Forward Testing and the Limits of Demo

A backtest tells you the rules would have worked. Forward testing tells you something different and equally necessary: whether they work on data nobody has ever fitted them to, and whether you can actually execute them. Those are two separate questions, and a demo account answers the first one much better than the second.

What demo leaves out

A demo account reproduces the platform faithfully and the experience barely at all.

The mechanical differences matter too, and they all run in the same direction:

• Fills are optimistic. Demo servers usually fill at the quoted price, with no slippage and no requotes.
• Spreads are often the advertised ones, not the widened ones you meet at 22:00 or around a release.
• Your size is imaginary, so nothing is tested about liquidity at the size you intend to trade.
• Demo balances are unrealistic. Trading a $100,000 demo when you will fund $2,000 tests a strategy you cannot run.

Sample size before conclusions

How long to forward test is not a question about weeks. It is a question about trades, and the answer comes from your frequency.

The bottom rows are uncomfortable and honest. A strategy taking two trades a week genuinely needs months before its results mean much, and that is a reason to weight the backtest more heavily rather than a reason to conclude early from ten trades.

• Thirty trades is a minimum for a sanity check — enough to catch a strategy that is obviously broken, not enough to confirm an edge.
• A hundred is where expectancy starts to mean something, as Track 5 established.
• A low win rate needs more of both. A 30% win-rate method can easily produce thirty trades containing four winners or fifteen.

Conditions matter alongside the count. A hundred trades in one quiet trending month has tested the strategy against one month, and the honest description of that result is "it survived a trend", not "it works".

Small live size as a middle step

The gap demo leaves — execution under real pressure — is closed by real money, and the size needed is much smaller than people assume.

Trade the minimum size your broker allows, or use a cent account. At 0.01 lots on EUR/USD a 30-pip loss is $3. That is small enough not to matter and, crucially, not zero — and the difference between $3 and $0 turns out to be most of the psychological difference.

What to measure at minimum size

1. Deviation rate. What share of trades departed from the rules in any way? Under 5% is a working process; above 20% means the problem is not the strategy.
2. Fills versus expectation. Your actual slippage against what the backtest assumed.
3. Frequency. Did the strategy produce roughly the number of trades the backtest implied? A large shortfall means you are filtering trades the rules would have taken.
4. Expectancy in R, which is comparable across sizes and therefore comparable with the backtest.

Point 1 is the reason this stage exists. A strategy with a positive expectancy and a 30% deviation rate is not being traded, and moving to full size makes that more expensive rather than less.

The abandon criteria, set first

Before the test starts, write down what result would make you stop. Afterwards is too late, because by then every result has an explanation.

Four numbers, decided in advance:

• The drawdown criterion protects against a broken strategy, and it has to exceed the backtest's worst drawdown or you will abandon a working method during a normal streak. Track 5's drawdown lesson is the reference.
• The trade-count minimum protects against abandoning too early, which is the more common error. Most strategies are dropped inside twenty trades, where the result is indistinguishable from noise.
• The deviation criterion is the honest one. If you did not follow the rules, the test measured nothing about the rules.

Then let the test run to its own conclusion. The value of pre-set criteria is entirely in the moments when you want to override them, and a forward test stopped early because it felt wrong has produced the one result that means nothing at all.