This article synthesizes the supplied warnings and examples about revising backtest entries, exits, adjustment points, profit targets, and other parameters. It distinguishes outcome-driven loss filtering from potentially educational nearby-parameter testing without treating every rule revision as curve fitting.

  • Distinguish investigating nearby parameters from changing a rule specifically to remove an unfavorable historical result.
  • Assess what an improved backtest does—and does not—establish about a revised strategy.
  • Recognize how discretion, excessive specificity, and repeated tweaking complicate the interpretation of historical results.
  • Preserve the limited scope of hypothetical, strategy-specific, and speaker-specific examples when reviewing a backtest.

Identify the purpose of the revision

The central warning concerns why a rule is changed. In the supplied examples, moving an entry, exit, or profit target specifically to exclude a known loss improves the historical record by construction, but does not by itself establish a better strategy. [1][5][12][13]

  • Moving an exit by one adjacent day solely to omit an adverse event does not, on its own, show that the revised exit is better. [1]
  • In the hypothetical bull-trade example, changing entry from 65 to 64 days to expiration because the sampled 65-day entry lost filters the result without materially changing the trade. [5]
  • In the supplied profit-target example, reducing the target from $5,500 to $5,000 specifically to remove a losing period is characterized as curve fitting. [12]
  • A rule made more specific than its available justification may be fitted to past data. [13]

Separate a cleaner history from evidence of improvement

Historical tuning can suppress observed losses without preventing future ones. The supplied claims caution that repeated tweaking, retrospective revision, or engineering an unusually consistent backtest can create confidence that the evidence does not warrant. [2][3][8][9]

  • The speaker warns that an M3 backtest engineered to win every tested month since 2011 may be form-fitted and should not be expected to preserve that record in live trading. [2]
  • Repeatedly modifying adjustment rules until historically lucky moves produce attractive results fits the strategy to past conditions and can create unwarranted confidence about future trading. [3]
  • The source describes a retrospective practice in which backtests are revised after a winning period so that the strategy appears to have won throughout its history. [8]
  • Future conditions may resemble the past without repeating it exactly, so filtering past losses does not prevent future losses. [9]

Use nearby-parameter tests as investigation, not repair

Nearby-parameter testing can be educational when it reveals sensitivity around an adjustment point. The tension is that selecting only the winning variant can obscure why the trade loses and turn investigation into result filtering. [6][10][11]

  • The speaker recommends testing several delta thresholds near an adjustment point; mixed results may be informative but should not be treated as proof that the trade has been fixed. [6]
  • Narrowly changing an adjustment delta to filter unfavorable results can create an unrealistic view of the strategy and hide the dynamics behind its losses. [10]
  • A parameter chosen through backtesting can be fitted to past conditions; the speaker contrasts this with selecting delta or gamma from accumulated experience and observed market behavior. [11]

Examine displaced risk and discretionary variation

Changing a universal threshold after one severe historical case may relocate rather than eliminate the vulnerability. Interpretation is further complicated when discretionary gray areas allow different outcomes under otherwise similar trades and rules. [7][4]

  • After one double-maximum-loss backtest case, moving a universal adjustment point does not eliminate the tail event: a similar move could occur just before the new threshold. [7]
  • The same threshold change can introduce a different vulnerability elsewhere in the strategy. [7]
  • Tweaking rules until a backtest wins may not indicate future results when gray-area decisions can lead traders using otherwise similar trades and rules to different outcomes. [4]

Key takeaways

  1. Ask whether a revision investigates strategy behavior or was selected specifically because it removes a known loss; the latter does not by itself establish improvement. [1][12]
  2. Treat an increasingly attractive historical record cautiously when it emerges from repeated tweaking or retrospective revision. [3][8]
  3. Nearby-threshold comparisons may teach sensitivity, but choosing the winning variant can amount to filtering rather than fixing. [6][10]
  4. When reviewing a threshold change, consider whether the adverse event could occur near the new boundary or whether a new vulnerability appears elsewhere. [7]
  5. Preserve strategy and decision context: discretionary gray areas can produce different outcomes even under otherwise similar trades and rules. [4]

Review questions

A revised exit omits one adverse historical event. What can the revised backtest establish by itself?

It can show how that revised historical sample looks, but moving an adjacent exit solely to exclude the event does not by itself establish that the exit rule improved the strategy. [1]

How should a reviewer interpret several nearby delta tests when only some variants win?

The variation may be educational about threshold sensitivity, but the winning variants should not be mistaken for a repaired trade because they may merely filter out a loss. [6]

Why might moving a universal adjustment point after one severe backtest loss be an incomplete response?

The tail event may occur just before the new threshold, and the changed rule may introduce another vulnerability elsewhere. [7]

Why does a backtest tuned to win consistently not establish comparable future performance?

The result may be fitted to historical conditions, while future conditions may resemble but not exactly repeat the past. [2][9]

What additional interpretive problem arises when a strategy contains discretionary gray areas?

Different traders can make different gray-area decisions and obtain different outcomes even when they otherwise use the same trade and rules. [4]

Evidence index

Canonical source claims used in this guide. Open a session link to verify the underlying passage at its original timestamp.

[1]Moving an exit by an adjacent day solely to exclude one adverse event from results does not by itself establish that the revised exit rule improves the strategy.

The exit days and adverse move are hypothetical examples.

The sources do not establish that adjacent exit windows are universally equivalent.

[2]The speaker warns that an M3 backtest engineered to win every tested month since 2011 may be form-fitted and should not be expected to retain that result in live trading.
[3]Repeatedly tweaking adjustment rules until past lucky moves produce attractive backtest results fits the strategy to historical conditions and can create unwarranted confidence about future trading.
[4]Tweaking a backtest's rules until the strategy wins may not indicate future results because discretionary gray-area decisions can produce different outcomes even when traders otherwise use the same trade and rules.
[5]Changing a bull-trade entry from 65 to 64 days to expiration solely because one sampled 65-day entry lost would filter out an unwanted result without materially changing the trade.
[6]When backtesting near an adjustment point, the speaker recommends testing several delta thresholds; if only some variants win, that may be educational but should not be mistaken for fixing the trade because it may merely filter out a loss.
[7]Moving a universal adjustment point after one double-maximum-loss backtest case does not eliminate the tail event; the same move can occur just before the new threshold, and the rule change can introduce a different vulnerability elsewhere.
[8]The speaker warns that people may revise backtests after a winning period so the strategy appears to have won throughout its history.
[9]Tuning a strategy on past backtests can filter out historical losses without preventing future losses, because future conditions may resemble but will not exactly repeat the past.
[10]Filtering unfavorable backtest results by narrowly changing an adjustment delta can create an unrealistic view of a strategy and obscure the dynamics that make it lose.
[11]A strategy parameter selected through backtesting can be fitted to past conditions; the speaker contrasts that process with choosing a delta or gamma value from accumulated experience and observed market behavior.
[12]Changing a backtest profit target from $5,500 to $5,000 specifically because that change removes a losing period is curve fitting.
[13]The speaker warns that making a strategy rule more specific than the available justification may amount to fitting it to past data.