Forums  > Trading  > How high a Sharpe is considered "good?"  
     
Page 2 of 4Goto to page: 1, [2], 3, 4 Prev Next
Display using:  

liquidity peddler


Total Posts: 60
Joined: Aug 2007
 
Posted: 2007-10-02 19:08
As someone who trades mostly his own money from a home office, maybe I can offer the piker perspective.

What I worry about and cannot quantify is : how will the performance of my strategies evolve?

Will it wax, then wane, then wax - in which case what I care about is the periodicity. If one were to measure the sharpe ratio with a trailing measurement period less than the periodicity, it would be bobbing up and down like a cork.

Or will it be strong, then suddenly fall off a cliff? In which case what I care about are developing earning warning signals.

Or will it wax, then very slowly wane and dissappear.

With these three "life cycle of a strategy" scenarios in mind, I track the sharpe ratio of my strategy historical performance and backtest, as instrumentation to guess which bigger picture I am in.

nodoodahs


Total Posts: 227
Joined: Sep 2007
 
Posted: 2007-10-02 22:01

I'm even more of a piker.  Oh, bother.

My level of concern with "strategy performance evolution" is directly related to the complexity of the strategy.  In other words, I don't have much concern, because I don't (and this may be heresy on this board) ever contemplate opening up my 15+ year old differential equations textbook to derive an idea.

One thing that I do, however, is compare the performance of the stock strategies to whether the S&P 500 was in EZ Trend up versus down, and whether the 10YT yield was in the same.  I have noticed that the relative outperformance of some strategies waxes and wanes with the performance of the overall market, i.e., more outperformance when the market is not doing well overall.


I haven’t seen a beatin’ like that since somebody stuck a banana in my pants and turned a monkey loose.

LowDD


Total Posts: 142
Joined: Feb 2009
 
Posted: 2009-02-12 19:50

So, Aaron.  I very much appreciate this post on Sharpe ratios (10/2007).  Your thoughts are innovative, useful, simple.  I think the same way about sharpe ratios (1 to 2 being the interesting range) and you point out the math which fits my intution. But it does seem to be missing something. 

I sometimes ask the question "what is the problem with Sharpe ratios?"  Probably the general answer I get most... which is fine, helpful, in the right direction is: "the tails".  This is a logical focus for people who deal with derivatives, arb trades, etc. My area of focus is different so I tend be looking for: "the MSCI World shows an pretty good Sharpe ratio during certain 18 month periods"  For example, from May 03 to May 06 the MSCI World EM index returned >40% annual at a vol of ~20%.  This is a sharpe of close to 2.  Yet has no bearing on skill versus random selection.

It then follows that the first problem might be that it does not separate alpha from beta.   One element of this is market neutrality... but long MSCI EM / short DAX probably had an even better sharpe! So it follows next that it does not give us any sense of diversity.  That is -- diversity of trades, ideas, etc on 2-sides of a "beta" is what separates skill from luck.  

Or to summarize sharpe is only helpful under a bunch of common-sense portfolio constraints meant to control beta and divesify themes / ideas which together are meant to minimize short-term drawdown potential.  Make sense?  What do you think?


NeroTulip


Total Posts: 996
Joined: May 2004
 
Posted: 2009-02-13 13:48

Reading this thread from the start, I saw this comment from FDAXHunter, which I find interesting:

Note that some people throw around Sharpe Ratios of way above 10. These are always very constrained in the amount of actual profits they can generate. Sure, you can have a Sharpe Ratio of 25, but if you consider that the most you can do is generate a million dollars a year with it, that's not how you build a scalable business. Scalability is never to be underestimated and indeed, is more important to the business as such than pure risk-adjusted performance.

Let's say I am not interested in building a scalable business right now, but rather in getting the highest return on capital. Obviously it would make sense to use these strategies first, until I reach their maximum capacity, and then introduce stategies with lower returns and higher capacity as my capital grows.

