An idea that came to me today... Orders in the book are either placed by humans or machines. 1) Is it the case that the majority of machine placed orders are cancelled before being executed whilst the majority of human placed orders rest in the market until they are filled? 2) Can you assume that the human placed orders will have more market impact because they are more likely to be filled? 3) Can you assume that orders in the book that have been there longer have been placed by humans, whereas ones that have been there for a shorter time have been placed by machines (because machine placed orders are often cancelled quickly)? 4) Is it possible to get the age (i.e time in seconds) of each order in the book, and therefore calculate which side of the book has more human placed orders and make some assumption about price movement? Any thoughts?
I have always been amused by speculators who "race" large stacked orders away from the market when the DOM ladder appears unbalanced and biased. "Real" money doesn't cancel a big order when the price uptics towards them, and "real" money will hit a large bid or lift a large offer if the price and quantity meets their objective and they need to get them done. Furthermore, "real" money will iceberg orders and scale buy or sell.
order flow analysis is of a really, really short timeframe. and most real orders don't appear in the book any more, but there're a dozen of spoofed orders. and cancelled ones. I really doubt you can build an edge from watching orders, but good luck.
1). Yes 2). Yes but not for that reason 3). No. Most long dated orders are machines holding queue position. With major exceptions is some markets (gasoil) 4)varies by exchange. Timestamps of individual orders is useful but it's unlikely to be in your feed because either the exchange doesn't support it or your provider doesn't preserve it.
Interesting link. That guy is saying the opposite of me - human orders have less alpha and algorithms should seek them out and take the opposite side of the trade. He also suggests using the timestamp to differentiate between human orders and machines. I assumed that machines don't have any alpha - they tend to execute market making strategies rather than directional strategies, although I guess a market making strategy has to have an idea of whether the market is about to tick up or down so that it's not left with an adverse position. Maybe in some markets human orders contain more alpha than other markets. I'm thinking about commodity markets that are used by commercials rather than more popular markets (ES?) used by day traders.
It answers most of your questions, thought the implications are not quite what you had guessed. If they can identify informed "human" flow they will try to avoid it. They will pay for retail order flow, that tells you something.