Dilemma in modeling idiosyncratic high-frequency noise

Discussion in 'Automated Trading' started by bluelou, Jan 5, 2009.

  1. No offense vikana but the same question would then go to quieting the noise in a car.

    What's the value of trying to figure out what is causing the noise in your car . . . simple deal with it and move forward.

    Noise should not be apparent in our charts. It causes confusion and it is a distraction. Find out what is causing it and eliminate it if you can so you can essentially eliminate the confusion and the distraction.

    Now some traders are deaf to the noise of a chart and that is understandable. They don't have any idea of what they are looking at in the first place so the idea of quieting that which they can not see is ridiculous. Lots of screen time is the only way you will ever realize that the pesky noise actually exists.

    I find it odd that seasoned traders don't think that common sense should be applied to how we build our charts. IMHO the same set of standards should apply to real life as apply to trading.
     
    #21     Jan 6, 2009
  2. I guess it's a matter of preference. I prefer to spend my time understanding underlying structure, and build a simple statistical model of the residual noise. I'm not sure it's worth focusing on the noise, but I could be wrong, of course.
     
    #22     Jan 6, 2009
  3. Acumen

    Acumen

    My guess is your idiosyncratic noise is a result of your particular “detrending” function. Change your function and you will create a new noise signature. Make your trend slower, and your noise should be the same but scaled up, make it faster and it should scale down.

    On the other hand, it could be explained by support and resistance levels. Plenty of traders use trend lines to set up support and resistance, bouncing the price back and forth in between until a large player comes in and changes the game. You could just be identifying a channel, problem is when it bucks the trend, it will likely do so in a big way. There’s no easy answer on how to trade it.
     
    #23     Jan 6, 2009
  4. Hook N. Sinker previously alluded to this question: Are you sure that your detrending algorithm is pulling out all of the signal?

    Or:
    Perhaps you started your study with a definitional preconception that wasn't entirely accurate. You may have to adjust your definition of the 'residual' as you progress. You started by likening it to noise. However, my feeling here - and I have nothing to back this up - is that if the residual is structured enough to have a 'signature', then a relationship can be drawn between the residual (or portion thereof) and the 'signal'.
     
    #24     Jan 6, 2009
  5. bluelou

    bluelou

    ProfLogic,

    You made a very good point here...
    "1. The noise exists in the chart YOU use to view a market."

    To address that issue in particular I calculate the residual on both a tick chart and a time-based chart. Since I'm trading intraday I use 4 min or smaller bars on my time-based chart.

    Though my focus is on the tick-based chart, using both the tick- and time-based charts helps to confirm that the residual has structure beyond just a few hundred ticks.

    I disregard the noise at lower frequencies b/c it's not particularly useful in this case.
     
    #25     Jan 6, 2009
  6. Oh, don't get me wrong, I absolutely do not focus on it. I elimininated it.
    Like a good mechanic I dug around until I found what caused it and dumped it like an annoying girlfriend. :D
     
    #26     Jan 6, 2009
  7. bluelou

    bluelou

    BLB,
    Yes, there is a relationship b/t the signal and the residual. And there should be, otherwise I wouldn't have much to work with. But the relationship isn't a misspecification error.
     
    #27     Jan 6, 2009
  8. Engineers and mathematicians suck at trading. Move on to something that you will be able to utilitize your skills more efficiently. Don't over analyze the markets, which are full of noise.
     
    #28     Jan 6, 2009
  9. Time based bar charts typically have volume bars displayed on the bottom. Now lets take constant volume bar charts. What if you could display "time bars" on the bottom where time was the length of time each volume bar took to complete. That would be a way to add back the time information missing from a chart showing only constant volume bars. This would represent the "noise" you are describing isolated in the time bars along the bottom. Question is, it is really noise? I use the time it takes for volume bars to complete in my strategy and I don't find that data to be noise.
     
    #29     Jan 6, 2009
  10. OK. I was hung up on your usage of the term 'noise'. Can you develop a model that uses the residual to trade across multiple markets?
     
    #30     Jan 7, 2009