Guest Post: Why Did The “Flash Crash” Occur?
Submitted by Vahid R. Riasati, Ph.D.
Why did the “Flash Crash” occur?
Often and in many conversations the reason is side-stepped via “Perfect Storm” arguments. This warrant’s a brief discussion. One point of view is based on the economic and market factors and behavior which is the basis of many opinions. This involves deductive reasoning that can often be traced back to the deductive reasoner ’s expertise: ‘…we see the world as we are not as it is…’. Looking at the print that goes-by et cetera… The fact is HFT does not rely on traditional events and sector trends and performances, except in very sophisticated systems and event then in secondary considerations, also, most HFT programs do not currently incorporate emotions as decision cycles are reviewed; hence, emotions, sector trends, and traditional events arguments become mute.
HFT algorithms often rely on some simple technical indicators. They can be triggered by a variety of events, but generally involve normalized price volume relationships; like velocities and accelerations over these independent variables, so one would observe price per volume and price per volume squared. The important factor that should be the focus of most discussions is not the particular reason the HFT algorithms send huge orders into the market via flashes of time that incorporate these order volumes in tens or hundreds of microseconds, rather, the issue that should be discussed is how these orders are processed and filled? Recently, at the, IQPC’s “Next Generation Algorithmic Trading Strategies Summit,” I witnessed a flurry of discussion on the topic of “Flash Crash” and the desire to prevent this-type of event from ever occurring again: for various apparent reasons. I found much of the discussion addressed many related themes relevant to the “Flash Crash,” and provided an anecdote that may be interesting to a broader audience.
I began my career in Optics with Professor Richard Fork who held the world record at the time on the shortest optical pulse ever invented: 15 femto-Seconds. Dr. Fork’s light pulse occurred so fast that every molecular dynamic illuminated by it would look frozen. Hundreds or even thousands of these pulses could illuminate a microscopic scene and provide a static picture. They occurred so fast that the dynamics of the microscopic scene did not change prior to the completion of light pulse train in the medium. This is essentially what occurred during the “Flash Crash.” The Brownian motion that describes the price of stocks essentially froze in the perspective of the fast flashes of orders that encompassed the market and the particular stocks’ dynamic behavior during the “illumination.” The correlation of the process with itself reached unity and the equilibrium status of the floating stock price was breached. This circumstance could have occurred under a variety of different scenarios and one can expect that due to the principle of large numbers, it will occur more often than is expected with normal statistics.
Essentially, because the number of orders in the market have increased these so called perfect storms, rare in pre-HFT days, will occur with more ‘regularity.’ However, from my vantage point, I believe the solution to this problem is available and rests in fragmentation of orders and the filling of these orders through unobservable pools of demand that are undetectable by the market. A large order is often used by the HFT as a signal to trade further. The HFT is often ignorant about the reason the large order is occurring it merely measures its attributes and initiates a new order due to its physical occurrence. There exists well-established techniques that segment well-characterized signals and fragment them into other signals that are small in magnitude when detected at the receiver at any instant in time, and sum to a miniscule number, simply by the way they are arranged and characterized relative to each other. This keeps any un-intended receiver/’predatory program’ from detecting and monitoring the orders/signals. All this occurs with negligible effects relative to the user and her intended receiver: the investor and the market that fills the order.
What was really interesting to me was that the Markets/ Wall-Street already appear to have a system in place that can enable the spreading of the order signals. This system was described earlier in the IQPC conference in a talk given by Dr. Robert Barnes, Managing Director of UBS Investment Bank, to be published next month in “The Handbook of World Stock, Derivative & Commodity Exchanges.” Dr. Barnes and others can develop methodology that fragments the orders, the key here is that the fragmentation must be performed in appropriate settings and distinct methods focused to enable the signals’ un-delectability. Under appropriate construction, Dr. Barnes and associates will in fact render the issue of “Flash Crash” events mute by removing the factors that caused its development. The law of large numbers will essentially be segmented into many venues, or alternatively, the short time associated with the large order events is segmented into spatial dimensions that encompass their magnitudes and render their time behaviors undetectable by the HFT algorithms.
Many algorithm designers may dislike this approach, as it will make the algorithm designer’s life much more difficult. However, I would point out that events such as the “Flash Crash,” could undermine the markets and take away their efficiency and believe this should be guarded against as much as possible.