Complete Fibonacci Harmonic Strategy and Backtesting
Lesson 12 of 12. Design a reproducible test, assess expectancy and identify overfitting before risking capital.
1. Write the hypothesis
Example: after a confirmed four-hour bullish impulse, a 61.8–78.6% retracement overlapping a validated Gartley PRZ and 15-minute bullish structure break predicts a positive net expectancy. This is a hypothesis, not a finding.
2. Specify the rules
Define swing detection, permitted ratios and tolerance, minimum PRZ overlap, trigger, stop, exit, maximum holding time, trading hours and instrument universe. Write them before seeing outcomes.
3. Avoid look-ahead bias
A pivot requiring two candles on the right becomes known only after those candles close. Backtest the signal from that later timestamp, not from the historical pivot itself.
4. Sample design
Split historical data chronologically into development, validation and out-of-sample periods. Avoid repeatedly tuning on the final test set. Include multiple volatility regimes and realistic spreads.
5. Measure outcomes
Track trade count, win rate, average win/loss in R, expectancy, drawdown, exposure, slippage sensitivity and time in trade. A high win rate can still lose money if average losses are large.
6. Worked expectancy
At 42% wins, 1.6R average gain and 1R average loss, gross expectancy = 0.42×1.6−0.58×1=0.092R. If total average costs equal 0.12R, net expectancy becomes −0.028R.
7. Robustness tests
Vary ratio tolerance, pivot sensitivity and spreads slightly. A robust idea should not depend on one exact setting. Compare against a simple trend-following or random-entry baseline.
8. Forward practice plan
Paper trade the frozen rules for a defined sample and log every signal including no-trades. Review rule compliance separately from profit. Educational patterns cannot promise income.
Knowledge check
- Which entry, stop and exit rules must be fixed before a backtest?
- How will you account for spreads, slippage and rejected pattern candidates?
- Which out-of-sample result would make you abandon the strategy?
Answer using the worked example above. Record your calculations and one reason the setup could fail before revealing any chart outcome.
Worked numerical example: Forward-test decision log
Pre-register a rule such as: only take a candidate if B and D ratios satisfy the documented template, a close confirms the trigger, and projected reward/risk after costs exceeds 1.5. Record all eligible signals, including losers and missed fills. A small favorable sample is not evidence of a reliable edge.
Practice and answer
Show worked answer
Treat the 60% retracement as one sample observation in a backtest. Record the pattern label, leg anchors, transaction costs, entry trigger and invalidation; include rejected patterns to avoid cherry-picking only attractive outcomes.
Independent worked analysis: Backtest the rule, not the drawing
Hypothetical worked example: Imagine 80 historical setups, 36 wins and 44 losses. A 45% win rate can be profitable with sufficiently large average wins, but a win rate alone is not expectancy.
Calculation and decision exercise: With average gross win of 2R and average gross loss of 1R, gross expectancy is 0.45×2 - 0.55×1 = +0.35R per trade. Subtract commissions, spreads and slippage before drawing conclusions.
What this example cannot establish: Separate training and holdout periods. Changing pivot tolerances after seeing the results is data snooping.
Evidence checklist for this specific lesson
- Write down the precise inputs used in the backtest the rule, not the drawing example, including their units and observation dates.
- Distinguish what can be computed directly from what must be inferred about future prices or policy.
- Test the opposite interpretation using a clearly stated alternative assumption, not a revised chart drawn after the outcome.
- Document transaction costs and the maximum hypothetical loss before treating any market view as an executable strategy.
Applied analysis: testing the lesson in context
The following exercise develops the specific decisions involved in complete fibonacci harmonic strategy and backtesting. It is a hypothetical training workflow, not a prediction or an instruction to place a live trade. Keep the market, time zone, instrument, observation window and data source fixed before evaluating the idea. Separate what the chart shows from the interpretation assigned to it; two analysts can label the same sequence differently without either label being a verified market mechanism.
Case step 1: 1. Write the hypothesis
Start with the lesson's 1. write the hypothesis component. Source observation: Example: after a confirmed four-hour bullish impulse, a 61.8–78.6% retracement overlapping a validated Gartley PRZ and 15-minute bullish structure break predicts a positive net expectancy. This is a hypothesis, not a finding. Record the exact candle or pivot at which this information first became available. If confirmation depends on a later close, mark that later close as the earliest legitimate decision point. A visually compelling retrospective drawing is not sufficient evidence that the setup was identifiable in real time.
Now construct a counterexample for 1. write the hypothesis: imagine the price initially behaves as the lesson suggests but then fails to follow through. Write the invalidation condition as an observable price or time event, not as a vague change in confidence. Compare two annotated screenshots, one before the proposed confirmation and one after it. Explain what genuinely changed in the available evidence. For a trading-system study, record spread, slippage, position size, and whether the proposed stop could have been filled at the assumed price.
Case step 2: 2. Specify the rules
Start with the lesson's 2. specify the rules component. Source observation: Define swing detection, permitted ratios and tolerance, minimum PRZ overlap, trigger, stop, exit, maximum holding time, trading hours and instrument universe. Write them before seeing outcomes. Record the exact candle or pivot at which this information first became available. If confirmation depends on a later close, mark that later close as the earliest legitimate decision point. A visually compelling retrospective drawing is not sufficient evidence that the setup was identifiable in real time.
Now construct a counterexample for 2. specify the rules: imagine the price initially behaves as the lesson suggests but then fails to follow through. Write the invalidation condition as an observable price or time event, not as a vague change in confidence. Compare two annotated screenshots, one before the proposed confirmation and one after it. Explain what genuinely changed in the available evidence. For a trading-system study, record spread, slippage, position size, and whether the proposed stop could have been filled at the assumed price.
Case step 3: 3. Avoid look-ahead bias
Start with the lesson's 3. avoid look-ahead bias component. Source observation: A pivot requiring two candles on the right becomes known only after those candles close. Backtest the signal from that later timestamp, not from the historical pivot itself. Record the exact candle or pivot at which this information first became available. If confirmation depends on a later close, mark that later close as the earliest legitimate decision point. A visually compelling retrospective drawing is not sufficient evidence that the setup was identifiable in real time.
Now construct a counterexample for 3. avoid look-ahead bias: imagine the price initially behaves as the lesson suggests but then fails to follow through. Write the invalidation condition as an observable price or time event, not as a vague change in confidence. Compare two annotated screenshots, one before the proposed confirmation and one after it. Explain what genuinely changed in the available evidence. For a trading-system study, record spread, slippage, position size, and whether the proposed stop could have been filled at the assumed price.
Evidence record and independent assessment
Create a small research log specifically labelled “Complete Fibonacci Harmonic Strategy and Backtesting”. For each candidate observation, store its date, instrument, session, timeframe, setup definition, entry eligibility, invalidation level, expected costs, and actual subsequent path. Include candidates that never triggered, rather than selecting only attractive examples. Compare at least two market regimes and note whether the interpretation changes when a nearby swing anchor or session boundary is chosen. If it does, classify the result as sensitive to analyst discretion rather than robust evidence.
Finally, write a three-part conclusion about complete fibonacci harmonic strategy and backtesting: (1) which elements can be verified directly from historical prices, (2) which elements are hypotheses about participants or future behavior, and (3) what observation would cause you to reject the hypothesis. The exercise is complete only when a second reader could reproduce your markings using the written rules. Do not infer a profitable edge from a single worked example; that requires a separately designed, cost-aware and out-of-sample evaluation.