"Fred Smith" wrote in message
...
"XIRR() __function__ should not return a rate that fails to cause the
(X)NPV to be close to zero."

I agree. When this happens, XIRR should return #NUM instead (as Help
states), instead of a fictitious number.
[....] Sorry if I offended you with my previous post. It was unintended.

No offense taken. I was just frustrated because I had tried to make this
very point several times in two threads, but I was not making myself
understood.

Anyway, I'm glad to see we are on the same page now. Thanks for posting
back.


----- original message -----

"Fred Smith" wrote in message
...
"XIRR() __function__ should not return a rate that fails to cause the
(X)NPV to be close to zero."

I agree. When this happens, XIRR should return #NUM instead (as Help
states), instead of a fictitious number. This would make it easier to
determine the required fix, which is almost always a better guess. When
XIRR returns a fictitious number, the fix isn't nearly as obvious.

Sorry if I offended you with my previous post. It was unintended.

Regards,
Fred.


"JoeU2004" wrote in message
...
Clarification....

"JoeU2004" wrote:
the Excel documentation disagrees with you, not in silence,
but by explicit instruction. It states: "In most cases you do
not need to provide guess for the XIRR calculation".
[....]
"The rate of return calculated by XIRR is the interest rate
corresponding to XNPV = 0"

Arguably, these may or may not be product "specifications" per se.

Most user documentation these days is written by someone who has an
English degree, and sometimes not even that :-(. Gone are the days when
documentation was written by the product developer, or at least by a
"tech writer" who used to develop similar products, who had access to
source code or hardware drawings, and who was even expected to "test"
his/her documentation against the product ("what a concept"!). (I'm
talking about at large corporations like MS.)

So someone might reasonably argue that the Excel Help pages are not
product "specifications" per se, but simply one writer's interpretation
of how he/she thought the product works. Hopefully, that is based on
internal documentation (true specifications) written by the product
developer. But sometimes, it is the result of the writer's own usage of
the product.

But usually, the product developer is responsible for reviewing the
writer's documentation. And usually, the mistakes that get by are errors
of omission, not assertive statements that the product this or that. In
my experience, end users tend to treat the word of the user documentation
as a product specification. (I am speaking from painful experience :-<.)

All that aside, I think it is common sense that the XIRR() __function__
should not return a rate that fails to cause the (X)NPV to be close to
zero. And again, I am not saying that the internal __algorithm__ should
not converge on such invalid rates. It might; that's the nature of the
beast. I am simply saying that any invalid rates found by the algorithm
should be hidden from the user, substituting an error condition instead.

I hope my intention are clearer.


----- original message -----

"JoeU2004" wrote in message
...
"Fred Smith" wrote:
The silver lining to their laziness is that you and I
get to pontificate on the machinations of XIRR.

<g


As we both agree, MS itself could have programmed
XIRR to pick a different guess, and try again.

Where in my "pontification" do you think I said anything like that. I
explicitly did not disparage the XIRR approximation algorithm. On the
contrary, I wrote: "I presume that it is working as well as might be
expected".


Why can't users figure out there's a reason XIRR has a guess parameter?

Because the XIRR() function fails to let them know when they should. If
you are saying that we should always provide "guess", well, the Excel
documentation disagrees with you, not in silence, but by explicit
instruction. It states: "In most cases you do not need to provide
guess for the XIRR calculation".


Newton-Raphson can't be blamed for poor programming on the part of MS.


I explicitly said as much, too.


Why didn't MS do a better job of programming XIRR?
Because they're programmers.

Spoken by someone who apparently knows nothing about computer
programming, at least not professionally. I was a professional
programmer and system architect for 35 years. Part of my
responsibility, in effect, was to specify what can expect of
programmers.


It meets the spec

I have quoted Excel's specs. I think it is clear that you are
incorrect. It says: "The rate of return calculated by XIRR is the
interest rate corresponding to XNPV = 0". Don provides an example where
that specification is not met: the rate of return calculated by XIRR is
not a rate that causes the XNPV to be zero.

