In general Cauchy distribution doesn't have standard deviation defined, though it should be possible to calculate it for a given interval.

This is the formula that I'm trying to use:

PDF for Cauchy is

The mean for Cauchy is 0, so taking 0.4 as a scale and [-3;3] as interval we get an integral

Which gives a value of 0.6174 = (0.7858)^2

But when I generate any number of Cauchy values with scale 0.4 (I was trying to generate 10,000 and 10,000,000 numbers - it didn't matter) and filter-out all the numbers which absolute value is greater than 3 it always gives me SD = 0.8198 which is completely different.

Any ideas on what am I doing wrong? Sorry to ask such a silly question.

  • 4
    $\begingroup$ You are looking at a truncated Cauchy distribution. That means you need to recalculate the normalizing constant for the PDF. That is, you need to divide the untruncated pdf by $P( -3 < T < 3)$. Once you use this new normalizing constant, your formula for the standard deviation should work out. $\endgroup$ – Cliff AB Apr 27 '18 at 5:12
  • $\begingroup$ Could you please clarify what you mean by "normalizing constant"? $\endgroup$ – andkorsh Apr 27 '18 at 5:22
  • 2
    $\begingroup$ A constant that you multiple the PDF by to insure that it integrates to 1. See here. $\endgroup$ – Cliff AB Apr 27 '18 at 5:31
  • $\begingroup$ Unless you normalize to make the distribution a true truncated Cauchy it is not appropriate to call the integral a variance. $\endgroup$ – Michael R. Chernick Apr 27 '18 at 5:35
  • $\begingroup$ Actually by "truncated Cauchy distribution" I've found a paper which gives another PDF (even without PI in it) and by integrating this I've managed to get the required number. Thanks a lot! projecteuclid.org/download/pdfview_1/euclid.bjps/1291387776 $\endgroup$ – andkorsh Apr 27 '18 at 5:46

Cliff helped me a lot and I've found a nice paper about truncated Cauchy distribution https://projecteuclid.org/download/pdfview_1/euclid.bjps/1291387776

Solving the integral for symmetrical interval [-A; A] we get:

$$\delta = -\frac{\gamma (arctan(\frac{A}{\gamma})*\gamma-A)}{arctan(\frac{A}{\gamma})} $$

$$MAD = \frac{\gamma*ln(\frac{\gamma^2+A^2}{\gamma^2})}{2*arctan(\frac{A}{\gamma})}$$

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.