But this sounds almost too good to be true... If these opportunities do exit, some people should be trading them on their PA and getting rich rather quick. After all, with a sharpe of 10 you can easily produce returns in the 300% area, unless your returns are negatively skewed. So why do I see no evidence of this happening? Is it just that these people are low profile, is it negative skew, or is it a lack of capital?

I suppose it would be easy to raise capital for such strategies and exploit them fully. But what would be the typical profit split between the trader and his financial backer?


Inflatable trader

jungle
Chief Rhythm Officer
CSD LLC
Total Posts: 3162
Joined: Jul 2004
 
Posted: 2009-02-13 14:49

If these opportunities do exit, some people should be trading them on their PA and getting rich rather quick.

Perhaps a little difficult to build a high frequency market making system for the PA that you can run while you're out doing the day job...


AndyM


Total Posts: 2308
Joined: Mar 2004
 
Posted: 2009-02-13 15:21

Day jobs are so 2007, though...


If you owe the bank a million, it's your problem; if you owe them a hundred million, it's their problem, and if you owe them 10 billion, it's the taxpayers' problem.

lys


Total Posts: 49
Joined: Apr 2008
 
Posted: 2009-02-13 15:35
But this sounds almost too good to be true... If these opportunities do exit, some people should be trading them on their PA and getting rich rather quick. After all, with a sharpe of 10 you can easily produce returns in the 300% area, unless your returns are negatively skewed. So why do I see no evidence of this happening? Is it just that these people are low profile, is it negative skew, or is it a lack of capital?

Actually, without doing any calculation, I reckon that a sharpe of 2 is already enough to produce returns over 300% (it'd run a bit hot though), thing is you might need more infrastructure that you can conjure at home...

Also I wonder. What kind of evidence would such a thing leave behind? It's not like you'd read it anywhere, or even worse, day trading forums are full of people claiming that kind of success (which naturally we dismiss)...

/lys

NeroTulip


Total Posts: 996
Joined: May 2004
 
Posted: 2009-02-13 16:33

@ Jungle: You could find such an opportunity while working at your day job developping trading systems for a large fund. You could then decide it is of no use to your employer because it is not scalable, but you may want to ditch the day job and trade it on your PA. Of course, to go on your own you need some capital and infrastructure, which is not easy, and you may think keeping the day job is a more attractive option.

@ lys: Can you elaborate on how you go from a sharpe of 2 to 300%+ returns? My own back of the enveloppe calculation is that a sharpe of 10 says your returns are 10 times the stdev. Take 3 stdev as the capital required to back your trading (in practice this is probably way too agressive), and your returns are 10 stdev i.e. 10/3 = 333% of your capital. Anecdotically, I know of one fund that has returned 200%+ with a sharpe of 5, so this is the right order of magnitude.


Inflatable trader

LowDD


Total Posts: 142
Joined: Feb 2009
 
Posted: 2009-02-13 23:59

@ NeroTulip -- Aaron Brown's post that I referenced previously does a good job explaining why Sharpe=10 and Sharpe=2 are not that far apart.  I think his reference is basically, if we think of them as trying to say "what is the chance this happened randomly?" in the sharpe=2  case it is 99% and the sharpe=10 case it is 99.99%.  Either way there are sufficient unknowns that there really isnt too much difference between these outcomes.

To the more general point of "low capacity / ulta high-sharpe ratio strategies" -- I just dont see it.  Show me all the $100M funds with consistent 30% returns?  I actually feel like once you can systematize true high-quality sharpe ratio stuff.  That is a process/mindset you can repeat across different markets, regions, or new algos.

Further, stuff that looks like it has ulta-high sharpe ratios over a local period (3 months to 3 years) is typically arbitrage stuff that is running classic blow-up / liquidity risk.  In the fullness of time, that "10 sharpe ratio" all too often ends in negative total returns to capital.

