由买买提看人间百态

topics

全部话题 - 话题: kurtosis
1 (共1页)
w******o
发帖数: 3259
1
来自主题: Statistics版 - Q about Kurtosis distribution
两个variable,
一个skewness 是-.836 , kurtosis是.235,
一个skewness是-.224, kurtosis 是-.653.
sample size 是27.
请问这kurtosis代表的是什么?从这些distribution的数据来看能不能证明两个
variable是independent的呢?skewness -.224和-.653是significantly different
from each other的吗?怎么证明呢?-.836和.235呢?
万分感谢!
g******0
发帖数: 1165
2
来自主题: _pennystock版 - 最后一贴 TA FA大杂烩
Wall Street Networking
Business - Investing
Description:
Inception: Feb 7th, 2007
Find over 100 links in this group to help you with jobs/internships,
Economics, Bloomberg tutorials, excel spreadsheets, fund manager interviews,
interview questions, learning about the industry (investment banking or
trading? You decide), developing contacts & learning how to make money in
the markets.
_______________
Wall Street Networking Fan Page
http://tinyurl.com/d3kszq
Blackberry Addicts group
http:/... 阅读全帖
q********u
发帖数: 53
3
来自主题: Quant版 - 问一个time series的问题
1) calculate the kurtosis. Testing the hypothesis of kurtosis is 3 given
the data size. We can determine (3) non normal case.
2) I guess single plot we can get the delta_x we can get the volatility
should be a time trend. We can determine (2) vol is a time trend
3) investigate the epsilon. epsilon should behave like white noise.
Investigate the ACF which should be get the gaussian series for lag>=1 the
series is within confidence band.
Any other thoughts?
i*****r
发帖数: 1302
4
来自主题: Quant版 - 问个关于CF VaR的问题
cornish fisher VaR其实是个泰乐展开式,只有在分布接近于normal的情况下才有作用,
这个是原paper上写的,所以有时候在kurtosis很大, skewness偏正的时候会算出来是正
的,计算上是没错,但是常识上不make sense. 有什么办法调整么? 比如把kurtosis截断
在4或者5之类的
p*****k
发帖数: 318
5
(plz leave baozi to others)
it's usually hard to see on a linear scale - try the log-plot here:
http://en.wikipedia.org/wiki/Kurtosis#Kurtosis_of_well-known_distributions
your intuition is correct for Gaussian case: wider->fatter tail.
however, there is an additional intersection for a non-zero-kurtosis
p.d.f. with Gaussian p.d.f.
r****7
发帖数: 5
6
来自主题: Quant版 - 电面 高盛ISG in IMD
一个preliminary phone interview screening.这个组的全称叫investment strategy
group在pwm下面。电话打了半小时左右,没有任何behavioral。面试官从HR那里拿的我
的简历是
居然半年以前的。。。莫非他们的database是有半年的延滞?我让面试官在电话上等了
2分钟,我
给他发个更新版的。简单过了下简历后就是如下的technical的问题了:
1. 给出定义:VaR, CVaR, Skewness, Kurtosis。算他们有什么意义,衡量什么。
2. Suppose a return follows N(a,b), what is the 95% VaR?
3. Theoretically, give the formula of CVaR given the probability density
function of f(x).
4. What is the skewness and kurtosis of N(a,b)?
5. X(u)=(Integral from 0 to u) exp(-t) dB_t,... 阅读全帖
p*s
发帖数: 71
7
来自主题: Statistics版 - approximate CDF problem
is it possible to use mean, standard deviation and skewness to approximate
CDF of a skewed distribution? I know we can use Johnson curve, but it need
kurtosis, if kurtosis is set as zero then the parameters for Johnson curve
often can not be solved. Does anyone know other ways?
Thanks,
q******s
发帖数: 7469
8
来自主题: Military版 - 加拿大这个国家到底有多烂?
加拿大的收入分布曲线从kurtosis 到skewness都比美国小。以兑换率衡量比美国低,
以生活水平衡量和美国不差。当然加拿大的红脖子和美国一样洪波,只是数量小很多
t***s
发帖数: 163
9
来自主题: Investment版 - [LINK]Risk: What Exactly Is It?
