What's your Robot Sharpe Ratio?

bigdatascience

Junior member
Messages
13
Likes
0
I am interested to know what sharpe ratios are being traded based on their backtest analysis.

Best,

A
 
I am interested to know what sharpe ratios are being traded based on their backtest analysis.

Best,

A

Why ?

For maybe 90% or more retail traders its a meaningless ratio - OK great for multi million dollar accounts and comparing investments - but get the feeling not too many member here dealing with 10's of millions etc etc

Drawdowns - RO1 - efficiency ratios - etc etc - but surely Sharpe ratio's are only of interest for big money investors trying to make a choice based on 3 -10 years of investments - and even then the ratio is misleading

Only my view - but good luck - I hope some members can oblige

Good Trading

Regards


F
 
Why ?

For maybe 90% or more retail traders its a meaningless ratio - OK great for multi million dollar accounts and comparing investments - but get the feeling not too many member here dealing with 10's of millions etc etc

Drawdowns - RO1 - efficiency ratios - etc etc - but surely Sharpe ratio's are only of interest for big money investors trying to make a choice based on 3 -10 years of investments - and even then the ratio is misleading

Only my view - but good luck - I hope some members can oblige

Good Trading

Regards


F


Why?

Admittedly professional reasons:

1. Is Sharpe considered relevant to members, your answer is clearly in the negative.
2. If it is being used to back test models and verify potential then I am interested in feedback about it's effectiveness.

Best,

A
 
Sharpe ratio could be useful. It depends a lot on the strategy that you are trying to design. As has been pointed out, there are other things that one cares about when designing a strategy like drawdown depth and duration. One serious downside of the Sharpe ratio is that it equates volatility and risk, which are not equivalent or necessarily related!

In my experience you're getting somewhere when your sharpe after transaction costs estimates is north of 2. Just my 2 cents.
 
I am interested to know what sharpe ratios are being traded based on their backtest analysis.

Best,

A

See here
http://qoppac.blogspot.co.uk/2015/03/simulating-my-futures-system.html

I get a SR of about 0.90. This is based on an out of sample backtest (real performance has been much better, but the last 12 months have been exceptionally good for my kind of trading).

Anyone using an in sample SR is likely to be overestimating what they can achieve in reality. SR of 2.0 after costs are unlikely in my opinion.
 
I am interested to know what sharpe ratios are being traded based on their backtest analysis.

Best,

A
I'd be interested to know how you could calculate a Sharpe Ratio for backtest analyses. Even using ex-post (rather than ex-ante) you still have a major issue coming up with the portfolio std. dev. unless you treat each asset being tested as a portfolio in it's own right which then rather misses the point of using Sharpe in the first place.
 
I'd be interested to know how you could calculate a Sharpe Ratio for backtest analyses. Even using ex-post (rather than ex-ante) you still have a major issue coming up with the portfolio std. dev. unless you treat each asset being tested as a portfolio in it's own right which then rather misses the point of using Sharpe in the first place.

Er... I'm probably missing some subtle point here, but to spell out the calculation:

Every day in the backtest we calculate the mark to market p&l of each asset we're trading. Add them up to get the portfolio return for that day.

Put together the series of daily portfolio returns. Sharpe is obviously mean divided by standard deviation.

This approach is best if your standard deviation is roughly stable. So you have to keep the maximum amount of capital at risk, rather than rolling up profits as you might do in reality.

Does that make sense, or perhaps you could clarify what the issue you have is?
 
I have an issue with Sharpe being calculated as part of a backtest operation. Even using ex-post, which you are which uses realised returns rather than expected returns (ex-ante), using it over a number of related assets in this manner will tend to under-compensate for non-normally distributed returns. I'm not even going to get started on my number one concern regarding selection bias. If you're keen to associate Sharpe with your system, perhaps you could look at the Deflated Sharpe Ratio? It will be less flattering to the outcome, but will give you a more accurate assessment of it.
 
I have an issue with Sharpe being calculated as part of a backtest operation. Even using ex-post, which you are which uses realised returns rather than expected returns (ex-ante), using it over a number of related assets in this manner will tend to under-compensate for non-normally distributed returns. I'm not even going to get started on my number one concern regarding selection bias. If you're keen to associate Sharpe with your system, perhaps you could look at the Deflated Sharpe Ratio? It will be less flattering to the outcome, but will give you a more accurate assessment of it.

Sure combining a large number of uncorrelated non-normal distributions will produce a normal one. Is this flattery? I'm not sure. If you end up with a normal portfolio then using the Sharpe Ratio (SR) is an appropriate thing to do. I suppose if you combined a lot of massive negative skew strategies, then the final distribution would be hiding a lot of potential downside if everything went to zero at once.

Its fair to say that you should look at the SR of the components of your portfolio; since if you have one big winner and lots of losers, then you probably don't have as much statistical confidence as your raw SR implies.

Its also fair to say that Sharpe Ratio will make negative skewed strategies look better than they really are, and that looking at skew and other performance metrics is probably a good one. However the skew of returns on individual assets is positive as you'd expect from a trend following strategy, and on the total portfolio, is slightly positive. So this is less of an issue for me personally.

Selection bias is a broader issue around backtesting and overfitting I guess, which would be true of any performance metric. However I am relatively careful about backtesting my stuff; if you read my blog I used a robust method entirely out of sample. I still think the SR is overstated, perhaps by 25%, due to some residual selection bias, and one off asset price movements that won't be repeated.

I have a conservative expectation that I'll make a SR of 0.5; translating to a half kelly risk target of 0.25. Since I've made a SR of about 2.0 in the last 12 months with real trading, a one in forty year event if my SR is really 0.5, this is probably pretty conservative.

But its still reasonable to ask what SR people expect, without getting into a debate that SR is rubbish. It is good to look at individual parts of your portfolio and other performance metrics other than SR. Can we take that as given, and stick to the original question?

I think the main point here is that people who expect a SR of 2.0 and size bets accordingly are on dangerous grounds.
 
Some things may need clarifying...

What sharpe ratio are you measuring? When i say that a sharpe of 2 means you are making good progress I am talking about the annualized figure. There are many ways to measure this though. One thing that I do as part of the analysis process is to look at the sharpe ratio on a per trade basis since it varies though time and use those variations to simulate streaks of good and bad luck for my strategy. I find that this is a very good way to set expectations for how the strategy is performing. Given a certain number of trades I can tell statistically if things are working as the "should" (according to initial research).

The most valuable information that comes from this is that I can tell how the strategy is performing relative to "expected" (average) performance.



I do not size bets based on sharpe ratio or the kelly criterion since those things are not in line with my objectives. I care much more about the depth and duration of drawdown than about getting maximum compounded growth (at maximum volatility).
 
Top