Another interesting point I'll add to the mix is that Millennium Partners, a $15B heavy stat arb firm has long-term performance of 15% annual returns at 5% volatility.  Now across most major hedge funds those numbers are pretty spectacular, but thats a Sharpe of 2. And they are running ~100 different trading groups.  If we factor in the 5% mgmt fee and 25% inventive fee then the gross return sharpe ratio is closer to 3.  But if that is the pooled product of 100 different groups it appears that average underlying sharpe ratio must be closer to 1. 

 


ccooper


Total Posts: 13
Joined: Nov 2006
 
Posted: 2009-02-14 04:57
I know a medium-high-frequency trader who had a Sharpe ratio of 2.5 in 2008, with gains of +300%. He is liquidity-constrained, though, and believes that his annual income has topped out at the low 8 figures. He therefore doesn't have a fund, and doesn't manage OPM. What would be the point?

NeroTulip


Total Posts: 996
Joined: May 2004
 
Posted: 2009-02-14 18:10

@ LowDD: a Sharpe of 2 and 10 both say that your edge is significant, but I was thinking about the second use of the Sharpe ratio as a measure of return on capital.

@ ccooper: The issue then is how to reinvest the profits. Assuming your guy can take 5m and generate 15m of profits, and that this is the maximum amount he can extract from the market with this strategy, what can he do with his capital the following year? That is a problem many people would like to have, but a problem nonetheless... Obviously he is going to reinvest 5m in his strategy, but that leaves 15m to play with. The options are to invest in his second-best idea (he needs to do more research, and that may return much less), to find other traders to manage his capital, or to invest in something unrelated (real estate, starting another business, keeping it in treasuries, etc)

The advantage of the second option is that he can invest in something he understands well, and that is hopefully uncorrelated with his own trading. He also knows that at some point he only had 500k to his name and could have taken some outside capital to fully exploit the opportunity, so there must be other traders in that position now, and he can provide them with some capital. At least that is why I am interested in investing in other traders.


Inflatable trader

london


Total Posts: 302
Joined: Apr 2005
 
Posted: 2009-02-15 21:45
{Blatant namedrop starts}

Bob Litterman spoke last week at BBG in NY. I hung around after the talk when he was very generous with his time for the few starstruck geeks left. He made a comment pertaining to this thread:

"Whenever a researcher shows me an in-sample Sharpe, I halve it and would be happy to see that in live performance."

{Blatant namedrop ends}

aaron


Total Posts: 746
Joined: Mar 2006
 
Posted: 2009-02-16 00:07
LowDD kindly posted my answer to the "Sharpe ratio of 10 or 25" question. I'd add that the 300% return calculation assumes your prime broker agrees that your SR is 10. If that were true, everyone would pile into this strategy and the SR would fall.

Moreover, I'm not sure I believe that very high SR strategies exist. Suppose you found a strategy that produced LIBOR+100 bp every year for the last 40 years with a standard deviation of 4 bp. That's an SR of 25. But if there's just one chance in 50,000 of losing all the money, the SR falls to 2. No one can possibly know the chance of a large loss is less than one in 50,000. Certainly 40 years without seeing it is no evidence.

The situation changes if the strategy returns LIBOR + 50% with a 2% standard deviation. Now it takes a 1 in 44 chance of disaster to bring the SR down to 2. But it's hard to believe something like this would be unnoticed.

Then, almost by definition, very high SR strategies have to be ones other investors dislike. They might take a lot of work, so you have to worry about your hourly wage given the limited capacity. They might appear crazy, so you cannot raise outside capital for leverage. They might be distasteful, so your profit has to be weighed against reputational problems or personal dislike.

As to LowDD's question, I think the biggest problem with SR is that people use historical estimates rather than forward-looking measures. At the very least, you should resample the historical data and think about the underlying economics. You should know from whom you are making money.

But I have faith in a well-computed SR. I think it's the best single measure of investment performance for liquid, high-capacity, low-work, financeable strategies. I agree with LowDD that it assumes sensible diversification, drawdown control and risk management in general.

bluelou


Total Posts: 68
Joined: Jan 2009
 
Posted: 2009-03-05 07:00

Aaron,

I apologize in advance for the remedial nature of my question but could you give an example of how you're associating probabilities w/Sharpe ratio.

