Crashes and High Frequency Trading: An evaluation of risks posed by high-speed algorithmic trading

Abstract

We present a partial review of the potential for bubbles and crashes associated with high frequency trading (HFT). Our analysis intends to complement still inconclusive academic literature on this topic by drawing upon both conceptual frameworks and indicative evidence observed in the markets. A generic classification in terms of Barenblatt’s theory of similarity is proposed that suggests, given the available empirical evidence, that HFT has profound consequences for the organization and time dynamics of market prices. Provided one accepts the evidence that financial stock returns exhibit multifractal properties, it is likely that HFT time scales and the associated structures and dynamics do significantly affect the overall organization of markets. A significant scenario of Barenblatt’s classification is called “non-renormalizable”, which corresponds to HFT functioning essentially as an accelerator to previous market dynamics such as bubbles and crashes. New features can also be expected to occur, truly innovative properties that were not present before. This scenario is particularly important to investigate for risk management purposes. This report thus suggests a largely positive answer to the question: “Can high frequency trading lead to crashes?” We believe it has in the past, and it can be expected to do so more and more in the future. Flash crashes are not fundamentally a new phenomenon, in that they do exhibit strong similarities with previous crashes, albeit with different specifics and of course time scales. As a consequence of the increasing inter-dependences between various financial instruments and asset classes, one can expect in the future more flash crashes involving additional markets and instruments. The technological race is not expected to provide a stabilization effect, overall. This is mainly due to the crowding of adaptive strategies that are pro-cyclical, and no level of technology can change this basic fact, which is widely documented for instance in numerical simulations of agent-based models of financial markets. New “crash algorithms” will likely be developed to trade during periods of market stresses in order to profit from these periods. Finally, we argue that flash crashes could be partly mitigated if the central question of the economic gains (and losses) provided by HFT was considered seriously. We question in particular the argument that HFT provides liquidity and suggest that the welfare gains derived from HFT are minimal and perhaps even largely negative on a long-term investment horizon. This question at least warrants serious considerations especially on an empirical basis. As a consequence, regulations and tax incentives constitute the standard tools of policy makers at their disposal within an economic context to maximize global welfare (in contrast with private welfare of certain players who promote HFT for their private gains). We believe that a complex systems approach to future research can provide important and necessary insights for both academics and policy makers.