Let me add one more category, market making or pseudo-market making, which while not available most of the retail traders, have worked. My strategies purely automated. It is about 65% pseudo-market making and 35% trend following. At a high level, my system mostly put on a pair of orders (buy and sell) based theorical prices, and if it happens to catch a trend, the system will let the trend to ride itself out (hedge the exposure if necessary). Obviously being MM-variant strategy, the system would trade very often. My system's per trade profitability is very little.
Stepher> Only one timeframe calculates the mean. Rufus> Thats interesting. Sounds like you do a lot of stat arb. Is it mostly Level 2 screening for figuring out possible misprices? Or do you do pairs trading at a micro timeframe?
I do neither. In the purest sense, I statistically model the market micro-structure (the book, the fills, etc), combine the model with an automatically adjusted theoretical valuation engine. Pure stat arb is not something with my level capitalization can afford. I used to have have a stat arb program running when I was with the hedge fund, but it took up fairly large a mount of capital. I also do very little pairs trading, only using external instruments to hedge out certain undesirable exposures. The core of the system is nothing overly exciting, just your plain vanilla market making algorithm tuned so that I am not put on direct two sided markets (so that I am not deemed a MM in the exchange's eyes), based from a theoretical value. Once I take a favorable position, then the position is only exited when the statistcal properties associated with the "trend" (it is not a trend, obviously, since I don't use TA) disappears. My strategy is not very new nor exciting, it is just a combination of MM and market behavior analysis. So naturally, I have do a lot of volume in order to generate decent returns.
Which timeframe? I've read papers that suggest there are several distinct timescales where most structure exists and that the relative weight of each timescale itself varies over time.
I personally use hourly timeframes for trading the model on the currency pairs I follow. I could go into more micro timeframes however the spread would become a real issue then. These papers sound interesting. Any links where to get them?
Thanks for the detailed explanation. Sounds like a good strategy. So basically you try to remain on the correct side of the theoretical market value and indirectly use trend components to exit? Pretty interesting stuff.
To take profit, yes, more or less that is the philosophy. The market making side of the strategy is just so that I have a mechanism to consistently get good fills, being that I basically make a few ticks on the profitable trades, good fills would make a decent amount of difference. Of course, using a MM strategy means that I will get a lot of fills, and therefore volume.