Spread, swap and latency: the three silent costs of gold

Costs you never see on the chart

Charts show price. They do not show what you actually paid to participate in it. On gold, the difference between a well-executed and a poorly-executed identical strategy is routinely larger than the strategy's entire edge, and none of it appears in a candle.

There are three costs, they behave differently, and they compound in the same direction: against you. Measuring them takes an afternoon and it is the single highest-return afternoon most automated traders will ever spend.

Spread: variable exactly when it matters

Advertised spreads are averages taken during calm liquidity. Gold's spread expands sharply around the London open, the New York open, monetary policy releases and the daily rollover window — which is to say, precisely when systematic strategies are most likely to be transacting.

• —Log the spread every minute for a full week using a simple recorder script, not a vendor's marketing page.
• —Compute the median and the 95th percentile separately. The 95th percentile is your real planning number.
• —Compare rollover-window spread against midday spread. A 10x expansion is common and is worth avoiding entirely.
• —Repeat the measurement quarterly. Broker pricing changes without announcements.

Swap: the cost of patience

Swap is charged for holding overnight and is triple-charged on one weekday to account for the weekend. On gold, swaps are frequently negative on both sides, which means simply existing in the market costs money regardless of direction. For a strategy that holds positions for days, swap can consume a double-digit percentage of gross profit.

Latency and fill quality

Latency is the delay between your terminal deciding and the broker acting. It matters less than most people believe for strategies holding hours or days, and far more than most people believe during volatile releases, where a 300-millisecond delay on gold can mean several dollars of adverse movement per ounce.

• —Measure round-trip latency to your broker's server, not to a generic ping target.
• —Track realised slippage per trade as a running average — it is more informative than latency itself.
• —Watch requote and rejection frequency. A broker that rejects during volatility is a broker that costs you the trades that mattered.
• —If you host on a VPS, choose one physically near the broker's execution servers rather than near yourself.

Building a real cost model for your own account

Costs are not a footnote; on a strategy that trades frequently they are the difference between a profitable year and a flat one. The only way to know your true cost per trade is to measure it on your own account, at your own broker, during the hours you actually trade — not to read the spread advertised on a landing page.

Measure, then multiply

• —Log the spread at order time for every trade for one month.
• —Add commission per round turn and any swap actually charged.
• —Divide the total by the number of trades to get true cost per trade.
• —Multiply by expected annual trade count — that is your annual cost hurdle.

Once the annual figure exists, the comparison between brokers becomes arithmetic rather than opinion, and the difference between a good and a mediocre execution venue usually turns out to be worth more than any parameter change you were considering.

Where latency actually matters, and where it does not

Latency is the most over-discussed and least understood cost in retail automation. For a strategy holding positions for hours, a hundred milliseconds is irrelevant. For a strategy modifying stops around a data release, the round trip to the broker is the difference between the level you intended and the level you received.

The three latencies worth separating