OFX Import Matching Problem

[Thomas Baumgart]

On Freitag, 21. Juni 2019 22:55:29 CEST Brendan Coupe wrote:

> I'm running a week old build from the 5.0 branch on Fedora 29.
> 
> When I download my savings account transaction using online banking
> the paycheck frequently matches with a very old paycheck. This results
> in the splits being way off.
> 
> This happens when the amount of the new paycheck is not very close to
> the most recent paycheck which has been happening a lot lately due to
> reimbursed business expanses.
> 
> On the import tab of the ledge settings I have tried setting "Match
> transaction within days" from 7 days (paycheck is weekly) to 30 days
> and the same thing happens. KMM is definitely matching transactions
> that are much more than 30 days old. In fact the transaction that it
> matched was only $0.01 closer to the new transaction than the previous
> paycheck (difference was $8.29 versus $8.30). The transaction it
> matched is over 18 months old. It appears to be ignoring the "Match
> transaction within days" setting. it's simply matching the transaction
> from the same payee that is closest in value.
> 
> I'm pretty sure this is fairly new behavior but I'm not sure if it
> started with the initial version of KMM5 that I used or more recently.

This probably goes back to a change I made in January this year: 

   https://cgit.kde.org/kmymoney.git/commit/?id=447213e04d6e7ab9022caeb5c258800625036967

which added the part of choosing an ancient transaction based on the smallest difference in amount whereas before it only used old transactions if the amount was identical. 

Here's what I found in the code (which perfectly explains what you encounter):

In case the payee name has been found, the following will take place:

      // Fill in other side of the transaction (category/etc) based on payee
      //
		// [...]
		//
      // We'll search for the most recent transaction in this account with
      // this payee.  If this reference transaction is a simple 2-split
      // transaction, it's simple.  If it's a complex split, and the amounts
      // are different, we have a problem.  Somehow we have to balance the
      // transaction.  For now, we'll leave it unbalanced, and let the user
      // handle it.

For the category to be found, the first thing is to check if the payee has a default category assigned. If yes, it is taken and we're done. If not, all transactions for that payee in the account will be searched backwards. Note: no date filtering here, which certainly is the cause of the behavior you encounter. The algorithm then works as follows:

          // if there is more than one matching transaction, try to be a little
          // smart about which one we use.  we scan them all and check if
          // we find an exact match or use the one with the closest value

The scan works backwards with the last one being the default. So we have at least one transaction for that payee, and in case of multiple the one with the least difference in amount will be selected. Then we continue with:

                // in case the old transaction has two splits
                // we simply inverse the amount of the current
                // transaction found in s1. In other cases (more
                // than two splits we copy all splits and don't
                // modify the splits. This may lead to unbalanced
                // transactions which the user has to fix manually

The point is, that we are not talking about 'matching' at this point but automatic categorization of the imported transaction. Matching happens in the next step when KMyMoney tries to figure out if you already have the said transaction on file (entered manually for example). And it is for that matching that the interval is used, but not the automatic categorization happening in the step before. Matching actually means merge two transactions (the one on file and the imported one) into a single one. This is not what is happening for you and what you certainly don't want with older transactions.

I am not sure at this point what happens, if I increase the matching period beyond one month and another salary payment comes in and it matches. It is certainly not detected as a duplicate but does it match the transactions? I honestly don't know and have never tried.

Why did I implement the feature as it is: I receive two payments with very different amounts from the same payee each month and they differ in categories. One of the amounts varies each month and the other one is fix (we talk salary and reimbursement here as well, but I receive them in two payments). The old behavior was always wrong, because taking the last payment from that payee as categorization base is certainly false and only worked when there was no reimbursement (which means I received two salary payments in a row). So for me, a matching period of a few days is OK, but for the categorization I probably need a few months. The default to take the last one on file if nothing else was found is probably a good decision.

Would a new setting to limit the search for transactions to do the auto categorization help here? What would best describe it and what would be a neat name for it?

Any ideas, anyone?

-- 

Regards

Thomas Baumgart

https://www.signal.org/       Signal, the better WhatsApp
-------------------------------------------------------------
Software and cathedrals are much the same –
first we build them, then we pray. -- Sam Redwine, 1988
-------------------------------------------------------------
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 868 bytes
Desc: This is a digitally signed message part.
URL: <http://mail.kde.org/pipermail/kmymoney-devel/attachments/20190622/0daffd2f/attachment.sig>