Performance Summary Reports
Posted: Oct 16 2006
Hi All,
Have taken a look at the MC strategy performance reports and i like what I see. However, there are elements that I'd like to see included, and which may already be roadmapped.
The statistical output of the report, IMO, can be improved. I see the sharpe ratio is there but I don't see useful things like T-tests, correlational stats and nonparametric statistics, e.g. runs test.
T-test results are certainly a good thing to include I think. They're good for things like:
- The probability that total profit from a set of individually measured and recorded trades could be greater than a chosen threshold resulting from chance.
- Evaluating samples of daily, monthly or yearly returns of a portfolio.
- Determine the probability of future performance (if current conditions prevai) being above a particular threshold.
Correlational stats would help establish the statistical significance of a particular test run. Inferential statistics are the best tool here I believe...
And statistical analysis like the runs test would help to establish whether a strategy is serially dependant. Once identified that can be used to help with optimization.
I can go on but my knowledge of statistics is limited. My point here is that insightful use of statistics in the performance report can help greatly to improve the usefulness of the performance report in the repetative process of system creation.
Thx
D
Have taken a look at the MC strategy performance reports and i like what I see. However, there are elements that I'd like to see included, and which may already be roadmapped.
The statistical output of the report, IMO, can be improved. I see the sharpe ratio is there but I don't see useful things like T-tests, correlational stats and nonparametric statistics, e.g. runs test.
T-test results are certainly a good thing to include I think. They're good for things like:
- The probability that total profit from a set of individually measured and recorded trades could be greater than a chosen threshold resulting from chance.
- Evaluating samples of daily, monthly or yearly returns of a portfolio.
- Determine the probability of future performance (if current conditions prevai) being above a particular threshold.
Correlational stats would help establish the statistical significance of a particular test run. Inferential statistics are the best tool here I believe...
And statistical analysis like the runs test would help to establish whether a strategy is serially dependant. Once identified that can be used to help with optimization.
I can go on but my knowledge of statistics is limited. My point here is that insightful use of statistics in the performance report can help greatly to improve the usefulness of the performance report in the repetative process of system creation.
Thx
D