How to Backtest a Trading Strategy Manually

Manual backtesting is walking historical bars forward under a frozen rule set and recording what the rules would have done — without coding an engine. It is slower than an automated backtest and more honest than scrolling a finished chart hunting confirming examples.

For discretionary NSE traders, that middle ground matters. Your “strategy” is often visual: a breakout definition, a pullback location, a time filter. If you cannot yet express it in code, a bar-walk is how you discover whether the idea even fires often enough to bother practising.

Bar-walk vs chart replay vs coded backtest

• Coded backtest — best when rules are fully specified; answers “did the rules work?” across large samples fast.
• Manual bar-walk / backtest — you advance history and log rule-based outcomes; still about the rules, but human-paced.
• Chart replay — emphasises whether you would have decided under uncertainty; overlaps with manual testing when you enforce the same rule sheet.

Use manual backtesting to stress the definition. Use replay to stress the decision-maker. Confusing the two is how traders “validate” a strategy they never actually followed.

The manual bar-walk protocol

Print or pin a one-page rule sheet before you touch a chart. Freeze it.

Step 0 — Freeze the rules (one page)

• Market universe (e.g. liquid NSE large-caps or a specific index chart) — practice universe, not a buy list
• Timeframe
• Session filter (e.g. no new risk in first 15 minutes) if any
• Entry conditions (must / must-not)
• Invalidation / stop placement rule
• Exit rules (target, time stop, trail — pick and write)
• Position sizing in R (1R risk per idea)
• What you will not trade

If the sheet needs a paragraph of “feel,” you do not have a testable strategy yet — you have a mood.

Step 1 — Choose the sample before looking

1. Pick a date range and instrument list in advance.
2. Prefer consecutive sessions or randomised starts over “famous move” days.
3. Write the planned sample size (sessions or expected setups) before starting.

Step 2 — Walk bars forward only

1. Hide future candles. No scrolling right for clues.
2. At each bar, ask only: do the frozen rules trigger?
3. If yes, log entry, stop, targets as the rules prescribe — not as you wish.
4. Advance until exit rules complete the trade; then continue.
5. If rules are ambiguous on a bar, log ambiguous and stand aside. Ambiguity is data: your rules are underspecified.

Step 3 — Log fields that survive ego

• Date/session tag (or blinded ID if dates are hidden)
• Setup tag
• Entry / stop / exit
• R-multiple result
• Rule followed? (Y/N) — if N, exclude from “strategy” stats
• Notes: one line max

Step 4 — Review once per block, not after every loss

Finish the pre-committed sample. Then summarise: trade count, R expectancy (descriptive only), max losing streak in the sample, % ambiguous bars, % rule breaks. Do not rewrite rules after three losers mid-walk — that is overfitting in real time.

Overfitting traps (manual edition)

Manual testers overfit differently from quants, but just as eagerly:

• Rule mutation mid-sample — “I’ll also allow entries on wicks” after missing a winner. Start a new sample if rules change.
• Venue shopping — dropping names that “don’t trend” until the journal looks clever.
• Outcome-based annotation — redrawing the level that “was always the level” after the bounce.
• Tiny flattering windows — testing only a strong bull month for long-biased rules.
• Ignoring frequency — a beautiful R on eight trades a year is a hobby, not a process you can practise daily.
• Conflating compliance with edge — following rules that lose steadily is still useful information; do not “fix” them into a win rate.

When to stop tweaking

Stop mutating rules when any of these are true:

• You have changed definitions more than twice on the same sample
• Ambiguous tags dominate the log — specify or abandon
• Trade count is too small to discuss edge seriously — see how many trades you need to test a strategy
• You are improving the story faster than you are collecting out-of-sample walks

A sane next step after one honest manual sample: freeze rules again, run a fresh date range, then move residual uncertainty into decision practice — candle-by-candle replay and a finite block like the 300-trade experiment. Replay environments (including Replay Trader) help you keep future bars hidden while you accumulate those decisions.

What manual backtesting cannot certify

It cannot certify live profitability, fill quality, or your behaviour when rupees move. It can certify whether your written rules are clear enough to apply without debate, whether setups appear in liquid NSE contexts you care about, and whether you are honest enough to log losers without editing the past. That is enough value for a tool that costs nothing but attention — provided you refuse to turn the walk into a highlight reel.

Educational content only. Not investment advice. Not a recommendation to buy or sell any security. Securities markets involve risk — see SEBI investor education on key risks.