Quote from goldenarm:
The problem is: how do you write a system that can identify the prevailing market tone/sentiment? This is a fairly subjective assessment that must take into account a wide variety of data (i.e., volume, volatility, VIX, TICK/TRIN, stochastics, global indicators, economic news, world events, etc. ad nauseum).
Well, I think the real question becomes, how parsimonious can you make your model while still retaining its statistical significance? Yes, I can find plenty of measurements for volatility, but I might only need one to get me 90% of the way there. If we can identify a vector of market characteristics, and identify simple metrics to determine each characteristic, we could explain the market condition.
Quote from goldenarm:
Right now the overall market tone is bearish and trading conditions are choppy and volatile.
Here you describe the market condition. All of which can be quantitatively identified, too, if you ask me. Downward sloping moving averages, VIX readings, trend to noise ratios ... all descriptors. And all pretty standard for a bear market -- fear is more volatile (and powerful) than confidence. It makes fundamental sense, then, to trade in a manner that suits this condition.
Quote from goldenarm:
A good system should know to stay away from swing and trend following systems and instead rely on short-term intraday microtrends (i.e., momentum scalping) for stocks with good volume and intraday ranges (i.e. AAPL). This type of trading will provide consistent (albeit smaller) profits regardless of the type of market you're in.
And here you describe the niche system that can trade this environment...
Quote from goldenarm:
Momentum scalping can then be supplemented by trend following systems once we find ourselves in a good trending environment
And finally, you describe yet another niche system that will work in another environment.
That is exactly my point. If we can quantitatively describe the environment, we can better identify which strategy to use!
But, I am coming around to the idea that we don't have to identify the market condition through any external program.
Imagine 10 trading systems, each designed to trade a specific niche market.
When I turn them all on for the first time, they begin paper trading. Some will win, some will lose. The ones that win consistently turn themselves on, and trade live. Eventually, when the condition of the market changes and their bias no longer aligns, they will start to lose. Eventually, in a statistically significant manner (not within normal win/loss variance, perhaps?) At which point, the system goes back to paper trading.
In this way, you don't need a separate identifier of market condition ... you let the systems self regulate.
The downside of a system like this is the potential draw-downs you take before a system recognizes that it is out of its normal variance. To avoid this, you would need a separate system whose sole purpose is to identify which niche strategy to employ...