z****g
发帖数: 1978 | 1 The calculation of ACF an PACF assumes the variance of the innovation is
constant.
when there is dynamics in variance, ACF/PACF itself is wrong. If you still
STICK to
PACF/ACF, we call it overfitting
+ |
|
s****n
发帖数: 489 | 2 没接触过。现在要在10天内做出来一个project
做一个250+的data的analysis paper
要求能出来model.要有分析。比如acf and pacf分析
还要有residual analysis等等
我现在是盲人过河，想找个类似的paper用自己手头的数据
跟着做一下。
可我找了半天都没找到合适的paper. （需要牵涉到acf,pacf,aic,
forecast)
有大侠有类似的吗。推荐一个。大包子伺候 |
|
P*******9
发帖数: 9700 | 3 来自主题: _FantaSoccer版 - 龙黑哥哥 呵呵，就是用这三个工具来选择 ARMA(p,q) 里的p和q
比如说，如果true model是 MA(q), 用ACF就能判断q, 就是看acf 第一个非常小的auto
correlation对应的lag减去1.
如果true model是AR(p), 用PACF就可以.
如果是ARMA(p,q)就比较麻烦，得用EACF,这有个网页
http://www.uc.edu/sashtml/ets/chap7/sect28.htm
有介绍怎么做
另外,matlab里面可以画 ACF 和PACF |
|
|
s******s
发帖数: 11 | 5 I know it's partly an art, so just show me your sense of beauty please.
1 The determination of p-lag of AR. Many ways. Has there been any criterion
for specific cases? Or which do u suggest? PACF? AIC & SBC? Or something else?
2 Suppose the lag=L, then you estimate AR(L) and save the residuals E. Then
you use the Box-Pierce test to confirm that the E is uncorrelated. What's the
criterion? (Easy, but I just forgot)
3 If by any standard, E is uncorrelated, why did the guy write:" Now construct
L- |
|
z****g
发帖数: 1978 | 6 Poor basic knowledge
Let's start from the very beginning of basic time series.
AR type models, they deal with the conditional mean.
ARCH type model, they deal with the conditional variance.
So they describe structure of the different momentum of the unit innovation
.(if
you don't know what is a unit innovation, you should do a serious review)
ACF and PACF only give information of the mean part. Then after you figure
out the
structure of conditional mean, if you still find the innovation not the |
|
c****e
发帖数: 1842 | 7 should be ACF/PACF, i guess. |
|
|
|
l*******z
发帖数: 108 | 10 您好，我也觉得是AR1.但是不知道，判断的准则，什么拖尾，结尾，该怎样判断呢？
怎么的是拖尾，怎样的算是是截尾？ |
|
f******n
发帖数: 640 | 11 【 以下文字转载自 Statistics 讨论区 】
发信人: facedown (不要脸了), 信区: Statistics
标 题: 求大牛们一道题啊，给点思路哈
发信站: BBS 未名空间站 (Thu Nov 7 20:43:13 2013, 美东)
给了2组数据，分别是2个基金的最近20个时间的回报率(两组都是时间序列)。用什么办
法来评估这两组基金的表现和风险呢？
是否投资其中一个基金还是两个全部投资呢？
谢谢了
第一个问题的话，首先来比较一下平均回报率和方差,然后看了看两组数据的acf和pacf
发现这两组数据就是white noise啊
第二个问题的话 看看两组的数据的correlation 发现corr=-0.15,那既然两组的平均回
报率都是为正 是不是都可以投资呢？
谢谢了啊 |
|
u******r
发帖数: 61 | 12 SAS-ETS
直接出结果。
不是很难啊！
看acf/pacf 是 哪个阶段 cut - off or tail-off ,然后找出周期放到你的model里。
找书看是正解。 |
|
m****e
发帖数: 255 | 13 Sorry no Chinese input
you need additional variables to prove/disprove the data is faked.
For example, if you have the time stamp of each event, you can do time
series analysis. PACF and ACF graphs will also help.
The problem is very hard if there is no protocol for analyzing the data. |
|
c*********8
发帖数: 608 | 14 系统是这样工作的，customer发了一个请求，系统返还一个确认message. 这是第一个
event. 第二个就是customer回复确认message完成注册。
1）前后event 理论上都是人产生的，系统就记录了时间。现在有种推测是server有什
么优化程序来平衡系统负荷。
2）我有每个event 的timestamp, 用ACF/PACF看些什么呢。一共两个event,怎么构建序
列？以第一个event发生时间为x,对应的delay时间为y? |
|
m****e
发帖数: 255 | 15 Recode the events as
X_1 is the time of request confirmation. x_1_i's are the observations of
Event (type) 1.
X_2 for time of registration confirmation.x_2_i's are the observations of
Event (type) 2.
You indeed have multiple events indexed by i.
Let Y be the length of delay, ie, Y=X_2-X_1.
y_i=delay of event i.
The index i should be the order of the events.
First plot Y again index i and see if there is some trend.
