I see why sharpe of 3 is so desirable

[Palindrome]

I trade futures, directionally, place bets on where it will go. Average hold between 2 hours and 2 days or so. Long short. I trade about 10 different markets (energy, currencies, metals, equity indexes. etc)

It took me 8 years to learn this, 8000 hours or so.

 

[SabreMan]

I trade futures, directionally, place bets on where it will go. Average hold between 2 hours and 2 days or so. Long short.

More proof that short term trading gives the best Sharpe ratio and are superior to longer term trading for smaller accounts (below $10 million).

 

[rallymode]

More proof that short term trading gives the best Sharpe ratio and are superior to longer term trading for smaller accounts (below $10 million).

Not sure any proof is needed. Assuming you have alpha, the higher the frequency the higher the sharpe(daily). IMO, sharpe is useless for short term traders. It's all about $PnL.

 

[Palindrome]

Rallymode,
I hear ya but is it more impressive comparing:

2million account generating $500,000 in profit
or
700k account generating $500,000 in profit

I know PNL is all that matters in reality, but it has it's limitations in comparing results.

 

[drm7]

I only have a very basic understanding of stats but I think i am understanding why a Sharpe ratio of 3 and above is so desirable.

My current system i am developing (day trading system) has a sharpe ratio of 2 .
It makes 50% a year with a standard deviation of 25% (50:25= sharpe ratio of 2, im ignoring the risk free rate)
As i said my understanding of stats is pretty basic but i think this means that 99.73% of my annual returns should fall between -25% and +125% (50%+/- 75%). As 3 standard deviations of 75% each side of the mean.

Now if my system had a Sharpe ratio of 3, the same 50% a year return but with only 16% stddev. Then the 99.73% range for the yearly returns would be 0% to 100% a year. That would mean no losing years, ever!

On the other hand a sharpe ratio of 1.0 would be quite bad eg 50% a year with 50% stdev.
This means my returns could fall anywhere between -100% and +200% a year.
So you would have to use a really low risk size with a sharpe of 1.0

Another useful tool of the Sharpe ratio is position sizing. It is mathematically linked to the optimal Kelly bet formula. The "continuous Kelly leverage" is calculated by Sharpe/STDDEV. So, a 50% return/25% STDDEV, K = 2/.25 = 8x leverage to maximize lifetime log return of the account. Of course, it is only a theoretical number, and the slightest deviation from your return record can blow you out. You also have to maintain the leverage "continuously" i.e. - buy into a higher equity curve and sell down into a lower equity curve. The drawdowns at full Kelly are huge as well.

To extend your example:

K (sharpe = 1) = 1/.5 = 2
K (sharpe = 3) = 3/.1667 = 18 (!!!)

Ernie Chan explains this in his book Quantitative Trading (and also this blog post)

http://epchan.blogspot.com/2006/10/how-much-leverage-should-you-use.html

 

[SabreMan]

So, a 50% return/25% STDDEV, K = 2/.25 = 8x leverage to maximize lifetime log return of the account.

Cool, thanks for that. Im using 3x leverage at the moment to boost my returns to over 150%, but that is based on the max leverage IB will give me. At other brokers i could go closer to x8 leverage if i was starting out small.

The drawdowns at full Kelly are huge as well.

I think the idea is you decide how much max drawdown you want and then use kelly on that.
So if you have 1million and only want to risk 250K, you would assume your equity is 250K and use close to full kelly on that. That way 750K is safe.
At least that is what Ed Thorp said in Hedge fund market wizards.

 

[SPYAlgoTrader]

I would feel uncomfortable trading anything less than 4 or 5.. takes too long to see if the edge is still working or not.

that's correct: high sharpe systems show a straight uptrend, once that's broken it usually indicates a system stopped working.

 

[sle]

Not sure any proof is needed. Assuming you have alpha, the higher the frequency the higher the sharpe(daily). IMO, sharpe is useless for short term traders. It's all about $PnL.

That and beating the costs. There are so many things out there where you can make a few hundred K a year with a Sharpe of 3, but scaling it to something meaningful like 5-10 bucks never seem to work.

Sharpe is pretty misleading, all things considered. You can have all sorts of artifacts due to the fact that it penalizes any discontinuity, so big positive days can easily lower it.

 

[fordewind]

I would feel uncomfortable trading anything less than 4 or 5.. takes too long to see if the edge is still working or not.

Playing with the equity curve?

 

[cruisecontrol]

Playing with the equity curve?

I'm not sure what you mean by this. To elaborate on why high sharpe is important to me: it is because sharpe indicates statistical significance of my edge(s). Especially if the edge is developed through data mining instead of some kind of fundamental economic analysis, you need a way of having confidence in the pattern. It is similar to a scientist looking for a p-value below a certain threshold. Without a high sharpe, a backtest is about as useful all those studies that say chocolate reduces your risk of cancer, while the other half of studies found the opposite.