231 reputation
19
bio website
location
age
visits member for 3 years, 3 months
seen Jul 27 '13 at 3:00

I am a Fellow at the Leonard Davis Institute of Health Economics and an MD/PhD candidate at U. Penn. (Perelman School of Medicine and the Wharton School). I have a particular interest in computer-intensive statistical methods and bringing state-of-the-art methods to applied problems. I also teach the Statistical Programming Workshop series for incoming doctoral students.


Nov
23
awarded  Autobiographer
Jul
10
revised Is there such a thing as “sell-off risk” in bond funds?
escaped the $
Jul
10
comment Is there such a thing as “sell-off risk” in bond funds?
From both of your most recent comments, I can see I'm still not being particularly clear. I just edited to try to focus the discussion on the very specific thing that the cited authors seem to mean by "sell-off risk".
Jul
10
revised Is there such a thing as “sell-off risk” in bond funds?
added 934 characters in body
Jul
9
comment Is there such a thing as “sell-off risk” in bond funds?
@AlexeyKalmykov That is my question. From the original question, the claim is: "By this they apparently mean that shareholders in the bond fund who do not sell experience losses as a direct result of other shareholders selling their shares in a period of steep market losses."
Jul
9
comment Is there such a thing as “sell-off risk” in bond funds?
@MattWolf "again there is no difference, when people sell assets that comprise a portfolio that you invest in then the asset itself as well as the portfolio will suffer" - I gave an example of my understanding of how this would work. Can you give a counter-example?
Jul
9
comment Is there such a thing as “sell-off risk” in bond funds?
@AlexeyKalmykov "Even if you sell with other shareholders you are still exposed to market impact risk" -- I don't care about the sellers; I care about those who continue to hold the fund.
Jul
9
revised Is there such a thing as “sell-off risk” in bond funds?
Added example
Jul
9
comment Is there such a thing as “sell-off risk” in bond funds?
@MattWolf Please keep comments/answers specifically focused on the damage to continuing shareholders from others selling. The inflation panic stuff is pretty far off topic. Of course selling when the market is down is going to cost you money--it's the difference between holding a fund and holding the bonds directly that interests me.
Jul
8
asked Is there such a thing as “sell-off risk” in bond funds?
Jun
26
awarded  Commentator
Jun
26
comment Why is an inverted yield curve a problem?
@TomAu That's the theory, yes. That explanation does not matter for market participants seeking a crystal ball but (if true) matters for academic theory. For market participants, the problem is the usual one that the world reacts to all known information and beating the market becomes difficult/impossible again.
Aug
8
awarded  Yearling
Nov
22
answered Why is an inverted yield curve a problem?
Oct
25
revised How many explanatory variables is too many?
added 144 characters in body
Aug
15
comment How many explanatory variables is too many?
The bullet point after "within the range of the data" means don't extrapolate. That's different than out-of-sample. Key point there is does not affect $\hat{y}$ ....
Aug
15
comment How many explanatory variables is too many?
"The holdout sample won't necessarily catch multicollinearity" is precisely right. And irrelevant for prediction. A random Google search confirms, with among others "If interest is only in estimation and prediction, multicollinearity can be ignored" public.iastate.edu/~alicia/stat328/Model%20diagnostics.pdf
Aug
15
comment How many explanatory variables is too many?
The point is that you arrived at humidity down/uptown by trying out a bunch of regressors and chancing on one that fit that dataset. Therefore, the holdout sample is unlikely to show the regressor as being good, and you'll see the problem. If it does work in the holdout sample, maybe you've just found a new regressor that actually does work. It doesn't matter whether it's causal or not--for prediction covariance is good enough as long as it consistently covaries.
Aug
15
comment How many explanatory variables is too many?
You want to know how it performs out-of-sample. Therefore you agree in advance that when developing your model you will only use, say 2/3 of your data. That tricks nature into giving you 1/3 of your data as out-of-sample data that you actually can observe. This is a common technique in predictive applications, and it works quite well when you've got a lot of data--as is likely the case here.
Aug
14
comment How many explanatory variables is too many?
This is a good start, but ultimately it's still in-sample. If what you're interested in is prediction, it's hard to beat actually predicting out-of-sample and seeing how good it gets. Since you don't care about the coefficients just the results, avoiding multicollinearity isn't really the answer here.