Towards better protection from superiority by increased precision in detection: The case of financial markets with informed and high-frequency trading

This study progresses towards the improvements and developments of measurement methodologies for informed trading and high-frequency trading (HFT) in financial markets. First chapter suggests an unbiased computation methodology for the broadly implemented PIN (Probability of Informed Trading) (Easley et al., 1996) measure. Multiple initial parameter sets are strategically determined by a developed clustering algorithm. Besides, computing time is reduced to one seventeenth. In the next chapter, a new measure, Multilayer Probability of Informed Trading (MPIN) is proposed. This rests on demonstrating that only 3.59% of empirical datasets have single information type as assumed by PIN model. Number of information layers in the model is not constant but exogenously determined by a clustering algorithm searching for separate Skellam distributions in daily numbers of absolute order imbalances. Respective solutions are provided for extended computational problems. While PIN estimates are heavily biased for simulations with multiple layers, MPIN estimates are bias-free regardless of number of layers. Mean quarterly MPIN (PIN) estimate is 31.9% (23.7%) for 361 stocks listed in Borsa Istanbul (BIST) between Q1 2007 and Q1 2014. Third chapter measures the extent of HFT in BIST by examination of 243 million messages in 85 million orders. Initially, the methodology suggested by Hasbrouck and Saar (2013) is applied. Next, an extended methodology is proposed and implemented. This is based on the linkage of same-sized orders of a stock if they have messages (submission, modification or cancellation) arriving within a low latency of a second. While this measure attributes around 6% of all orders to HFT, this rate is twofold for large orders and portfolio/fund firm orders.