How we test
Every change to the Velocity Portfolio EA goes through the same tests before it ships. Most ideas fail them. This page explains the tests and lists some of the ideas that did not make it.
The test setup
| Software | MetaTrader 5 Strategy Tester, one backtest at a time. The MT5 optimiser is never used. |
|---|---|
| Account | $10,000, leverage 1:100, gold (XAUUSD) on the M15 chart, every setting at its default. |
| Data | Darwinex real ticks, IC Markets every tick and real ticks, and TMGM. Three brokers, so a result cannot come from one broker's price feed. |
| Execution | MT5 random delay switched on, so fills are not perfect. Spread, commission and swap are included. |
| Dates | 2020 to the latest data available. Every test runs to the newest date, so a strong or weak last month cannot be left out. |
| Drawdown | Quoted as MT5's "Equity Drawdown Relative", the worst drop from a high, including open trades. |
The rules for a change
Write the pass rule first
Before the first test runs, we write down what result counts as a pass. It does not change after the results come in.
It must work on every feed
A change has to improve the results on IC Markets and Darwinex together. Better on one and worse on the other is a fail, because that usually means it fits one broker's prices.
It must hold up on data it has never seen
We also test on older data that was never used to build or tune anything. A change that only works on the years it was built on is rejected.
Check every year, not just the total
One great year can hide several bad ones. We read the result year by year, and pay most attention to the weakest year.
Record the result either way
Passes and fails are both written down with their numbers, so a failed idea is never quietly tried again.
Some ideas that failed
A neural network to predict direction
It could not call direction better than chance once trading costs were included.
24 versions of the London open breakout
Several looked good on recent years. None held up on older data.
About 20 Nasdaq strategies
One passed every test and became Pulse. The rest were dropped.
More weight on the strongest strategy
More profit on paper, but it doubled the loss in the weak years.
A "safe" grid strategy
The safety layer worked. The strategy underneath did not make money.
Letting one strategy block another
It cut the worst drop by less than 1% and cost more than half the profit.
What testing cannot tell you
A backtest shows how the rules would have traded in the past. It cannot promise the future. Live results will differ because of your broker's spreads, slippage, and how the market changes. That is why there is a free demo: run it on your own broker before you trade real money.
The full MT5 reports behind the published numbers are on the reports page.