longinvest wrote:Hi LadyGeek,
The ROUND()s were added by design in the spreadsheet. I round percentages to the nearest .1% and all $ withdrawal amounts to the nearest $1000.
This was done in response of being accused of false precision. We are modeling the withdrawal of thousands of dollars out of a portfolio using historical data that doesn't even provide for enough accuracy (transactions costs, spread, taxes, etc.). If rounding a few dollars out or in (during withdrawals) breaks the plan, then there's no plan. Note that I keep more precision tracking portfolio returns (I round to the nearest 1$, much more precise than withdrawals).
The Backtesting sheet cannot model the withdrawal out of pocket money (a few thousand dollars), only out of a portfolio (hundred of thousand dollars). I could do as you suggest, but I somehow like the current intentional imprecision; it is a philosophical message to VPW users.
Now I see your intentions, which are clear. To ask this in another way - what is the minimum portfolio amount where the rounding has no effect? I think this occurs above $600,000.
Philosophical messages are fine, but they should be documented. I would not have encountered this "feature" if I did not try these lower portfolio values. Can you perhaps restrict the minimum investment value to avoid these discrepancies?
The users need to know that accuracies are reduced below a certain threshold. It's very difficult to see the effects, especially when the Constant Dollar Withdrawal threshold starts cutting off earlier than the others - is it real or the effects of truncation?
An idea: Instead of a fixed "0" or "3" rounding error, replace it with a cell that can change depending if we want rounding enabled (0 or 3 place precision) or not (14 places, perhaps). Then, just change the referenced cell's value. I'll take a look at this (and will share the result if it looks promising). BTW, I'm using MS Excel for this round of testing.