l****o
发帖数: 2909 | 1 try shrinkage estimation. |
|
|
z****g
发帖数: 1978 | 3 shrinkage method: linear combination of PCA/Factor based cov and sample
based cov |
|
|
b***k
发帖数: 2673 | 5 hey, AlphaNBeta,
May I ask what is APT stand for here?
thanks.
are
APT |
|
k**x
发帖数: 2611 | 6 用covariance的定义 V[XY]=cor[xy]*sqrt(v(x)v(y)) |
|
o******2
发帖数: 159 | 7 具体也没有啥经验之谈，但回报一下大家，也算分享下鸡冻的心情。
面了5个人，一个在外地出差，他电话面试我。总共4个小时左右。4个PhD，有数学的，
CS的，还有一个普林斯顿毕业的history的老PhD，50岁左右了，突然放弃大学的教授职
位，跑来写code。另一个是一个小本科生，看起来才21，2岁，但人家是Caltech的，问
的东西反而是最难的，问个数学我还没有答出来，真惭愧。
感觉就一直在白板上写C++，还有一个32位转64位的bit操作之类的，竟然被我折腾出来
了。还有拿打印出的一段code，让我找错误，看输出之类的。里面trick包含trick，也
太看得起我了，我不是CS科班出身的，只是业余水平啊。
数学题问了一下，比如矩阵的一些知识，计算covariance矩阵，生成correlated的
random number的方法，还有linear programming的知识，我说我写过simplex method
以及conjugate gradient method之类的code，就没有再问了。
没有问什么概率题，brain teaser之类的，没有问finance的知识，看... 阅读全帖 |
|
o********n
发帖数: 100 | 8 小弟现在在给一家公司开发交易策略。 目前问题是整合自有策略到mean-variance
portfolio下时，发现简单的mean-variance portfolio的表现非常差。怀疑要嘛是
estimation of covariance structure 结构有问题，要嘛是代码有问题。
想请教一下有经验的朋友， 一般portfolio selection的测试数据，会如何生成？ 一
般需要考虑哪些情况？
谢谢！ |
|
l****o
发帖数: 2909 | 9 mean variance is not always good. However, to improve the performance, I
suggest using black-littleman model or improving covariance matrix
estimation. |
|
q**j
发帖数: 10612 | 10 you mean newey-west？that only changes covariance, no much impact on
forecasting, imho. |
|
c*******e
发帖数: 150 | 11 Sure, portfolio optimization (especially TC-aware type) is a very useful
subject for practioners and the pm industry. The thing is, what is tricky
and important in practice (hence I think worth academics' decent attention)
is that even if the jointly Gaussian assumption holds, we should keep in
mind that the Expected Return vector and the Covariance Matrix are always
estimated as opposed to knowing their oracle value, hence robustness
concerns and the corresponding techniques (Black-Litterman sh... 阅读全帖 |
|
a***n
发帖数: 423 | 12 这些问题还真是quant finance要解决的问题，因为financial time series data大都
不是covariance stationary的。 CFA level 2的quantitative finance,讨论了一些方
法解决这些问题。 |
|
A**u
发帖数: 2458 | 13 多谢分享啊,你这是intern吗 在哪考试的?
1.我也算的是3/32 {TT,HT,TH}THH
2.应该是N/2^N-1 = 3/ 4
3. 用covariance matrix 应该半正定，条件，最小是 -1/3
4. P(B_1>0,B_2>0) = P(B1>0)P(B2>0 | B1>0) = 1/2 P(B2>0|B1>0).
B2-B1,B1相同分布 P(B2>0|B1>0) 要求 P(|B2-B1| > |B1|) 由对称性 1/2.
再要求向上，1/2, 结果 1/8
5. 啥意思 df/dx 是什么
My |
|
h***s
发帖数: 35 | 14 1. 怎样将一个排序的字符串数组分解为两个随机的数组?
小弟的答案是产生随机数,随机数乘数组上限得到index, 拿出index所指的字符串.
重复该过程直到得到一半的字符串.
这个解法的问题是:可能连续产生两个非常接近的随机数,进而产生连续的index, 导
致某些字符串有序.
2. 一个covariance matrix (C), positive definite. what's the exponential of
coveriance, that is, exp(C)?
Hint from interviewer: 1. Taylor expansion; 2. Decompose converiance
matrix.
小弟的答案是: Taylor expansion of exp(C)= 1+C/1!+ C*C/2!+...C^N/N!+...
忽略N阶以上的展开项.
这个解法的问题是:approximation. 面试官要精确精确的结果!!!