一篇2003年的老文章
http://www.indexfunds.com/articles/20030808_Risk.htm
Variations on Standard Deviation
Skewness & Kurtosis
The Risk of Taking Risk
Tracking Error Risk
The Risk of Abandoning the Plan
Risk is Real
The Risk of the Maverick
The Risk of a Short-Term Outlook
The “Bearer of Bad Tidings” Risk
t***s
发帖数: 163
10
【 以下文字转载自 Quant 讨论区 】
发信人: meever (Life\\\\\\\'s a struggle), 信区: Quant
标 题: 苦闷, portfolio optimization 问题求助
发信站: BBS 未名空间站 (Mon Sep 10 14:44:07 2007)
我是一个学portfolio optimization方面的phd学生. 本来一直对quant很感兴趣, 做了
两年多的research越来越觉得自己没这个天份. 很想请求板上的高人指点指点.
从markowitz到后面一些复杂的方法, 近几年paper里面的我都读了一些, 自己跟着做了
一些, 但是好多我读到的都是很多的理论, 最后的检验却少的可怜, 都是很少的data,
或者很少的几个stock在做. 我自己做的结果更是令我很丧气. 回到最基本的东西, 我
发现历史上的return跟未来的return联系很小, 只有volatility似乎是有联系的, 其它
从mean,到skewness 到kurtosis到很多其它statistics都很难找出联系. 那么大家根据
过去来opti
m********0
发帖数: 2717
11
The thing is the it's not a graph for any stock. I just make it with the
following R code,
require(TTR)
ret<-cumsum(rnorm(252,sd=0.036))
ret2<-exp(ret)
p<-50*ret2
ma20<-SMA(p,20)
ema20<-emaTA(p,20)
ema50<-emaTA(p,50)
plot(p, type="l")
lines(ema20, type="l",col='green')
lines(ema50, type="l",col='red')
and I used lognormal distribution for the returns, textbook assumptions
for
a big group of models.
skeptic(as exists everywhere), yes, I borrowed this idea from ET. Please
don
't rush to criticize,... 阅读全帖
m********0
发帖数: 2717
12
I was over-reacting, sorry.
I agree with you consistency is the key. And I always believe in consistency
makes ATM.
Market evolves, a lot strategies works well does not work any more.
Meanwhile, there are more strategies been created, more systematic framework
, more computing power to search patterns, even patterns like short X every
trading day around XX:XX(maybe combined with market up/down for that opening
). I have a few candidates in my mind.
And there are different *market inefficiency* i... 阅读全帖
i*u
发帖数: 299
13
来自主题: Stock版 - Weekly Market Review 201202
• Small cap outperformed large cap; both bonds and stocks
experienced unusual correlation.
• S&P500 stocks performed stronger than the NASDAQ composite
internally.
• RSMCs continue to indicate a short-term stock market peak.
• Offensive stock lost steam after pre-earnings rally; financials,
materials and industrials gave up the most on Friday.
• All nine sectors look weak from historical forward performances;
exposure may come from the healthcare stocks l... 阅读全帖
d*********3
发帖数: 277
14
来自主题: Immigration版 - TSC 485 处理时间统计分析
基于本版注册的2012到现在一共236个样本。
Variable Dates_between_FP_and_AD
Sample size 236
Lowest value 5.0000
Highest value 143.0000
Arithmetic mean 41.6229
95% CI for the mean 38.1658 to 45.0799
Median 35.0000
95% CI for the median 30.0000 to 36.5397
Variance 726.6785
Standard deviation 26.9570
Relative standard deviation 0.6476 (64.76%)
Standard error of the mean 1.7547
Coefficient of Skewness 1.4944 (P<0.0001)
Coefficient of Kurtosis 2.0110 (P=0.0002)
D'Agostino-Pearson test
for Normal distribution reject No... 阅读全帖
d*********3
发帖数: 277
15
来自主题: Immigration版 - TSC 485 处理时间统计分析
统计表明LIn and SRC ttest p value = 0.61, no significant difference
Variable Dates__between_FP_and_AD_LIN
Sample size 201
Lowest value 11.0000
Highest value 151.0000
Arithmetic mean 40.3085
95% CI for the mean 36.5369 to 44.0801
Median 33.0000
95% CI for the median 31.0000 to 35.0000
Variance 735.3244
Standard deviation 27.1169
Relative standard deviation 0.6727 (67.27%)
Standard error of the mean 1.9127
Coefficient of Skewness 1.9843 (P<0.0001)
Coefficient of Kurtosis 4.5839 (P<0.0001)
D'Agostino-... 阅读全帖
G***a
发帖数: 27294
16
来自主题: loseweight版 - 为毛大家都觉得跑步减肥?