All I was saying is: XIRR() should return an error when its algorithm
computes a discount rate that does not cause the NPV to be (close to)
zero. That does not disparage the algorithm. It is to be expected that
the algorithm will miss the mark sometimes, as you say.


----- original message -----

"Fred Smith" wrote in message
...
Joe,

Glad you were able to help Don. My comments a

As is usual, a more reasonable guess solved the problem. Why can't
users figure out there's a reason XIRR has a guess parameter? Because
they're users -- they just want it to work. Why didn't MS do a better
job of programming XIRR? Because they're programmers. It meets the
spec, so why put in extra work.

My bet is that over 90% of the errors in using XIRR can be fixed by
using a guess of 10% when the total cash flow is positive, and -10%
when it's negative. In fact, I always use a guess of
10%*sign(sum(cashflows)). It would have been so simple for MS to do the
same, but unfortunately, they didn't.

Newton-Raphson can't be blamed for poor programming on the part of MS.
Any calculation of the ROI for irregular investments must use iteration
to find the result. There are several iteration algorithms available;
NR just happens to be the fastest, which is why everyone uses it.

As it's been explained to me, it's like finding the edge of the lake
from a starting point on the shore. You calculate the tangent of the
curve you are on as the next starting point. You continue until you
find the lake. Unfortunately, if you start in a hollow, you go off in
the wrong direction, and never converge towards the lake. In this
situation, you need to pick a different guess. It doesn't have to be a
*better* guess, just different (using my analogy, somewhere out of the
hollow).

As we both agree, MS itself could have programmed XIRR to pick a
different guess, and try again. They could easily have calculated a
guess based on the ratio of the total positive cash flows to the total
negative. But they didn't.

The silver lining to their laziness is that you and I get to
pontificate on the machinations of XIRR.

Regards,
Fred

"JoeU2004" wrote in message
...
Don acknowledged in email that the -400% that he observed was due to
human error.

Nonetheless, for the cash flow stream that he sent me (below), XIRR()
returns about 0.0000002980%. That is obviously incorrect, since
XNPV() returns about -43,338.86, which is not even close to zero.

The root cause is the need for a "guess" argument (Don did not have
one), and a correct "guess" argument at that.

Reducing the irregular daily cash flows to monthly net cash flows (see
below), we see a series of negative cash flows (investments) and one
positive cash flow (return). Since the sum of the investments
(about -136K) is much larger than the return (about 96K), we expect a
negative IRR. So I tried a "guess" of -1%, and XIRR()
returned -24.83%. Plugging the exact IRR into XNPV() using the daily
cash flow, the result is indeed close to zero. QED.

Don still wonders, reasonably, why XIRR() returns such an obviously
incorrect rate when the "guess" was missing.

I invite people like Fred Smith and Myrna Larson to comment on the
behavior of the Newton-Raphson method of approximation, or whatever
method Excel might use. We all know that these methods can take a
wrong turn under adverse conditions. But some details might be
interesting.

In any case, I would call this a defect in the XIRR() implementation.

No, I am not disparaging your beloved approximation method. I presume
that it is working as well as might be expected.

Instead, I am disparaging the witless MS programmer who does not seem
to know how to write an "if" statement just before exiting the XIRR()
function, of the form: "if there has not been an error and the XNPV
with the last result is not close to zero, return an error, per the
function specification".

I am referring to the XIRR Help page, which states: "If XIRR can't
find a result that works after 100 tries, the #NUM! error value is
returned". Now, we might argue about what "close to zero" means
exactly. But I am sure that -43,338.86 does not meet anyone's
reasonable definition. So clearly, the XIRR() result does not "work",
even it was found in fewer than 100 iterations.

If XIRR() had returned the #NUM! error, as it should in this case, Don
might still have been perplexed. But at least his question would have
been much less mysterious; probably something to the effect of: "how
in the heck am I supposed to know what ``guess`` should be?", and
"why does Excel need this, but my HP 12C does not?". We've dealt with
such questions before.

The following is Don's original data and my monthly net cash flow
approximation.

