Article/Paper - "Filtering in Finance"
- Thread starter Equalizer
- Start date
huh?you have to kidding
looks like martian script to me
will that help me make money?
looks like martian script to me
will that help me make money?
FredBloggs
Guest
exactly!
why dont they just say buy when $ > last $
sell when $ < last $
Been there. Done that. Doesn't work. Those of us who REALLY know Kalman filtering tried that over 20 years ago. In fact, a correspondent from the B-Team who shall remain nameless told me Kalman himself tried it. If you wonder how he did, he still works for a living.
Preaching to the Kalman choir, the fundamental problem is that there is no measurement noise and no noise in the "signal". So you don't need a Kalman. Further, there is no way to create the required physical model because price doesn't behave "physically". If it did you could just calculate as many derivatives as you need for accuracy and project forward. Tried that, too, hahahaha! There ARE underlying market mechanics that you can model mathematically that affect price, but pure price don't cut it. Fucking mathematicians. There are days I wish I wasn't one. Just led me ass-tray for years in the markets.
Preaching to the Kalman choir, the fundamental problem is that there is no measurement noise and no noise in the "signal". So you don't need a Kalman. Further, there is no way to create the required physical model because price doesn't behave "physically". If it did you could just calculate as many derivatives as you need for accuracy and project forward. Tried that, too, hahahaha! There ARE underlying market mechanics that you can model mathematically that affect price, but pure price don't cut it. Fucking mathematicians. There are days I wish I wasn't one. Just led me ass-tray for years in the markets.
Thanks for the link.
nitro
Umm, the unscented kalman filter is definitely not linear. While they aren't all that useful for simple price prediction they can be *extremely* effective for training neural networks. I've used ekf/ukf trained neural networks in speech recognition software I wrote about 2 years ago and I must say it is much more effective than simple gradient descent and back-propagation.
Quote from kut2k2:
Written by mathematicians for other mathematicians, not for real practicioners of trading. And it's another attempt to force a square peg into a round hole: another linear solution applied to a nonlinear problem.