rossry

• rossry 28 Jul 2020 22:59 UTC

  4 points

  on: Del­e­gate a Forecast

  When will air travel from New York to Hong Kong no longer re­quire ar­riv­ing pas­sen­gers to self-quaran­tine?

• rossry 10 Jun 2020 14:50 UTC

  9 points

  in reply to: ESRogs’s comment on: ESRogs’s Shortform

  It’s worse than that. If there weren’t any shares available at your bro­ker for you to short-sell in the mar­ket, you should con­sider it likely that in­stead of pay­ing 0.4%/​day, you just are told you have to buy shares to cover your short from as­sign­ment. This is an ab­solutely nor­mal thing that hap­pens some­times when it’s hard to find ad­di­tional peo­ple to lend stock (which is hap­pen­ing now).

  (Dis­claimer: I am a fi­nan­cial pro­fes­sional, but I’m not a fi­nan­cial ad­vi­sor, much less yours.)

• rossry 25 May 2020 14:50 UTC

  1 point

  in reply to: quanticle’s comment on: What are ob­jects that have made your life bet­ter?

  Strictly speak­ing, they’re not both lap­top charg­ers, but lap­top/​phone/​USB-C charg­ers. So two of them are use­ful on the road for charg­ing lap­top and phone si­mul­ta­neously.

• rossry 24 May 2020 18:02 UTC

  −1 points

  on: What are ob­jects that have made your life bet­ter?

  Not a phys­i­cal ob­ject, but the Cloud9 IDE (now ab­sorbed into Ama­zon’s AWS suite) for pro­gram­ming work. If your work fits into a ter­mi­nal plus text ed­i­tor (which it prob­a­bly does), then mak­ing the ac­tual hard­ware be a cloud server in­stead of a lap­top that can run out of charge is a big win, and be­ing able to ac­cess your “real” ma­chine from differ­ent in­ter­face ma­chines is some­times use­ful.

  For the in­ter­face lap­top it­self, I’ve been very, very happy with a Google Pix­el­book (which I got af­ter many, many satis­fied years with the origi­nal Chrome­book Pixel), but that de­pends on whether you have tasks out­side the browser, ter­mi­nal, or text ed­i­tor.

• rossry 24 May 2020 17:41 UTC

  2 points

  on: What are ob­jects that have made your life bet­ter?

  Separate travel toi­letries from at-home toi­letries. (The biggest win is not hav­ing to un­pack ex­actly when you get home tired.) Similarly, sep­a­rate travel phone/​lap­top charg­ers from at-home charg­ers, for same rea­son.

  I haven’t yet gone all the way to a sep­a­rate set of travel clothes, but would like to, one of these years. The ⁸⁰⁄₂₀ ver­sion is mak­ing sure to lay out 1-2 full sets of clothes be­fore go­ing on the long trip.

  (For refer­ence, I spent maybe five weeks of 2019 trav­el­ing, though nat­u­rally 2020 has been much less than that.)

• rossry 24 May 2020 17:24 UTC

  4 points

  in reply to: Three-Monkey Mind’s comment on: What are ob­jects that have made your life bet­ter?

  Ahem: a fifth lap­top/​USB-C charger. (One each for my couch, desk, and bed­room; two stay packed in my travel lug­gage.)

  h/​t to Zvi for mak­ing this sug­ges­tion in Dual Weild­ing, un­der the gen­eral head­ing of More Dakka.

• rossry 19 May 2020 12:28 UTC

  1 point

  on: English Bread Regulations

  The first two reg­u­la­tions have refer­ence prices for wheat that differ by 50%. How far apart in time were they is­sued?

• rossry 12 May 2020 13:43 UTC

  1 point

  in reply to: Wei_Dai’s comment on: Tips/​tricks/​notes on op­ti­miz­ing investments

  I’m be­ing un­nec­es­sar­ily oblique in the above com­ment, for which I’m sorry.

