Why Data Quality Matters in Trade Analytics

A trading dashboard never measures your trading directly. It measures a set of records— one row per trade, carrying a symbol, direction, volume, open and close times, prices, swap and commission. Every metric on top of that is arithmetic over those rows, so a flawed row set produces flawed metrics with complete confidence: nothing on the screen looks broken, the numbers are just quietly wrong.

The damage is rarely random, either. Record-keeping faults tend to push statistics in flatteringdirections, because the trades most likely to go missing are the ones from losing days — exactly the rows an honest review needs most.

Most corrupted trade analytics trace back to a small set of causes. Each one enters the record set differently, and each leaves a different fingerprint:

Duplication has the most recognisable signature of the six. Import the same history twice and a 412-trade record with $3,180 net profit becomes 824 trades and $6,360 — yet the win rate stays at 54% and the profit factor does not move, because every trade now sits on both sides of every ratio. Doubled totals with unchanged ratios is the tell; duplicated rows also share ticket and position IDs, which makes the cleanup mechanical.

Timezone faults are subtler because no individual row is wrong — only the grouping is. MetaTrader stamps every trade in the broker’s server time, so a trader in New York (UTC−5) whose broker runs GMT+3 is eight hours adrift. A trade closed at 01:40 Tuesday server time actually happened at 17:40 on Monday local time: group by server-time days and the “Tuesday” column absorbs Monday afternoon’s results. Daily P/L, trades-per-day and weekday breakdowns all need one declared timezone basis before they mean anything.

Broker suffixes mark account types or price feeds, not different markets. An account that traded one pair through a broker account change can end up with 38 trades on EURUSD.m, 21 on EURUSD and 14 on EURUSD.pro — three per-symbol lines of 38, 21 and 14 trades where there is really one line of 73. Each fragment is a small sample, so per-instrument win rates and averages turn into noise. Normalising suffixed variants to one canonical symbol restores the real sample size.

The mirror-image fault is blending: where suffixes split one thing into three, missing tags merge three things into one. Two EAs left on magic number 0 plus a handful of manual trades produce a single combined line in which no strategy can be judged on its own. Tagging conventions and how to keep manual and automated results apart are covered in the guide to separating manual and EA trades; the data-quality point is simply that the tags must exist before any per-strategy number can.

A partial close is not an error — the rows are accurate — but it changes the unit of measurement. Close a single 1.50-lot position in three 0.50-lot steps and one decision becomes three history rows: row-level statistics can score it as two wins and a loss, the trade count inflates, and average volume drops to a third of what was actually risked. The platform mechanics behind those rows are covered in the partial close guide; for analytics the remedy is to group rows by position ID so that statistics count decisions, not fills.

Currency mixing is the quiet cross-account version of the same unit problem. A €640 quarter on a EUR-denominated account plus a $710 quarter on a USD account is not “$1,350” — the two figures are in different units. Compare accounts in percentage terms against their own balances, or convert both to one base currency at a consistent rate before adding anything together.

Most data-quality faults are easier to diagnose from the statistic that looks odd than from the raw rows. Working backwards from the symptom:

Several of these faults exist only because a human re-types what the platform already knows. Manual journalling introduces transcription errors — a price typo, a buy recorded as a sell, swap and commission fields skipped — and manual importing introduces gaps and double imports. A journal that syncs read-onlyfrom your own MetaTrader account reads closed trades, swaps and commissions exactly as the broker’s ledger recorded them, from one source, once — which removes the missing-trade, duplicate and transcription categories more or less at the origin.

None of this requires special tooling — it requires a handful of habits applied consistently:

• Reconcile monthly: compare your journal’s closed-trade count and net profit with the platform’s own account statement for the same period.
• Keep one import source per account — never merge overlapping exports.
• Declare a timezone basis for daily and weekday groupings, and convert timestamps to it before grouping.
• Normalise symbol suffixes to one canonical name per instrument.
• Give every EA a unique magic number, and tag manual trades so strategy groupings stay clean.
• Group partial-close rows by position ID when you want decision-level statistics.
• Pick one basis for cross-account comparison — percentages, or a single base currency — and stick to it.

The payoff is dull but real: when the records are right, an ugly statistic can be trusted as a fact about your trading rather than dismissed as a possible artefact of the bookkeeping — and that is the whole point of keeping the numbers.