SERVING THE QUANTITATIVE FINANCE COMMUNITY

 
User avatar
evansjtriceris
Topic Author
Posts: 7
Joined: July 14th, 2002, 3:00 am

Normal distribution > 1 for small sigma?

September 13th, 2006, 6:27 pm

Has anyone run across this problem before?If the normal PDF is defined as 1/ [ sqrt(2*pi) * sigma ] * e^.... , why is the PDF mis-behaved for small values of sigma?For example - assume mu = 0 and we evaluate the PDF for x=0 (eg: e^... =1).Then, assume sigma is small - 0.01 etc. -- the PDF evaluates to 39.89 (same result occurs for the NORMDIST func. in Excel)... But, by definition, no value of the PDF can be > 1 as the integral from -inf to +inf = 1The references I've checked all say sigma must be strictly greater than zero -- but, isn't 0.3989... actually the lower limit?After a day of trying to figure this out, I'm sure there's an extremely simple answer... Thanks,Jeff.
 
User avatar
Rez
Posts: 440
Joined: May 28th, 2003, 9:27 pm

Normal distribution > 1 for small sigma?

September 13th, 2006, 6:52 pm

There isThe CDF has to be below one not the PDF ...CDF(x) = integral[ PDF(u) du, u= -infinity to x]K.
 
User avatar
evansjtriceris
Topic Author
Posts: 7
Joined: July 14th, 2002, 3:00 am

Normal distribution > 1 for small sigma?

September 13th, 2006, 7:08 pm

Right.. but doesn't that imply the PDF would have to take negative values somewhere for the CDF to sum to 1?Eg: if the PDF for sigma = 0.10 @ x=0 is 39.89, the PDF has to be negative somewhere else to pull the sum back below 1... but this creates negative probability events..
 
User avatar
iadams
Posts: 25
Joined: September 5th, 2006, 10:55 am

Normal distribution > 1 for small sigma?

September 13th, 2006, 8:24 pm

No: it's an integral not a sum.As sigma tends to 0, the PDF becomes narrower and taller in the middle, but its integral over the Real domain remains at 1, and the PDF becomes a Dirac delta function.
 
User avatar
evansjtriceris
Topic Author
Posts: 7
Joined: July 14th, 2002, 3:00 am

Normal distribution > 1 for small sigma?

September 14th, 2006, 11:28 am

Except that doesn't work... for example... ( 1/ { sqrt(2pi) *sigma } ) * e^ ... Evaluate for mu=0, x=0, so e^.. = 1Then, as sigma -> 0, 1/ (sqrt(2pi)*sigma) tends to infinity
 
User avatar
spv205
Posts: 478
Joined: July 14th, 2002, 3:00 am

Normal distribution > 1 for small sigma?

September 14th, 2006, 12:43 pm

as iadams said its an integral not a sumThe PDF of a discrete random variable cannot be higher than 1, since effectively your width is 1, thats not the case for a continuous random variableyou have infinite height for an infinitely small width so the area you get is 1.so the Probability that x is between +/- 1/2 for instance is still finite
 
User avatar
iadams
Posts: 25
Joined: September 5th, 2006, 10:55 am

Normal distribution > 1 for small sigma?

September 14th, 2006, 2:59 pm

Try here. You'll find reference to the Normal function about 2/3 down the page, but read the top section explaining the concept first.The PDF is very large across a very small width, resulting in a constant integration result of 1. Remember that this is a continuous function.
 
User avatar
croot
Posts: 104
Joined: July 23rd, 2006, 8:30 pm

Normal distribution > 1 for small sigma?

September 17th, 2006, 12:20 am

I spy a post in the wrong section...This is the place for seasoned professionals' technobabble. Newbies, read (and learn) only!
ABOUT WILMOTT

PW by JB

Wilmott.com has been "Serving the Quantitative Finance Community" since 2001. Continued...


Twitter LinkedIn Instagram

JOBS BOARD

JOBS BOARD

Looking for a quant job, risk, algo trading,...? Browse jobs here...


GZIP: On