少来这套。
你们统计师好好教学生们把normality, kurtosis, skewness, 全给test了再出来
用t
哦,对啦,别忘了什么时候用non-parametric
别让我整天看些垃圾paper,垃圾般的数据处理,看到内分泌不调,我就谢天谢地
了。
PS:我们打了n圈了,lz想好自己要test什么了吗?
g****g
发帖数: 1828
17
来自主题: WaterWorld版 - Normal distribution
In probability theory, the normal (or Gaussian) distribution, is a
continuous probability distribution that is often used as a first
approximation to describe real-valued random variables that tend to cluster
around a single mean value. The graph of the associated probability density
function is “bell”-shaped, and is known as the Gaussian function or bell
curve:[nb 1]
f(x) = \tfrac{1}{\sqrt{2\pi\sigma^2}}\; e^{ -\frac{(x-\mu)^2}{2\sigma^2}
},
where parameter μ is the mean (location of the pe... 阅读全帖
Y*H
发帖数: 1582
18
It doesn't really matther in my opinion.
If you really want to compare the 3 exams, I don't see any overlapping
between MFE and MLC.
You are required to know some basic knowledge about stock price for C (2
short lessions in ASM), and this is the basis for MFE. Knowing all the
materials in MFE could save you 2-3 hours for C.
You are required to know basic statistical funcations for both MLC and C,
such as skewness, kurtosis, conditional variance, etc. Both exams also
cover hazard functions, cu
c*w
发帖数: 4736
19
来自主题: Computation版 - skewness in sas?
en... maybe you are talking about kurtosis.
for skewness, it's third moment, so skewness of standard normal is 0,
because standard normal is symmetric.
m****r
发帖数: 14
20
我是一个学portfolio optimization方面的phd学生. 本来一直对quant很感兴趣, 做了
两年多的research越来越觉得自己没这个天份. 很想请求板上的高人指点指点.
从markowitz到后面一些复杂的方法, 近几年paper里面的我都读了一些, 自己跟着做了
一些, 但是好多我读到的都是很多的理论, 最后的检验却少的可怜, 都是很少的data,
或者很少的几个stock在做. 我自己做的结果更是令我很丧气. 回到最基本的东西, 我
发现历史上的return跟未来的return联系很小, 只有volatility似乎是有联系的, 其它
从mean,到skewness 到kurtosis到很多其它statistics都很难找出联系. 那么大家根据
过去来optimize未来的目的到底是什么呢? 或许我看到的完全是错误的. 但这让我很受
打击. 导师让我做的东西都要求对几百个fund的data做实际检验, 结果试了很多中方法
, 结果都是很随机的, 没有出现一边倒的现象, 也就是说某种"神奇"的方法beat其它方
法的. 当然就挑个别的类型的fund在个别的阶段
b***k
发帖数: 2673
21
☆─────────────────────────────────────☆
meever (Life\\\\\\\'s a struggle) 于 (Mon Sep 10 14:44:07 2007) 提到:
我是一个学portfolio optimization方面的phd学生. 本来一直对quant很感兴趣, 做了
两年多的research越来越觉得自己没这个天份. 很想请求板上的高人指点指点.
从markowitz到后面一些复杂的方法, 近几年paper里面的我都读了一些, 自己跟着做了
一些, 但是好多我读到的都是很多的理论, 最后的检验却少的可怜, 都是很少的data,
或者很少的几个stock在做. 我自己做的结果更是令我很丧气. 回到最基本的东西, 我
发现历史上的return跟未来的return联系很小, 只有volatility似乎是有联系的, 其它
从mean,到skewness 到kurtosis到很多其它statistics都很难找出联系. 那么大家根据
过去来optimize未来的目的到底是什么呢? 或许我看到的完全是错误的. 但这让我很受
打击
i*****r
发帖数: 1302
22
比如我要约束skewness, kurtosis是多少,应该怎么做?
b***k
发帖数: 2673
23
☆─────────────────────────────────────☆
meever (Life\\\\\\\'s a struggle) 于 (Mon Sep 10 14:44:07 2007) 提到:
我是一个学portfolio optimization方面的phd学生. 本来一直对quant很感兴趣, 做了
两年多的research越来越觉得自己没这个天份. 很想请求板上的高人指点指点.
