Neural Networks Revisited

[Jerry030]

Conventional wisdom says it was because too many people were using it. I remain skeptical. There are other possibilities.

But unlike RSI, we're talking algos that can adapt.

When two chess programs that are the same play against each other at the same strength, do they draw every time. Not at all...I least not when I did this experiment as a teen.

There are to many unknowns. Algos controlling more bankroll many adopt different strategies than Mom and Pop's algo. Etc.

Keep in mine we are talking about the most advanced--hidden and known--algos out there.

And don't forget about the concept of the self-fulfilling prophesy. If everyone has the same algo. And all of them buy at the same time. Wouldn't they cause the market to rise, and therefore be correct in being long.

Chaos.

Now that you mention it there would be two effects. In stocks everyone agreeing would cause prices to rise and in time create a huge bubble...which would after a bit burst like housing in 2008. In zero sum markets like forex and futures if everyone is going long the models run by the big players with billions to play with would likely discover a strategy with a huge shorting that would cause the optimistic crowd to have their stops hit and stop loss selling would further depress prices and make the shorts very rich very fast.

 

[userque]

Now that you mention it there would be two effects. In stocks everyone agreeing would cause prices to rise and in time create a huge bubble...which would after a bit burst like housing in 2008. In zero sum markets like forex and futures if everyone is going long the models run by the big players with billions to play with would likely discover a strategy with a huge shorting that would cause the optimistic crowd to have their stops hit and stop loss selling would further depress prices and make the shorts very rich very fast.

Maybe it'll be like when the unstoppable force meets the immovable object.

 

[maninmoon]

Maybe it'll be like when the unstoppable force meets the immovable object.

Are you running your system across a range of Stocks?

 

[userque]

Are you running your system across a range of Stocks?

No. Just GDX (for NUGT/DUST) and SP500 (for Retirement Account). Just those two require many hours per trading day to run.

 

[maninmoon]

No. Just GDX (for NUGT/DUST) and SP500 (for Retirement Account). Just those two require many hours per trading day to run.

OK thank you. One issue I grapple with is that it is relatively "easy" to create a model (simple heuristic - not NN) for a particular Stock. When I try to apply the same model to a range of Stocks it often does poorly. I usually assume that the Stock it worked for was just fitted. Do you have any concerns/thoughts on this?

 

[userque]

OK thank you. One issue I grapple with is that it is relatively "easy" to create a model (simple heuristic - not NN) for a particular Stock. When I try to apply the same model to a range of Stocks it often does poorly. I usually assume that the Stock it worked for was just fitted. Do you have any concerns/thoughts on this?

No concerns. The kNN uses historical data, not 'black box' NN computations. The system also does internal backtesting everyday on an ensemble of 'individual' kNN's. (NN do, however, process their outputs.)

The fact that my kNN's use actual historical data, rather than 'computed' data (NN's), is the reason, I believe, it will always out perform NN's and the like. NN's etc. are trying to make order from the chaos. KNN's mearly present historical chaos as the forecast for future chaos.

Curve fitting is another thing I believe is misunderstood by the masses.

My models embrace curve fitting. Curve fitting works if done on the 'quantum' level, rather than the 'macro' level, so to speak.

 

[Simples]

No concerns. The kNN uses historical data, not 'black box' NN computations. The system also does internal backtesting everyday on an ensemble of 'individual' kNN's. (NN do, however, process their outputs.)

The fact that my kNN's use actual historical data, rather than 'computed' data (NN's), is the reason, I believe, it will always out perform NN's and the like. NN's etc. are trying to make order from the chaos. KNN's mearly present historical chaos as the forecast for future chaos.

Curve fitting is another thing I believe is misunderstood by the masses.

My models embrace curve fitting. Curve fitting works if done on the 'quantum' level, rather than the 'macro' level, so to speak.

Heresy! Thou shallt neva curve fit!

Joking aside, curve fitting can be a good tool if you know the function of your fitted curve

Would you call your implementation adaptive or optimizing, and on what type of "period" (number of bars, 1/2 historical data, all historical data, etc.) is such based on? This could mean something for the "lag factor" (adaptiveness to changing market conditions), but can also mean statistical significance for that specific instrument / time series (strategy/tactic optimizing).

 

[lucysparabola]

Good luck, and keep up posted!

thanks - supposed to have a beta on Mon/Tues. pass or fail, i'll update here for the good of science lol.

 

[userque]

Heresy! Thou shallt neva curve fit!

Joking aside, curve fitting can be a good tool if you know the function of your fitted curve

Right, but in the case of financial markets, we don't know the function. Yet, I find curve fitting crucial to my system.

Would you call your implementation adaptive or optimizing,

You forgot one: Adaptive and Optimizing. I call it both.

...and on what type of "period" (number of bars, 1/2 historical data, all historical data, etc.) is such based on?

No set number. The system decides.

This could mean something for the "lag factor" (adaptiveness to changing market conditions), but can also mean statistical significance for that specific instrument / time series (strategy/tactic optimizing).

If I understand you correctly:

There's no lag. For example: If you wait for the MA cross-over; you lag. But if your signal is to wait for when the MA acts as though it will cross-over; you don't lag.

The first thing I have to do when trading an instrument is to build what I call 'the matrix.' The system essentially goes back in time and generates output from all of the kNN's in the ensemble. This is the data that is updated daily and, will be fed into the NN.

The system 'tunes' itself to the instrument. It has no preconceived notions.

Also, it doesn't use any traditional indicators. Only my custom indicators...price action. It also is aware of the dates for each bar.

 

[Simples]

Right, but in the case of financial markets, we don't know the function. Yet, I find curve fitting crucial to my system.



You forgot one: Adaptive and Optimizing. I call it both.



No set number. The system decides.



If I understand you correctly:

There's no lag. For example: If you wait for the MA cross-over; you lag. But if your signal is to wait for when the MA acts as though it will cross-over; you don't lag.

The first thing I have to do when trading an instrument is to build what I call 'the matrix.' The system essentially goes back in time and generates output from all of the kNN's in the ensemble. This is the data that is updated daily and, will be fed into the NN.

The system 'tunes' itself to the instrument. It has no preconceived notions.

Also, it doesn't use any traditional indicators. Only my custom indicators...price action. It also is aware of the dates for each bar.

Yes, but I guess the system uses all historical price data in order to "tune in" to price action. That means if those historical prices become obsolete because of some sort of change in the instrument / time series / market, the system could become "falsely tuned" or lagging to some degree? What happened 10 years ago might never be applicable now or in the future, but what is happening the last 3 months could be key.

Back to "Machine Learning". I admittedly don't have much experience with state of the art ML, but the concept itself seems to have a bad name. Machines don't really "learn", at least not in the sense humans do, yet. They can take data input, process that and spit it to output, something akin to what humans also seem to. But the processing itself is usually not very "deep" and lack broader understanding of context and awareness (though, this is true for humans as well!). What can be done, is using pattern recognition, statistics, weighting, etc. to search for solutions to questions / problems. It might seem to be "learning", but really, is just a slightly more advanced form of data transformation. One key ingredient to human learning, is to be able to "unlearn", to discard falsifiable knowledge. Probably there are other key ingredients which can provide more intelligent capabilities to machines, or at least tools that can be used to reach more flexible results when facing pure CHAOS .