b***k 发帖数: 2673 | 1 ☆─────────────────────────────────────☆
QuantHR (quantile) 于 (Tue Aug 21 14:39:52 2007) 提到:
谁了解吗?就业行情,老师质量等,我看课程安排还可以。
http://www.msfinance.ch
多谢。
☆─────────────────────────────────────☆
xvgsfx (pains) 于 (Wed Aug 22 05:14:23 2007) 提到:
估计只能考虑当地的需求情况了
这个地点挺好的,学校再当地也好
剩下的就是当地业界的需求
去欧洲的华人论坛,瑞士论坛看看把
☆─────────────────────────────────────☆
CavHawk (GO|!) 于 (Wed Aug 22 06:47:36 2007) 提到:
非常好,号称是euro的top1,gs等会去招聘。
需要去eth参加面试,会问sde的问题。
但是网站上面列出的一些过去的学生毕业后的工作,感觉一般。
☆─────────────────────── |
|
b***k 发帖数: 2673 | 2 ☆─────────────────────────────────────☆
iambear (iambear) 于 (Tue Nov 27 23:52:39 2007) 提到:
我用的是matlab,用的第三方软件EVIM,和splus那个EVIS一样的东西.
但是在用那个gpd的时候,老是出error msg.换一下threshold就出问题,比如0.02就好的
,换0.15就error. 或者nextreme换一下就出问题
因为我的是return,所以就先*-1变成loss,然后threshold也加个负号,这应该都不是关
键,error msg显示是optimization中出问题,改天可以贴上来.
☆─────────────────────────────────────☆
QuantHR (quantile) 于 (Thu Nov 29 08:42:11 2007) 提到:
你把threshold设置过高,你的数据不够去拟合GDP ML,所以optimization出了问题,
检查一下你的return是不是小于0.15的很少。
☆───────── |
|
b***k 发帖数: 2673 | 3 ☆─────────────────────────────────────☆
sbtim (15#) 于 (Wed Sep 5 10:41:11 2007) 提到:
因为一个有限的采样数据中存在outlier,
为了得到更为可靠的distribution的location, scale估计值,
想采用bootstrap方法,
只是不确定这样产生的simulation samples, 会
不会使得估计出的statistics (mean, median, variation, quantiles)
与初始sample得到的statistics估计值有很大偏差?
请大家不吝赐教。
多谢了先。
☆─────────────────────────────────────☆
arnaud (Prrrrrf) 于 (Wed Sep 5 11:18:51 2007) 提到:
bootstrap方法要在现有样本基础上重新抽样,这个resampling过程有很多研究,根据数
据特征,可以分别采用传统bootstrap(iid),moving blocks boot(depe |
|
b***k 发帖数: 2673 | 4 ☆─────────────────────────────────────☆
wandafish (A fish names Wanda) 于 (Mon Mar 31 11:43:11 2008) 提到:
刚拿到国内一个offer, 公司是挺好的,一家大基金公司。 基本工资只有RMB 20k/
month, 我在国外工作了差不多6年了。这个工资也太低了点。人事部的说,奖金很高啊
,我们这里流动率很低啊。。。有谁知道国内情况的,给个大概情况吧。
我的情况: 首先我不是做投资的,所以大家听说的天文数字的bonus 跟我没有关系。
但具体别的职位的 bonus是怎么样的?
☆─────────────────────────────────────☆
jelty (who am I?) 于 (Mon Mar 31 12:49:01 2008) 提到:
太低,把你当学生了
☆─────────────────────────────────────☆
QuantHR (quantile) 于 (Mon Mar 31 14:55:22 2008) 提到:
具体要 |
|
l******i 发帖数: 1404 | 5 判断normal光看quantile distribution不行,得用formal test来证明,例如Jarque-
Bera Test。
我觉得楼上的truncation说法的很对,展开到infinity说不定就normal了。 |
|
s***e 发帖数: 267 | 6 My hunch is, it might be possible. If so, then it could be to extend the
idea so that we are comparing to numbers which are 33% and 66% quantiles of
the proposed distribution instead of the median, etc. I could be wrong, need
more thinking on this. |
|
s***e 发帖数: 267 | 7 Yes quasi-MC can reduce the variance faster, and it is not a MC method
although the name contains MC.
I think for one dimension you can split the quantiles and take equal spaces.
For higher dimension you may consider the hilbert curve method to fill the
space:
http://en.wikipedia.org/wiki/Hilbert_curve
For your problem with dimension = 20, the computation could be expansive.
sequence |
|
x********9 发帖数: 31 | 8 You might approximate the binomial by a Gaussian. The variance is 25, so the
standard deviation is 5. So -2*\sigma is approcimatively the 47% quantile |
|
y****d 发帖数: 432 | 9 ★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
前面说明:
需要的童鞋请到我的签名档的博客查找!谢谢!发E-mail太累了!
