TraderCoachTraderCoach
← All posts

Backtest Assumptions: Why Evan Rejected Monday’s Live Signal

Two men reviewing stock market data on a tablet, pointing at charts.

AlphaTradeZone

A polished backtest can show how a strategy behaved under its historical assumptions. It cannot tell you whether Monday’s spread, opening gap, liquidity, or decision timing will match the conditions that produced that curve.

At 9:31 on a Monday morning, Evan sat at his kitchen table in Chicago with a cooling mug beside his laptop and the backtest still open on a second screen. The equity curve rose cleanly from left to right. His rules had passed the checks he knew to run: entries, exits, position size, stop distance, and a defined maximum drawdown.

The first live signal arrived minutes after the open.

It called for an entry near Friday’s closing range. The market had already moved beyond that range. Evan could approve the order, reduce the size, wait, or reject it. His backtest had assumed an entry price available at the signal. Monday offered a different price, and a larger position risk than the one his chart implied.

If he approved the original order, the loss at his planned stop could exceed the amount he had written down as acceptable. If he chased the move with a tighter stop, ordinary price movement could remove him before the trade had room to work. The bad ending was simple: one “validated” trade could break the risk limit he thought the system respected.

He paused the order.

That pause was the useful result of the test. The historical curve had earned attention. It had not earned automatic execution.

Educational content, not financial advice.

A backtest measures the rules and assumptions you gave it

Historical testing can help a trader inspect a defined idea. It can show how a rule set responded to past price data, where losses clustered, how long drawdowns lasted, and whether the logic produced results worth studying further.

The result depends on the assumptions inside the test.

An entry rule may use a closing price that could not have been known until the candle ended. A strategy may assume every order fills at the displayed price. It may ignore spread, partial fills, commissions, overnight gaps, halted trading, changing liquidity, or the time required for a person to review a signal. Each choice can change the curve.

Evan had chosen a simple execution assumption because it made the test easier to compare across periods. The choice was reasonable as a starting point. Monday showed him the cost of treating that starting point as a live-trading promise.

A backtest also reflects the market regimes included in its data. A rule that performed through one set of trends, volatility levels, and trading hours may face a different environment next week. Historical results can support a hypothesis. They do not remove uncertainty.

The entry price a backtest assumed matters most at the moment a real order becomes possible. That is when a clean chart turns into a choice with actual position risk.

Monday exposed the assumptions Evan had not written down

Evan reread the test settings before touching the queued signal. The system used a fixed percentage stop and position size based on the assumed entry. It had no rule for a price gap between the signal and the available market price. It also did not account for the time he needed to inspect the order.

Those missing rules were not small implementation details. They determined whether his stated risk limit still applied.

He wrote three additions to his plan:

  • Reject a signal when the available entry price moves beyond a defined distance from the tested entry.
  • Recalculate position size using the actual entry and planned exit level before approval.
  • Record the reason whenever a signal is delayed, resized, or rejected.

The signal remained on screen while he did the math. The revised position size was smaller than the test would have used. Evan chose to reject it because the revised trade no longer matched the setup he had tested.

Nothing dramatic happened. No trade executed. That was the point. A disciplined rejection can preserve a risk rule when the market no longer offers the conditions the rule required.

Test out of sample, then make execution assumptions visible

A stronger process separates the data used to build a strategy from data used to evaluate it. The first period helps shape the rules. The later period asks whether those rules still behave reasonably without further tuning.

This does not prove a strategy will work in the future. It reduces the temptation to keep adjusting rules until they fit every past move. The research context here carries the same warning: choices about out-of-sample validation can affect how robust reported results appear.

Keep the test record concrete. Note the market, timeframe, entry definition, exit definition, fees included, assumed fills, position-sizing rule, and maximum drawdown calculation. Then list the conditions the test does not model.

A useful question is: “What would have to be true on Monday for this historical trade to happen as tested?” The answer may include a particular price, enough liquidity, a stable spread, and immediate execution. Once those conditions are visible, you can decide which deserve a rule and which make the trade unsuitable.

Approval creates a final risk check at the point of execution

An approval gate does not make a signal correct. It gives the trader a point to compare the live order with the tested setup before money is committed.

For Evan, that meant checking the actual entry, stop distance, position size, and total amount at risk. A signal could be approved when it still fit his plan. It could be resized when the plan allowed a smaller position. It could be rejected when the trade had changed too far from what the backtest measured.

That review also protects against a quieter error: treating a historical drawdown figure as a future limit. Maximum drawdown describes the path the strategy took in the data tested. A live drawdown can exceed it, especially when execution differs or conditions change.

Later that afternoon, Evan added the rejected signal to his trading journal. He saved the quoted entry, the live price he saw, the revised risk calculation, and the reason he declined the trade. By Friday, the cleanest line on his screen was no longer the equity curve. It was the written rule that prevented him from pretending Monday was already part of the past.

Sources (1)
  1. doi.orgBacktest Overfitting in the Machine Learning Era: A Comparison of Out-of-Sample Testing Methods in a Synthetic Controlled Environment

TraderCoach

Nokware is an approval-gated AI trading assistant for crypto and stocks: the AI generates and queues trade signals, and a human approves or rejects each one before anything executes — you always keep the final decision, and it never trades unsupervised.

Try TraderCoach

Comments

No comments yet.