Dynamic instabilities from information annihilation in speculative markets
Final Report Abstract
How and why stock prices move is a centuries-old question still not answered conclusively. From an economic viewpoint, prices should reflect available information and residual fluctuations should be almost unpredictable. Empirically, while price changes are indeed hard to predict, several other state variables exhibit a very long memory. Examples include the direction of market orders, which remains correlated over entire days. Price-change magnitudes (“volatility”) even exhibit positive correlations for years, as well as extremes that threaten entire economies. These “stylised facts” are similar to critical states in other complex systems, and challenge the notion of markets operating close to stable, efficient equilibria. Understanding such dynamics is of great importance for practitioners, as well as for exchanges and regulators interested in improving market efficiency and stability. In this project, we studied how the balance of opposing market forces shapes price dynamics on time scales from milliseconds to one day. More precisely, we investigated the interaction of two types of buy- and sell orders: “passive” limit orders, which represent offers for possible future trades, and “aggressive” market orders, which immediately trigger transactions. These market orders tend to impact prices: statistically, a buy (resp. sell) market order pushes the price upwards (resp. downwards). Therefore, the aggregate price impact of the market orders in a short time interval has been previously theorised to be a linear function of the imbalance of buy- and sell market orders. It was noticed, however, that this picture must be incomplete. Otherwise, the aforementioned long-ranged order-flow correlations would lead to equally correlated price changes. As a part of this project, we quantified how order-flow imbalance moves prices. We found a highly nonlinear, saturating relationship, which can be described by a universal master curve. Its scaling is largely determined by order-flow correlations. Surprisingly, we found that extreme market-order imbalance does not lead to large price-changes. To the contrary, it is observed when the price is “pinned” to a particular level by an opposing flow of limit orders. Therefore, these episodes represent a temporarily perfect coordination of market participants with different goals. Prices move only when there is sufficient balance in the local order flow. This also implies that the more surprising a trade is given the preceding order flow, the more it impacts the price. We found this behaviour on all timescales between few milliseconds and the length of an entire trading day, and for all instruments in a very diverse sample. We furthermore investigated how well our findings can be reproduced by commonly used kernel-based models. We found that the classification of trades as price-changing versus non-price-changing can explain the price impact nonlinearities and short-term dynamics to a very high degree. We also solved, several long-standing technical issues for model calibration and -testing. To this end, we also developed new spectral estimators for two- and three-point cross-correlations in time series. These methods make estimation of three-point correlations feasible even in very long time series and are potentially applicable in arbitrary contexts. In an upcoming publication, we also investigate the phenomenon from a nonlinear mechanics perspective. We found that the combination of long-range order-flow correlations and price efficiency, as well as the nonlinear price impact can be explained in a micro-founded dynamical modelling framework based on first principles. This model furthermore reveals a mechanism for the breakdown of the balance of marketand limit orders, which can lead to extreme price changes even on the daily scale. In conclusion, our results provide substantial new insights into the dynamics of financial markets. We revealed how the adaptation of market participants to the predictable behaviour of others can shape price dynamics on all intra-day timescales. In addition to these empirical results, we contributed to the literature on models used by practitioners. Our results have implications for academic research, trading institutions, market operators, and regulators alike.
Publications
- Balancing information - from sticks to speculative markets. In Market Microstructure Paris, 6 2016
Felix Patzelt
- Nonlinear price impact from linear models. Journal of Statistical Mechanics: Theory and Experiment., 12:123404., 2017
Felix Patzelt and Jean-Philippe Bouchaud
(See online at https://doi.org/10.1088/1742-5468/aa9335) - Universal nonlinear features of intra-day price impact. In Market Microstructure London, 2017
Felix Patzelt
- Universal scaling and nonlinearity of aggregate price impact in financial markets. Physical Review E., 97(1):012304, 2018
Felix Patzelt and Jean-Philippe Bouchaud
(See online at https://doi.org/10.1103/PhysRevE.97.012304)