  What I mean is, in a tax­able ac­count, you have the op­tion to donate win­ners and har­vest cap­i­tal losses on losers. In a post-dona­tion in­vest­ment ve­hi­cle like a DAF, you don’t have that op­tion­al­ity. (Com­pared to a tax­able ac­count, your treat­ment on win­ners also comes out to no cap­i­tal gains tax, but your treat­ment on losers is worse, with no har­vest­ing losses.)

  (not tax or in­vest­ment ad­vice)

• rossry 11 May 2020 10:40 UTC

  2 points

  in reply to: Jonas Vollmer’s comment on: Tips/​tricks/​notes on op­ti­miz­ing investments

  It’s worth men­tion­ing that this is gen­er­ally a bad idea in the US tax regime (de­spite be­ing triv­ially easy), be­cause the op­tions for han­dling cap­i­tal gains and losses differ­ently mean you can some­times do bet­ter with pre-dona­tion in­vest­ments than with post-dona­tion in­vest­ments.

  (I’m a fi­nance pro­fes­sional, but no one’s tax or in­vest­ment ad­vi­sor, much less your tax or in­vest­ment ad­vi­sor.)

• rossry 22 Apr 2020 15:23 UTC

  3 points

  on: Could city de­sign im­pact spread of in­fec­tions?

  Hav­ing lived in New York (but only hav­ing vis­ited LA), the differ­ence in city de­sign that is im­me­di­ately salient to me is the pres­ence/​ab­sence of the Sub­way.

  Ac­cord­ing to MIT health economist Jeffrey Har­ris, the sub­ways seeded the mas­sive coro­n­avirus epi­demic in New York City:

  New York City’s mul­ti­ten­ta­cled sub­way sys­tem was a ma­jor dis­sem­i­na­tor – if not the prin­ci­pal trans­mis­sion ve­hi­cle – of coro­n­avirus in­fec­tion dur­ing the ini­tial take­off of the mas­sive epi­demic that be­came ev­i­dent through­out the city dur­ing March 2020. The near shut­off of sub­way rid­er­ship in Man­hat­tan – down by over 90 per­cent at the end of March – cor­re­lates strongly with the sub­stan­tial in­crease in the dou­bling time of new cases in this bor­ough. Maps of sub­way sta­tion turn­stile en­tries, su­per­im­posed upon zip code-level maps of re­ported coro­n­avirus in­ci­dence, are strongly con­sis­tent with sub­way-fa­cil­i­tated dis­ease prop­a­ga­tion. Lo­cal train lines ap­pear to have a higher propen­sity to trans­mit in­fec­tion than ex­press lines. Re­cip­ro­cal seed­ing of in­fec­tion ap­pears to be the best ex­pla­na­tion for the emer­gence of a sin­gle hotspot in Mid­town West in Man­hat­tan. Bus hubs may have served as sec­ondary trans­mis­sion routes out to the periph­ery of the city.

• rossry 11 Mar 2020 22:44 UTC

  10 points

  in reply to: Vaniver’s comment on: Why don’t sin­gu­lar­i­tar­i­ans bet on the cre­ation of AGI by buy­ing stocks?

  Higher var­i­ance is worth avoid­ing (un­der stan­dard as­sump­tions), but I for one was sur­prised by how lit­tle ad­di­tional var­i­ance one takes on by al­lo­cat­ing, say, 10% of one’s port­fo­lio to a sin­gle ar­bi­trary bet. In this com­ment I bal­l­parked it at maybe an ex­tra 0.5% var­i­ance.

  That said, al­lo­cat­ing one’s en­tire port­fo­lio this way ba­si­cally re­quires a re­jec­tion of the stan­dard risk-bud­get as­sump­tions.

  (Dis­claimer: I’m a fi­nan­cial pro­fes­sional, but I’m not any­one’s in­vest­ment ad­vi­sor, much less yours.)

• rossry 9 Feb 2020 4:58 UTC

  1 point

  in reply to: willbradshaw’s comment on: Some quick notes on hand hygiene

  I’m not sil ver, but as a ca­sual coro­n­avirus watcher (in part be­cause I live sig­nifi­cantly closer to af­fected ar­eas than most), my in­stinc­tive doubts are mostly 1 and 3. What num­bers are you us­ing for those to base the claim “ankify­ing this is prob­a­bly among the most valuable things you could ever use it for.”?

