A0314
Title: On the evaluation of intraday market quality in the limit-order book markets: A collaborative filtering approach
Authors: Makoto Takahashi - Hosei University (Japan) [presenting]
Takaki Hayashi - Keio University (Japan)
Abstract: A methodology for evaluating the liquidity of individual stocks in the high-frequency domain is considered by utilizing a framework from recommender systems that have become ubiquitous in our daily lives. In general, it is not necessarily easy to evaluate the ``true" liquidity of individual stocks. In particular, evaluating liquidity over a shorter term with high-frequency data can be challenging for many stocks due to the increasing sparsity of observations. Since stocks that have exhibited similar behavior in the past are expected to perform so in the future as well, one can expect that collaborative filtering, which is the main approach of recommender systems, can work effectively for the liquidity ``estimation" problem. Specifically, we adopt a regression-based latent factor model (RLFM), hybrid-type collaborative filtering. It has a hierarchical structure designed to address the so-called ``cold-start" problem in the recommender systems literature. As a result of the empirical analysis using high-frequency limit-order book data from the Tokyo Stock Exchange, various characteristics that characterize liquidity were identified from the estimated regression coefficients obtained by fitting the RLFM to the training dataset. In the meantime, there was room for improvement of the methodology regarding the accuracy of liquidity prediction.
