Markets are bursting at their seams with financial data. Much of the data that researchers are now mining is called Level I, II, and III data; this data comprises information on orders, executions and cancellations across different price levels. Until recently, such data was scarce. Today, it is more accessible, but still little understood. The not-for-profit Big Data Finance 2015 conference taking place at NYU Courant on March 6, 2015, will present selected techniques and results of big data analytics applied to Finance.
This article briefly explores a data phenomenon. The relationship between returns and order placement and cancellations is examined and the following findings are discussed:
- Limit orders and cancellations do not significantly move prices. Instead, changes in prices significantly affect new limit order placements and cancellations.
- During the minute when prices go up (down), more limit buy (sell) orders are cancelled than placed
- The minute after an up (down) move, more limit sell (buy) orders are cancelled than placed
- No statistically-significant dynamics can be detected past these two minutes
First, a chicken and the egg question: does order placement and cancellation impact prices or do prices impact order placement and cancellations? My latest research shows that markets work both ways, but with slight caveats: market orders tend to drive prices, while prices tend to impact limit orders and their cancellations. By way of a quick refresher on order types, a market order is the order to buy or sell at the best price available at the time the order is placed. A limit order is the order to buy or sell at a specific price. The limit order is executed only when it becomes the best-priced order in the market.
At the minute-by-minute level, the arrival of limit orders or their cancellations do little to move the prices the following minute or beyond. However, price changes themselves do impact the relative changes in limit orders, yet in unexpected ways. Counter-intuitively, during minutes when the market price rises, limit buy orders tend to be cancelled and new limit sell orders tend to be placed. However, during the minute following a rise in market prices, new limit buy orders rush to the markets, while many limit sell orders are cancelled. During the minutes when the market price falls, limit sell orders are cancelled and new limit buy orders come in. Following minutes with a drop in the market price, new limit sell orders are placed and more limit buy orders are cancelled. No statistically-significant relationships can be detected before or after the change in market prices.
The dynamics on the buy and sell front are not symmetric, however. In the S&P 500 ETF (NYSE: SPY) every 1-minute 1 bps (0.01%) rise in the mid-quote price (average of best bid and best offer) is accompanied on average by net cancellations of 500 shares in limit buy orders (new limit buy order cancellations less new limit buy orders across all price levels) with 99.99% statistical confidence. Every 1-minute 1 bps drop in mid-quote price, however, is contemporaneous to net cancellations of just 26 shares in limit sell orders. No statistically-significant changes in net limit sell (buy) orders can be detected when prices rise (fall).
Following every 1-minute mid-quote price rise of 1 bps new buy limit orders exceed buy limit order cancellations by 150 shares with 97% significance over the following minute. A 1-minute drop in the mid-quote price of 1 bps induces a net addition of 20 more shares than new limit order arrivals in the following minute. No persistent relationships between net order additions, cancellations and mid-quote price movements can be detected beyond the two-minute range discussed above.
The limit order activity is persistent and sizeable on the bid (buy) side of the limit order book, and much smaller on the offer (sell) side. This suggests that most of the observed and recurrent activity at 1-minute intervals is generated mostly by long-only investors.
Furthermore, most of the observed activity is likely due to execution orders, broker or internal algo-driven, and here is why. As the price rises, the limit-order traders are competing to stay at the top of the queue, cancelling their old limit orders and placing new ones, closer to the latest market price. The observed 1-minute delay, however, suggests that much of the limit order action may potentially be driven by humans, and not lightning-speed machines, defying the common opinion of who places and then cancels the orders in today’s fast-paced markets.
Additional inferences are too numerous to mention in this article – come to the Big Data Finance Conference to learn more!
Irene Aldridge is Managing Director, Able Alpha Trading, LTD., and AbleMarkets.com, and author of High-Frequency Trading: A Practical Guide to Algorithmic Strategies and Trading Systems (2nd edition, Wiley, 2013).