Diversification is said to be the "one free lunch" in investing. In principle, by maintaining a portfolio of non-perfectly-correlated assets, you can increase risk-adjusted returns: either increase returns while keeping volatility the same, or decrease volatility while keeping returns the same, or any combination. See Modern portfolio theory on Wikipedia.
This tool provides backtesting and optimisation of model portfolios.
Loading assets…
* Asset total return history begins after the start of the selected range. Before that date, this asset is excluded and the other selected asset weights are scaled up to reach 100%.
† Asset has no total return history on or before the selected end date, so it is disabled.
Loading chart…
Why is my mean ann. return sometimes positive when CAGR is negative? The mean annualised return is the arithmetic mean of annualised weekly returns, while CAGR only depends on start and end equity. A path with many small positive weeks and one very large loss can still show a positive mean annualised return while CAGR is negative.
Leverage risk: use the leverage slider to see how quickly risk can rise. Even with funding disabled, leveraged strategies can underperform unleveraged ones, due to volatility drag and rebalancing.
Risk of ruin: this simulation does not model margin calls, forced liquidation, or position limits, and it allows negative equity. Real-world leveraged outcomes can therefore be worse than shown here.
Sharpe vs realised wealth: in principle, a higher Sharpe ratio indicates a better risk-adjusted strategy. In practice, you "can't eat a Sharpe ratio" and may care more about CAGR or mean annualised return. You also can't scale risk-adjusted returns for free: leverage typically incurs funding costs and introduces additional implementation risk.
Costs and taxes are excluded: this simulation ignores trading fees, bid/ask spread, slippage, borrowing fees above the risk-free rate (if selected), and all taxes. Real-world returns would therefore be lower than shown.
Lookahead bias: tuning parameters based on historical results will overstate what could have been achieved in reality, because those choices implicitly use future information. A better sanity check is to split history into two halves, optimise on the first half, then evaluate unchanged settings on the second half. But note that if you then change anything at all and re-check, you reintroduce lookahead bias. In real life you don't get to rerun the last 20 years just because you didn't like your portfolio's performance.