Paper Trading vs. Live: Why Paper Trade First
Many trading bots go live too early. Here's why paper trading comes first, what it can and cannot tell you, and and what we learned from our own simulation.
Bottom line: Backtests flatter strategies. Paper trading is a closer test, but it is still a simulation. Treat it as a way to find bugs and test risk controls before real money is involved.
The Problem: Going Live Too Fast
Browse any trading bot marketplace and the pitch is often the same: deploy quickly, start earning, backtested results you can trust. Here is what those pitches tend to leave out:
- Backtests don't capture slippage on market orders
- They don't model API rate limits or downtime
- They don't account for latency between signal and execution
- They can't capture the psychological pressure of real money
- They can't predict bugs that only appear under live load
So teams deploy, something breaks, they lose money, and then they work out what went wrong. Paper trading exists to move that discovery earlier.
What Paper Trading Is For
In paper mode, the full stack runs as it would with real money, except that orders are logged instead of sent to an exchange. Our own bot was built that way, with a critic agent that could veto trades, circuit breakers, and Telegram alerts, and it never traded real capital.
- Find bugs before they cost money. Integration problems, bad assumptions, and risk-rule gaps show up in simulation.
- Test risk controls. Circuit breakers and veto layers need to be exercised before you rely on them.
- Model execution costs. Paper trading is only useful if it includes realistic slippage and fees rather than assuming perfect fills.
- Keep records. Log every decision from day one, so any result can be tied to a run record.
What Paper Trading Cannot Tell You
Paper trading does not capture real slippage, liquidity impact, or live execution, and it does not reproduce the stress of losing real money. A positive simulated result over a short period is not evidence of a durable edge.
Our Own Result, With Context
Our Paper Bot, Simulated
Feb 26 - Apr 8
Simulated period, 2026
Net loss
Both books ended below $10,000
Paper
Simulated, not live
Simulated paper trading, not live results. Not financial advice. Trading carries risk of loss, and past or simulated performance does not predict future results.
Was it impressive? No. The ledger shows the run ended net negative, and we did not treat it as evidence of an edge. The numbers and data caveats are in the retired-project case study. The bot has since been retired; read why we retired it.
When Is It Reasonable to Go Live?
Whether and when to go live depends on your own risk tolerance and evidence. Reasonable gates include a positive risk-adjusted result over a meaningful sample of trades, a tolerable maximum drawdown, no unresolved critical bugs, and risk controls that have been tested. Start with small size, and monitor for deviation from your paper results. We never reached the point of trading live capital.
Practical Lessons
- Infrastructure is most of the work. Logging, fallbacks, rate limits, and alerts matter more than any single strategy.
- An LLM is better as a brake than as a predictor. Use it to spot contradictions and queue uncertain trades for review.
- Keep a run record. If you cannot tie a number to a record, you should not publish it.
Related Reading
Our paper trading bot has been retired. Read why, and the design behind it.