First Officer · Strategy Testing and Uncertainty · About 30–40 minutes
Safety rule: Never use a later candle to justify a decision supposedly made earlier.
Your mission today
By the end you can run a chronological manual backtest and record every valid and rejected trade with costs.
Before you start
Finish C06.5. Choose a single demo symbol and a fixed set of rules. Use a replay tool or printed candles that hide the future.
1. A backtest asks what the rule would have done
A backtest applies a fixed trading rule to earlier market observations. Think of checking a fire-drill plan against old building layouts. It may reveal weaknesses, but it does not guarantee that tomorrow behaves like yesterday. A manual backtest means a person advances historical candles in order and records decisions without knowing the later outcome. Start with one exact instrument, its contract specification, a declared timeframe and a specific historical period. Write the entry, stop, exit, skip and cost rules before opening the replay. Include the time zone and chart data source. Survivorship bias can appear when you select only symbols or periods that remain attractive in hindsight. Avoid starting with a dramatic winning chart and working backward. Use an independently chosen chronological sample. A valid research log should include every setup offered by the rules, including the boring losers and correctly rejected candidates.
2. Calculate the proposed loss before simulating a result
Imagine a fictional EURUSD long that becomes eligible only after a completed M15 bar closes above 1.08500. The next executable Ask is assumed to be 1.08520; the planned Bid stop is 1.08420. That is 10 pips of price risk. At an invented $1 per pip, a stop fill at the intended level would lose −$10 before fees or worse execution. If a separately modelled round-trip commission costs $0.50, the result would become −$10.50. A later executable Bid of 1.08670 could instead give +15 pips, or +$14.50 net after the same commission, in a different imagined outcome. Neither result can be selected just because it looks nicer. Record which level was reached first under the simulation's stated execution assumptions. If both stop and target appear inside one candle and their order cannot be established, label the outcome AMBIGUOUS.
3. Freeze the exact signal timestamp
A common error involves look-ahead bias, where information from the future affects a claimed past decision. Imagine a swing-high rule requiring two completed bars on the right of a candidate pivot. The pivot cannot be confirmed until both later bars close. A charting tool may draw its marker above the earlier pivot candle, but that visual position is not the time the information became available. Write the signal's first knowable timestamp, not merely the bar containing the extreme. Then advance one complete candle. Record whether the rule says TRADE, WAIT or SKIP at that time. Use the correct quote side. A Bid-candle close is not automatically the Ask available to buy. Record actual known spread if historical quotes exist; otherwise declare a justified cost assumption and test how sensitive results are to it.
4. Make an audit ledger that includes rejections
Use one line per offered setup, not just one line per placed trade. Include setup identifier, date, market, timeframe, eligible closing time, intended entry, stop, size, estimated cash loss, permitted or rejected status, reason, exit and total costs. A rejected setup is one that fails a prewritten criterion such as maximum spread or an affordable stop. Do not delete it because it would have won afterward. Add the original rule version as a field, because mixing two evolving sets of rules makes the totals hard to interpret. If a data gap hides the next executable quote, record NO VALID FILL rather than inventing the price. If your platform uses different contract multipliers over the sample, document each period separately. Your goal is a reproducible record that another student can challenge, not an impressive screenshot.
5. Calculate results in cash and in risk units
An R-multiple expresses a trade's net result divided by its original planned cash risk. Suppose three fictional completed trades use $10 planned risk each and net results of −$10.50, +$14.50 and −$10.50. Their net R values are −1.05R, +1.45R and −1.05R. Add them: −0.65R across three trades, or about −0.217R per trade. Even though one profit was larger than one loss, the sequence loses overall. Do not quote gross pips while ignoring charges. Track maximum peak-to-trough account decline, losing streaks and the number of trades your rules actually offered. A small sample cannot establish a dependable advantage. When you make a rule change, label the new version and evaluate it separately, including data not used to design the change.
6. Finish with a falsifiable conclusion
After twenty or more chronological teaching cases, produce three paragraphs: What exactly did the rule permit? What did the complete ledger show under stated costs? What information is missing? A falsifiable claim is one that evidence could prove wrong. 'My method works when the market is good' is not precise enough to test. 'Across these fixed dates and assumptions, the net result was below zero' can be checked. Your conclusion should also say which situations the sample did not include and which fills were ambiguous. Do not tune the stop after a loss and quietly recompute the same historical records. A fair next step is to preserve the first sample as development work and move to a genuinely unseen period or a forward paper test. The worksheet is educational research, not a live trading recommendation.
Common mistake
A learner marks a pivot at its turning candle and treats it as known immediately. The confirmation required later bars, which makes the early entry impossible. Fix the test by recording the first timestamp the rule could have fired.
Plain-English glossary
- Backtest: applying fixed rules to earlier observations.
- Look-ahead bias: using information that was unavailable at decision time.
- R-multiple: net trade result divided by initial planned cash risk.
- Ambiguous fill: an outcome not reliably inferable from the available price sequence.
- Falsifiable: capable of being disproved by specific evidence.
Try it yourself
Backtest 20 consecutive eligible EURUSD cases in replay. Freeze the strategy version first. For every case log the first knowable signal time, quote assumptions, planned loss, net R, exit path and any skipped or ambiguous outcome. Preserve the entire offered set.
What you can now do
- State a repeatable rule and its main limitation.
- Recompute the key loss or test statistic using correct units.
- Keep an auditable record of accepted and rejected decisions.
Next lesson
C07.2 · Forward Paper Testing. Take the same frozen rule into decisions made without historical replay.
