From backtest to paper trading

Backtest, forward test, and live trading

A backtest applies rules to historical data. Every market event in the sample has already happened, which makes it easy to adjust a disappointing rule and try again. That feedback creates overfitting risk.

A forward test freezes the rule and records what it would do as new observations arrive. Paper trading adds simulated orders and portfolio bookkeeping, often through a broker’s practice environment. Live trading uses real capital and real execution.

Each step answers a different question. The backtest asks how the rule would have behaved under a historical model. Forward observation asks whether the unchanged logic continues to behave as expected. Paper trading asks whether the operational process can turn decisions into simulated orders. Live trading introduces actual fills, fees, risk, and human pressure.

Freeze the rule before moving forward

Write down the ticker universe, data source, indicator formula, parameters, entry and exit events, signal timestamp, intended order time, costs, position size, and risk limits. Give the specification a version and date.

If you change RSI from 14 to 12 after three weak paper trades, the original forward test has ended. The new version may be reasonable, but it begins a new experiment. Keep both records instead of rewriting history.

TickRun saved results can preserve a backtest’s settings and outputs. Also retain the exact market-data snapshot because a future corrected file can change the result.

Translate a daily closing signal into an action

TickRun calculates signals from saved daily bars. A position change on row t first earns the next close-to-close return. Real operations need a more precise bridge: when is the final bar available, when will you read it, what order type will you use, and what price will count as the paper fill?

If the rule needs the official Close, it cannot know that value before the close. A simple forward protocol could calculate after the session and record a next-session order. The realized gap between the signal reference and paper fill should be measured rather than silently treated as zero.

Do not change the timing after seeing whether the next open was favorable. The execution convention is part of the strategy.

Keep a decision and execution log

For every expected action, record:

• signal date and the data used;
• indicator values and rule version;
• intended side, quantity, and order type;
• submission and simulated fill times;
• reference price, fill price, and difference;
• fees or estimated spread;
• current position and cash;
• exceptions, missing data, and operator errors.

Record “no action” checks too. Otherwise the log contains only memorable trades and cannot show whether the process ran consistently.

What paper trading still gets wrong

A simulator may fill an order when the market merely touched its price. It may ignore queue position, limited size, partial fills, halts, rejected orders, spread changes, and market impact. Some practice platforms use delayed or simplified data. Small liquid orders may be approximated reasonably; large or illiquid orders can be far less realistic.

Paper capital also changes behavior. A simulated drawdown does not create the same fear as losing savings. There is no temptation to override a signal because real money is at stake—or, if you do override it, the reason may not match live pressure. Paper success is evidence about process, not proof of live profitability.

Reconcile model and observation

After each action, compare the forward record with what the backtest convention expected. Differences can come from data timing, corporate-action handling, rounding, missed sessions, spread, or an ambiguous rule.

Separate implementation error from market performance. If an indicator was calculated incorrectly, repair it and document the affected records. If the rule was followed correctly and lost money, do not call that a software bug. A strategy can be implemented perfectly and perform poorly.

Re-run the historical engine only when the saved dataset is deliberately updated. Do not let newer data leak into the original training period and then present the whole curve as an untouched result.

How long should a forward test run?

There is no universal number of weeks or trades. A daily rule that trades twice per year cannot be assessed from one month. A fast rule may create more observations, but those trades can still belong to one market regime.

Set a review horizon before starting, using both calendar time and a minimum opportunity count. Continue recording failures and quiet periods. Evaluate whether the rule generated the expected kind of behavior, whether execution assumptions were realistic, and whether risk remained within the predeclared limits.

Questions before considering real money

• Can another person implement the written rule without guessing?
• Did the frozen logic survive genuinely new observations?
• Were paper fills based on a realistic and consistent convention?
• Did costs and gaps resemble the backtest assumptions?
• Were drawdowns emotionally and financially tolerable?
• Is the amount at risk small enough that a total loss would not threaten essential finances?
• Are broker, tax, regulatory, and recordkeeping obligations understood?
• Is there a defined way to stop the process for operational failure?

Passing this list does not turn a strategy into investment advice or guarantee profit. It only shows that basic research and operational questions have been addressed.

A sensible TickRun workflow

1. Run defaults and understand every signal.
2. Add plausible transaction costs and inspect the equity curve.
3. If optimizing, study nearby parameters and record every trial.
4. Validate the frozen candidate on a later period.
5. Save the result and archive its data snapshot.
6. Write a separate forward-execution protocol.
7. Observe new sessions without modifying the rule.
8. Compare expected and paper execution at the planned review date.

Continue learning

Read how to construct a proper backtest, how chronological validation works, and what historical cost models leave out. Use the TickRun manual for saving and reopening research results.