Don's cash flow (my apologies for the poor alignment; there are too
many to adjust manually):

3/27/2006 -3,994.58
4/1/2006 871.42
5/1/2006 871.52
5/3/2006 -9,728.80
6/1/2006 870.51
6/2/2006 -4,864.40
7/1/2006 870.04
7/5/2006 -4,864.40
8/1/2006 869.06
8/3/2006 -4,864.40
9/1/2006 868.53
9/6/2006 -4,864.40
10/1/2006 868.53
10/4/2006 -4,924.67
11/1/2006 867.95
11/6/2006 -4,966.81
12/1/2006 866.79
12/4/2006 -4,966.81
1/1/2007 866.19
1/8/2007 -4,966.81
2/1/2007 866.19
2/5/2007 -4,966.81
3/1/2007 910.86
3/5/2007 -4,864.40
4/1/2007 910.20
4/5/2007 -4,864.40
5/1/2007 909.34
5/3/2007 -4,864.40
6/1/2007 908.50
6/4/2007 -4,864.40
7/1/2007 908.02
7/5/2007 -4,864.40
8/1/2007 907.87
8/3/2007 -4,864.40
9/1/2007 906.91
9/4/2007 -4,864.40
10/1/2007 905.77
10/3/2007 -4,924.67
11/1/2007 905.17
11/5/2007 -4,966.81
12/1/2007 904.60
12/6/2007 -4,966.81
1/1/2008 904.60
1/7/2008 -4,966.81
2/1/2008 904.79
2/6/2008 -4,966.81
3/1/2008 942.92
3/6/2008 -4,864.40
4/1/2008 941.97
4/3/2008 -4,864.40
5/1/2008 940.92
5/5/2008 -4,864.40
6/1/2008 940.48
6/2/2008 -4,864.40
7/1/2008 941.24
7/2/2008 -4,864.40
8/1/2008 941.24
8/4/2008 -4,864.40
9/1/2008 940.68
9/5/2008 -4,864.40
10/1/2008 942.05
10/6/2008 -4,924.67
11/1/2008 944.31
11/3/2008 -4,966.81
12/1/2008 945.74
12/3/2008 -4,966.81
12/28/2008 92,520.92

Monthly net cash flow approximation (my apologies for reversing the
columns):

-3,123.16 4/1/2006
-8,857.28 5/1/2006
-3,993.89 6/1/2006
-3,994.36 7/1/2006
-3,995.34 8/1/2006
-3,995.87 9/1/2006
-4,056.14 10/1/2006
-4,098.86 11/1/2006
-4,100.02 12/1/2006
-4,100.62 1/1/2007
-4,100.62 2/1/2007
-3,953.54 3/1/2007
-3,954.20 4/1/2007
-3,955.06 5/1/2007
-3,955.90 6/1/2007
-3,956.38 7/1/2007
-3,956.53 8/1/2007
-3,957.49 9/1/2007
-4,018.90 10/1/2007
-4,061.64 11/1/2007
-4,062.21 12/1/2007
-4,062.21 1/1/2008
-4,062.02 2/1/2008
-3,921.48 3/1/2008
-3,922.43 4/1/2008
-3,923.48 5/1/2008
-3,923.92 6/1/2008
-3,923.16 7/1/2008
-3,923.16 8/1/2008
-3,923.72 9/1/2008
-3,982.62 10/1/2008
-4,022.50 11/1/2008
-4,021.07 12/1/2008
92,520.92 1/1/2009


----- original message -----

"Don Kline" wrote in message
...
I have a XLS file I can send you. I don't know how to attach the XLS
file to
the discussion. Please advise how I can get this to you.


"JoeU2004" wrote:

"Don Kline" wrote:
I have a series of monthly transactions over several years.
I have used the XIRR function to find the ROR for each month,
each quarter, and each calendar year.

That's a neat trick, since XIRR only returns the annualized rate. I
hope that is what you mean. But I wonder if you are misinterpreting
the results from XIRR.


The client now wants a rate from inception. My expectation
was I would get an annual rate for the entire period. However
I feel that a -400% ROR is not accurate.

Is there anything different I need to do when the period is
greater than one year?

Yes. But I cannot tell you what that is because I'm not a
mindreader. If you post an example, you might get some help.