The indicator called On-Balance Volume should work if adverse selection (and Akerlof's market for lemons) were true. However to my disappointment I couldn't find any predictive power in it.

Yeah, I think this assumption is ignoring a lot of other reasons people buy and sell. It could be because they have limited buying power and want to buy something else so they sell current holdings. Some people sell because they don't want overnight exposure, some buy on a garbage tip from a friend or newsletter. Big firms do "window dressing" where they're buying after the stock moved up to have the appearance they didn't miss the boat. The point is, many people buy for bad or neutral reasons not related to inside info that you just can't know about.

@Justin - I agree there are a lot of reasons why people buy and sell. I write/trade algos with very short time horizons (often in trades for 10 seconds or less) and so I am primarily concerned with the potential for follow through on a block trade.

If it is due to adverse selection, I likely want to be more aggressive about getting out whereas if it is just a large order, the trade exit logic may well be quite different.

@Cameron - what time frame were you looking at when considering if there was any predictive power in those indicators? I am only concerned about the next few seconds, a minute max. I have other mechanisms that will kick in past that time frame.

All profit oriented traders must deal with adverse selection at some point, by averaging out to a net profit (even if they dont think of it in those terms). Dealers especially constantly worry about adverse selection because they have to keep balanced, 2 sided order flows.

Not all participants are in the markets to profit (e.g. utilitarian traders). It follows that the strategy a profit-oriented trades should serve utilitarian traders. As a sub-minute HFT, i assume you average out quite quickly. Who are you serving?

Usually price discovery where anyone can be a taker requires 2 steps. Placing a trade to see the sentiment. And then covering that trade (immediate positive selection) or placing a new trade to avg out later...even a few seconds later. Mathematically that's quite aggressive!

@Jon, yes, given the short time horizon of my trades, adverse selection is an issue from me in much the same way as it is for a dealer. I am not sure if you count hedges as a utilitarian trade or not but it does qualify as one that may not have a direct profit motive.

I do average out given that I haven't detected what the algo classifies as adverse selection. There are certain tell tale clues as to whether the order is intended solely as a directional entry or exit or whether it is due to market information my algo doesn't have or track (adverse selection).

I am not sure what you mean when you ask who I am serving but in general I am serving my PnL. I have certain support/resistance levels (or setups) that I have years of statistics on and where I know the characteristics of what a good trade looks like at every stage of price movement. For those that do not meet the criteria of a good trade, the algos are very aggressive about exiting the trade.

utilitarian = people who come to the markets to solve a problem that originated outside of the markets. Usually uninformed and don't care about profit. Profit motivated = speculators/dealers

who provide a service or take on risk specifically to profit. Technology providers are profit motivated, but do not take on any risks associated with holding any inventory in any market.

I equate 'negative selection' or adverse selection in derivatives markets to a supermarket having 'loss leader' physical products. You offer products that may lose money by themselves but you will also sell products that make a large enough profit so that you are at a net gain. So with derivatives, part of price discovery is taking a position in the market and then (aggressively) covering the position later with net profit via averaging. You have to have A LOT of confidence that you will be able to net your position out. Plus the ability to adjust for both expected and UNEXPECTED volatility.

Scott, I spent years watching OBV on 5min and 10min charts. It didn't work as far as I could tell. But perhaps you might find something in 10 - 30 second bars but it would have to be an active stock for it to have any chance of working.