Obviously you can't accumulate cent by cent. You can't even safely accumulate by quarter. Epsilon is too large to do that. I calculate cumulative pnl using std::fma, then multiply AUM with that and round to cents. It's good enough for backtesting, and it shaves a bunch of seconds off the clock.
I see - I guess it's a financial modelling program or similar where the quantities don't represent precise values of money. I was imagining some kind of accounting-like app that would need to be reconciled with real-world balances.
Comments
I have used single precision floats in my latest project just to disprove this baloney.
You are using 32 bit floats to represent money?
Does your project correctly calculate $300,000.00 + $0.01, (or even just correctly represent the value $300,000.01) and if so, how?
Obviously you can't accumulate cent by cent. You can't even safely accumulate by quarter. Epsilon is too large to do that. I calculate cumulative pnl using std::fma, then multiply AUM with that and round to cents. It's good enough for backtesting, and it shaves a bunch of seconds off the clock.
I see - I guess it's a financial modelling program or similar where the quantities don't represent precise values of money. I was imagining some kind of accounting-like app that would need to be reconciled with real-world balances.
Yes, we work with futures and the largest transaction is around 1000 qty. It is so rare that most banks have a fat finger alert for such quantities.