How to follow gold and read a currency pair — XAU/USD, EUR/USD and the points your history is written in

A pair is a price, not a prediction

Every FX quote is two currencies separated by a slash. The first is the base — the thing being bought. The second is the quote — the thing you pay with. EUR/USD at 1.0850 means one euro costs 1.0850 dollars; if the number rises, the euro got stronger and the dollar got weaker. USD/JPY at 150.00 means one dollar buys 150 yen; if that number rises, the dollar got stronger. Gold uses the same grammar with a metal code: XAU/USD is one troy ounce of gold priced in dollars (XAG is silver, XPT platinum, XPD palladium). Crosses such as EUR/GBP or GBP/JPY are two dollar legs combined — EUR/GBP is simply (EUR/USD) divided by (GBP/USD) — which is why a cross can move when neither of its two legs has moved much. In MetaTrader the tickers are usually written without the slash (EURUSD, XAUUSD), but some brokers label gold GOLD, XAUUSD.m or GOLD.silver. The ticker is cosmetic; the specification behind it is not.

Where "points" come from — and why brokers disagree

A point is one unit in the last decimal place of the quote. How many decimal places there are is a property of the symbol, not of the market: a five-digit EUR/USD quote moves in steps of 0.00001, a three-digit USD/JPY quote in steps of 0.001, and gold is usually quoted to two decimals (0.01) but by some brokers to three (0.001). The classic pip is the fourth decimal on EUR/USD — 0.0001 — which is ten five-digit points, and the word survives from an era when quotes had four decimals. None of this matters to the market and all of it matters to your history file, because a result recorded as "85" is meaningless until you know whether it meant 85 points of 0.00001 or 85 points of 0.01.

The arithmetic that turns a price into a result

Points first, money second. A buy of EUR/USD at 1.08432 closed at 1.08517 made 0.00085, which is 85 points at a 0.00001 point size. A sell of the same pair at 1.08517 closed at 1.08432 also made 85 points, because a short profits when the price falls. A buy of gold at 2648.35 closed at 2651.20 made 2.85, which is 285 points at a 0.01 point size. Money per point is point size times contract size times lots: on a typical EUR/USD contract of 100,000 base units and a 1.00 lot, one 0.00001 point is about one dollar; on a typical gold contract of 100 ounces and a 1.00 lot, one 0.01 point is also about one dollar. Both contract sizes vary by broker and by account type, which is why the only safe source is the symbol specification in your own terminal — Market Watch, right-click, Specification — rather than a number you read on a forum.

From points to money: what one point is worth

A point is a price step. Money per point is contract size times point size, and that product is denominated in the quote currency — which is the detail that catches people out. On the two symbols most copy-traders start with, the arithmetic is short enough to memorise. A standard EUR/USD contract is 100,000 euros, and 100,000 × 0.00001 = 1.00, so one point is about one dollar on a 1.00 lot and 100 points is about 100 dollars. A standard gold contract is 100 troy ounces, and 100 × 0.01 = 1.00, so one point is again about one dollar on a 1.00 lot and a 285-point gold move is about 285 dollars. That coincidence is why traders loosely say “points” when they mean “dollars per lot” on these two symbols — and why that habit becomes expensive the moment they change symbol.

Typical retail specifications, shown so the multiplication is visible — not a quote for your account. Contract size, point size and the currency a point is paid in are broker settings; some brokers also offer “micro” and “cent” accounts that shrink every figure by a factor of ten or a hundred.

The JPY row is the one to study, because it breaks the coincidence twice. One point on USD/JPY is 100 yen, not one dollar, and the dollar value of those 100 yen moves with the exchange rate itself — so a “100-point stop” on USD/JPY is a different amount of money every day, in a way that a 100-point stop on EUR/USD is not. Indices and crypto break it a third time by varying both the contract and the point size, which is why “one point” is meaningless there until you have read the specification.