For instance, how does a 1/50,000 risk of ruin turn a SR of 25 to an SR of 2?

Similarly, is this the type of risk you were referring to in your 2/07 posting where you said "What kind of probability would you look for?  I think a reasonable answer is something like 2% to 15%."?  Why do you say that 2% to 15% is reasonable? 

Thanks in advance for your help,
Lou

 


Je suis ce que je suis, et c'est tout ce que je suis -Popeye

aaron


Total Posts: 746
Joined: Mar 2006
 
Posted: 2009-03-07 22:22
You put me in mind with an exchange I had with my first boss. I said I was sorry for having made a mistake. She replied, "That doesn't help anything." I said, "Nevertheless, I think there's a place for apologies in this world." She said, "I don't know about that one way or other other, but I do know there's a place in this company for people who don't need to offer them." We didn't get along but, like the Boy Named Sue, I benefited from having worked for her.

You have a strategy that returns an expected LIBOR+100 with a standard deviation of 4 basis points. Subtract the risk-free rate (LIBOR) from the expected return and get 100 bp. Divide by the standard deviation (4) and get a Sharpe of 25.

Now suppose the strategy has 1 chance in 50,000 of a -100% return. That reduces your expected return by 10,000 bp / 50,000 = 0.2 basis points, so it's LIBOR + 99.8. It increases your variance of return by (10,000 bp)^2 / 50,000 = 2,000 bp^2. Your old variance was 16 bp^2, so your new standard deviation is (2,016)^0.5 = 45 bp. Your new Sharpe = 99.8 / 45 = 2.

aaron


Total Posts: 746
Joined: Mar 2006
 
Posted: 2009-03-07 22:35
Sorry, I forgot your second question.

Other people might have different opinions about the 2% and 15%, but I think most people will agree with the logic. When you do a backtest on a strategy, there is some level of significance (probability that a strategy with zero excess return could generate the results) for the backtest to be an argument in favor of the strategy. If that level were, for example, 50%, there is no evidence from the backtest that the strategy has excess expected return. That doesn't mean it's a bad strategy, sometimes you want to make the investments that lost money in the past. It does mean that you can't argue it will work in the future because it worked in the past.

I said 15% was the maximum significance at which I would put any weight at all on the backtest. Someone else might say 20% or 5%, but there is some level.

As the significance declines, the persuasiveness of the backtest increases. But not indefinitely. At some point, which I said was 2%, the pure statistical worry declines to insignificance compared to other worries that invalidate the entire exercise. Maybe you've computed the historical returns incorrectly, or failed to account for some cost or risk, or maybe you're looking at data-mined results. I know people can data mine to any desired level of significance, and I find it difficult to believe there are really good opportunities out there that are both completely obvious and basically never lose money. So a significance of 0.01% doesn't impress me more than a significance of 2%. I like both numbers, but I treat them equally.

bluelou


Total Posts: 68
Joined: Jan 2009
 
Posted: 2009-03-07 23:54

Aaron,

Both of your posts make sense to me now.  I was missing the 10000 bps /50000 = 0.20 bps. 

I understand what you're saying w.r.t. p-values. You're more concerned about some type of testing bias than you are about exactly where you are in the tail of the distribution.  I would tend to agree.  I would think 10% or so would be close enough to get started given everything else that could be wrong.  Hell, I'd take 20% if you could replicate the results on a different instrument at a different point in time.

Thanks,

Lou

 


Je suis ce que je suis, et c'est tout ce que je suis -Popeye

bluelou


Total Posts: 68
Joined: Jan 2009
 
Posted: 2009-03-08 00:06

FWIW,

Here's an excellent paper I found a few years ago, "Why Most Published Research Findings are False", about how researchers manipulate hypothesis tests.

