing risks. The approach identifies several discrete types of market and limit orders, and models the hazards of their arrival, execution, and cancellation in continuous time. The hazard rates are allowed to depend semiparametrically on time since the last observable event and on a linear index of covariates characterizing the past history and current market conditions. The model can incorporate various patterns of unobserved heterogeneity due to time varying market conditions. To the extent that the structure of competing risks closely replicates the dynamics of traders’ activity, the approach can help researchers of market microstructure and practitioners better understand and interpret the behavior of market participants. Moreover, some empirical results might be strong enough to trigger interest among traders and their sponsors, who are intrinsically interested in developing dynamic strategies to square existing positions.

The first chapter applies the competing risks methodology to analyze the timing and interaction between the Deutsche Mark/U.S. dollar quotes and trades in the Reuters D2000-2 electronic brokerage system. Estimation of this model generally supports empirical evidence from previous research. In particular, the composition of the order flow is found to be very sensitive to the state of the limit order book and the trading history. The direction of past trade has strong predictive power for future activity of buyers and sellers in the market. There is some evidence of an adverse information effect due to non-trading that manifests itself in the negative dependence of aggressive order arrival rates on time since the last observed event. Finally, traders tend to submit and cancel their orders more aggressively immediately after changes in the electronic order book.

The second chapter studies in detail the problem of nonparametric and semiparametric estimation of competing risks when the sample of observed durations is highly skewed. In this situation, which is fairly common for high-frequency financial data, the fixed bandwidth kernel estimator often fails to detect fine peaks of the hazard functions on the left and leads to extremely volatile hazard estimates in the right tail of the duration range. The paper evaluates the large and finite sample performance of two alternatives, the k-nearest neighbor and the combined duration scale deformation–kernel estimation techniques. The asymptotic properties derived for the k-nearest neighbor estimators in the covariate-free case are also investigated for a number of semiparametric competing risks specifications. The effectiveness of the duration scale deformation method will be evaluated for artificial and real financial datasets.

The third chapter (in progress) evaluates the quality of forecasts based on the competing risks model of a limit order market. Preliminary results indicate an ability of the model to capture clustering of the buyer- and seller-initiated events and predict the direction and scale of the high-frequency fluctuations in tick-by-tick foreign exchange data. Forecasting performance of the semiparametric competing risks model will be compared to alternative models applied to high-frequency financial data.