Every trading system eventually reaches a point where the logic is in place, the parameters are set, and the temptation to start trading real money becomes hard to resist. This is exactly the moment to slow down. Before committing actual capital, there's a step that can't be skipped: paper trading — running the system against live market data without real money on the line. This post covers how I built paper trading into the system, what it actually revealed, and what paper trading can and can't tell you.
How Paper Trading Was Built Into the System
The core design principle for the paper trading mode was simple: the live trading logic and the paper trading logic had to follow the same code path as far as possible. If the paper trading module was written as a completely separate system, all the bugs and edge cases that only appear under live conditions would remain hidden until real money was on the line. That defeats the purpose.
In practice, this meant using a single configuration switch to toggle between modes. Everything from signal detection to entry logic to position tracking ran identically in both modes. The only branch was at the very end: in live mode, the system sends an order to the brokerage server; in paper mode, it records the current market price as the hypothetical fill and updates the simulated position internally. All the logic before that final branch — the pattern recognition, the grid stage calculations, the exit condition checks — was shared.
What Paper Trading Actually Found
The first paper trading session produced no trades at all. Not one entry over an entire trading day. The initial reaction was "maybe the market just didn't offer the right setup today." The second session produced the same result.
After the third consecutive blank session, it was time to look at the logs. The issue turned out to be a bug in the initialization logic — a variable that was supposed to hold the initial price was set to zero when the system started tracking a new stock. When the algorithm calculated the price range as a percentage of this reference price, dividing by zero or near-zero produced an absurd result (a calculated range of 100% or more), which caused every stock to immediately fail the range validation check and get dropped from tracking. The pattern recognition stage never ran at all.
This bug would have been difficult to catch in live trading because it produced no error message and no crash — just a system that appeared to be running normally while silently failing to identify any opportunities. Spotting it required reading through logs on a day when the expected behavior (at least one entry) hadn't occurred. Paper trading made it possible to investigate without any financial cost.
What Paper Trading Can't Tell You
Paper trading is not a complete substitute for live trading, and treating it as one leads to overconfidence. A few specific limitations are worth understanding before relying on paper trading results.
Market impact. In paper trading, your orders don't touch the order book. In live trading, especially in less liquid stocks, a real order can move the price — particularly on the sell side, where the system might be trying to exit quickly. The fill price the paper system assumes may be impossible to actually obtain at the size and speed required.
Latency effects. Paper trading records the current market price at the moment the signal fires. Live trading introduces network latency, order processing time, and queue position at the exchange. In fast-moving situations, this gap between the paper fill and the real fill can be significant.
Psychological difference. Paper trading carries no financial risk, which means it also carries no psychological pressure. The experience of watching a live position move against you, knowing that each tick represents real money, is categorically different from watching a paper position do the same thing. Discipline that holds up easily in paper mode sometimes doesn't survive the transition to live.
The right way to use paper trading is as a validation layer for system logic — not as proof that the system will produce the same results in live markets. If the system fails in paper trading, that's a definitive signal that something is wrong. If it succeeds, that's a necessary but not sufficient condition for live deployment.
Today's Investing Insight — Backtesting vs. Paper Trading
Backtesting runs a strategy against historical data to see how it would have performed in the past. Paper trading runs a strategy against current live data without real money. Both are simulation methods, but they have different failure modes. Backtesting is prone to overfitting — designing a strategy that fits historical data so well that it performs poorly on data it hasn't seen. Paper trading avoids overfitting on past data, but a short paper trading window may not expose the full range of market conditions the system will eventually face. Using both together — backtesting for initial validation, paper trading for live-environment verification — provides better coverage than either alone, while recognizing that neither guarantees live performance.
---
This post documents a personal journey of building an algorithmic trading system and is not a recommendation of any specific strategy. Paper trading results do not guarantee equivalent performance in live markets, and all investment decisions and their outcomes are the sole responsibility of the investor.
Comments
Post a Comment