173 reputation
7
bio website
location
age
visits member for 1 year, 3 months
seen Feb 13 at 1:01

Jan
20
asked How do you determine whether a time series is stationary with reasonable confidence?
Jan
20
asked ARIMAX model's exogenous components?
May
6
asked What do you think is the trick making ARMA/ARIMA a good method for forecasting?
Apr
28
comment Length of time series and likelihood estimation
@mpiktas OLS estimates conditional likelihood, and it is not accurate for time series with rather limited sample size, thats where exact likelihood should be used to represent MLE instead of using OLS.
Apr
28
asked Length of time series and likelihood estimation
Apr
23
comment Efficient numerical methods to estimate ARMA models?
@Glen_b: I mean since solving the problem involves the computation of Hessian (typical), which is N^2 in size, so the evaluation of functions (N) is not likely to be the bottleneck of the performance when Hessian is computed analytically.
Apr
23
comment Efficient numerical methods to estimate ARMA models?
@Glen_b: Thanks, but since it is rather straight forward to compute the Hessian for ARMA with normal errors, I think the computation of likelihood is trivial in such cases...
Apr
23
asked Efficient numerical methods to estimate ARMA models?
Feb
6
asked In general which runs test is more powerful?
Jan
22
comment Modelling market mode (trending vs about to reverse)
@gung I dont discount miracles and I am not agree with jim collins (I have not validated his claim to be honest), but since when successful/profiting strategies get published or at least until they become under-achieving?
Jan
22
comment Modelling market mode (trending vs about to reverse)
Well, it can, althrough maybe not the way you thought, also, your example is flawed, since we see trends from a collections of thousands of indivial stocks, that demonstrated not much less significant trends comparing to the trends of each indivial stock, and there is very strong correlation between each stock in the market, whilst if trends are made purely by chance like you implied, then the collection of thousands or more stocks should not show any serious "trends", and indivial stock's trend should NOT correlated to each other the way we see in the market.
Jan
22
awarded  Commentator
Jan
22
comment Is there a concept called “contribution” in statistics?
Well, thats maybe the answer, actually it can be seen from the report that the authors's knowledge in statistcs are not very deep, anyway if there is not some technique term that i happen to be unwared of, then thats OK.
Jan
22
comment Is there a concept called “contribution” in statistics?
Well, it is an internal tech report, the so-called contrubtion are listed in one of the tables there, one "contribution" for each xi that used to develop the linear model (y=Ax+b), and the numeric value of these "contributions" varies from 0.0% to some <20%, almost look like some correlation factors or the co-efficient of the linear model, but get a percentage-representation.
Jan
22
comment Is there a concept called “contribution” in statistics?
@drknexus, kind of, they simply say it is the contributions of some xi to y (the outcome they tried to model).
Jan
22
asked Is there a concept called “contribution” in statistics?
Jan
22
comment Modelling market mode (trending vs about to reverse)
I wouldnt jump to the conclusion so quick, even if you can prove that stock price time series is i.i.d (which is obviously not), it still doesnt mean you cannot develop better investment strategy that beat the market with probability one.
Jan
22
awarded  Tumbleweed
Jan
20
comment Test for independent but not identically distributed time-series?
@cardinal Well, I am mostly interested in finding a test to test time-series data against a zero-mean and independantly distrubted stochastic processes (for instance, a random variable at time t: Xt~N(0, f(t)), where f(t) is some non-constant function), practical experiences shows BDS and runs tests works poorly in such cases, as the authors of these methods admitted.
Jan
20
asked Test for independent but not identically distributed time-series?