Suppose you have 2 trading strategies which give you a positive edge today to make money. One day eventually, they will lose their edge but you don't know when. Strategy 1 ---------- Slightly above 50% win-rate. When you win, you win 1R. When you lose, you lose 1R, unless some accident happens or the security makes a huge unexpected gap. Strategy 2 ---------- Win-rate is way below 50%. You lose most of the time. However, when you win, you win 2R. When you lose, you lose 1R. Which strategy would you prefer, assuming that that the returns in dollar terms is about the same? I would like to ask the elite traders here with practical experience. Which strategy would you prefer and why?
Depends how far below 50% strat 2 is....IT COULD have a higher E[]. As @d08 said, look at some other metrics... I'd go with the Sortino ratio as opposed to the sharpe, as it only looks at volatility of negative returns. Sharpe penalizes both upside and downside vol.
drawdown = risk CALMAR ratio built to evaluate Hedge Funds and CTA programs is a good metric for this. Trailing 36 month return divided by max draw down during the period. It's a real-world return vs risk metric. I personaly prefer this to sharpe/sortino as real-world realised risk is not the same as theoretical modeled risk.
Assuming each strategy has the same expected value per trade AND the same number of trades per time period, I would pick the one with the highest winning percentage, as the Kelly Criterion allowable bet size would be higher. The Kelly formula favors higher win % strategies, because these strategies would have a higher chance of hitting "escape velocity" before a big drawdown. Note that betting "full Kelly" is not recommended, as a) the drawdowns are HUGE and b) you will bust quickly if/when the strategy deteriorates. That is why I mainly use it to compare two similar strategies.
I'd be more interested in how you came up with these stats: did you use a random pool of data (e.g. randomly pick 100+ stocks from different sectors minute data from the last 20 years or create your own random or synthetic data or use currency or commodity data) examine how it handles recessions, how it would buy/sell stocks that loose 90% of their value still looks like a good algo? (99.9% of what an experienced algo trader might come up with will fail these tests... which is good so you won't deploy and loose $) now plug in the stock(s) you think about trading and backtest again random timeframes still beats S&P? yes: deploy and don't meddle with it no: (very likely) just buy SPY
strategy 1 breaks-even 50% at 1 to 1 so is a waste of time and likely be under with commission charges. I would need to know the hit rate of 2 below 50% can be 10% can be 49%