Orderbook foundations of price risk and liquidity

Most financial markets are nowadays organized in the form of electronic limit order book markets. The introduction of electronic trading has substantially changed the trading landscape over the last decade. Many high-speed activities on financial markets are not well understood and are subject of concerns to market operators and regulators. Our research will be the first project bringing together the limit order books (LOBs) of a stock market with those of derivatives written on this asset. Using empirical and statistical models we will investigate the vast amount of information provided by integrated stock and derivative LOBs. This information will be processed to measure and predict risks associated with volatility, liquidity and price jumps. We will address two major research questions: First, can we exploit detailed LOB data of stocks and corresponding options to improve the measurement and prediction of volatility? Second, how does LOB information spill over from equity and derivative markets and how can this information be used to better understand the determinants of volatility, liquidity and jump risks on the micro level? We will investigate how posted limit orders, i.e., offers to buy and to sell, contribute to volatility and how they can be used to predict future levels thereof. Derivative prices explicitly provide volatility expectations (called implied volatility) and we will compare these with estimates obtained directly from changes in stock prices (called realized volatility). Previous research has used transaction prices and the best buying and selling prices; we will innovate by using complete LOBs providing significantly more information. In such a setting we will discover how information is transmitted from option LOBs to stock LOBs (and vice versa) and thus identify the most up-to-date source of volatility expectations. We will moreover provide new insights into the microstructure of option markets by evaluating liquidity related to contract terms such as exercise prices and expiry dates. This will allow us to find robust ways to combine implied volatilities into representative volatility indices. We will identify those time periods when price jumps occur and will test methods for using stock and derivative LOBs to predict the occurrence of jumps. We will also model the dynamic interactions between different order types during a jump period. We will use databases which record all additions to and deletions from LOBs, matched with very precise timestamps. For stocks, we use the LOBSTER database which constructs LOBs from NASDAQ prices. For derivatives, we use the Options Price Reporting Authority (OPRA) database. Our research is the first to combine and investigate the information in these separate sources of LOBs.

We have constructed a comprehensive dataset of millisecond-stamped high frequency option trade and quote data disseminated by the Options Price Reporting Authority (OPRA). The dataset spans the first 8 months of 2015 and covers all option classes traded in 12 U.S. security exchanges and written on more than 3,500 equities, more than 500 exchange traded products and about 50 index-driven assets. This dataset is the most comprehensive and granular option dataset available to scholars. We provide a thorough empirical description of the high-frequency characteristics of option data that is based on a sample of more than 25 billion records. We further demonstrate the significant incremental information provided by high-frequency option data in creating reliable intraday risk-neutral return variation measures and in extracting and examining intraday changes in the risk-neutral density following important news events, which can not be obtained using daily option data. Using high-frequency transaction and quote data for a hundred of stocks from NASDAQ, we empirically document systematic and frequent fluctuations of return serial correlation patterns over very short intraday intervals. We associate such fluctuations with price discovery in a setting with incomplete information and learning. To formally address this hypothesis, we analyze a class of parametric models for high-frequency price dynamics which accommodate the features largely ignored in classic statistical models for financial prices, such as temporal pricing error correction and microstructural noise endogeneity. The model parameters suggest a convenient interpretation for the observed return variation and serial dependence which is aligned with the market microstructure theory. Therefore, our analysis links critical concepts from high-frequency statistics and market microstructure theory. Another significant methodological contribution of this study, is a thorough characterization of the parameter identification issues which arise for the entire class of considered models under certain regimes of price/return dynamics. To the best of our knowledge, some of identification challenges are novel to the literature and had not been analyzed before. This evidence might be very useful for more accurate assessment and interpretation of the related empirical results. These findings motivate the development of new estimation methods which retain identification, and we plan to develop and explore it in ongoing and future research.