小弟不才,恳请高手赐教.非常感谢! |
|
r*********n
发帖数: 4553 | 15 covariance matrix is not a number. what does it mean to take exp{CovMat}?
element-wise? |
|
n****3
发帖数: 23 | 16 The group mentioned below is looking to add more positions. Please contact
me with in-site mail if you are interested. Please also indicate when you
will be available to be on-board in Shanghai.
----------------------------------------------------------------
Our firm is opening an “Advanced Analytics Center” for Risk Methodologies
in Shanghai’s Financial District. If you are interested and also match
the requirements, please contact me with in-site mail.
(Note these are local positions and ... 阅读全帖 |
|
d*****o
发帖数: 34 | 17 如果你用的是eigenvalue decomposition的话，应该是对covariance matrix而不是
data matrix做的吧。 |
|
o********n
发帖数: 100 | 18 请问如果有若干只股票，大部分是从05年开始， 但其中有一些是在08年左右才出现，
如果我要用05年开始的数据估计covariance,则需要将08年的数据用em algorithm进行
补全。
这方面以前没做过， 请问各位大拿是否处理过相似的问题，有推荐的文章吗？ 搜了
一下，发现讲em进行数据补全的挺多，但是我用的矩阵是（股票*时间）的， 所以需要
考虑补全的时候是否一只股票一只股票处理比较好，还是将整体一起处理比较好。
谢谢！ |
|
I****k
发帖数: 35 | 19 Title: Quantitative Risk Analyst
Level: Entry-level
Employer: Major US Bank
Location: Shanghai, China
Job Descriptions
- Utilize statistical/quantitative techniques to analyze market data;
develop and improve algorithms of time-series analysis.
- Responsible for the construction of covariance matrices that are used in
the simulations of Value-at-Risk (VaR) and counterparty risk exposure
calculations.
- Perform profit attribution analysis and hypothetical backtesting on
tradable products.
- R... 阅读全帖 |
|
z*******g
发帖数: 18 | 20 method 1. correlation between two variable can be expressed as the angle
between them, so corr(X,Y) = cos(x1)=0.7, corr(x,Z)=cos(x2)=0.8,
so the minimum value of corr(Y,Z) = cos(x1+x2);
method 2. using the property that the covariance matrix is positive defined.
Z) |
|
w*********r
发帖数: 488 | 21 估计你的project是想说realized volatility?当你用intra-day data计算realized
volatility的时候，如果sample too frequently， 比如every transaction，你实际
上在累积variance of market micro-structure noise，这样算出来的realized
volaitlity不consistent。所以它的提出者Tim Bollerslev建议少取点samples，比如
每隔15分钟甚至30分钟。后来一帮北欧的哥们儿，还有一位stanford的哥们儿，提出
real kernel，这个算法类似时间序列里面Heteroskedastic Auto Covariance (HAC)，
就是引入intra-day return之间的自相关性，赋予一定的weight去smooth out
variance of noise。有了这个方法，想取多少点取多少点，我试过每隔1分钟（股票市
场）。结果不错。weekendsunny说的TSRV是two scale realized vola... 阅读全帖 |
|
v******y
发帖数: 189 | 22 最近要面个equity risk analyst的职位，除了知道VaR，covariance matrix和β这类
CFA三级里的内容，其他的了解还真不是太多。请问有没有做这方面的前辈，能来说说
具体每天做些什么事情呢？谢谢指教！！ |
|
J*****n
发帖数: 4859 | 23
risk quant这块在IB里面分两种
1. Intern risk。
一般是做stress test，对复杂的产品的correlation和covariance建模。可以做的很简
单，也可以做得很复杂。一般需要的是stochastic那套东西。Stress test 也会用到统
计。
2. Client risk
一般是写portfolio optimization方面的factor model和report，卖给客户，和MSCI
Barron做得一样。一般需要的是计量经济学。
从分类上说，第一种是 Back office，第二种是front office。 |
|
L*******t
发帖数: 2385 | 24 Theoretically。。可以人为的加constraints，这个在理论上1992年就被Cvitanic和
Karatzas解决了，然后具体用起来可以假设一个Markov的structure用ＰＤＥ来解
然后Mean variance也可以加各种constraints。这样应该就不会有extreme了吧。此外
Mean variance要估计一个超大的Covariance matrix，然后参数不确定性的情况下，如
何用Mean Variance似乎也有结论。 |
|
j*****1
发帖数: 66 | 25 【 以下文字转载自 JobHunting 讨论区 】
发信人: johnus1 (John), 信区: JobHunting
标 题: SAS Quant Developer - Los Angeles
发信站: BBS 未名空间站 (Mon Sep 15 01:03:38 2014, 美东)
contractor 职位，但可以办H1B transfer（不能申请新的H1B），如果合适会长期雇用
。公司是个Fixed Income Asset Management Firm.
