Strategies and how one is selected

1 · Strategy lab (weekly)

For each asset class, 121–168 recipes (entry style × timeframe × stop/target × break-even) are replayed trade by trade on 45 days (5- and 15-minute bars) to a year (crypto) of eToro's own prices. Each trade pays eToro's spread and fees, plus overnight financing on CFDs. Entry is at the next bar's open, a bar that touches both the stop and the target counts as a stop, and a gap through a level fills at the bar's open.

2 · Chosen on older data only

Each market's history is split in half by date. A recipe must pass every one of these on the older half only:

• at least 25 trades on the older half
• a per-trade t-statistic of at least 2 on the older half
• a profit factor of at least 1.1 on the older half
• profitable on at least 60% of the markets tested (older half)
• a win rate of at least 55% on the older half

Among those that pass: among recipes passing every older-half rule, the highest older-half return per market per month.

3 · Checked on newer data

The chosen recipe is then checked once on the newer half, which played no part in choosing it. It must pass all of these, or the asset class is set to stay in cash:

• at least 10 trades on the newer half
• a positive average trade after costs on the newer half
• a profit factor of at least 1 on the newer half
• a win rate of at least 55% on the newer half

Stricter checks (demo accounts)

Many recipes are tried and markets move together, so a per-trade t-statistic overstates confidence. Demo accounts trade a recipe only if it also clears a multiple-testing threshold on a day-clustered t-statistic makes money in a rolling walk-forward test where each fold is chosen without seeing its own data, and stays profitable on the newer half after an assumed market-order slippage (indices 0.01%, fx 0.005%, commodities 0.02%, stocks 0.03%, crypto 0.1% per side). Real-money accounts use these checks once the operator has reviewed them.

A t-statistic threshold is a screening rule, not proof of an edge: many recipes are tested, trades overlap in time and markets move together, so ordinary per-trade t-statistics overstate confidence.

4 · Switching

The lab re-runs every 7 days (the admin can run it sooner). A class switches recipe only when a new run picks a different one; open trades keep their own stop, target and time stop. In the classic best-fit rule, each market re-picks among 10 classic strategies every 6 hours using the previous 600 bars, and the in-app backtest replays that walk-forward: each block of 100 bars uses only the bars before it.