I couldn't get the upload to work so here's the link.  The paper is available for free on the web.

http://medicine.plosjournals.org/perlserv/?request=get-document&doi=10.1371/journal.pmed.0020124&ct=1

-Lou


Je suis ce que je suis, et c'est tout ce que je suis -Popeye

cordura21


Total Posts: 228
Joined: Aug 2009
 
Posted: 2010-01-02 17:42
Just wanted to add to this great thread: from a study quoted on Mebane Faber's site, here's a chart of CTA's SR against the years of reporting. Interesting, even supposing that CTA's have less scalability issues than big funds.




Vespertilio homo est cientificus

LowDD


Total Posts: 142
Joined: Feb 2009
 
Posted: 2010-01-05 02:21

This fails on a number of accounts:

1. Database of "actively" reporting CTAs is always biased to exclude the few 4 SR successful funds after 10 years.  Say you were "that guy" and had done it for 10 years... would you continue to report to this database?  Multi-SD unlikely.  Say I was "that guy" and launched last year at $1B... would I report to this database?  Again, v unlikely.

2. Going back 20yrs+ is silly.  First "careers" are not that long in this business.  A good 5-10 years and some large chunk of people end up choosing to spend the rest of their life on a boat. Second, think about the state of markets 20 years ago (1990).  You did not really have electronic trading as we know it today. You did not have instantaneous news and events on single stocks.  You did not have well developed short markets (so alpha was less controlled). In the mid-to-late 90s a number of "day-trading" types who were the earliest adopters of short-term info trading made 1000% annual returns (for 2-5 years) as the mainstream market adjusted from "calling your broker".

3. I can name a solid handful of 10-20-30 y.r., sustainable ~3 SR funds.  The generic firm models are all pretty similar.  The fact that one cannot, means one is not really looking very hard.   Why not ask, "What are the top 5 performing long-term funds, and what are their Sharpe Ratios?"

Classic lazy academics.  Go share a nobel prize with somebody (sorry).


FDAXHunter
Founding Member

Total Posts: 8335
Joined: Mar 2004
 
Posted: 2010-01-05 08:15
LowDD: I can name a solid handful of 10-20-30 y.r., sustainable ~3 SR funds.

Please do.

The Figs Protocol.

quantz


Total Posts: 237
Joined: Jan 2009
 
Posted: 2010-01-05 10:18
+1

3 sharpe for 30yrs? Confused
i'm guessing anyone with that kind of record would be pretty rich ...
i'm furiously googling away on this topic right now ...

LowDD


Total Posts: 142
Joined: Feb 2009
 
Posted: 2010-01-05 17:11

@FDAX - C'mon its silly pointing out the obvious to people.  A quick top-of-my-head ranking might be: RTC, SAC, Millennium, Moore.  Those should all be 2-4 SR, 15-30 yrs.  Interestingly, the generic firm structures and approaches are mainly the same. 

I could breakout other categories, like credit funds or long-biased funds with 3-4 information ratio.  And like I said, there are a larger handful of 7-15 year stories than 15+.  Thats a long time.  Say you start your fund firm at 35, you have to be 50+ and still working in the same structure.  For most, retirement is tempting at some point.


PeregrineFalcon


Total Posts: 28
Joined: Feb 2007
 
Posted: 2010-01-05 17:36
"A quick top-of-my-head ranking might be: RTC, SAC, Millennium, Moore. Those should all be 2-4 SR, 15-30 yrs. Interestingly, the generic firm structures and approaches are mainly the same. "

Do not know about SAC but Moore is not in the 2-4 SR range last I checked. They are around 1.7 annulised. As they are a global macro fund, which historical do not break 2 SR. Moore and Millennium's structure is also not the same - Millennium runs *independent* books, and are more driven by systematic trading.

mib


Total Posts: 354
Joined: Aug 2004
 
Posted: 2010-01-05 17:40
I have data for Millenium at hand - they are under 2 SR over 10 and 12 years

Head of Mortality Management, Capital Structure Demolition LLC
Previous Thread :: Next Thread 
Page 2 of 4Goto to page: 1, [2], 3, 4 Prev Next