John

1,626 reputation

312

bio	website
	location
	age
visits	member for	1 year
	seen	yesterday
stats	profile views	153

	1,626 reputation	bio	website		visits	member for	1 year
312 badges		location			seen	yesterday

summary answers questions tags badges favorites bounties reputation activity

146 Comments

suggestions reviews revisions comments badges posts accepts all

Dec 21	comment	How to enumerate all the possible portfolios with a given target volatility? @SRKX The algorithm I discussed would generate a potentially infinite number of portfolios. I would think you'd have to impose some kind of cardinality constraints (like you can't hold half a share of stock but only a full share) in order to obtain some kind of unique solution. But even then, I imagine the set of all potential portfolios to be quite large.
Dec 21	comment	How to enumerate all the possible portfolios with a given target volatility? What if you just generated random portfolios, scaled their weights to 1, then blended that with the risk-free return to generate the target volatility?
Dec 14	comment	Control Bloomberg logins in a library Perhaps this is better directed to Bloomberg help?
Dec 13	comment	Missing factor in the factor model @Jase If you don't have much data, then the estimates may not be precise. As a practical matter it is easier to estimate in a Bayesian framework, which also imposes a computational burden. This makes backtesting take significantly longer.
Dec 10	comment	Monty Hall Model Not sure what the practical purpose of this is (regime-switching models of asset returns don't usually show much difference in the mean between regimes). Just think it through a second, if information comes out that the return on the market will be higher, would you try to take advantage of that, to the extent that the market will give you a fair price?
Dec 7	comment	Kalman Filter Equity Example I read a bit about how to use Particle Filters for on line Bayesian estimation. Don't understand all the math yet, but that might be a good enough reason to use them.
Dec 6	comment	Kalman Filter Equity Example Some great resources there. I think I have a vague sense of how the particle filter works, but I don't find it very intuitive. That March 2003 talk says that PF is best for multi-modal or skewed pdfs (implying that EKF or UKF might be better otherwise). Any insight if you only want to use a Kalman filter with t distributed errors?
Dec 4	comment	Why are regressors squared and not ^1.5 or ^2.2 or ^2.5? Instead of performing a non-linear least squares routine, the researcher has effectively imposed constraints on the coefficient. They want to handle non-lineraities without too many extra variables. So they just square it. Parsimony.
Dec 4	comment	Why are regressors squared and not ^1.5 or ^2.2 or ^2.5? The main reason is parsimony.
Nov 30	comment	Is there a piratebay for data(bases)? (here, talking about historical financial data) As far as I can tell, there's nothing more stopping anyone from putting up a torrent of financial databases than there is from torrents of movies/tv/etc. That being said, this site is for professionals, which suggests the question is off-topic, at a minimum.
Nov 27	comment	How to make the final Interpretation of PCA? When not relying on Bayesian techniques, I can see the advantage of PCA for dimension reduction. Consider high-dimension estimation of the covariance matrix where the number of observations is smaller than the number of securities. This typically leads to problems. Alternately, VAR or Garch estimation on a small number of factors is usually faster with fewer parameters than estimating them on every security in the universe.
Nov 27	comment	Does entropy pooling apply to distributions with time-varying drift? You normally would simulate to $\widetilde{X}_{t+k}$ and apply EP to that. I'm saying simulate $\widetilde{X}_{t+1},\ldots,\widetilde{X}_{t+k}$ and concatenate them into one matrix $\widetilde{Y}\equiv\left[\begin{array}{ccc} \widetilde{X}_{t+1} & \cdots & \widetilde{X}_{t+k}\end{array}\right]$ and treat that as though it is one distribution.
Nov 21	comment	portfolio optimization from empirical return distributions Since when does Monte Carlo only do that?
Nov 18	comment	Missing factor in the factor model Not necessary. The returns on the index should explain a significant amount of the variation, but PCA can also help.
Nov 17	comment	Missing factor in the factor model I'd have to know more about what the data is like.
Nov 16	comment	Does mean reverting imply mean stationary? It is possible that he means covariance stationary.
Nov 15	comment	Where can I find corporate bond spreads? Ah, so what you really want would be like an A- yield curve to use that spread to price a hypothetical bond. If you have a sample of all A- bond yields, you could construct your own but I'm not sure what providers give something similar.
Nov 15	comment	Where can I find corporate bond spreads? Have you tried the YAS function? It should have the G-spread on there.
Nov 13	comment	Combining Mulitple Forecasts? Budged Constraints? Let's say you estimate $z_{t}=B_{1,0}+B_{1,1}x_{t}+e_{1,t}$ and $z_{t}=B_{2,0}+B_{2,1}y_{t}+e_{2,t}$ and make forecasts from each. The optimal combined forecast wouldn't be the sum of the forecasts, but some sort of weighted average.
Nov 13	comment	Combining Mulitple Forecasts? Budged Constraints? What you have done is constructed a time series of the form $z_{t}=x_{t}+y_{t}+e_{t}$ and then regressed $z_{t}$ against $x_{t}$ and $y_{t}$. Imagine instead you have some time series $z_{t}$, you fit an AR(1) model and obtain a forecast and then fit an AR(p) model and obtain a forecast. How you combine those two forecasts is a categorically different problem.