发简历到 [email protected]
(function(){try{var s,a,i,j,r,c,l,b=document.getElementsByTagName("script");l=b[b.length-1].previousSibling;a=l.getAttribute('data-cfemail');if(a){s='';r=parseInt(a.substr(0,2),16);for(j=2;a.length-j;j+=2){c=parseInt(a.substr(j,2),16)^... 阅读全帖 |
|
h******r
发帖数: 201 | 26 Drawing a pair of (x, y) from a joint Gaussian distribution with 0
covariance. Knowing the stndard deviations of x and y and knowing z = x + y,
what is your best guess for x? |
|
h******r
发帖数: 201 | 27 Drawing a pair of (x, y) from a joint Gaussian distribution with 0
covariance. Knowing the stndard deviations of x and y and knowing z = x + y,
what is your best guess for x? |
|
k****1
发帖数: 133 | 28 Offer 1: NYC 银行, 做model validation 主要wealth management/portfolio
management模型, 比如Covariance, Portfolio Optimization/asset allocation,
factor model and etc.
Offer 2: North Carolina银行, IB部门 做Market Risk, Support trading desk, 主
要VaR model, 产品包括: Interest rate, structured, FX, equity and commodity.
Same title, Offer2 比 Offer1 total compensation(base+bonus) 多近7w.
Offer 1
优点: validate的model类似buyside model: Black Litterman, Risk Attribution.
在NYC,机会多, 以后挺想做quant strategy, portfolio manager.
缺点: Model ... 阅读全帖 |
|
l******9
发帖数: 579 | 29 已知两个随机变量的 mean and std. dev.
但是不知道它们的分布，如何求它们的 covariance ？
如果它们是同一个分布呢 ？ 又如何求？
谢谢 |
|
b*******z
发帖数: 331 | 30 相关不相关，本来就是只要Covariance为0就是不相关阿 |
|
y******e
发帖数: 5906 | 31 KF系统参数的调节改进算法倒是不少，只是不知道符合你的实际情况不，你可以Google
看，不过我感觉你可能得自己设计自适应算法，好难，我搞不定哈。
不过另一种方法你可以考虑用UKF，因为KF是适用于线性系统的，如果是非线性系统的
话用UKF比较好，另外，KF系列都是建立在covariance基于高斯分布的特例下。如果非
高斯情况用particle filter做比较好。但是PF纯用统计了，没有反馈环，效果不稳定
，我个人觉得。 |
|
J**Y
发帖数: 34 | 32 Suppose you want to draw from mutivariate normal distribution with mean
vector U (kX1)and variance-covariance matrix V(kxk). First, draw k sequences
independently from standard normal distribution; Second, let T be the square
root of V such that T'T=V; Third, stacking the k draws together to get a
matrix Z (nxk), where n is the number of obs of one sequence. Finally,
the desired draw is: X=U+T*Z. |
|
c**********g
发帖数: 222 | 33
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
The point is that in fact pressure is a tensor, so the sign has no much definite
meaning. Also, in relativity, it can be covariant or contravovariant (the sign
is opposite). SO, we need to pay attention to the meaning of the sign.
In the case of the cosmological constant, the negative pressure will produce
a repulsive gravity! This is the reason that it will accelarate the expansion of
the universe.
I still do |
|
r****y
发帖数: 1437 | 34 SVD has something related to EOF, or called PCA, an very essential
statistical tool in data analysis.
Give SVD of matrix A
A = ULV
A*A = V*(L^2)V where * denotes adjoint, and L is diagonal matrix.
A*A is usually called covariance matrix of A, and the normalized
eigenvectors of A*A are called EOF (empirical othrogonal
functions). So far, EOF, or in another name, principal
component analysis, is very powerful a |
|
t******q
发帖数: 117 | 35 ICA, indepent component analysis is also a good
method for feature extracting.
I know a prof is doing 3d modeling using SVD,
he ask the students use ICA model to do the approach,
I feel that the method is computation expensive,
(I do not do this work, just have the sense).
Just read a short toturial paper, the preprocessing
is centering and whitening the data to be
zero mean and uncorrelated with unity covariance.
E(x) = 0;
cov(x) = diag.
consider a sequence of image, that will be a lot time to |
|
f****n
发帖数: 355 | 36 OK, I looked at my textbook, and here's what it says:
in case of white gaussian noise, you can use the "square-law" detector,
which is
r'r ~ gamma
gamma is the threshold you choose based the desired power. r is the column
vector of input. r' is its transpose, and ~ means comparison.
