Want to use this thread to discuss very general trading ideas and see what we can learn from one another... and maybe even create new knowledge in the process...
to get the ball rolling...
A short primer on systems
Systems are basically models. Whatever system you are using basically builds a model of trading... We aren't really interested in explaining how the forex market works; what we care about more is if the model has out-of-sample predictive ability...
What is meant by out-of-sample predictive ability? It is contrasted with in-sample predictive ability. A system may have very high win-rate based on backtest, but do very poorly in real-time... This happens due to over-fitting. A better way to evaluate systems will be to build the system using information from say with data from 2001 to 2010 and then evaluate it on data from 2011 to 2012... If it is profitable then it probably works...
Discretionary traders are also using models. They have constructed some mental model of the market and then applies it to output parameters like risk, expected direction, knightian uncertainty... More advanced traders will have some filtering rule in their mind to decide when to apply which model, hence further reducing knightian uncertainty/ model uncertainty.
Most systems fail because they are pluck out of thin air... someone then 'validates' it with 100 days of historical data and declares it a good system. That's nonsense... But even if it works, traders still shoot themselves in the foot by using too much leverage, over trading etc. etc.
The fact that EAs don't usually work is because the market is just too damn hard to predict... To beat the freaking market, high frequency traders have to apply complex genetic algorithms and machine learning techniques to build systems... The computing power needed to create a model based on historical data is INTENSE, let alone if you were trying to build, select and use the best model in REAL TIME...
Humans can do better because our brain is more powerful than any computer out there.
- Neuron switching time ~ 0.001 second; human neuron are pretty darn slow compared to transistors which can turn on and off at 10^-9 of a second.
- Number of neurons ~ 10^10; we now have more transistors than we have neurons in the brain.
- Connections per neutron ~ 10^4; connections are very expensive in computers, so around 10. Computers are obliterated here.
- Scene recognition time ~ .1 second, around 100 inference steps...
In other words, the human brain manages to recognise a scene with only 100 steps. We don’t know how to write such a simple program. The answer is that there’s a lot of parallel processing going on in the brain.
So there's no question that the best trader in the world probably beats a computer at trading, because he is constantly evolving his trading systems based on evaluation of market condition and how his systems are performing..
to get the ball rolling...
A short primer on systems
Systems are basically models. Whatever system you are using basically builds a model of trading... We aren't really interested in explaining how the forex market works; what we care about more is if the model has out-of-sample predictive ability...
What is meant by out-of-sample predictive ability? It is contrasted with in-sample predictive ability. A system may have very high win-rate based on backtest, but do very poorly in real-time... This happens due to over-fitting. A better way to evaluate systems will be to build the system using information from say with data from 2001 to 2010 and then evaluate it on data from 2011 to 2012... If it is profitable then it probably works...
Discretionary traders are also using models. They have constructed some mental model of the market and then applies it to output parameters like risk, expected direction, knightian uncertainty... More advanced traders will have some filtering rule in their mind to decide when to apply which model, hence further reducing knightian uncertainty/ model uncertainty.
Most systems fail because they are pluck out of thin air... someone then 'validates' it with 100 days of historical data and declares it a good system. That's nonsense... But even if it works, traders still shoot themselves in the foot by using too much leverage, over trading etc. etc.
The fact that EAs don't usually work is because the market is just too damn hard to predict... To beat the freaking market, high frequency traders have to apply complex genetic algorithms and machine learning techniques to build systems... The computing power needed to create a model based on historical data is INTENSE, let alone if you were trying to build, select and use the best model in REAL TIME...
Humans can do better because our brain is more powerful than any computer out there.
- Neuron switching time ~ 0.001 second; human neuron are pretty darn slow compared to transistors which can turn on and off at 10^-9 of a second.
- Number of neurons ~ 10^10; we now have more transistors than we have neurons in the brain.
- Connections per neutron ~ 10^4; connections are very expensive in computers, so around 10. Computers are obliterated here.
- Scene recognition time ~ .1 second, around 100 inference steps...
In other words, the human brain manages to recognise a scene with only 100 steps. We don’t know how to write such a simple program. The answer is that there’s a lot of parallel processing going on in the brain.
So there's no question that the best trader in the world probably beats a computer at trading, because he is constantly evolving his trading systems based on evaluation of market condition and how his systems are performing..
Build good relationships with others.
