9-23-2006
Welcome and hope you find these files of use to you
This zip files includes the following:
AA Report plus
pictures
11/08/06
Fred Tonetti-
The equity curve of course gives
us a basis for calculating performance metrics ...
Let's deal with the standard ones first ...
If we set the AA Beg / End Dates to our range of trades i.e. 8/5/2004 &
10/18/2006 respectively and click backtest and then report we see the following
performance metrics ...
CAR = 36.92
MDD = 5.79
UI = 1.82
UPI = 17.29
If we click Explore to get the AFL's internal calculations of the same standard
performance metrics, we get in the last line of AA Explore Output ...
CAR = 36.852
MDD = 5.791
UI = 1.824
UPI = 17.241
Exact ? No ... Close enough ? ... In my book, yes ... You can fool with this
sort of stuff forever trying to get it to match AB Internals or other statistics
generated elsewhere if you like but to me this is sufficient.
Go back and look at the equity curve in the previous post and think about what
SHOULD happen to the performance metrics if you look at them in the context of
only where buy signals are active ...
MDD is the easy one ... It will be the same, Right ? ... The equity curve does
not go up or down when you are not invested so it should only have the flat
periods removed from it ergo the same peaks and valleys. See the equity curve in
the previous post and imagine it without the flat periods.
CAR should increase because you have the same postive return over a shorter
period of time.
Although it's not obvious UI should probably decrease because of its formula and
as a result UPI should probably increase ...
So now look at the results of the Explore again in terms of the WI ( While
Invested ) perf metrics columns ...
Productivity afl
plus pictures
11/13/2006
Ken CLose-
OK, finally back to post
some results.
Fred: thanks, I thought I looked in the unnamed.afl file above but must have
clicked on the original one, as I did nothing with it.
Thanks for acknowledgement of CARWI--that was what I thought was needed from the
very start, even if I did not know what to name it.
I am going to post the comparisons I have done as well as the code file I
produced. Mine has routines to bring in standard FT signals. Plus the comparison
shows values from the Gansz trade file--they are in pretty good agreement.
I am unable to calculate Standard Dev and Sharpe during Buy or During Sell, but
what the heck...
EEVEMA.afl 12/15/2006 Tom Tsao-
FT-sig_test2.afl
& FT-Utilities8.afl
12/15/2006 Dave
Howarter-I have
given a lot of thought to MDDE, the maximum drawdown from the time a signal went
to buy state, since I saw it in Fred Tonetti’s IO. I think it is a good measure
for an intermediate term signal of how well I will sleep when following the
system.
For instance, consider the attached plot of LZOEX timed with Fubar5x7 from start
of LZOEX data on 3/19/01 to 12/13/06. The Mdd has been –11.96%, but the MDDE was
–2.81%. This means at some point the equity curve went down by 11.96%, but the
most it went down below where the signal entered was 2.81%. Thus the market took
back 11.96% from the money it had given me, but it never went down by more than
2.81% from where I put my money in. I look at the MDDE as how much of “my” money
the market might take and the difference between MDD and MDDE as being how much
a fickle market might take back from the money it had given me. For me, if I can
get a CAR of 34.05% with a risk of losing 2.81% from where ever I put money into
the pot, I would be happy..
I did not have a tool to do a quick measurement of MDDE, so I decided to create
one. Possibly it can be done with Trade, but I saw this as a difficult task and
saw an easy way to do it with AmiBroker. So I created the attached file that
does the following:
1. Test using any signal that is stored in the FT data base, specifying the
signal name via a parameter
2. Specify the time period of the test via the Automatic Analysis (AA) settings.
3. Measure MDDE, MDD, CAR, Ann (the last one year’s return), and a few other
things.
4. Make plots to visually show the equity curve with and without signals, the
MDD and MDDE points and other points of drawdowns that were less than the worst
case.
5. Add optional HoldMinDays to simulate fund early redemption fee rules, the
same as Trade does, using an optional parameter
6. Add an optional penalty if the signal goes to sell before the early
redemption fee time period, the same as Trade does, using an optional parameter.
The reason I could do this easily in AB was that I borrowed all the MDDE
calculations and graphical plots from Fred Tonetti’s IO plots and the signal
import code from Bruce Robinson’s FT-Ultilities8 code.
There are instructions in comment lines in the beginning of my .afl file on how
to run the code and how to interpret the graphical plots.
Bruce’s FT-Utilities8.afl is “included” into my code, so must be present in the
AB “formulas/include” directory. I looked around to make sure that it was still
available in some archive and could not find it on FT-Talk or AB-Yahoo. It had
been posted on FT-Talk, but apparently the older archives have been deleted. I
discussed this via e-mail with Bruce and he said there had been no changes to
the file so I should post it again, which I am doing.
FT-sig_test2.afl
12/15/2006 Dave
Howarter-I
think I am being punished by the Gods of Humility for bragging about my
HoldMinDays code, because I have already found a bug in it. It failed to extend
some trades to meet the HoldMinDays. The attached version fixes that.
FT-sig_test2.afl
12/18/2006 Dave
Howarter-I
found a bug in the attached file I posted a few days ago that caused the
MinHoldDays to not stretch trades sometimes the first time a backtest was done
after the fund name or signal name was changed. The attached code fixes the
problem.