More questions

[Greg Stark]

Benoit Grégoire <bock@step.polymtl.ca> writes:

> You would need a different script for each bank (and change them each time a 
> bank decide to rework their web interface).  That's an almost impossible 
> race...  

Well you only need one user of each bank to send in a script, i was picturing
a kind of contrib directory full of such scripts. But that's why I was leery
since these scripts would be handling acct#s and passwords.

> However, we do need a way to have gnuash be invoked as a helper application 
> from a web browser, so you at least don't have to go fishing for the file you 
> just downloaded.

I don't know. I would still have to go through several pages to get my bank
account statement, then another few pages to get my credit card statement. I
would only get around to it occasionally as I do now. It would be much
smoother if I could go to gnucash and choose "update statements".

> Well, it does, but only when looking for duplicates.  You are talking about 
> the account matching for autobalance which currently only use memo and 
> description.  I never thought about checks, which don't have a payee (at 
> least not electronically), so their memo and description is always the same.  
>
> However, I don't know how we could match on amount without wrecking havoc on 
> the matching.  Or even how gnucash would know when to match on amount.  It 
> has no way to know that the exact string match it just found is actually a 
> check (Well I suppose if a check number is present, but I don't know if all 
> online systems support that.   OFX does, but that info may not always be 
> sent).    
> 
> But tell me, you must write a LOT of checks for this to be a problem ;)  Humor 
> aside, I hate to end mails without a solution, sorry.  

No, I'm talking about deposits. I actually write hardly any checks, I use
interac for nearly everything. (Hi btw, I'm in Montreal too.) However all my
deposits just show up as ATM deposits exactly as if I had deposited cash. The
only way to match them is by amount.

Somehow it would have to notice that "Visa" in the description always means a
credit card payment regardless of amount whereas "Deposit" is a mostly
worthless symbol without a matching amount.

The naive approach is to check for not just one exact match for several. If it
searches and finds five exact matches that are all for the same account then
it has a match. If they're for five different accounts then it should try to
narrow the field by looking for more evidence.

The more sophisticated approach would be to throw a bayesian filter at all the
data attached to the transaction. That could conceivably even pick up on
things like "transactions on thursdays are more likely to be paychecks". But
it's probably overkill.

-- 
greg