In a previous blog post I described the use of virtual postings to track accidental personal/family expenses. I've always been uncomfortable with that, and in hledger 1yr I outlined a potential scheme for finally addressing the virtual posting problem.

separate journals

My outline built on top of continuing to maintain both personal and family financial data in the same place, but I've decided that this can't work, because the different "directions" (or signs) of accidental transactions originating from either the family or personal side can't be addressed with any kind of alternate view on the same data.

To illustrate with an example. A negative balance in family:liabilities:jon means "family owes jon". A coffee bought by mistake on the family credit card will have a negative posting on the credit card, and thus a positive one on the liabilities account. ("jon owes family"). That's fine.

But what about when I buy family stuff on a personal card? The other side of of the transaction is also going to have a positive sign, so it can't be posted to family:liabilities:jon: it would have to go to somewhere else, like jon:liabilities:family. Now I have two accounts which track versions of the same thing, and they cannot be combined with a simple transaction since they're looking at the same value from opposite directions (and signs).

Back when I first described the problem I was using a single journal file for all my transactions. After moving to lots of separate journal files (in hledger 1yr), it's become clearer to me that I don't need to maintain the Family and Personal data together, at all: they can be entirely separate journals.

getting data between journals

When I moved to a new set of ledger files for 2023, I needed to carry forward the balances from 2022 in the form of "opening balance" transactions. This was achieved by a report on the 2022 data, exported as CSV, and imported into the 2023 data (all following the scheme outlined by fully-fledged hledger.))

Separate Personal and Family journals need some information from each other, and I can achieve that in the same way as for opening balances: with an export of the relevant transactions as CSV, then imported on the other side. HLedger's CSV import system is flexible enough that we can effectively invert the sign of liabilities, addressing the problem above.

Worked example

We start with an accidental coffee purchased on the family card (and so this belongs to the Family ledger)

2022-08-20 coffee
    liabilities:creditcard           £ -3
    liabilities:jon:expenses:coffee  £  3

I've encoded the expense category that the Personal ledger will be interested in (the last bit, expenses:coffee) as a sub-account of the liabilities category that the Family ledger is interested in1 (the first bit, liabilities:jon). When viewed on the Family side, the expense category is not interesting, and we can hide it with HLedger's alias feature2:

    alias /^liabilities:jon(.*)$/ = liabilities:jon

It then looks like this from the Family side:

2022-08-20 coffee
    liabilities:creditcard           £ -3
    liabilities:jon                  £  3

This transaction (and others like it) are exported via

hledger reg -f family/2023-back.journal liabilities:jon: -O csv \
        jon/import/family/liabilities.csv

(The trailing colon on liabilities:jon: is important here!)

In the resulting CSV file, the running example transaction looks like

"55","2022-08-20","","coffee","liabilities:jon:expenses:coffee","£ 3.00","£ 3.00"

This is then converted into a journal file by hledger import. The rules file for the import is very simple: the fields date, description, account1 and amount are taken as-is; account2 is hard-coded to liabilities:family. The resulting transaction looks like

2022-08-20 coffee
    liabilities:jon:expenses:coffee  £  3
    liabilities:family               £ -3

Before this journal is included by the main one, we have to adjust the expense account, to remove the liabilities:jon: prefix. The import rules can't do this3 , so we use another journal file as a go-between with another alias rule:

    alias /^liabilities.jon:/ =

This results, finally, in the following transaction in the Personal ledger:

2022-08-20 coffee
    expenses:coffee                  £  3
    liabilities:family               £ -3

avoiding double-counting

There's one set of transactions that we don't want to export across this divide, and that's because they're already there: any time I transfer money from myself to the family accounts (or vice versa) to address the accrued debt, the transaction is visible from both my family and personal statements. To avoid exporting these and double-counting them, I make sure those transactions don't post to an account matching the pattern used in the hledger reg report. That's what the trailing colon is for: It ensures I only export transactions which are to a sub-account of liabilities:jon, and not to the root account liabilities:jon itself: which is where I put the repayment transactions. I could instead use a more explicit sub-account like liabilities:jon:repayments or similar, since the trailing colon is quite subtle, but this works for me.

Wrap up

I've been really on the fence as to whether the complexity of this scheme is worth it to avoid the virtual postings. The previous scheme was much simpler. I have definitely made some mistakes with it, which didn't get caught by the double-entry rules that virtual postings ignore, but they're for small sums of money anyway.

On the other hand, a lot of the "machinery" of this already existed for getting opening balances between calendar years, and the gory details are written down and hidden inside the Makefile. I also expect that I will continue to see advantages in having Family and Personal entirely separate, as they can each develop and adapt to their own needs without having to consider the other side of things every time.

It's a running experiment, and time will tell if it's a good idea.


  1. This scheme was originally suggested to me by Pranesh on Twitter (described in dues), but I discounted it at the time because of the exact arrangement they suggested, not realising the broader idea might work.
  2. I've hand-waved one problem with using hledger aliases here. If we use them as described, to hide the Personal expense details, we need them to not be applied when performing the CSV-generating report. Therefore, in practise I have them in a front-most family/2023.journal file, which imports the data from another family/2023-back.journal, and the CSV export is performed on the backing journal with the data and not the alias.
  3. HLedger import rules can't manipulate the fields from the CSV a great deal, but one change I proposed and started hacking on would allow for this: to expose Regexp match-groups as interpolatable tokens: https://github.com/simonmichael/hledger/issues/2009.

Comments