• rossry 29 Dec 2019 2:38 UTC

  1 point

  on: [Link] Watch­men trilemma

  In what sense are you us­ing the word “trilemma”? I’m ei­ther not fa­mil­iar with the us­age or miss­ing a big mes­sage of the post.

  (The com­mon defi­ni­tion of “trilemma” I’m most fa­mil­iar with pre­sents three desider­ata, of which it’s pos­si­ble to achieve at most two.)

• rossry 28 Dec 2019 2:13 UTC

  3 points

  on: Med­i­ta­tion Re­treat: Im­moral Mazes Se­quence Introduction

  I, too, am ex­cited.

• rossry 11 Dec 2019 13:51 UTC

  2 points

  in reply to: Pattern’s comment on: Bayesian examination

  How­ever, the post as­sumes that 1) there is (or should be) one cor­rect an­swer, 2) which is of the form: (1, 0, 0, 0) or a per­mu­ta­tion thereof, and 3) the ma­te­rial is in­de­pen­dent of the sys­tem (does not in­clude prob­a­bil­ity, for ex­am­ple).

  Th­ese are as­sumed for the sake of ex­pla­na­tion, but none are nec­es­sary; in fact, the scor­ing rule and anal­y­sis go through ver­ba­tim if you have ques­tions with mul­ti­ple an­swers in the form of ar­bi­trary vec­tors of num­bers, even if they have ran­dom­ness. The cor­rect choice is still to guess, for each po­ten­tial an­swer, your ex­pec­ta­tion of that an­swer’s re­al­ized re­sult.

• rossry 23 Nov 2019 6:03 UTC

  1 point

  in reply to: leggi’s comment on: Open & Wel­come Thread—Novem­ber 2019

  just be­cause “I don’t want to see more of this” doesn’t mean it’s up to me to in­fluence whether any­one else can see it.

  I feel like this proves more than you want. For ex­am­ple, is it up to you to in­fluence whether some­one sees more of some­thing, just be­cause you want to see more of it?

  Similarly, it’s also helpful to get a rea­son for up votes, but en­forc­ing that a rea­son be given can re­duce the amount of in­for­ma­tion-ag­gre­ga­tion that will oc­cur, on some mar­gins. What jus­tifies an asym­me­try be­tween how we ag­gre­gate pos­i­tive in­for­ma­tion and how we ag­gre­gate nega­tive in­for­ma­tion? Or would you also ar­gue that up votes should come with rea­sons?

• rossry 13 Nov 2019 14:10 UTC

  2 points

  in reply to: John_Maxwell’s comment on: Open & Wel­come Thread—Novem­ber 2019

  I mean a weighted sum where weights add to unity.

• rossry 12 Nov 2019 13:53 UTC

  9 points

  on: An op­ti­mal stop­ping paradox

  You need an ex­po­nen­tially in­creas­ing re­ward for your ar­gu­ment to go through. In par­tic­u­lar, this doesn’t prove enough:

  Since at each mo­ment in time, you face the ex­act same prob­lem (lin­early in­creas­ing re­ward, α-ex­po­nen­tially de­cay­ing sur­vival rate)

  The prob­lem isn’t ex­actly the same, be­cause the ra­tio of (lin­ear) growth rate to cur­rent value is de­creas­ing over time. At some point, the value equals $\beta /\alpha$ (is the right ex­pres­sion, I think?), and your marginal value of wait­ing is 0 (and de­creas­ing), and you sell.

  If the ra­tio of growth rate to cur­rent value is con­stant over time, then you’re in the same po­si­tion at each step, but then it’s ei­ther the St. Peters­burg para­dox or worth­less.

• rossry 12 Nov 2019 13:28 UTC

  2 points

  in reply to: John_Maxwell’s comment on: Open & Wel­come Thread—Novem­ber 2019

  Sorry, I’m writ­ing pretty in­for­mally here. I’m pretty sure that there are senses in which these ar­gu­ments can be made for­mal, though I’m not re­ally in­ter­ested in go­ing through that here, mostly be­cause I don’t think for­mal­ity wins us any­thing in­ter­est­ing here.