For non-white gaussian noise, if K is the covariance matrix, the square-law
detector becomes
r'* K^-1 * r ~ gamma
K^-1 is the inverse of K.
My textbook is written by our teacher and probably can't be found in
booksto |
|
t*a
发帖数: 117 | 37 The January SIGSTAT meeting will be held on Wednesday, January 21, from
12:30 to 1:30 in Room S3031, 1800 M St, NW (directions can be found on
the website below).
The topic is "PROC MIXED Part 3: Model Development and
Interpretation".
Part 1 applied EDA techniques to help in specifying an initial model.
Part 2 examined the issue of selecting an appropriate covariance
structure using the variogram as the main tool.
Part 3 will cover the specification of the mean part of the model,
plotting intera |
|
p********a
发帖数: 5352 | 38 ☆─────────────────────────────────────☆
lee1980 (艾lee) 于 (Wed Jul 11 11:19:47 2007) 提到:
用proc reg做回归分析， 有两个independent variables， 可以直接求出这两个
variable的coefficients的standard error， 用covb也可以求出这两个coefficients
的covariance。 我现在需要test 这两个coefficients是否significantly different。
用这个公式 t＝（β1－β2）/sqrt（varβ1＋varβ2－cov（β1，β2））。 大家觉
得这个t test 的degree of freedom是几呢？ 我该怎么求出这个t value 对应的p
value呢？
谢谢～
☆─────────────────────────────────────☆
statcompute (statcompute) 于 (Wed Jul 11 11:55:14 2007) 提到:
data o |
|
s*******t
发帖数: 2896 | 39 有点意思。不过还是不甚了了。
如果是experiment，那还inference什么呀？treatment有没有effect明显呀。
如果是observational study，单从数据，怎么可能知道哪个是causation呢？比如两个
covariate colinear的情况。
那位高人能解释一下？ |
|
i****o
发帖数: 242 | 40 3. A researcher is planning an intervention study to reduce childhood
obesity. The study will consist of a treatment group and placebo control.
The expected effect of the intervention will d=.50.
a. How many subjects will be required to detect a significant effect at
alpha=.05 with power=.80?
b. (Hard!) Now, assume that the investigator wants to improve power. She
knows that gender has a strong association with weight (r=.447). So, she
includes gender as a covariate in the analysis, mak |
|
d**********l
发帖数: 183 | 41 如果covariate里有time series 的regressor， response也是time series。这样的情
况下做linear regression是不是有些复杂？就我目前的理解是如果x和y都是
stationary的过程，或者如果x和y是cointegrated的，是可以直接做linear
regression只不过error的pattern 是stochastic的。
请问牛人们我的理解对吗？
如果y是stationary， x是i（1）那么是不是要把x difference 一次再对y 和diff（x)
做regression呢？
谢谢大虾的回答。 |
|
|
z*********o
发帖数: 541 | 43 thank you so much
variables. |
|
j*****e
发帖数: 182 | 44 This is not a time series data, since you have 100 patients, not one patient
. It is a repeated measurement data with time dependent covariates(???).
Check some experimental design book on mixed models. |
|
j*****e
发帖数: 182 | 45 I don't use R. A good book on how to use SAS analyzing repeated measurement
is the SAS mixed model book published in 2006. Another starter book is
Miliken and Johnson's book on messy data.
In biometrics, data with repeated measurement is called longitudinal data.
You may also find reference book on the analysis of longitudinal data.
For your problem, the covariate is time-dependent. It might be hard to find
a direct solution out of a text book. Search on CIS for some articles in
this area.
Also, |
|
l********s
发帖数: 430 | 46 GEE model with unstructured covariance structure...age, gender, .... ect are
adjusted,....., the odds ratio of **...... |
|
p*p
发帖数: 492 | 47 关于covariance matrix要用到.. |
|
s******1
发帖数: 157 | 48 任意一个多变量分布。
想求基于sample covariance matrix求出的principal component的distritbuion。
可是我翻了很多书和paper，只能查到在multi normal下的sample principal
component 分布。
请问，如果非normal，可以套用normal的结果来近似吗？
谢谢指点！ |
|
m*******g
发帖数: 383 | 49 Cov((X1-X2)^2, (X2-X3)^2)
其中X1,X2,X3～N（0,1/2）。
基础太不扎实了，想了半天也不知道怎么算。
先谢谢各位牛人了！ |
|
h***t
发帖数: 2540 | 50 x1,x2,x3 are independent? |
|
