A0598
Title: Modeling lead-lag effects using bivariate Neyman-Scott processes with gamma kernels
Authors: Takaaki Shiotani - Graduate School of Mathematical Sciences, The University of Tokyo (Japan) [presenting]
Abstract: The lead-lag effect refers to the correlation that occurs with a time difference between 2 time series data. It has been explored in financial engineering because it can be applied to statistical arbitrage or understanding market structures. Broadly, there are two approaches: one focusing on the correlation structure of prices and the other on the correlation structure of transaction/order times. The latter approach, which is thought to be robust against microstructure noise, is considered. An intuitively understandable model is proposed utilizing a multivariate Neyman-Scott point process model equipped with gamma kernels to model the lead-lag effect. Furthermore, an efficient computational method is developed for parameter estimation through quasi-likelihood. Numerical experiments are conducted to check the proposed estimator's performance. In addition, high-frequency Japanese stock data is analyzed to verify that the model effectively explains the correlation between pairs of stocks. Research provides a new point process model for lead-lag estimation that is highly interpretable and usable within a realistic computational timeframe. Additionally, the method can be applied to phenomena beyond financial data, such as social media and earthquake data.
