Diversification is holding several exposures whose results do not move together, so that a bad period in one is offset by an ordinary period in the others. Done properly it reduces the size of drawdowns and the variability of returns while leaving the average return roughly where it was. It is the one thing in trading that improves a result without asking the market for anything.
What does not diversify
Trading more instruments in the same family does not. Ten currency pairs that all include the dollar are one bet on the dollar. Five stock indices are one bet on equities. Adding instruments spreads the same risk across more names and adds costs.
Running the same strategy on more timeframes does not diversify much either, since a breakout on the four-hour chart and a breakout on the daily chart are usually the same move.
What does
Different markets whose drivers differ, currencies, equities and gold, help, though their correlations rise in stress. Different kinds of strategy help more. A strategy that buys pullbacks in a rising market and a strategy that fades weekly extremes in a ranging one make their money at different times, and their bad periods rarely coincide. A trend follower and a mean-reversion system are close to opposites in what they need, and holding both is the most reliable smoothing available.
The arithmetic
Two strategies with equal returns and equal maximum drawdowns of 20 per cent, run together at half size each, do not produce a 20 per cent drawdown unless their worst periods coincide. If their results are uncorrelated, the combined drawdown is typically closer to 14 per cent, and the combined return is the average of the two. The return per unit of drawdown has improved by about 40 per cent, from nothing more than not putting all the eggs in one basket.
Strategy A trades stock index pullbacks and has a bad year when equities fall. Strategy B trades weekly currency reversals and is indifferent to equities. In the year equities fell 20 per cent, A lost 14 per cent of its allocation and B gained 6 per cent of its allocation. A combined account at half each fell 4 per cent. A trader running only A sat through 14. Both traders held the same A. One of them also held something that did not care.
A multi-strategy EA is diversification in one program, and whether it delivers depends on whether its strategies are actually different: different markets, different logic, different timeframes, and, on the record, results that do not rise and fall together. An EA that runs six variations of one breakout rule on six pairs is one strategy with six names.
The test is in the record. If a vendor can show the strategies' results separately, and they have their bad months in different places, the diversification is real. If every strategy lost in the same month, they were the same bet.