从markowitz到后面一些复杂的方法, 近几年paper里面的我都读了一些, 自己跟着做了
一些, 但是好多我读到的都是很多的理论, 最后的检验却少的可怜, 都是很少的data,
或者很少的几个stock在做. 我自己做的结果更是令我很丧气. 回到最基本的东西, 我
发现历史上的return跟未来的return联系很小, 只有volatility似乎是有联系的, 其它
从mean,到skewness 到kurtosis到很多其它statistics都很难找出联系. 那么大家根据
过去来optimize未来的目的到底是什么呢? 或许我看到的完全是错误的. 但这让我很受
打击
c****y
发帖数: 3592
24
比如我给个mean,stdev,skewness,kurtosis.能simulate出来么?
顺便问一下,那个johnrand是什么用?
谢谢!
j****r
发帖数: 286
25
you may need more info to define a random variate
suggest you google 'johnson's translation'

比如我给个mean,stdev,skewness,kurtosis.能simulate出来么?
顺便问一下,那个johnrand是什么用?
谢谢!
j*******d
发帖数: 55
26
ratio of an expected value and an std. deviation, 就是: mu/sigma, 数学上有
什么术语表达吗?我只知道skewness 和kurtosis,不知道这个一次项的叫什么,或者
根本就没有这么个术语?谢谢。
b***k
发帖数: 2673
27
☆─────────────────────────────────────☆
fishdaddy (无) 于 (Mon Apr 7 12:30:01 2008) 提到:
先考了几个BS model和volatility的基本问题
然后当场看SPY的IV curve和open interest curve
解释为啥call和put相同strike的IV不一样
解释为啥某些地方不smooth,有小小的fluctuate
解释为啥大部分地方vol skew看起来是convex的,但有些地方不是convex的
☆─────────────────────────────────────☆
Jadeson (Jadeson) 于 (Mon Apr 7 13:06:04 2008) 提到:
解释为啥大部分地方vol skew看起来是convex的,但有些地方不是convex的
这个问题,原来我看过,也有相关的解答,但问题在于那个解答是手写的,实在难以辨
认。哪个高手给翻译一下呢?
http://www.wilmott.com/blogs/kurtosis/enclosur
i*****r
发帖数: 1302
28
来自主题: Quant版 - 问一个maltab optimization的问题
就是说minimize kurtosis instead of variance?
i*****r
发帖数: 1302
29
来自主题: Quant版 - 问一个maltab optimization的问题
但是Kurtosis有个问题,刚刚test的
如果数全部相同比如[1,1,1,1],kurt就是无穷大了
还有[1,1,1,2]和[1,1,1,3]和[1,1,1,100]的kurt都是相同的...
i*****r
发帖数: 1302
30
来自主题: Quant版 - 问个关于simulation的问题?
如何simulate correlation是0.6的两条数列?
还有simulate指定skewness kurtosis的数列?
s***e
发帖数: 267
31
来自主题: Quant版 - 问个关于simulation的问题?
1.如果没有其他条件,只是要求population correlation是0.6, 可以用normal和
cholesky分解:
X 是2维standard normal, A=[1, 1/3; 1/3, 1],则Y=A*X的两个元素满足条件。
2.skewness是measure是否对称的parameter。normal的变量的skewness都是0。没有其
他限制可以用一个简单的不对
称分布构造,比如bernoulli(p). 给定skewness可以计算p的值,就是要解一个二次方
程(可能需要rescale)。kurtosis
是measure分布的tail是否light/heavy的parameter。可以用laplace distribution或
者bernoulli来构造。
N******r
发帖数: 642
32
skew risk = skewness
smile risk = kurtosis
m**********4
发帖数: 774
33
来自主题: Quant版 - 电面 高盛ISG in IMD
I find many questions (seeming related to statistics) very uninteresting (
Like memorizing kurtosis for Gaussian distributions)...I bet many stat phd
don't know it
Could you share your background with us?

