Information theory, Communication theory in trading ?
- Thread starter jacksmith
- Start date
Same answer we gave to your question about DSP ("Digital Signal processing in trading?"), putz! You need to put a little communication theory to work and read the trading classics. It ain't changed since the first ideas were committed to print over a century ago. Everything has long since been revealed. You just have to BELIEVE it! What part of the "mechanistic fallacy" didn't you get?
Anybody wanna guess the next question?
"Has anyone used XYZ theory in trading?"
I vote for XYZ = "quantum chromodynamic theory."
Feel free to contribute your own guesses.
"Has anyone used XYZ theory in trading?"
I vote for XYZ = "quantum chromodynamic theory."
Feel free to contribute your own guesses.
Old Man think failure to use definite article (der, el, il, le, the) mean you not occidental. Thread here
http://www.elitetrader.com/vb/showthread.php?s=&threadid=154220
tell why be accidental you get response. Crean up act and tly again.
http://www.elitetrader.com/vb/showthread.php?s=&threadid=154220
tell why be accidental you get response. Crean up act and tly again.
Yes of course people use concepts from communication theory in trading. Consider it a "problem in estimation" and consult the works of Norbert Wiener et al for inspiration.
I find that such things do provide an edge BUT should be considered crude tools for ugly problems. Wield them like a meat-axe, not a scalpel.
Be prepared to fail (i.e. find no exploitable edge) in your first 40 or 50 investigations. If it were easy, professors and grad students would already be doing it.
I find that such things do provide an edge BUT should be considered crude tools for ugly problems. Wield them like a meat-axe, not a scalpel.
Be prepared to fail (i.e. find no exploitable edge) in your first 40 or 50 investigations. If it were easy, professors and grad students would already be doing it.
Estimation and signal detection in the presence of noise works well with the underlying assumption that the channel is STATIONARY. That is, the stochastic processes do not migrate over time.
Optimal channel equalization (matched filtering), maximum likelihood detection, and sophisticated coding are the dominant themes of communication theory and optimal receiver design. They all hold water in application because the intended engineering systems operate in reasonably stationary stochastic environments.
The same cannot be said of financial markets. Hence, the famous drift and volatility coefficients used in the PDEs that model financial time series are always annotated as time-varying.
IMO, these are two completely separate worlds with little overlap.
rt
Optimal channel equalization (matched filtering), maximum likelihood detection, and sophisticated coding are the dominant themes of communication theory and optimal receiver design. They all hold water in application because the intended engineering systems operate in reasonably stationary stochastic environments.
The same cannot be said of financial markets. Hence, the famous drift and volatility coefficients used in the PDEs that model financial time series are always annotated as time-varying.
IMO, these are two completely separate worlds with little overlap.
rt
adaptive filters are used primarily as equalizers (matched filters) to provide the detector with optimal signal to noise ratio.
The tap weights of the filter adapt in real time to provide the "inverse" of the band-limited channel to the extent the length of the filter allows. It's usually an accuracy vs. hardware efficiency tradeoff.
From my days in engineering, the key assumption that facilitates system design is that the system will be 1) linear, and 2) time-invariant. 2) is relevant in this case. Without time time-invariance, the frequency response of the channel will be constantly drifting, and the filter will never adapt to a stable matched configuration. The signal detection will be absolutely crummy, even when sophisticated coding has been applied to the signal.
In wireless, the channel is not isolated, as is the case with certain other applications. Problems such as high-power interferers, adjacent to the small, desired signal in the spectrum pose "receiver sensitivity" challenges. In addition, cellular receivers face the problem of multipath, where a radio signal can propagate through multiple geometric paths, resulting in time-delayed echoes on the receiver-side. However, these application-specific challenges never change the fact that a transmitted radio signal is ALWAYS received as a sinusoid, and the spectral characteristics of wireless channels are well-known.
This is NOT the case of financial time series, where the power spectrum is constantly shifting in time due to changes in volatility, drift value, and also, very importantly, poisson events.
rt