1,142 reputation
321
bio website
location Vienna, Austria
age
visits member for 2 years, 3 months
seen Oct 10 at 14:06

Oct
7
comment Determine $E[W_p W_q W_r]$
For a normal distributed rv $X ~ N(\mu,\sigma)$, $E[(X-\mu)^3] = \mu^3+3\mu\sigma^2$ (just search for moment normal). In our case, $\mu=0$ and $\sigma = p$. Alternatively, you can calculate it by hand (several times integration by parts) or via the moment-generating function.
Oct
1
comment quadratic programming portfolio optimisation
Hm, thats why I wasn't sure this is the right explanation because $x$ should be $x = [0.8,0,0.5,0,-0.3,0]$ in this case (splitting $x$ up in a positive and a negative part). Now, $x(1:3)+x(4:6)$ is the original portfolio and $x(1:3) - x(4:6)$ its component-wise absolute value.
Oct
1
comment quadratic programming portfolio optimisation
Its not linear because $|-x|\neq-|x|$ which would hold for a linear function.
Oct
1
comment quadratic programming portfolio optimisation
Please provide some more information about the constraints the example employs. There are techniques to reformulate optimization problems that simplify those but I doubt thats the problem here. Does your example use a different solver than matlab? Have you tried solving the example your way and compared the solutions? Further more: If you look at F and multiply with vectors $(x,-x)^T$ and $(x,-x)$ from both sides, you will miss the factor $1/2$. Probably the author moved into the $c$ but without any more information its really hard to say...
Jul
23
comment reference question about portfolio optimization
Although I voted to close this question: You are recommending the Pfaff book. I havent read it yet, is it good? Just a small comment though: #4 is also weritten by Bernd Sherer and those are commercial software packages. #3 Is basically a cookbook for the corresponding R package and treats rather advanced optimization problems (compared to classical MVO).
Jul
17
comment Handling Missing values in stocks returns when estimating the co variance matrix
@user3481555 Well if you delete the stock you simply reduce your investment universe. My intuition says that this induces at least some bias because the stocks with shorter history had a reason to enter the index later on. On the other hand, If you throw away the all the data where at least one stock has a missing value you might run into dimension problems (if you estimate a covariance matrix for hundrets of assets you want to search for "dimension reduction" or "shrinkage" or "robust" in quant.SEs search function). I will see if I can come up with the paper I mentioned...
Jul
2
comment Implementing A 50/50 Prediction Model Strategy
@emcor The 50% rate is only the percentage of right classifications. Just imagine an algorithm that gets the lottery numbers right 50% of the time. See also the comment by Joshua Ulrich
Jun
26
comment economic facts that causes the financial time series to be heavy tailed
For EVT, the more puzzling fact is that the "real" heavy tails do not scale with time...
Jun
23
comment What Is A Good Success Rate Using Machine Learning For A Beginner?
I think this question is definitely appropriate here. I am no expert on this topic, but I think if you use models like that you aim for a success rate that is well above $50\%$ with the amount depending on the cost of implementing the trading signal so that you place bets with an expected value >0. Further more, you would want to add some additional margin (for estimation and modelling error or changing environment for example). It definitely depends on the cost of getting in and out of the position. Thats why there can't be a single target number.
Jun
11
comment PDF Calculation by Fourier Inversion of Characteristic Function for Affine Intensity Process in Matlab
Have you tried the matlab-function ifft? It seems to make more sense than doing the quadrature explicitly. Depending on the implementation, you might have to normalize your vector afterwards though.
Apr
25
comment derive black scholes greeks
I suppose $q$ is a continuous yield dividend and $r$ the risk-free rate. But be careful, $N^\prime$ is most likely not the differential operator here! It is the probability density function of the standard normal.
Apr
15
comment how to extend lognormal model so that $\sigma$ is correlated to $\mu$?
@athos: I guess what you will end up with in the end will something like the "market price of risk" (the linear relationship you mentioned) and the notion of a risk-neutral measure. Please do not mix up correlation and simply a functional relationship. The correlation of deterministic quantities is always $0$. If you really want to pour time into this: This seems like a dead end to me.
Mar
31
comment Understanding the derivation of a ML-estimator
Could you clarify the first equation in (1)? There are some brackets missing I suppose? And why the notation $(\mathbb{1}^\prime \mathbb{1})^{-1}$? Isn't that just $1/T$?
Mar
21
comment Scaling of a transition matrix
@BrianB Nice!! Now that I read it, the Schur decomposition looks like the natural approach for this...
Mar
21
comment Scaling of a transition matrix
@Gerard Torrent OK I think I am missing something: Why should the eigenvalues of $D$ be nonnegative or even real? Why should all fractional powers exist (like 1/12 for example)?
Mar
19
comment What are the equation that gives hurst exponent of value >0.7 and <0.3?
I edited my answer.
Feb
19
comment What R-packages for SOCP problems are there?
@WilmerEHenaoH Why not? My problems are neither time-critical nor large scale...
Jan
21
comment Book on market microstructure
OK. Now you made me curious. Will check out your book...
Jan
13
comment Cointegration tests
The link appears to be dead.
Dec
18
comment Calculating and interpreting cumulative returns is R
@SBS Probably there is a require(TTR) missing. ROC() calculates discrete returns and you convert back with the exponential. You should correct that. Further more, if you model the price directly, try to reduce the sigma. You could also try something like seq(100,100.95,by=0.05) just to get started...