The quant guys have been discussing the 'flash crash' .
Wilmott Forums - A Wild Ride in US Market Today - Human Error or Rise of the Machines?
Quoting a couple of replies in part:
"The deeper-level problem is the interaction of large orders and HFT. To the extent that a signficant fraction of the buyers of Wadddell's flow of sell orders were HFTs, those HFTs did not absorb Waddell's stream of small sell orders as much as they immediately turned around and submitted another sell order. That is, Waddell might sell 100 contracts, wait a short time, and submit another 100 contracts. But in the presense of high levels of HFT activity, Waddell's second 100 contract sale might occur when the HFT is trying the resell Waddell's first 100 contract sale. Thus, 200 contracts of Waddell's sell order are sitting in the market (but Waddell thinks it's only responsible for 100 of those contract, the other 100 contracts are the HFT trying to resell what it bought from Waddell). Repeat this process in the presence of lots of HFTs and a long-stream of suborders and most of the selling pressure in the market might be "caused" by one large order that has accumulated and recirculated. They called this a "hot potato" effect and, at one point, there were 27,000 trades in 14 seconds but the actual net trade was only 200 contracts."
"My suspicion is that neither Waddell nor the order-splitter could not tell that it was influencing the markets and they assumed they were selling into adequate liquidity -- Waddell's algorithm saw lots of trading and assumed that trading volume implied liquidity. But the high trading volumes probably arose from HFT churning on the price volatility rather than a true depth of buyers absorbing the 75,000 contact order. In other words, Waddell thought the market had an adequate depth of buyers to absorb the 75k but, in reality, those buyers were HFT that were re-selling those contracts an instant later. "