  Some notes, though: (still in a fairly in­for­mal mode)

  My in­tu­ition that the only way to com­bine the two es­ti­mates with­out in­tro­duc­ing a bias or as­sumed prior is by a mix­ture comes from treat­ing each es­ti­mate (treated as a ran­dom vari­able) as a true es­ti­mate plus some idiosyn­cratic noise. Then any func­tion of them yields an ex­pres­sion in terms of true es­ti­mate, each re­spec­tive es­ti­ma­tor’s noise, and maybe other con­stants. But “un­bi­ased” im­plies that set­ting the noise terms to 0 should set the ex­pres­sion equal to the true es­ti­mate (in ex­pec­ta­tion). Without mak­ing as­sump­tions about the ac­tual dis­tri­bu­tion of true val­ues, this needs to just be 1 times the true es­ti­mate (plus­maybe some other noise you don’t want, which I think you can get rid of). And the only way you get there from the noisy es­ti­mates is a mix­ture.

  By “as­sem­bly”, I’m propos­ing to treat each es­ti­mate as a larger num­ber of es­ti­mates with the same mean and larger var­i­ance, such that they form equiv­a­lent ev­i­dence. In­tu­itively, this works out if the count goes as the square of the var­i­ance ra­tio. Then I claim that the nat­u­ral thing to do with many es­ti­mates each of the same var­i­ance is to take a straight av­er­age.

  But they’re dis­tri­bu­tions, not ob­ser­va­tions.

  Sure, for­mally each ob­server’s pos­te­rior is a dis­tri­bu­tion. But if you treat “ob­server 1′s pos­te­rior is Nor­mally dis­tributed, with mean $G_1$ and stan­dard de­vi­a­tion ${\sigma }_1$” as an ob­ser­va­tion you make as a Bayesian (who trusts ob­server 1′s es­ti­ma­tion and cal­ibra­tion), it gets you there.

• rossry 11 Nov 2019 14:01 UTC

  3 points

  in reply to: John_Maxwell’s comment on: Open & Wel­come Thread—Novem­ber 2019

  Ah, okay. In that case, here are a few at­tempts to ground the idea philo­soph­i­cally:

  1. It’s the “prior-free” es­ti­mate with the least er­ror. See that un­bi­ased “prior-free” es­ti­mates must be mix­tures of the (un­bi­ased) es­ti­mates, and that bi­ased es­ti­mates are dom­i­nated by be­ing scaled to fit. So the best you can do is to pick the mix­ture that min­i­mizes var­i­ance, which this is.

  2. It ac­tu­ally is the point that max­i­mizes the product of like­li­hoods (equiv­a­lently, the joint like­li­hood, since the es­ti­mate er­rors are as­sumed to be in­de­pen­dent). You can see this by re­mem­ber­ing that the Nor­mal pdf is the in­verse ex­po­nen­tial quadratic, so you max­i­mize the product of like­li­hoods by max­i­miz­ing the sum of log-like­li­hoods, which hap­pens where the log-like­li­hood slopes are each the nega­tive of the other, which hap­pens when dis­tances are in­versely pro­por­tional to the x^2 co­effi­cients (or the weights are in­versely pro­por­tional to the var­i­ances).

  3. There’s a pseudo-fre­quen­tist(?) ver­sion of this, where you treat each es­ti­mate as an as­sem­bly of (higher-var­i­ance) es­ti­mates at the same point, no­tice that the count is in­versely pro­por­tional to the var­i­ance, and take the to­tal pop­u­la­tion mean as your es­ti­ma­tor. (You might like the mean for its L2-min­i­miz­ing prop­er­ties.)

  4. A Bayesian in­ter­pre­ta­tion is that, given the im­proper prior uniformly dis­tributed over num­bers and treat­ing the two as in­de­pen­dent pieces of ev­i­dence, the given for­mula gives the mode of the pos­te­rior (and, since the pos­te­rior is Nor­mal, gives its mean and me­dian as well).

  Are any of those com­pel­ling?