Is it insane enough? Reminds me of when I placed an order for 2 fully operational straight jackets with "Merrill Lynch Global Equity Derivatives" on the back. Pretty reasonable price too.

the real question is: does data mining actually work?!

Clearly, data mining works, but sometimes the result is "I can't find anything".-Will Dwinnellhttp://will.dwinnell.com

QuoteOriginally posted by: erstwhileIs it insane enough? Reminds me of when I placed an order for 2 fully operational straight jackets with "Merrill Lynch Global Equity Derivatives" on the back. Pretty reasonable price too.Good for you. Are you a fan of Pauli?"We all agree that your theory is crazy. The question which divides usis whether it is crazy enough". -- Bohr to PauliPhysics quotes

so, what technique is systematically superior to data mining (not counting illegal activities).

(we) data miners say: if any behaviour has a "law of nature" behind and if you have historic (i/o) data, which represent it well, we create (approximative, predictive, interpretable, computational) models from this data automatically. The right bouquet of methods shall be derived from the structure of the problem (continuous flows, event driven, patterns and features,...). If the behaviour is influenced by the knowledge of its representing model, you better simulate?If you want to get insight in the structure of your data you might want to apply unsupervised learning methods, like clustering, Self Organised Maps (SOM),...Taking this in account a PhD thesis might be quite challenging?An overview on software :kdnuggets

QuoteOriginally posted by: exneratunrisk(we) data miners say: if any behaviour has a "law of nature" behind and if you have historic (i/o) data, which represent it well, we create (approximative, predictive, interpretable, computational) models from this data automatically. The right bouquet of methods shall be derived from the structure of the problem (continuous flows, event driven, patterns and features,...). What you say is true, on a good day. Let's be honest, though: if the relationship between predictors and target is too complex, if the historic data is too noisy or has too many missing values, if the relationships are too transitory, etc., then in many real-world cases data mining comes up with nothing of interest.-Will Dwinnellhttp://will.dwinnell.com

predictor, yes 100% correct. But you can cope with this, by cross model testing or what have you. But if you look into complex customer behaviour or analytic customer relationship modeling, you often run into "self fulfilling prophecy" situations, which questions the approach itself.This might be the challenge for dm/ml applications in quantitative finance?

predictor:sorry for the late answer, been away for a few daysIn credit scoring the state of the art techniques are all related to data mining techniques: logistic regression, neural networks, support vector machines,...these techniques are all more related to statistics than to mathematics:econometrics,...

How about automated trading? Do the current models used for algorithmic trading involve data mining techniques ? Anyone here uses GP, NN or SVM for learning from past data and forecasting ? Or is this a thing of the past ?

I've written my own packages for extensive high frequency data mining.. recurrent neural nets, unscented kalman filter, cointegration, generative topographic mappings, K-nearest neighbor, you name it.QuoteOriginally posted by: nsmHow about automated trading? Do the current models used for algorithmic trading involve data mining techniques ? Anyone here uses GP, NN or SVM for learning from past data and forecasting ? Or is this a thing of the past ?

QF laws:IB = legacyHF = fun