

FlashChat  Actuarial Discussion  Preliminary Exams  CAS/SOA Exams  Cyberchat  Around the World  Suggestions 
DW Simpson Global Actuarial & Analytics Recruitment 

Thread Tools  Search this Thread  Display Modes 
#51




Quote:
__________________
It was a surprisingly short journey from Robert E Lee to Dr Seuss.  @iowahawkblog But the bloody heirloom ensures the last laugh.  TaNehisi Coates. First of all, I think it’s very hard to untangle the motives of this shooter. As we’ve seen in a whole range of incidents with mass shooters, they are, by definition, troubled. [...]So I think the danger, as I said, is that we somehow suggest that the act of a troubled individual speaks to some larger political statement across the country. מַרְבֶּה נְכָסִים מַרְבֶּה דְאָגָה 
#52




Quote:
Quote:
In some ways, the decimal representation of numbers (a la mainframe/Cobol) had its advantanges when doing Dollars and Cents. Otherwise, "totals may not agree  due to rounding."
__________________
Carol Marler, "Just My Opinion" Pluto is no longer a planet and I am no longer an actuary. Please take my opinions as nonactuarial. My latest favorite quotes, updated Apr 5, 2018. Spoiler: 
#53




Quote:
The very concrete example was doing a weighted average (xA + yB)/(A+B), where x and y are rather small (and of similar magnitude to each other) and A and B are rather large (and of similar magnitude to each other). We were finding a loss of significant figures because the calculations were being calculated as above. When you multiply and divide numbers of vastly different magnitudes, you start losing sig figs. And this can propogate, of course  we're doing 50year runs (at least), and we want to show inforce liabilities running out, not persisting due to weird calculation stuff. We thought of various fixes, but the one that seems to be the most promising was to calculate p = A / (A + B) and then do the weighted average as xp + y(1p), which would keep all the numbers closer together in magnitude (p is in a reasonable range, like 20 %  80%). Another one that a friend told me about was something like x*y*A, where x & y are both very small and A is very large. x * y could give you underflow, so you change it to x * A * y. When I did a high school internship with a stats prof, I had to do stuff like this, as he had me program (in Fortran) discrete probabilities, such as the Poisson distribution, and he wanted me to calculate it exactly, not just do various continuous approximations. Thing is, if you try to calculate where k is large and lambda around 10 (say), you can get in trouble really quickly. So I started with e^{\lambda} and did a loop where on each step I multipled by lambda / i, where i was marching down from k to 1. Worked really well. So that's in avoiding an overflow problem. 
#54




The problem of precision goes way beyond Excel, and few people understand floating point calculations well enough to be aware of the issues. If you absolutely have to work with exact precision, then you've no choice but to program it yourself using the GMP library (or similar).
Insofar as Excel's precision, if I am ever concerned about the results, I fire up Mathematica and check the calculations. 
#55




OK, I take it back. The problem isn't with floating point. It's with the fact that our money is in decimal fractions. We should go to a binarytype system for amounts less than $1.00. Let's see:
Half dollar. . . check Quarter . . . . check Bit * . . . . . . check * As in "two bits" of course. It would be worth $0.125 under the present system. I envision it as being a lot like our current dime, but sufficiently different not to cause the confusion that some dollar coins produce. . . This new approach gets rid of the annoying pennies. It also does away with nickels  that might not be a bad thing either. Of course the stock market recently went through a huge conversion in the other direction.
__________________
Carol Marler, "Just My Opinion" Pluto is no longer a planet and I am no longer an actuary. Please take my opinions as nonactuarial. My latest favorite quotes, updated Apr 5, 2018. Spoiler: 
#56




Spreadsheets: Formulas for Success
http://www.cfo.com/article.cfm/11773...todayinfinance Quote:

#57




"Corporate managers are probably more willing to discuss their sex lives than the state of spreadsheet controls within their firms, one of the organisers said. "
The people on this board may feel the same way.
__________________
Carol Marler, "Just My Opinion" Pluto is no longer a planet and I am no longer an actuary. Please take my opinions as nonactuarial. My latest favorite quotes, updated Apr 5, 2018. Spoiler: 
#58




Ratio of threads discussing sex life to threads discussing spreadsheet controls on this forum (estimated): 78,457:1
__________________
Visit thing's web empire: Printable Sudoku Puzzles & 12 by 12 Sudoku & 6 by 6 Sudoku & Nurikabe 
#59




Maybe we should ask for a new forum. . . "Sex, Lies, and Spreadsheet Controls"

#60




Quote:
* . . . and fantasies, wherever they appear, such is in Cyberchat surveys.
__________________
Carol Marler, "Just My Opinion" Pluto is no longer a planet and I am no longer an actuary. Please take my opinions as nonactuarial. My latest favorite quotes, updated Apr 5, 2018. Spoiler: 
Tags 
machine learning, predictive analytics, speadsheet error, spreadsheets 
Thread Tools  Search this Thread 
Display Modes  

