It depends heavily on two things: how many symbols you record and how fine the resolution is. Tick-level recording across a wide universe grows much faster than one-minute bars on a short watchlist — the difference is more than an order of magnitude, so a single number would mislead you. Practical approach: record your intended universe for one full session, check the folder size, and multiply. That gives you a number that matches your actual setup rather than an average that matches nobody's.
The real question is what you'll ever do with three months of tick data. I recorded a mountain of it once, opened it twice, and both times I just squinted at candles anyway.
Learned this the hard way back when drives were measured in single digits, the file grows quietly until the day it doesn't. Set yourself a retention window now, not when the disk yells at you.
Add a checklist item: check free space weekly, and decide in advance what gets deleted first. A recorder filling a drive at the wrong moment is the kind of boring failure that ruins an otherwise fine day.
I once had a recording job die silently because the disk was full and I didn't notice for weeks. Went looking for a backtest window and there was just... nothing there. Not advice, but check that the thing is still writing.
Curious what you actually want the data for, Pete. I found I was hoarding it partly to feel prepared, and the hoarding itself was the comfort, not the analysis.
Start narrow!! Record the handful you actually trade, run it a week, look at the folder size, then extrapolate. Way better than guessing a drive size today!
Dumb question maybe — if the plan is eventually to train something on it, does tick data actually help or is minute bars enough? I keep reading that more granular = better but I honestly don't know if my little models would even use it. Might just do MomoQueen's week test on 5 symbols and see what the folder says before I get ambitious.