觉得有价值的话可以顶一下,以便更多的人看到!谢谢!
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
本合集包含内容(总有你想要的吧?!呵呵):
A Basic Course in Probability Theory
A Course In Probability Theory.djvu
A First Course in Statistics for Signal Analysis
A Handbook of Statistical Analyses Using R
A Handbook of Statistical Analyses using SPSS
An Introduction to Probability and Random Processes
An Introduction to Probability and Statistical Inference
An Introduction to Probability Th... 阅读全帖 |
|
L*******t 发帖数: 2385 | 10 经典的Markowitz之类的我做过一些基于quantile risk measure和copula的研究,把
Mean variance frontier往上面移了那么一点点,您说的经典模型是指哪些?
Dynamic portfolio choice的东西,简单的说,就是先figure out最优的terminal
capital是什么形式,然后假设complete market,去replicate这个最优的capital。
当然incomplete market的东东很好玩我在研究中。Markowitz的模型会产生puzzle
在dynamic portfolio choice的情况下有些puzzle会被解决。
街上都用啥模型啊?我孤陋寡闻,埋在书堆里太久了 |
|
m******t 发帖数: 273 | 11 Hi,
I need to do data fitting to find the distribution of a given data.
I need to find the pdf funtion of the distribution.
I can use data fitting functions in matlab and python.
It looks like a truncated gamma.
But, how to find the paramters of the distribution ?
What if the data cannot fit the truncated gamma well ?
The QQ-plot (qunatile-quantile) show that it is not a good fit for truncated
gamma.
How to find the distribution parameters such as alpha (shape), beta (scale)
for the truncated g... 阅读全帖 |
|
mw 发帖数: 525 | 12 中午跟个以前同事吃午饭,谈到今年equity的行情,说是特别好。‘基本上total comp
’都是半个米连。身心遂受到巨大刺激。谁给说说这话靠谱不,要是谁有个quantile就
好了 |
|
m*****e 发帖数: 207 | 13
If no calculator is allowed, then use the 68-95-99.7 rule.
1.3435 is between 1 and 2 sd, so the corresponding quantile
shd be between 84% and 97.5%.
hence the probability is between .025 and .16 QED |
|
s******d 发帖数: 303 | 14 The reason is I would like to calculate mean, SD and quantiles for each
dataset. I am not sure if "proc means" can do mean in a subset of a dataset. |
|
g******n 发帖数: 339 | 15 α 是一个预先设定的type I error阈值。这个α-level决定了一个decision rule,使
得当null hypothesis 为真的时候被拒绝的概率不超过α。β和1-β(power)当然也
和这个α-level有关,因为这个decision rule也决定了β和1-β。但是,这种关系不
是你所说的这种简单关系。比如说,你要做一个t test:u=0 vs. u>0. 你选择α=0.05
,表示你希望type I error不超过0.05。这个决定了你的decision rule:reject the
null hypothesis if the t test statistic t=x_bar/sd_err>t(1-α). where
x_bar=sample mean,
sd_err=sample standard error/sqrt(n),
t(1-α)=(1-α)upper quantile of the t-distribution with df=n-1.
那么,在这种情况下,你的power (1-β)又是什么?注意一般来说,这个power 是指
在al |
|
h***i 发帖数: 3844 | 16 那是不是说
standard bootstrap 和bootstrap from kernel density estimator with bandwidth
=n^{-1/2} 是一码事?
多谢,想破头了,实在是愚钝
这是for quantile variance
我翻了些paper, 比如 Angelis 和 Young的文章,没理出头绪 |
|
x***2 发帖数: 946 | 17 计算binomial模型,已知先验概率,然后有一个样本数很大的数据
比如,非常大的m, n
使得p^m * (1-p)^n非常小
即使做log transform 也还是极端数据,得到的quantile分布概率像 0 0 0 0 0 1 0 0
0
这样情况下,可不可以分割数据,做多次bayes?
还有没有比较好的方法? |
|
b*****n 发帖数: 685 | 18 R里面哪来的fivenumber?是summary吧。
matlab应该也有一些,像quantile什么的。 |
|
|
a********a 发帖数: 346 | 20 It seems p= gives quantile estimates? How is it related to
estimate the predicted survival time? Thanks. |
|
s*r 发帖数: 2757 | 21 要不你做quantile stratification
把数据分组之后,在每个组里运行
proc glm ;
model y=grp x1 x2 x3;
run;
看grp effect的average是不是要比在所有data里面运行同样model的估计值更接近true
value
感觉这样的comparison比较公平 |
|
|
|
a***g 发帖数: 2761 | 24 uiuc好的是不是都是搞quantile regression那一波的? |
|
y****d 发帖数: 432 | 25 ★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
前面说明:
需要的童鞋请到我的签名档的博客查找!谢谢!发E-mail太累了!
