The advice is everywhere. Before you risk real money, paper-trade for three months. Or six. Or a year. Use a simulator. Track your fills. See if you can run a profitable book on a hundred grand of pretend capital before you put a thousand of real capital into the same idea.
The advice is well-meaning. It is also answering a question almost nobody is actually asking.
What paper trading actually tests
A paper-trading account tests whether you can place orders, manage positions, and not blow yourself up on margin in an environment where the buttons and the data feed look real. Those are useful skills. They are operational skills — the kind you stop being bad at after a few hundred reps because the feedback loop is immediate. Wrong order type, wrong size, fat-fingered ticker, the simulator tells you within seconds.
That is not the question retail traders are usually trying to answer when they sit down with TradingView's paper account or Webull's simulator. The question they are usually trying to answer is am I any good at predicting what the market is going to do.
The simulator cannot answer that question. Not in a year. Not in five years. Not because the simulator is broken — because the question is wrong-shaped for it.
The thing that actually compounds is offline
Skill in markets, to the extent it exists for retail, lives in the predictions you make about specific things happening at specific times. NVDA earnings will disappoint and the print will gap down five percent. Oil holds 65 through quarter-end. The Fed cuts in September even though the dot plot says no. Those are claims about the world. They are right or they are wrong, and the world settles the question regardless of whether you put a trade on.
Paper trading captures the wrong half of that. It records that you opened a long NVDA position at $124.10. It does not record that you opened it because you thought the print would beat on data-center revenue, or because the chart looked good, or because someone you respect on the timeline said something convincing on Sunday night. When the trade closes — gain or loss — you are left with the dollars and no record of the bet. The thesis you were actually testing has evaporated by the time you'd want to grade it.
You can grade simulated dollars. You cannot grade simulated insight.
Without consequence, the data is contaminated
There is a second, quieter problem with paper trading. Without real money on the line, the decisions that get logged are not the decisions you would make under pressure. Position sizing drifts to the maximum the simulator allows. Stops get widened mid-trade because there is no actual cost to the widening. Conviction is performative because there is nothing being conserved.
The trader you are simulating in the paper account isn't you. It's a slightly braver, slightly sloppier version of you who has never been gut-punched by a six-figure drawdown. When that simulation tells you your win rate is 64% and your Sharpe is 1.8, the honest read is I have no idea what those numbers mean for the version of me that has to actually wire money.
This is the contamination that makes paper-trading data load-bearing for the wrong things. The numbers feel like evidence. They aren't.
What you actually want to know, before you risk capital
The question worth answering before you commit real money isn't can I run a P&L on a fake account. It's:
- When I form a view that something will happen, does the thing actually happen at the rate my conviction implied?
- Are my one-week catalyst calls more reliable than my three-month thesis calls?
- When I think a name is mispriced, am I right more often than I'm wrong — or am I just always opinionated?
Those are calibration questions. They are answerable, but only if you write the prediction down before the outcome is known and grade it after. And — load-bearing — they are answerable for zero capital at risk, because the prediction is the unit of analysis, not the trade.
The version of paper trading that compounds
Strip the trading from paper trading and you're left with the part that actually teaches you something: a written prediction, time-stamped before you knew the answer, with enough specificity that someone else could check whether you were right.
That is what a hunch journal is for. You log the read — NVDA prints, gaps down — with a conviction score and a horizon, and you do nothing else. No simulated dollars. No order. The record is the thing. Three months in, you have a hundred predictions you can grade against the world. Some were right. Some were dressed-up nothing. The ones that paid off are not random across the categories; the ones that didn't are also not random. That is the data you wanted from paper trading. And you got it without pretending to be a fund manager in a spreadsheet.
You don't need a simulator. You need a record.