Is the real issue finding the right balance between maximizing the Sharpe ratio and minimizing drawdown?

New comment
 

When facing the strategy tester, I think the most fundamental thing is maximizing the Sharpe ratio and minimizing drawdown. Since I don't want to get into an echo chamber, I'd like to ask you guys: what do you think about this?

 

About Sharpe - old posts:  and  
The thread: Sharp Ratio

for information

 

Good question, and the thread Sergey pointed to is worth reading.

I would push back gently on the framing though. Coming from the optimization side, the trap is that both Sharpe and drawdown are in-sample numbers. If you maximize either one on the strategy tester, you are usually not finding a better strategy - you are finding the parameter set that best fits this specific slice of history. The "best" Sharpe run and the lowest-DD run are very often the most overfit ones.

A few things that helped me more than chasing a single optimal pair:

1. Judge the parameter neighborhood, not the peak. A robust setting sits on a plateau where nearby parameter values give similar results. A tall isolated spike in Sharpe is a red flag, not a prize.

2. Split the data. Walk-forward or a simple out-of-sample window tells you more than any in-sample metric. A mediocre Sharpe that holds up out-of-sample beats a great one that collapses in-sample.

3. Separate the two problems. Drawdown is partly a strategy property and partly a risk-management property. A lot of what people try to optimize away in the tester (deep DD) is better controlled at runtime through position sizing and trade management, without touching the entry logic at all.

So the real issue is not the Sharpe vs drawdown balance itself - it is making sure whatever balance you pick was not just curve-fitted to the backtest. Optimize for stability across the data, then let sizing handle the drawdown you can live with.

New comment