Link : Journal of Finance

There are many microstructure models (asymmetric information models, inventory-control models)  that use BUY or SELL indicator associated with a trade as a variable for classifying other variables or use it as an exogenous variable for modeling. But the thing is that one needs to infer this variable from the trades and quotes data. The data feed from any exchange contains trades but one never knows whether the trade was in response to BUY order or a SELL order.

This paper contains what is popularly referred to as ,”Lee-Ready algorithm” for inferring trade direction. Why is it such a big deal? Can one not just look at the bid and ask quote near the time of trade and guess based on it? The paper starts off by explaining two specific issues with such a straightforward approach. They are

  1. Quotes are often recorded ahead of the trade that triggered them.

  2. Trades happen inside the bid ask spread.

The first issue is exchange specific issue. It is possible that in some exchanges such a thing never happens. However in the context of the paper where the authors analyze NYSE data,a hybrid market, there is an evidence of quotes being recorded ahead of the trade that triggered them. The second issue, though, is bound to come up in any sort of exchange data.

To address the first issue, the authors look at the empirical distribution of quote changes for a rather isolated trade across various trading days and infer that a 5 second time-delayed quote is a better quote than the current quote for a trade. The authors define something called “tick test”, which basically says that a trade is a BUY trade if it happens on a uptick or zero uptick and a SELL trade if it happens on a downtick or zero downtick. Using the time-adjusted quote and the tick test, the paper shows evidence of inferring trade direction.

To address the second issue, the authors suggest a combination of quote rule and tick rule. What’s quote rule ? Any trade that happens close to bid is a SELL trade and any trade that happens close to Offer is a SELL trade. For trades that are close to midpoint of bid and ask, the authors suggest tick rule, i.e. uptick or zero uptick is a BUY trade, downtick or zero downtick is a SELL trade.

So, the takeaway from this paper is an understanding of the widely used Lee-Ready algorithm for assigning trade direction to intraday data :

  • Time adjust the quotes if you see that quotes are recorded ahead of the trade that triggered them. The exact time adjustment to make depends on case to case basis.

  • Use a combination of quote rule and tick rule to assign trades that happen with in the bid ask quote.

Well not everything is great about this trade classification algorithm. Here’s a paper that suggests a better model based on the analysis of NASDAQ data.