oops - thanks for pointing out my missing dates definition. You are correct
On Tue, Sep 1, 2015 at 5:31 PM, Raul Miller <[email protected]> wrote: > What is dates? > > ]c=. +/ 0.02 < | pctChange d=. ((1999 < ". 4&{. every dates) # close) > |value error: dates > > Probably this... > > dates=: 0{"1 }. arr > ]c=. +/ 0.02 < | pctChange d=. ((1999 < ". 4&{. every dates) # close) > 354 > > I could muse at length about market volatility issues, but my short > form would be that the efficient market hypothesis is almost certainly > incorrect and that economists have had a tradition of putting too much > emphasis on the wrong issues. > > Still, > > (I. _.05 > pctChange periodClose) { (_10({.)\ dates) > > is interesting to look at > > I notice, for example, that 1987-10-23 is in there (and I think I > remember hearing some traders from Morgen Stanley saying they cleaned > up right around then, because the total valuation of the stock market > overflowed 32 bits but they had been using integers in APL, which mean > their implementation overflowed to floating point - which meant their > stock trading programs just slowed down a bit while everyone else's > crashed. And I think that that was right around when they said that > that had happened...) > > I imagine that there's bits of history surrounding each of these > episodes, and that each crash corresponds to someone "cleaning up". > > Thanks, > > -- > Raul > > On Tue, Sep 1, 2015 at 3:53 PM, Joe Bogner <[email protected]> wrote: > > I recently invested in an S&P index fund - great timing, I know, so I > > figured I'd let J help calm my emotional response to the recent market > > volatility. > > > > > > The questions on my mind are: > > > > 1. Is this level of daily volatility extreme? -- restated as how common > are > > 2% swings in daily close price (SOLVED) > > > > 2. Is it uncommon to have this much volatility over a two week period? -- > > restated as how common are streaks of 2% change in price across multiple > > weeks (SOLVED) > > > > 3. How long does it take to recover after a period over period change of > > -5% (SOLVED) > > > > I set out to answer these questions: > > > > Step 1: Let's get some data > > > > require 'web/gethttp' > > > > NB. beware linebreak > > csv=: gethttp ' > > > http://real-chart.finance.yahoo.com/table.csv?s=%5EGSPC&d=8&e=1&f=2015&g=d&a=0&b=3&c=1950&ignore=.csv > > ' > > > > arr=: > ',' cut each LF cut csv > > > > NB. closing price is 4th column > > close =: ". every 4{"1 }. arr > > > > > > QUESTION 1 - Daily volatility > > > > pctChange =: }: -&1@% }. > > > > NB. 708 days greater than 2% change from previous close > > (+/ 0.02 < | pctChange close) > > 708 > > > > NB. 4% of the time since 1950 > > (+/ 0.02 < | pctChange close) % (# close) > > 0.042852 > > > > > > NB. 354 times since 2000 > > ]c=. +/ 0.02 < | pctChange d=. ((1999 < ". 4&{. every dates) # close) > > 354 > > > > NB. which is 9% of time... > > c%(#d) > > 0.0898477 > > > > so +/ 2% swings is not exceptional, relatively uncommon but nothing to > > worry about > > > > QUESTION 2 - Volatility over a two week period > > > > NB. calculate the period close over 5 sessions > > periodClose=: _5({.)\ close > > > > NB. count of consecutive periods with more than 2% volatility > > NB. > http://www.jsoftware.com/pipermail/programming/2014-October/039985.html > > [ c=.+/ 2 < ([*+)/\.&.|. d=. (0.02 < | pctChange periodClose) > > 131 > > > > NB. 4% of the periods since 1950 had 3 or more weeks of 2% volatility > > c % (#d) > > 0.0396489 > > > > NB. since 2000 > > periodClose=: _5({.)\ ((1999 < ". 4&{. every dates) # close) > > > > NB. 50 times there have been 3 or more weeks of 2% volatility since 2000 > > [ c=.+/ 2 < ([*+)/\.&.|. d=. (0.02 < | pctChange periodClose) > > 50 > > > > NB. 6.3% of the time > > c % (#d) > > 0.0635324 > > > > QUESTION 3 - Given a decline of 10% in two weeks, how long does it take > to > > recover? > > > > NB. close at the end of 10 sessions > > periodClose=: _10({.)\ close > > > > NB. 59 times there's been more than 5% decline over 10 sessions > > [ +/ _.05 > pctChange periodClose > > > > > > These are some of the dates: > > > > ,. (I. _.05 > pctChange periodClose) { (_10({.)\ dates) > > +----------+ > > |2015-08-31| > > +----------+ > > |2011-08-10| > > +----------+ > > |2010-07-01| > > +----------+ > > |2009-03-11| > > +----------+ > > |2009-02-25| > > +----------+ > > |2008-11-12| > > +----------+ > > |2008-10-15| > > +----------+ > > |2008-09-17| > > +----------+ > > |2008-01-15| > > +----------+ > > |2003-01-28| > > +----------+ > > > > This is starting to look more concerning.... > > > > Now, to find how long it takes to recover: > > > > Here are 10 instances: > > > > 10 {. (I. _.05 > pctChange periodClose) > > 0 102 130 163 164 171 173 175 192 317 > > > > A crude attempt: > > > > NB. find # of 10 day periods between period before y and period closing > > above that close > > findRecovery =: 3 : 'y - >./ I. ((y+1) { periodClose) < ((i. y) { > > periodClose)' > > > > > > NB. number of weeks between the 5% decline over two weeks and the > recovery > > 2 * findRecovery every 10 {. (I. _.05 > pctChange periodClose) > > _ 22 10 2 8 28 72 116 252 16 > > > > > > 22 weeks to recover since the last time it was > 5% decline over two > weeks > > > > Graphical (from yahoo charts) > > http://pasteboard.co/k1aGw1i.png > > > > I guess I'm not feeling as great now :), but at least it's not that > > uncommon and in each case it recovered > > > > Hope this helps anyone looking at today's market and/or wishing to play > > with J > > > > Happy to take any feedback on the approach as well. > > ---------------------------------------------------------------------- > > For information about J forums see http://www.jsoftware.com/forums.htm > ---------------------------------------------------------------------- > For information about J forums see http://www.jsoftware.com/forums.htm > ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
