by kickass@road of trade I think the concept of statistical robustness in important in system developement. It all comes back to the to the assumptions you are making about your data. the problem of choosing/changing parameters is not a big issue in system developement as lutein believe, a system have to stable aross a range of parameters to be consider robust.so if a 5-20 moving average crossover works, we demand that a 4-30, or a 10-50 crossover works too..this is one of the fundamental requirement for a robust system. this minimizes the effects of optimization in the data. To most people who don't design system, that's it..., but there are many pitfalls in system developement, the process of developing system is similar to conducting a scientific experiment, your methodology determine the validity of your results. More important is containing the degrees of freedom in the system and determing the statisical significance of your results. the later uses a lot statstics such as z and t-score, bootstrap resampling, monte carlo, non-parametric tests, p-score to accout for data-mining and survivor biases.. I am just saying that is a lot of more to it then just changing parameters.. == These thoughts, analytic or synthetic, are optional for profitable trading. In the trading history, few of the big names attribute their success to this kind of elaborative knowledge. You don't think Jesse Livermore know all about this, do you? Trading is a simple matter. It's about trend, about risk, about psychology. While this synthesis thinking does not rule you out as a successful trader, it does not make you one either. -- "We have to understand the possible and positive relation between technology, culture and politics. They are not enemies. Tech is more than work and money. Culture is more than just some temples or paintings. Politics is not owned by the politicians or the media manipulation channels." ---Geert Lovink --

※ 發信站: 批踢踢實業坊(ptt.cc) ◆ From: 59.117.73.142 ※ 編輯: stasis 來自: 59.117.73.142 (03/08 00:04)