After spending a significant amount of time working on and evaluating intraday auto-trading systems, I'm losing faith that any of them are predictive.
But, let me qualify that by saying that I'm talking about indicator based strategies that use price and volume as inputs. So any combination of MACD, SMI, Aroon, Chaikin Money Flow, Accumulation/Distribution, OnBalanceVolume, MovingAverages, Ergodic, etc...
My intuition from playing with this is that there is no combination of those types of indicators that are reliably predictive. And here are my arguments:
1. All of those indicators are transforms on the base data set of price and volume. So, while they sometimes have fancy names or look different on the screen, mathematically they are all related. Therefore I argue that there are really only a handful of fundamental indicators and the rest are just variations. With some analysis we could probably identify the root indicators and just focus on those and stop considering the rest. I don't want to to do that analysis now, but intuitively I think it's correct. The point is, if there is any predictive power in this stuff, it could be found in just a few of the "fundamental" indicators. There is no exotic indicator that will be magically predictive, and no amount of searching for a magic indicator would be justified, because they are all derived from the same data.
2. There is no strong hypothesis for even "why" putting these together would be predictive. For most R&D projects, you have to have a reason to believe it can be done. The basic premise of an indicator based strategy would be "past prices and volume can predict future prices". And a related argument HAS to be, that markets are not random and have some exploitable structure. Then the question is, if markets can be exploited, can they be exploited just by looking at price and volume? And the next point is the killer to this argument...
3. If there were market structure exploitable by algorithms based on price and volume, it would already be done by the thousands of quant PhDs running exotic machine learning algos. My only counter-argument to this is that some market inefficiencies might be too small for the big firms to focus on and therefore can still be exploited by the small guy.
I know that somebody will mention trend trading (market regime filtered) which isn't intentionally trying to predict the future but instead just attempting to piggy-back onto prevailing trends. My argument is that this strategy is still trying to predict that the trend will continue based on past trends. It's like saying, the weather tomorrow will usually be like the weather today which often will statistically be true. The question is, can you generate alpha with a strategy based on this, and would it be more than just buying and holding (especially if you factor in taxes/risk)? You'd have to do +16% a year just to break even with by and hold after taxes. My intuition says no, but I'm really only basing that on the statistical failure of funds and retail traders to generate alpha.
Also, I don't want to say that psychological trading strategies can't work either. In fact, hot stocks that are being traded by manic gamblers might be one of the best areas of opportunity. There might be a small number of discretionary traders who consistently win when they compete against other retail traders. That's just not where my interests lie at the moment.
!!! SURPRISE !!!
So the above sounds defeatist, but maybe there is still hope? I have run across a few papers that show predictive power when using price, volume, and level2 (order book) information on short time scales. This is analogous to a physics based "weather" hypothesis. That with enough data and a good physics model, that you can predict the weather for short horizons of time. The more data you have, the longer your prediction horizon can theoretically be. This "order-flow" system says that we can predict based on the order book and …