The same arithmetic answers the risk question, run backwards. If a stop sits 200 points away and one point is worth one dollar on 1.00 lot, one lot risks 200 dollars and a 0.05 lot risks ten. Stop distance in points and risk in money are therefore the same number in two units, and a copy-trading journal that records points without a lot size cannot answer the only question that matters — how much of the account was actually on the line. The MT4 / MT5 Genetic Copy-Trade Lab keeps that link explicit: give it the cash value of one point and every result gains an account-currency row beside the points, so a holdout that reads “1,034 points” also reads as money per lot.

Gold is a currency pair with no country

Gold pays no interest and no dividend, so it is not priced like a company or a bond. It is priced as the stand-off between the dollar and everything else, and four forces do most of the work. First, the dollar itself: a stronger dollar makes gold more expensive for every non-dollar holder, so XAU/USD tends to fall as the dollar index (DXY) rises. Second, real yields — higher real interest rates raise the opportunity cost of holding a metal that yields nothing. Third, official-sector demand: central-bank buying and ETF flows are slow, large and trend-setting. Fourth, fear: geopolitical shocks and risk-off sessions move gold fast and unreliably. The practical consequence is that gold and EUR/USD often rise and fall together, and the shared driver is usually dollar weakness. That is one bet expressed in two instruments, not two independent confirmations.

What moves EUR/USD

EUR/USD is the rate differential plus the data that reprices it. When the market expects the ECB to cut faster than the Fed, the pair tends to fall; when it expects the opposite, the pair tends to rise. The releases that matter are the ones that move those expectations — inflation prints, PMIs, employment data and the central-bank decisions themselves. Behaviour through the day is fairly consistent: the Asian session tends to range while Europe is closed, and the day's directional move usually appears in the London afternoon, with the New York session either extending it or taking it back. Correlation with DXY is strongly negative and with GBP/USD strongly positive, so a EUR/USD chart read without the dollar index is half a chart.

The trading day, in UTC

Fix your clock to UTC before you read any chart, because every session boundary below is a UTC boundary. Sydney runs roughly 21:00–06:00 and is thin, gappy and expensive. Tokyo runs 00:00–09:00 and tends to produce ranges rather than trends. London runs 07:00–16:00 and is where most of the gold volume is fixed. New York runs 12:00–21:00 and brings the US data. The overlap, 12:00–16:00 UTC, is the deepest and most volatile part of the day for both gold and EUR/USD. Two fixed points are worth memorising: the London gold reference price is set at 15:00 GMT, and the big US releases land at 13:30 UTC (08:30 New York). Spreads also widen at the 21:00–22:00 UTC rollover, when swaps are booked — which is precisely when a copy-trade is most likely to be filled at a worse price than the signal assumed.

"Gold stock" is three different instruments

If you are following gold through a stock exchange rather than a terminal, you are probably watching one of three things and they are not interchangeable. Spot gold (XAU/USD) is the metal itself, traded around the clock with no exchange and no share count. A gold ETF is a share that tracks the metal's price minus an annual fee, so it is close to the metal with a slow leak. Gold-mining shares are an equity in a company that digs the metal: they add management risk, cost inflation, permitting, hedging and leverage to the metal, and they routinely fall while the metal rises. The three are correlated. They are not the same trade, and a backtest of one says almost nothing about the other two.

How to follow the move without fooling yourself

The routine that survives contact with reality has five steps. Fix your clock in UTC and know which session you are in. Read the pair together with its driver — gold with DXY and real yields, EUR/USD with DXY and the ECB/Fed rate path — and if they disagree, the honest answer is "no trade". Keep an economic calendar and mark the 13:30 UTC releases on your own charts, because that is when copy-trading is most likely to be late and most expensive. Record the spread and the execution delay at the moment you copy, because that gap is the difference between the signal's result and yours. And journal every trade with the features a backtest can actually learn: time, side, signal score, spread, copy delay and, if your EA can log it, the maximum favourable and adverse excursion. Then judge the system on data it was not tuned on, never on the trades you used to tune it.