If the observations are done manually, you may expect some seasonal pattern
or no... 阅读全帖 |
|
w**********y
发帖数: 1691 | 16 I think you didn’t get the spirit of the google interview. Based on the discussion with some friends interviewed by google statistical groups, they want an open mind, quick response ppl with a solid stat background and good skills in R(?)
U did so bad in Logistic regression…When you answered MLE (without any further info), the interviewer already knew your background is very weak…
There is no close form MLE. You need to either use Newton-Raphson or Iterative methods.
Open you mind..if they ask y... 阅读全帖 |
|
e*****u
发帖数: 337 | 17 看acf和pacf,或者用aic之类的criterion |
|
d******g
发帖数: 130 | 18 看acf和pacf,或者grid search,run 一个 loop看aic.看看那本Time series analysis
with example in R (Stoffer的书） |
|
l******n
发帖数: 9344 | 19 最近和同事争论一个问题，大家帮忙发表点意见。
有一个按时间序列排列的观察值，有几百个predictor,现在就像坐个简单地regression
model来预测这个观察值。找了些predictor，fit一下，结果发现residual的ACF，
PACF值不好。现在要fix这个问题，我建议重新去evaluate predictor,interaction之
类的,不过这样比较麻烦。一个同事就建议，直接加个lag variable。我觉得他的办法
肯定可以fix，不过完全是cheating。
大家有没有什么办法？ |
|
x****g
发帖数: 796 | 20 Need to figure out the meaning of following concepts:
acf: autocorrelation function (used to tell MA order)
pacf: partial autocorrelation function (used to tell AR order)
AR model
MA model
ARMA model
ARIMA model
Garch model
Just google 'ARIMA lecture notes', 'Garch time series' etc to find reading
materials.
Good luck. |
|
x****g
发帖数: 796 | 21 Need to figure out the meaning of following concepts:
acf: autocorrelation function (used to tell MA order)
pacf: partial autocorrelation function (used to tell AR order)
AR model
MA model
ARMA model
ARIMA model
Garch model
Just google 'ARIMA lecture notes', 'Garch time series' etc to find reading
materials.
Good luck. |
|
f******n
发帖数: 640 | 22 给了2组数据，分别是2个基金的最近20个时间的回报率(两组都是时间序列)。用什么办
法来评估这两组基金的表现和风险呢？
是否投资其中一个基金还是两个全部投资呢？
谢谢了
第一个问题的话，首先来比较一下平均回报率和方差,然后看了看两组数据的acf和pacf
发现这两组数据就是white noise啊
第二个问题的话 看看两组的数据的correlation 发现corr=-0.15,那既然两组的平均回
报率都是为正 是不是都可以投资呢？
谢谢了啊 |
|
A****F
发帖数: 1133 | 23 alpha, beta, sharpe, capture ratios.....?
pacf |
|
i*r
发帖数: 83 | 24 it appears to be just a moving average (8-wk window).
r package {nlme} look for gls fucntion, where you can specify the MA
parameter
gls(log(response) ~ \alpha + \beta_i*ith_day_of_week + + \beta*I(holiday),
correlatoin=corARMA(q=q)) # to get q-order moving average
you will need to determine the right value of order q using autocorrelation
plots such as acf() and pacf()
alternatively, you can try to calculate the 8-wk moving average using filter
()
tfilter = filter(log(response), rep(1/56, 56)... 阅读全帖 |
|
l***7
发帖数: 50 | 25 我有一组数据e1x1要生成ARIMA model.
44.7669 41.10659 43.70411 57.4064 47.21011
42.29574 44.65526 58.10271 50.38049 45.50983
47.45789 58.99774 53.02062 48.50563 52.39574
61.41254 56.42799 50.43014 57.19581 63.28712
59.45977 52.45281 59.88543 64.37775 60.75301
52.53201 61.68126 66.98495 63.95032 54.09921
64.57885 69.79277 67.75543 55.75373 65.93788
72.11951 70.56782 57.83367 67.67668 75.37291
72.30631 60.17343 69.73671 77.50877 74.87953
60.94993 72.871 79.53782 75.46017 62.1249
74.34622 81.8886 76.78092 63.... 阅读全帖 |
|
s***a
发帖数: 6258 | 26 来自主题: _FantaSoccer版 - 龙黑哥哥 还有，这两个data set分别要ACF PACF EACF 来model
怎么判断三个model哪个最好？也即判断model的标准是什么？
TA |
|
P*******9
发帖数: 9700 | 27 来自主题: _FantaSoccer版 - 龙黑哥哥 呵呵，不太理解用这三个东东来model是啥意思
这三个都是用来measure dependence的，比如PACF可以用来判断AR order, ACF可以用来
判断MA order, EACF 不知道啥意思， 等我放下狗 |
|
s***a
发帖数: 6258 | 28 来自主题: _FantaSoccer版 - 龙黑哥哥 是我看题看错了
要求的是根据这三个东东的结果来选择合适的model。那么也就是说如果fit PACF就是
AR model，fit ACF就是MA？那么怎么才能判断是不是fit呢？
炸机接班人果果是谁的马家？好博学啊。心心眼
用来 |
|
