Midas
Buy
Home / Methodology and data

Methodology and data

Every statistic Midas Grid publishes about gold grids comes from one source: a tick-by-tick replay of XAUUSD price history from 2 January 2025 to 23 October 2026, run through an engine that applies the rules of a martingale-style grid order by order. This page says where the ticks come from, what the engine does and does not copy from a real broker, and where it is wrong. Every number on the site that comes from it is labeled "in the model".

Where does the price data come from?

  • Source: the public tick history of the Dukascopy Swiss bank for XAUUSD (gold against the US dollar).
  • What is stored: the bid price and the time of every tick, to the millisecond. The data has no ask price.
  • Size: 467 trading days, about 143.7 million ticks, from 2 January 2025 to 23 October 2026. In this window the bid ranged from about 2,614 to about 5,597 dollars per ounce, with one very sharp fall in January 2026.
  • Ask price: the engine adds a constant 0.40 dollars to the bid. Real spreads on gold are wider at news releases and around the daily rollover, so this is a simplification (see below).
  • Gaps repaired: an early version of our tick base silently lost about a fifth of the ordinary trading hours when the downloader gave up after failed retries. We found it by comparing against direct requests, re-downloaded the missing hours and merged Sunday-evening ticks into Monday. Every result on the site is computed on the repaired base only.

How does the engine work?

The engine reads the ticks in order and makes every decision on a tick, never on a candle:

  1. Entry. When there is no open basket and the clock is inside the trading window, the engine opens the first order in the direction given by a simple moving-average rule on one-minute candles, using only candles that have already closed. We did not tune the entry rule; it is there so that the chain of baskets is repeatable.
  2. Grid. If price moves against the basket by one grid step, the engine adds the next order, larger by the chosen multiplier. A pause of 30 seconds between orders and a delay of 0.3 seconds from signal to fill are applied, and the fill price is taken from the tick history after that delay.
  3. Exit. The basket closes when the price returns to the break-even price plus the take profit. A stop-out is applied when it is due.
  4. Costs. Spread (0.40 dollars between bid and ask) and a commission of 0.07 dollars per lot for the round trip.
  5. Account. A cent account with leverage 1:2000, a minimum lot step of 0.01, and a broker stop-out when the margin level falls to 40%.

The result of a replay is a list of baskets with how deep they went (how many orders were open), how long they lasted, how much they made or lost and how much balance each of them needed in order not to be stopped out.

What do we call a "death", a "order" and a "basket"?

  • Basket (series): all the orders opened in one direction from the first order until the position is closed.
  • Order: one order of the grid. "Reached the fifth order" means five orders were open at once.
  • Death: a broker stop-out. In the model the account is then refilled to the original amount and the engine starts again from the next tick, so "deaths per year" is a count of stop-outs on a fixed deposit, not the share of runs that end in zero.
  • Needed balance: for each basket, the smallest balance that would have kept the margin level above 40% for its whole life.

How was the engine checked?

  • Against the broker's rules. The margin, stop-out and lot-step logic was checked by hand calculations and by automated tests.
  • Against a real terminal. The engine was calibrated against trade journals from a real terminal, and later versions replay individual real baskets to the second. A replay matches in the typical case, but a single basket can differ, and the match says nothing about the future.
  • Against look-ahead. An earlier line of our research was cancelled after we found a one-minute look-ahead in the way candle times were labeled. Since then the rule is that an entry may use only a candle that has fully closed.
  • Against the old tick base. All earlier results from the incomplete base were thrown away and recomputed.

A model is not a verdict. A replay can say what would have happened on past ticks under stated rules; it cannot say what your broker will do.

Where is the model wrong?

  • Constant spread. Real spreads widen at releases, at the rollover and in thin markets; the model keeps 0.40.
  • No slippage beyond a fixed delay. Real fills can be worse in fast markets, and some orders may be rejected or requoted.
  • Another broker's prices. Dukascopy quotes are not your broker's quotes; a stop-out can come a few cents earlier or later.
  • Simplified margin. One leverage and one stop-out level for the whole account; real brokers can change both by size, instrument or time.
  • No swaps, no weekend gap management beyond what is in the ticks.
  • One window, one trend. The sample is under two years long and contains a very strong rise and a sharp reversal of gold. A different period gives different counts of stop-outs. Counts of two or three deaths are not statistics.
  • Cap on precision. A death count per year is a noisy estimate; read it as an order of magnitude.

How to use the numbers

Treat them as a way to compare settings with each other: a deeper grid, a wider step or a smaller deposit changes the stop-out count in a direction the model shows reliably. Do not treat them as a promise. Trading with leverage carries risk: if the deposit is too small, a stop-out is possible. Size your deposit with a safety margin. Nothing on this page is individual investment advice.

Run it yourself

Questions about the method go through the support bot described on the support page.