Debugging a Strategy (Using Order History)

This guide shows how to use order-level data to diagnose why a strategy behaves the way it does, and how to fix it without guessing.

---

Why Order History Matters

Aggregate metrics (PnL, win rate, drawdown) tell you what happened. Order history tells you why it happened.

Order history exposes:

• When trades fired

• What triggered them

• How frequently logic re-evaluates

• Whether exits are intentional or accidental

If you don’t inspect order history, you’re debugging blind.

---

Step 1: Validate That Logic Fires Where You Expect

What to check

• Do buys and sells occur near obvious signal points?

• Are trades clustered around indicator crossings or thresholds?

Common failure

• Trades fire too early or too late

Likely causes

• Indicator periods too short or too long

• Using Compare when Crossover was intended

• Multiple evaluations per bar causing duplicate triggers

Fixes

• Replace threshold logic with crossover logic

• Increase lookback periods

• Add explicit exit conditions

---

Step 2: Look for Overtrading

Symptoms in order history

• Many small trades close together

• Rapid buy–sell–buy sequences

• High trade count with flat or negative PnL

Likely causes

• Conditions oscillating around a threshold

• No cooldown or state awareness

• Missing exit filters

Fixes

• Add confirmation logic using AND

• Widen thresholds (e.g., RSI < 28 instead of < 30)

• Add alternative exits with OR to avoid churn

---

Step 3: Identify Unintended Re-Entries

Symptoms

• Strategy re-enters immediately after exiting

• Back-to-back identical trades

Likely causes

• Entry condition remains true after exit

• Exit does not invalidate the entry condition

Fixes

• Add a separate exit condition that explicitly flips the signal

• Use a crossover instead of a static comparison

• Introduce a second condition to gate re-entry

---

Step 4: Diagnose Large Drawdowns

Symptoms

• A few large losses dominate PnL

• Drawdowns occur during sharp moves or events

Order history clues

• Trades entered late into moves

• Exits lag reversals

• Positions held through regime shifts

Likely causes

• Trend-following logic without risk exits

• Mean-reversion logic in trending regimes

• No stop-loss or profit-taking logic

Fixes

• Add protective exits (stop-loss or time-based exit)

• Add a regime filter (e.g., trend confirmation)

• Reduce position size before changing logic

---

Step 5: Compare Entry Quality vs Exit Quality

How to tell

• Plot order markers on price

• Visually inspect entry timing vs exit timing

Patterns

• Good entries, bad exits → profits given back

• Bad entries, good exits → churn and noise

Fixes

• If exits are weak:

  • Add explicit profit-taking logic

  • Exit on signal invalidation, not just reversal

• If entries are weak:

  • Add confirmation indicators

  • Delay entry using crossovers instead of thresholds

---

Step 6: Detect Logic That Only Works in One Regime

Symptoms

• Strategy performs well in one market

• Fails catastrophically in others

Order history clues

• Trades cluster in specific market conditions

• Long inactive periods followed by bursts

Likely causes

• Strategy implicitly assumes:

  • High liquidity

  • Smooth trends

  • Frequent oscillation

Fixes

• Test across multiple markets

• Add regime-dependent conditions

• Accept that some strategies are intentionally narrow

---

Step 7: Use “What Didn’t Trade” as a Signal

Sometimes the bug is silence.

Symptoms

• Very few or zero trades

• Strategy appears inert

Likely causes

• Over-constrained logic

• Conflicting conditions

• AND gate requiring incompatible signals

Fixes

• Temporarily remove conditions

• Test each condition independently

• Replace AND with OR to diagnose conflicts

---

Common Debugging Patterns

Symptom	Likely Issue	Fix
Too many trades	Threshold noise	Add crossover or confirmation
Large drawdowns	Missing risk exits	Add stop-loss / invalidation
Instant re-entry	Entry not invalidated	Add exit filter
Works once, then fails	Regime dependence	Add regime filter
No trades	Over-constrained logic	Simplify conditions

---

Debugging Mindset

• Change one thing at a time

• Use order history before changing parameters

• Assume the strategy is wrong, not the market

A strategy that fails loudly is a gift—it tells you exactly what assumption broke.

PredictBack is built to make those failures visible.