Here's What Happened on Thursday

Sunday, May 09, 2010

The trading of securities is a complete abstraction of the real world. Financial instruments place value on what is real and then trade between buyers and sellers. The inefficiency in finance occurs, will always occur, and sometimes dramatically occurs when financial instruments becomes divorced from the value of the underlying asset. Thursday's stock market crash was just such an event. Financial assets traded relative to other financial assets, not the actual value of the underlying.

The structure of our markets has changed over time. My friends, Joe Saluzzi and Sal Arnuk, over at Themis Trading summed it up well:
Today, the human specialist model has been replaced by an automated market maker model. Our market structure has evolved. It has evolved, not by design,
or a well-thought and reasoned plan, but it has evolved to cater to masters of expensive technology, deployed unfettered by participants whose only concern is to squeeze out every last picosecond and fractional cent before they move on to other countries’ markets and asset classes.
The New York Stock Exchange, in days of old, had a human specialist at the trading post for every traded stock on the exchange. These specialists were required to maintain an orderly market in the trading of shares throughout the day matching buying and selling orders. Should the sell-side or buy-side become overwhelmed in an individual company, the specialist had the obligation to take the other side of the trade, providing liquidity.

Now, stocks are maintained by designated market makers (DDMs). Each stock now has an automated DDM providing liquidity to the market. These DDMs fall under the heading of high frequency trading (HFT) strategies which now entirely dominate the equity markets. These strategies are enacted through "black boxes". Programmers write algorithms instructing a computer to buy and sell based on a pre-defined set of rules. HFT is estimated to produce approximately 70% of the equity market volume. These strategies claim to provide much-needed liquidity to the market.

One of the problems with these quant-based strategies is that they will always be victims of tail events. They are based on probabilities yet finance has consistently underestimated these tails. These tails are typically much fatter than financial statisticians model them to be. The market failed Thursday largely because of a structural issue. The market is dominated by algorithmic trading now as I have written about in the past (here and here and here). The problem, as we saw on Thursday, is that these strategies provide volume, not liquidity. The specialist model provided liquidity because it was there in times of stress. HFT is not there in times of stress.

It now seems consensus on the Street that the "fat finger" story is probably erroneous. I posted on Thursday, perhaps irresponsibly, about the possibility of a fat finger trade in PG triggering the sell-off. Now, that much of the dust has cleared, here's my estimation of what happened.

A lot of sell-side paper (institutional sellers) came into the S&P futures pit in Chicago. The market started to fall somewhat aggressively. The NYSE, seeing a huge sell-side imbalance, switched over to a "slow market" attempting to find buyers to match the selling orders rather than just knocking the market down. Unable to receive an immediate fill on these orders routed through NYSE, the algorithms quickly cancelled their orders and routed them through other electronic communication networks (ECNs). These other ECNs though, having lower liquidity, could not handle the influx of selling orders and many individual stocks and ETFs rapidly declined because of the lack of bids (limit order buys) on the book.

Add to this unforeseen problem between NYSE and ECNs a massive wave of selling from HFT trying to short with the downward momentum. Other HFT firms which have strategies typically working to keep spreads tight began to lose money as the tail event started occurring. Many of these HFT firms actually shutdown their computers during the crash. Tradebot Systems and Tradeworx are two examples of firms just choosing to shut off the computers completely in the midst of the storm. Tradebot often represents 5% of the equity market volume on a given day. So HFT, heralded for its liquidity-providing activity in the market, had several providers simply turn off when liquidity was needed most mixed with those that joined the selling only exacerbating the collapse.

The market essentially traded into a vacuum. There were no bids on the way down and then subsequently there were no offers on the way up. So, in extremely dramatic fashion, the market lost and regained nearly 700 points in just minutes. Here's a good note from Paul Kedrosky's Infectious Greed:
I wondered just how much money was made and lost during the glitch, so I looked at all PG trades done below $50. It was only 205,000 shares at an average price of 46.5, just about $15 below where P&G traded before the glitch. So the total lost in outrageous trades was at most $3 million or so, a drop in the bucket. (Broadening the range to $55 increases this total only modestly to about $4.4 million.)
Some major selling and then no bids whatsoever followed by no offers whatsoever, that was the story. Clearly, the NYSE didn't understand the possible consequences of initiating a slow market and DDMs horribly failed to maintain orderly markets. Other general HFT failed to provide liquidity as well, joining the momentum or simply turning off when needed most. This is a significant structural problem in the equity market.

Now, let's not fool ourselves into thinking that the selling is unjustified. After a non-stop 70% rally off the March 2009 lows seeing only two 9% pull-ins, the market is due for a correction. Sovereign debt problems in Europe are creating significant fears of contagion across the globe. The problem with Thursday is how it damages investor confidence and crushes any retail investor using stop-loss orders.
blog comments powered by Disqus