A consistent Sharpe above 3 with useful capacity is super hard to achieve. Years ago I helped create a quant hedge fund and I personally developed strategies that delivered a 3-year Sharpe ~2 fully market-neutral return to our investors. We won some hedge fund industry awards for that performance. It was damn hard to achieve several years ago and much harder to achieve today.
what was the capacity of that strategy? Is it still working? For 2-3 years of operation, is sharpe calculated from daily or monthly, which is quite different result?
I've always used Sharpe ratios calculated separately from daily, monthly and annual returns. Each time scale reveals different performance characteristics. If I had to pick only one I'd pick Sharpe ratio of monthly returns, annualized of course. I typically process daily returns though a moving-window convolution filter to give me monthly or annual returns sampled daily. The strategies I developed for a quant hedge fund years ago very likely don't work anymore. We had a few traders and quants working to improve those strategies for years. They made some good progress but then both the original and improved versions all started failing around the same time. I'm not really interested in redeveloping those old designs since they weren't very extensible. It was hard to incorporate and properly evaluate a large number of input factors working together - the common challenge AI / ML developers deal with. I use better approaches now, a decent ML algorithm and a good bit of ML supervision. Question of capacity is hard to address because one can trade larger size and endure a bit more market impact. For my current-day personal strategy development work: a rough estimate of a safe max capacity for each strategy is 20 FDAX contracts at 1:1 leverage to produce a 20% - 30% annual return with Sharpe ratios of 2.0 to 2.5. That gives a capacity of about $8m. This is all for each individual strategy design, on average. These are also market-neutral strategies that are very weakly correlated to each other. So its best to trade these as large portfolios of individual strategies. For a quant hedge fund I would want to trade a portfolio of 50 to 2000 strategies to give a total capacity of very roughly $1 to $10 billion with likely Sharpe ratios of 2.0 to 2.5.
The Sharp Ratio is an excellent ratio. But I think the Sortino Ratio is the better ratio. The reason is that the Sharp Ratio measures volatility in relation to profit. But even the sharp increase in the portfolio is considered high volatility and then gives worse Sharpratios than slowly increasing portfolios. The Sortino Ratio measures only the volatility of drawdowns and is therefore the better metric for evaluating the performance relative to the drawdown of a portfolio. The same problem arises with the standard deviation and the Ulcer Index.
Consider this system: Win rate 50%, winner size: 2, loser size 1. On average how many trades a month do you need to take or find with such a system to get a sharpe ratio of 3.0 over the long run? Answer: just 7 trades per month, or about 80 per year. So why dont more people have sharpies of 3? For automating trading, 50% win rate with that win/loss ratio (2:1) is hard to find, your win/loss ratio is going to be closer to 40% than 50%. And that will lower your sharpie from 3.0 down to 1.2! To get it back up to 3.0 you would need to find 500 trades per year not just 80. For manual trading where it might be more possible if you have a good feel for the market you trade, but most people cant emotionally handle losing 50% of the time.
Could you reach out to me as I like to further understand how you derived to the numbers and do calculations such as this to determine the effect of different win/loss ratios and win/loss percentages. Very interesting.
It is quite easy if you can do basic programming, you can just do Monte Carlo simulations with various win % and win/loss ratios. Then calculate your Sharpe ratio for various sample lengths, 50, 100, 250, 500, 1000 trades etc. There is maybe a formula that can calculate it without running a simulation, something like the Kelly ratio formula, but I just did a simple Montecarlo simulation.
Thanks for sharing on this. I am going to look into it further and see if I can derive into this Sharpe ratio. With limited knowledge at the moment, I am not sure what is considered a good ratio or range of ratios to be within, but I am sure once I can research further that I will come across this. Thanks again!
A Sharpe ratio of 2.0 and above is excellent. Not many hedge funds can do that, only the likes of Rentec Medallion can do it at scale. A ratio somewhere between 1.0 and 2.0 is probably more realistic for automated trading. If you are below 1.0, you might as well just buy and hold the SP500 or NDX. If you develop systems you have to be careful not to curve fit. Very easy to fall in to the trap of combining multiple systems together that have amazing Sharpe ratios on back data but are no where near as good in real world trading.