strategy
w********r
发帖数: 253
34
来自主题: Quant版 - 求救,关于interest rate的model
比如federal fund,我把1988年6月开始到2010年6月,每个月的interest
rate都拿到手,kurtosis大约有11,所以是fat tail,然后skewness是-1左右,所
以negative skewed。头们誓死不肯用normal distribution近似,所以小的我必须找出
合适的distribution。
大牛门帮忙啊!!!!
目的是为了predict下个月interest rate变动的99。95%confidence interval。
我本来是用time series建了个类似AR(2)的model,predict出来下个月的mean和std
,本来想说就可以算99。95%confidence interval了,可是头说下个月那个点也不能说是
normal,所以也不可以简单的mean+3。xx std就得到confidence interval。头说
interest rate现在很低,所以下一点的预测应该是上走的probability大于下降的,所
以每点的propability也应该是skewness的。。。
c*******g
发帖数: 71
35
大家好,我是新手,没有finance从业经验,第一次投. 周一电面,分享一下面试题并请教准备面
试该看的书。
1.we know past 5-yr performance of a fund, how to measure its
performance?
arithmetic mean, geometric mean, sd, (Interviewer: more), skewness,
kurtosis
, (Interviewer: more), benchmark, risk-free, (Interviewer: more) ....
2. a put option struck X=50, spot=100, T=30-day, describe the return
distribution. what measures should be used to measure risk?
3. spot=X, then forward pricing
4. give >4 possible situations why linear regression might be ... 阅读全帖
p********6
发帖数: 1802
36
红皮书看了几个小时,C++没看,统计没看=>superfriday, lz天赋极高。。。

kurtosis
K*****2
发帖数: 9308
37
他以为看个kurtosis和skewness就知道distribution了,开国际玩笑呢
z**f
发帖数: 6
38

in physics, if your system is nonlinear,skewness will exaggerate the
nonliner/nonsymetric part, which are the points far away from the average and
others. compare a system with all kinds of linear oscillations to a nonlinear
system, you may see the difference. to know exact physical meaning, you need
to add nonlinear terms in the model, and make your assumption
reasonable/result acceptable. kurtosis is different, but can be understand
similiarly.
if you don't have a model. you are shooting a
x**g
发帖数: 807
39
来自主题: Statistics版 - 请牛人帮帮忙
Proc means data = xxx skewness kurtosis ;
var yyy;
run;
其实不仅是分布,还有很多方面可以帮助你决定是否需要将数据作为非正态来处理。比
如说,数据的连续性,sample size, cluster data v.s. not clustering; 拟采用的
Model对数据非正态分布是否robust?
o******6
发帖数: 538
40
来自主题: Statistics版 - [合集] 请问个fat-tail的问题
☆─────────────────────────────────────☆
cumboy (cumboy) 于 (Thu May 1 17:50:44 2008) 提到:
我一直没搞明白fat-tail是啥样子,fat是哪里fat,和normal比起来,峰是高还是底? 如
果说是"尖峰厚尾",那应该是比normal的峰高了
但我看到书上写t就是fat-tail,但t的峰没有normal高,这似乎挺怪的,那就不是"尖峰厚
尾"了? t的kurtosis多少呢
☆─────────────────────────────────────☆
shac (敢犯中华者,虽远必诛) 于 (Thu May 1 17:55:11 2008) 提到:
fat tail就是tail的pdf比normal大
总的面积是1,tail的面积大了,中间就矮了

☆─────────────────────────────────────☆
cumboy (cumboy) 于 (Thu May 1 18:04:59 2008) 提到:
那"尖峰"是怎么得来的呢? 似乎总是"尖
o******6
发帖数: 538
41
☆─────────────────────────────────────☆
beauty84 (to be perfect) 于 (Fri Feb 20 21:56:35 2009) 提到:
例如用risk和return之间的关系说明等等
实在搞不清楚这两者之间的联系
请指教
☆─────────────────────────────────────☆
xiaobian (fisher) 于 (Sat Feb 21 00:21:54 2009) 提到:
google

☆─────────────────────────────────────☆
LamborghiniA (官仔骨骨 走路去紐約) 于 (Sat Feb 21 00:49:37 2009) 提到:
wiki
☆─────────────────────────────────────☆
beauty84 (to be perfect) 于 (Sat Feb 21 01:09:03 2009) 提到:
如果能直接google wiki出来 我就不在这里问了啊
☆────────
s*r
发帖数: 2757
42
skewness kurtosis as well as higher order moments
s********8
发帖数: 45
43
来自主题: Statistics版 - NEED HELP
大家好,知道这里都是学统计的,我这个夏天刚选修 一门统计课。 有个课题不会作,
我把问题 copy 到这里。希望大家帮帮我。先谢了。
Use the data set supplied (hypertension, diabetes, cholesterol in subjects,
including medication, weight, age, etc) in docsharing and using Excel,
provide the following for Frequency the variable(s) of your choice:
1 distribution of a variable and bar graph of the same variable
2 Descriptives of a continuous variable: mean, median, mode, skewness,
kurtosis, standard deviation
3 Cross tabulation of two variables
4 Comp
x***2
发帖数: 946
44
skewness接近0
kurtosis 0.77
峰比正太稍尖
shapiro-wilk显著
log transformation结果更难看
有什么办法吗?
