Reporting: weighted average price source

[Charles Day]

On Tue, Jul 8, 2008 at 5:59 AM, David G. Hamblen <dhamblen at roadrunner.com>
wrote:

> Charles Day wrote:
>
>> On Mon, Jul 7, 2008 at 8:53 AM, Derek Atkins <warlord at mit.edu <mailto:
>> warlord at mit.edu>> wrote:
>>
>>    "David G. Hamblen" <dhamblen at roadrunner.com
>>    <mailto:dhamblen at roadrunner.com>> writes:
>>
>>    > A few years back (v1.8x), I had problems with these absolute
>>    values, and I
>>    > patched report-utilities.scm,and commodity-utilites.scm so that
>>    the balance
>>    > sheet would balance.  In addition to completely removing all the
>>    numeric:abs,
>>    > I also had to do something about the division by zero in
>>    commodity-utilities
>>    > when there was a zero share balance.  I'm using the "Nearest in
>>    Time"  and the
>>    > problem went away in the 2.x updates.  If anyone's interested, I
>>    can dig up
>>    > my old postings.
>>    >
>>    > Anyway, I vote for getting rid of absolute values in bookkeeping.
>>
>>    It's not a question of book keeping.  It's a question of computing
>>    the share price.  GnuCash stores buys as a positive and sells as
>>    a negative.  As Christian pointed out before, if you buy x shares
>>    for $y and then later SELL x shares for $y then a non-abs weighted
>>    average gives you a price of 0/share!  Obviously this is wrong.
>>    The weighted average should be $(y/x) per share.   But without
>>    the ABS you get:
>>
>>      y * x + (-y)*x        xy -xy       0
>>      --------------  ==    ------  ==  --- = 0
>>         x + x                2x         2x
>>
>>    When you use abs() here you get the right answer.
>>
>>  The fundamental issue here is that I don't think that the average price
>  should include shares I no longer own.
>
> My tests (all using the Balance Sheet) indicate that share prices (the y's)
> are all positive and are calculated from each transaction (Total Buy/Total
> Shares). The Total Buy (the x*y product) and the Total shares (the x's) are
> both positive or both negative.  What seemed to be happening back in 1.8x,
> and is still happening in 2.4) was that when you have zero shares, the
> answer for the weighted average is irrevelant, but I have a situation where
> I sold all my shares (as in your example), but subsequently purchased a few
> more shares of the same stock.  The weighted average (without all the
> abs()'s) looks like this: (I can't wait to see what Thunderbird does with my
> ascii equations :-) )
>
>  y * x + y * (-x) + z * w               z * w
>  -----------------------------------    ==    ---------     ==  w
>          x + (-x) + z                          z
>
> With the abs(),  I get a nonsensical average price for my holdings in this
> asset.  In my specific test case, with the initial purchase/sell of 100
> shares at $5 (as in your example), and a subsequent purchase of 10 shares at
> $20, I get a weighted average price of $5.714, and an asset worth $57.14.
>  Using the "Nearest in Time" or the "Most Recent"  price source, I get an
> average share price of $20, and an asset value of $200.
> Without the abs(), you get 0/0 for the weighted average in your example, so
> that needs to be trapped.
>

So far as I can tell, this formula is the same as the new price source named
"Average Cost" that I added yesterday (r17266).


>
> The good news is that in 2.4, the Balance Sheet balances in all cases (not
> true in 1.8x).
>

Unless you have unrealized gains or losses on liabilities. I'm working on
that...

-Charles

    for sales.
>>
>>    So it's not a question of absolute values in bookkeeping.  It's
>>    a question of absolute values in computing share prices.  Not
>>    the same thing.
>>
>>
>> I agree with Derek. That is, the "weighted average" price source is not
>> intended to compute the cost of your holdings, but rather to compute the
>> volume-weighted average price of all buys AND sells. So it needs to keep
>> using the absolute value. However, it does need to be changed: it should
>> ignore exchanges with a zero "amount" in the split.
>>
>> So I think we need to add a new price source of "Cost" which computes
>> without absolute value and does include zero "amount" splits. That's where
>> taking a look at David's code might come in handy (though implementing this
>> seems like a pretty simple job: copy existing weighted average function, get
>> rid of absolute value, add report option).
>>
>> Whether to even keep the "weighted average" price source is a question
>> worth asking; I don't think it is a useful way to revalue current holdings.
>> Historical volumes don't seem relevant. Even if the volume weighting was
>> removed, the formula also weights quotes that are clustered within a small
>> period of time more heavily than quotes that are spread apart. In other
>> words, if I have 30 quotes from this month and one quote from last year, the
>> recent quotes practically drown out the quote from last year. If we want to
>> aim for a historical price average then using something like linear
>> regression seems like a better way to go. Could even add some time factors,
>> like getting the midpoint of the portion of the line covering the last 30,
>> 60, 90 days, whatever. But now I am brainstorming new price sources.
>>
>> I do think a volume-weighted average price can be useful; it's just more
>> appropriate for measuring your personal trading performance. Maybe there
>> should be a performance report designed around that.
>>
>> -Charles
>>
>>    derek
>>    --
>>          Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
>>          Member, MIT Student Information Processing Board  (SIPB)
>>          URL: http://web.mit.edu/warlord/    PP-ASEL-IA     N1NWH
>>          warlord at MIT.EDU <mailto:warlord at MIT.EDU>
>>   PGP key available
>>
>>
>>    Of course, then you need to make sure you re-apply the negative
>
>
>