On my target 50 signal the compound annualized rate of return in the statistics window is 39.5%. When you go elsewhere like the grid for example the rate of return is showing 29.8%. Can somebody here tell me how the different rates of return are calculated?
The problem isn’t the rate of return. The choppiness of your equity curve (Target 50) compared to the S&P curve isn’t worth the effort. Someone could way “outperform your performance” by doing a double beta of the S&P. Similar amplitudes and much better return with the current bull market.
I disagree with the comment that someone could way outperform this system with a double beta of the S&P. That statement would be true if alpha was negative. Look up the advanced statistics and you’ll notice a positive alpha for both daily and monthly returns. Moreover, beta for daily returns is (slightly) negative. The latter is an interesting characteristic that few profitable stock systems have. In that respect the system could be an interesting extension to an existing high-beta portfolio in order to make the portfolio more market-neutral. E.g. look at the profits the system generated during the S&P dip at the end February.
I do agree with the fact that the equity curve looks a little choppy, but with a Sharpe ratio of ~1.4 and a 35 week history, it’s not really bad.
What worries me more is that this is only 1 out of 4 systems launched by the vendor that shows a reasonable performance.
There are similar differences for other systems too. E.g. for extreme-os,
Grid: Ann. Return% 267.0%
Basic Statistics: Compound Annual % 156.6% over 783 days
Advanced Statistics, daily full: Annualized return (arithmetic extrapolation) 2.816
Advanced Statistics, daily full: Compounded annual return (geometric extrapolation) 1.561
My guess is that the small differences are a matter of updating and that the large differences are due to the type of return (compounded or uncompounded). The Grid seems to contain the uncompounded annualized return, while the basic statistics show the compounded annual return. The advanced statistics will probably not include todays account value, and that could be different for the basic statistics.
Your next question is probably “should these values not all be the same?” Yes that would be the ideal situation, but the server load would be a good excuse to compute the advanced statistics not too often.
I don’t know what the reason is to show the uncompounded annual return in the Grid (assuming that this is indeed the case). It would be more consistent to show the compounded version. OTOH, some systems do not compound, so perhaps this is a political decision to give these systems a chance to get on top of some list. Or it could just be a mistake.
Thanks Science Trader and Jules. Rosss you take the cake.
maybe, but true
The problem is that the annualized return for “target 50” shown on the system page simply cannot be true, whatever method (compounding or no compounding) or updating frequency is applied.
Let’s consider the best case scenario: compounding returns, using the peak equity attained in the last 30 days. Peak equity was 122044 and occurred less than 30 days ago (mid of March). So the max compounded annual return would be 1.22044^(365/237)-1 = 36%.
Using the current P/L of $20,159, I get the following numbers:
annualized compounded return: 1.20159^(365/248)-1 = 31%.
annualized uncompounded return: 0.20159*(365/248) = 29%.
For extreme-os, I get:
annualized compounded return: 6.79285^(365/783)-1 = 144%.
annualized uncompounded return: 5.79285*(365/783) = 270%.
It shouldn’t be too difficult to get these numbers consistent across the various tables (grid, advanced statistics and systems page), as the only inputs are total P/L and the number of days, which both seem to be updated on a daily basis.
Science Trader I understand your figures. Not sure how the error accured. I was kind of hoping Mathew would chime in here and explain it. Thanks
Science Trader I understand your figures. Not sure how the error accured. I was kind of hoping Mathew would chime in here and explain it. Thanks
that was what I was hoping for as well
I was looking at some of the other 100% relizum factor signals. TMG and QQQQ-trading. They also have 2 very different numbers. If there is a mistake it is not only my signal.
My explanation for it is that in, at least in the advanced statistics page, the CAR is computed with the number of days for which there is an equity value (henceforth called “trading days”), while the basic statistics table displays the number of calendar days. For extreme-os the computation is done with 738 trading days while the number of calendar days is 783. For target 50 the computation is done with 209 trading days while the number of calendar days is 248. With these numbers (738 and 209), using the account values of yesterday, you get exactly the reported results for the CAR on the advanced statistics page.
The problem of this is that the number of trading days is compared to 365, which is the number of calendar days in a year. This is incorrect: either both numbers should be the number of calendar days (e.g. 783 and 365) or both should be the number of trading days (e.g. 738 and the expected number of trading days in a year).
This problem was already identified a few days after the installation of the advanced statistics program, and my suggestion was to change 365 (which is a single parameter in the program) to the expected number of trading days in a year. I regret that it hasn’t be done yet. Perhaps there were problems that I do not know of, or perhaps it is a simple but unfortunate miscommunication between Matthew and me. Anyway, I agree that the CARs based on daily values in the advanced statistics page are inaccurate today. (I don’t know how the results on the other pages are obtained, but I expect that something similar happens there).
As for the consistency of the various pages, I do not completely agree with the suggestion that this is easy. It is true that the computation of the CAR requires only a few input numbers, but the advanced statistics page can only be updated by running the entire program and it will be based on yesterdays account value. So if the other pages are supposed to display more recent outcomes (I don’t know if that is the case), then they cannot agree with the advanced statistics page. But they can agree with each other.
Mathew are you ever going to correct this or at least explain how it is figured so we know wha is going on?
As far as the reason for the inconsistency: it is as Jules explains; there are two different sets of calcs, one compound and one simple. I will need to examine the number of days that are used in each calculation. I’ll try to get to this shortly.
Mathew in the stastics window my target 50 signal is now showing a 27% annualized gain. That is from what I understand compounded. On the grid target 50 is showing 33% and that is supposed to be simple. After looking at several signals I don’t understand why some of them have a higher compounded rate and others have a higher simple rate? When I look at others that have a simular simple rate they have a much higher compounded rate than target 50. Is this because you have audited and adjusted mine and not theirs? Thanks
Yes, I think so. I have implemented the software change recommended by Jules, but it may be 24 hrs before all system stats are recalculated using the new software.
I didn’t think long about this (so perhaps I shouldn’t post at all ;-)) but I think the difference that you describe can also be due to different ages of the systems. If A has 10% in one month and B has 120% in one year then A and B have the “simple” annualized but different compounded returns (because +10% every month is more than +120% after one year if you compound).
