[技术分析] How I predict?

greatwallmerlin

1) using Matched Pattern from 2000 stocks in last 5 years, matched the pattern of SPX over last 3 days,  found similar patterns.
2) Use all matched patterns and perform Markov Chain Monte Carlo simulation, burning-in of first 1000 simulation to reduce predict errors.
3） APPLIED different weight, most recent data with higher weight and perform Bayesian adjustment.
4) Perform Bootstrap simulation to get p values and prediction confidences.

Knowledge need: Bayesian Analyses, Linear model, Pattern analyses, MCMC simulation, Bootstrap simulation, SAS, JAVA, GRAPH, finance knowledge....

Each run cover 2000 stocks*average 30 matched patterns*10000 MCMC simulations * 10000 Bootstrap simulations.

If you have better suggestion, welcomed to suggest.

greatwallmerlin

The basic assumptions are:

1) MM will repeat their manipulations over history.
2) There are no sudden un-predicted big events happened.

greatwallmerlin

回复 2# DUH


    系统写了近4年， 最近比较完善了， 后台运行， 不需要我管了。

greatwallmerlin

大家这么快都变成秀才了， 俺还是童生呀

greatwallmerlin

这需要一定的 统计， IT,  数学， 金融知识， 外加对大盘的一些直觉， 没必要尝试。按你的理解去做就好了。

greatwallmerlin

回复 14# attm

I used daily to predict short term; Use weekly to predict mid-term.

It works for short-term, mid-term (Like I predicted oil = 40-43, GLD = 109 before using weekly pattern). Sometimes it works better, sometime it works worse, like today, 7+ points off upper bound, and Monday, 12 points off lower bound.

I did not tried long-term yet, since maximum holding time for me are half year, never more than 1 year, so I do not need long-term.

I think I provided enough information, others you need to figure out by yourself.

greatwallmerlin

回复 20# oldhorse


   马大大回归了， 多指教呀。。。。。。

greatwallmerlin

回复 22# oldhorse


    不是，保送去了北大数学系，先来美国去的康奈尔读的工程和计算机，后来不好找工作，就去nyu拿了统计，金融双学位。

greatwallmerlin

有，那说明2000和2008的情况快到了。

greatwallmerlin

回复 34# greatwallmerlin


    这样我的12核电脑就要运行1.5小时一个round，试过，没多一倍运行就增加20倍，买不起超级电脑，就将就了。

greatwallmerlin

回复 41# DUH


    是呀，林芳华呀，大牛人，当年让俺推导鼓被击打后的震动方程，我推导了一个星期，A4纸用了56页。

greatwallmerlin

回复 50# mitwkwk


    是推导，不是应用。

greatwallmerlin

回复 47# start168


  generalized linear model 不能用，任何parametric 统计方法都会把你带到无止境的穷途末路。

greatwallmerlin

回复 49# mitwkwk


是，也有不准的时候。
至于算法，肯定有优化的地方，也可以采用多机并联运行，那又是另外一个工程，现在管理自己的公司，还要管41个股票账号，没时间搞了。