觉得有价值的话可以顶一下,以便更多的人看到!谢谢!
★★★★★★★★★★★★★★★★★★★★★★★★★★★★★
本合集包含内容(总有你想要的吧?!呵呵):
A Basic Course in Probability Theory
A Course In Probability Theory.djvu
A First Course in Statistics for Signal Analysis
A Handbook of Statistical Analyses Using R
A Handbook of Statistical Analyses using SPSS
An Introduction to Probability and Random Processes
An Introduction to Probability and Statistical Inference
An Introduction to Probability Th... 阅读全帖 |
|
A*******s 发帖数: 3942 | 26 correct me if i were wrong. i think quantile regression is for minimizing
absolute deviation. |
|
t****r 发帖数: 702 | 27 你所说的问题在统计上是有这么个领域来解决这个问题的,叫做quantile regression. |
|
s*****r 发帖数: 790 | 28 It really depends on your assumption.
I guess you meant to control the accuracy of the estimated mean, 4 points
give you a wide confidence interval. if you want a narrower ci, you can
calculate a suitable sample size. for normal data, the estimate of the mean
and corresponding ci is:
xbar +- Z(1-alpha/2)*sd/sqrt(n)
where xbar is the sample average, Z(1-alpha/2) is the 1-alpha/2 quantile of
standard normal, sd is the standard deviation, and n is the sample size. |
|
a******n 发帖数: 11246 | 29 别总是和好的比,活得累不累啊。
你想想,你毕业时候GPA在系里有75% quantile么。
Organization, |
|
h**y 发帖数: 153 | 30 definitely above 75% quantile. probably above 90%! |
|
t****r 发帖数: 702 | 31 前面有人问bootstrap能不能更逼近真理。我不是这方面的专家,但是发表一点自己的简
介,希望和大家交流一下吧。
总体而言, 我觉得使用bootstrap不能说是逼近真理。但是有的时候,确实比不做boot
strap比更好,或者比使用单一样本更接近真理。大家都觉得bootstrap没有真正用处的
原因是觉得所有bootstrap重复抽样的样本都是从一个样本里出来的,所以用bootstrap
的效果不会比使用原来的样本好多少。这一点我也同意。
但是从另外一个角度来说,一个样本里面包含的信息是很丰富的,我们是否已经完全利
用了现有样本里面的信息呢?最简单的例子来说,一个样本,很多时候我们用就用samp
le mean来summarize样本信息,但是使用sample mean的时候又忽视了多少样本中原来的
信息呢? 比如各种quantile的信息之类。 类似的,换一个角度来说,bootstrap是在重
复地寻找原来样本中所含有不同的信息。我觉得这是为什么bagging和random forest的
方法之所以work的原因。事实上,bagging和random forest使用boots... 阅读全帖 |
|
a*******g 发帖数: 80 | 32 Hi, I used Dr. Harrell's rms package to make a nomogram.
Below is my code for nomogram and calculate total points and probability in
original data set used for building the nomogram. My question is how I get
the formula for calculating the survival probability for this nomogram. Then
I can use this formula to do validation by using other data set.
f1 <- cph(Surv(retime,dfs) ~ age+her2+t_stage+n_stage+er+cytcyt+Cyt_PCDK2 ,
data=data11,
surv=T, x=T, y=T, time.inc=5)
surv<- Survival(f1)
surv10 <- ... 阅读全帖 |
|
d***n 发帖数: 43 | 33 a<-c(0.2, 0.3, 0.4)
b<-c(4,0,8)
c<-c()
for (i in 1:3)
{
d<-rep(a[i], b[i])
c<-c (d, c)
}
Then sort the c vector and calculate the quantile. |
|
p********a 发帖数: 5352 | 34 ☆─────────────────────────────────────☆
tamuer (hoho) 于 (Fri Oct 21 20:41:07 2011, 美东) 提到:
前面有人问bootstrap能不能更逼近真理。我不是这方面的专家,但是发表一点自己的简
介,希望和大家交流一下吧。
总体而言, 我觉得使用bootstrap不能说是逼近真理。但是有的时候,确实比不做boot
strap比更好,或者比使用单一样本更接近真理。大家都觉得bootstrap没有真正用处的
原因是觉得所有bootstrap重复抽样的样本都是从一个样本里出来的,所以用bootstrap
的效果不会比使用原来的样本好多少。这一点我也同意。
但是从另外一个角度来说,一个样本里面包含的信息是很丰富的,我们是否已经完全利
用了现有样本里面的信息呢?最简单的例子来说,一个样本,很多时候我们用就用samp
le mean来summarize样本信息,但是使用sample mean的时候又忽视了多少样本中原来的
信息呢? 比如各种quantile的信息之类。 类似的,换一个角度来说,bootstrap是在... 阅读全帖 |
|
s*****1 发帖数: 9 | 35 You may consider using quantiles. If one score is over the 90th percentile (
or any critical value of your choice), it is very good. |
|
L********d 发帖数: 3820 | 36 我有一个比quantile discretization更好的算法
想合作么?