From an MT4 / MT5 report to a CSV the lab can read

Your terminal already holds a trade history; turning it into a file the lab can search takes four honest steps, and the fourth is the one most people skip.

1. Set the period before you export

MT5 keeps history in the Toolbox (Ctrl+T) on the History tab. Right-click it to choose the record type — Positions, Orders or Deals — and the period, then wait for the list to finish loading before exporting, because the export takes what is on screen and a long history is fetched from the server. MT4 does the same thing on the Account History tab: right-click, choose the period, right-click again.

2. Export the report

MT5: right-click → Report → Open HTML or Open XML (Excel). There is no CSV button in the terminal itself — the CSV comes from the spreadsheet you open the file in. MT4: right-click → Save as Report or Save as Detailed Report, which are both HTML. Either way you end up with a table you can save as CSV.

3. Choose Positions, not Deals

One position is one trade, but one position can be closed by several deals. Export the Positions view. If you export Deals instead, a single trade appears as several rows — an entry fill and its exit fills — and every number downstream is double-counted. This is also why the lab warns you when rows look identical: a duplicated position is the most common way a copy-trading backtest quietly inflates itself.

4. Add the columns the report does not have

A positions report gives you open time, close time, symbol, type (buy/sell), volume, open price, close price, stop loss, take profit, commission, swap and profit. That is enough to derive signed points from the two prices — give the lab the point size and it does the arithmetic — but it says nothing about why the trade was copied. Signal score, the spread at the moment of the copy, the delay between signal and fill, and the maximum favourable and adverse excursion all have to be logged by your own EA or copier. Without them the search can only learn a session filter and a direction filter, which is worth knowing before you interpret the result.

Two traps in the same paragraph. First, the timestamps in the report are in your terminal's clock, not necessarily UTC: either set the terminal to UTC or convert the column, because the session filter is only as good as the clock. Second, the profit column is in account currency and includes commission and swap; if you paste it as points you are mixing price movement with costs and financing, which is fine as long as you know that is what you did.

Reading the lab's output without fooling yourself

The results table has four columns for a reason: train baseline, train evolved, holdout baseline, holdout evolved. The first two describe the period the search was allowed to look at, so they flatter by construction. The comparison that carries information is the third against the fourth — the same filter, applied to trades it never saw. If the holdout evolved column is worse than the holdout baseline column, the filter did not work, and no amount of training-segment beauty changes that.

Read the metrics in this order. Trades kept first, because a filter that keeps three of thirty trades has not found an edge, it has found a small sample. Net points second, because win rate alone is a trap: taking many small wins and one large loss raises the win rate and destroys the account. Maximum drawdown and longest losing streak third, because the fitness function already penalizes both, so a genome that scores well has at least survived its own history. Profit factor last, as a summary rather than a verdict.

The seed stability table is the overfit guard. The identical search is run five times with five seeds; if the thresholds swing between seeds, the first answer was the seed, not the market. Treat "all five seeds positive" as weak evidence and "three of five" as noise. The warnings block is not decoration either: a disabled feature means the column was missing, incomplete or out of range, and a search that was never allowed to test spread cannot tell you anything about spread.

The honest next step is forward, not backward. Take the filter that survived the holdout, run it on a demo for a period you already have data for, and only then judge it on the trades that came after. The one thing that turns this from research into self-deception is re-running the search on the same period until a filter appears — every extra run spends the holdout, and there is only one.

From market to CSV: what a backtest can honestly learn

A search can only learn from columns that were knowable at the moment of the copy decision. Time gives it a session filter. Side gives it a direction filter. Signal score, spread and copy delay give it quality and cost filters. Maximum favourable and adverse excursion give it a basis for researching take-profit and stop-loss levels — with the honest caveat that excursions cannot reveal intrabar order, so a search that sees both a target and a stop touched on one trade assumes the stop came first. What it cannot use is anything learned after the trade closed: the final price, the final profit, the eventual outcome. A column filled in from the future will make a search look brilliant and be worthless, which is the single most common way a copy-trading backtest lies to the person who paid for it.