j******1
发帖数: 62
45
来自主题: Statistics版 - 请教大家:用SAS 做SEM出现的问题
请教大家一个SEM的问题:
我需要计算出Latent variable score
用sas proc calis做的model,以下是log中出现的问题:
WARNING: The number of observations 18 is not greater than the number of
variables 27.
WARNING: Corrected sample covariance matrix is not positive definite.
Multivariate kurtosis cannot be computed.
NOTE: Due to a sparse Jacobian the Hessian algorithm 11 will be used.
NOTE: GCONV convergence criterion satisfied.
NOTE: At least one element of the (projected) gradient is greater than
1e-3.
WARNING: The central paramete... 阅读全帖
m********o
发帖数: 427
46
是2d的离散的distribution,我想通过改变distribution的
偏度和峰度来改变这个分布的形状,不知道怎么做比较好?
多谢!
f*******i
发帖数: 8492
47
我用的是SAS 9.2,参照网上的例子,使用同样的命令行来运行不同的数据,但是输出结果的项目名称
和数量却完全不一样。
人家的输出结果,在“Centroid Hierarchical Cluster Analysis”下有如下几个项目
Variable,Mean,Std,Dev,Skewness,Kurtosis,Bimodality。
而我的输出结果,却只有如下四个项目Eigenvalue,Difference,Proportion,Cumulative。
在“Cluster History”, 人家有NCL,Clusters Joined,FREQ,STD,SPRSQ,RSQ和Dist。
而我的输出结果却只有NCL, cluster joined,FREQ, Norm cent dist,tie。

请问,这是怎么回事啊? 命令明明是一样的。
D******n
发帖数: 2836
48
来自主题: Statistics版 - 新手问个问题 (转载)
create a .vim directory under you home directory(there is a dot before
vim)
and then create a syntax directory under it
and then create a sas.vim file under the syntax directory
==============sas.vim======================
if version < 600
syntax clear
elseif exists("b:current_syntax")
finish
endif
syn case ignore
syn region sasString start=+"+ skip=+\\|\"+ end=+"+
syn region sasString start=+'+ skip=+\\|\"+ end=+'+
" Want region from 'cards;' to ';' to be captured (Bob Heckel)
sy... 阅读全帖
z**********i
发帖数: 12276
49
来自主题: Statistics版 - normality check
我用PROC CAPABILITY来CHECK NORMALITY.
看图形很NORMAL,但PVALUE 很小.这个到底是NORMAL DISTRIBUTION吗?
多谢!
Variable: RBMI
Moments
N 37667 Sum Weights
37667
Mean 0.05291262 Sum Observations 1993.
05967
Std Deviation 0.00865415 Variance 0.
00007489
Skewness -0.3560206 Kurtosis 0.
79177021
Unc... 阅读全帖
s*****G
发帖数: 1535
50
来自主题: Statistics版 - 问2个统计检验的问题
想请问统计大牛们2个检验的问题,都是1年的数据,大约250个。
1,我想验证我的parameters都是正态分布的,所以我算了数据的skewness,kurtosis,
发现并不是0和3,但是我画了qq plot 发现它们基本上是直线 y=x的这种,请问这种算
是基本符合正态分布么?
2,关于correlation, 我想检验2组数据是不适独立的,算了correlation, 基本上接近
0,但是怎么能证明他们是independent??? 用chi square??
谢谢啦!!!!!!!!!!!!
1 (共1页)