:)
低( |
|
g*******u 发帖数: 148 | 37 To date, Bayesian approaches can do almost everything that maximum
likelihood based methods can do, but the reverse is not true. For example,
in finance area, people had no idea of how to estimate complicated
multivariate stochastic volatility models until the introduction of Bayes
methods.
Bayesian methodology is simulation based, which means you can obtain the
whole sampling distribution as byproducts. This gives you much more
information such as higher order moments and quantiles. The main co... 阅读全帖 |
|
t******g 发帖数: 372 | 38 同意,国人median高, quantile 95% 以上也许不如老外 |
|
y****e 发帖数: 1012 | 39 大家好~
我最近在用R,感觉无从下手。我想做一个qqplot(qunatile-quantile),输入是一个
文件(两列,第一列是measured result,第二列是理论值,都是浮点数)。输出用
qqplot画出他们的近似程度。
现在我还是不知道怎么把input feed到qqnorm, qqplot里面。有达人能给写个R 的
script吗?
我请你吃饭~~~
谢谢! |
|
c*******h 发帖数: 51 | 40 To my knowledge, classify 4000 hsopital into 4-5 quantiles. Within each
group, rank the hospitals according to the cure rate. if the cure rate is
correlated to group size, the rank No.1 will be give to the hospital with
highest cure rate in largest group. give appropriate weight to each group.
you can rank all hospital. |
|
N*9 发帖数: 2829 | 41 有没有在model output里头显示r squared的syntax? |
|
N*9 发帖数: 2829 | 42 当然这个r squared是pseudo r2,stata可以显示出来,但sas不能自动显示 |
|
|
N*9 发帖数: 2829 | 44 嗯,自己已经算了,本想有现成的,output会漂亮一点 |
|
w******4 发帖数: 488 | 45 how about quantile regression? I suppose there are outliers for quality
score. |
|
w******4 发帖数: 488 | 46 How about quantile regression? |
|
c********h 发帖数: 330 | 47 Yes.
You first want the margin error. It's 0.01 for your case?
You also want a confidence level, say 0.95
You also want to know what statistic is interested in. I'm not quite sure
what you want to use, so let's just suppose it's some sample mean. The most
commonly used statistics in this situation is either a proportion or a
sample mean.
Then margin error = t(alpha) * s/sqrt(n),
where t(alpha) is the upper quantile of t-dist. s is the standard deviation
for your data, either know or use bootstra... 阅读全帖 |
|
z**k 发帖数: 378 | 48 My understanding is, for Adaboost M1, the loss function mean(-y*F) is always
strictly decreasing, but this is not the case for the following code. Can
anyone help?
I m following the example of Hastie ESL-II chapter 10.1.
sorry cannot type Chinese here. Thank you very much for help.
#================R Script====================
## Data using example given in T. Hastie, ESL, chapter 10.1
dta <- matrix(rnorm(20000), 2000, 10)
pred <- apply(dta, 1, function(x) sum(x^2))
y <- (pred > qchisq(0.5, 10))... 阅读全帖 |
|
m******t 发帖数: 273 | 49 【 以下文字转载自 Quant 讨论区 】
发信人: myregmit (myregmit), 信区: Quant
标 题: how to do data fitting to find distribution
发信站: BBS 未名空间站 (Sat Mar 15 11:02:05 2014, 美东)
Hi,
I need to do data fitting to find the distribution of a given data.
I need to find the pdf funtion of the distribution.
I can use data fitting functions in matlab and python.
It looks like a truncated gamma.
But, how to find the paramters of the distribution ?
What if the data cannot fit the truncated gamma well ?
The QQ-plot (qunatile-qua... 阅读全帖 |
|
m****D 发帖数: 686 | 50 apple面我也是跟下面差不多的题目~
还被面过一些具体的R function, 比如quantile~ 面试官都很positive,然后就没下
文~~
没有feedback,都不知道哪里没说好就被默拒了~~
variable
b_ |
|