[116] Monte Carlo history data randomizer gives overly optimistic results

Each time that I am using the history data randomizer in the Monte Carlo simulator, the results of all runs on the randomized history data are MUCH better than the original historical data itself. This can´t be right and there must be a flaw somewhere in that simulation. Other parts of the Monte Carlo similar work as excepted (give worse results than the original strategy, which is correct).



Thank you.

Attachments
Untitled01.jpg
(843.09 KiB)
Untitled02.jpg
(999.83 KiB)
  • Votes +6
  • Project StrategyQuant X
  • Type Bug
  • Status Fixed
  • Priority Normal

History

g
#1

geektrader

31.12.2018 03:02

Task created

g
#2

geektrader

31.12.2018 03:02
Voted for this task.
KL
#3

kainc301

31.12.2018 03:08
Voted for this task.
Rr
#4

Partizanas

31.12.2018 10:33
Voted for this task.
AC
#5

AC1962

01.01.2019 11:36
Voted for this task.
mp
#6

Michele

01.01.2019 12:09
Voted for this task.
t
#7

tnickel

02.01.2019 07:01
Voted for this task.
g
#8

geektrader

15.01.2019 22:41
Any chance that this is getting addressed for 118?
MF
#9

Mark Fric

27.01.2019 16:08
yes, it will be checked 
MF
#10

Mark Fric

29.01.2019 09:10

Status changed from New to Waiting for information

Attachment mc_hist_data.jpg added

I'm trying but I cannot reproduce this. In my tests when I run MC with Randomize history data it behaves as expected - see screenshot.


Perhaps it depends on strategy and/or settings, could you attach some strategy and Retest config that produces this problem?

g
#11

geektrader

29.01.2019 10:01
I don´t have that strategy anymore, unfortuantely, but you see in my screenshot that the test is complete non-sense. Maybe try with a bit longer data-horizon? My strategies are all done on 32 years, so it could be related to that. I will retest soon, no time at the moment unfortunately.
MF
#12

Mark Fric

29.01.2019 13:53

Status changed from Waiting for information to Fixed

I found the problem. It happened because in MC tests SQ always used Selected TF backtest precision and not precision of main test.
g
#13

geektrader

30.01.2019 02:30
Nice, that explains the overly optimistic results for sure!
m
#14

mabi

30.01.2019 08:00
Funny thing using real tick the results from MC tests is mostly better then on 1 min resolution ( which I geuss was selected time frame)
MF
#15

Mark Fric

30.01.2019 08:11
it's other way around. MC was simulated using Selected Timeframe, so for example H1 bars. Add it produced in some cases better resuts, than testing with higher - M1 or tick - precision.
m
#16

mabi

30.01.2019 08:22
+1

Votes: +6

Drop files to upload

or

choose files

Max size: 5MB

Not allowed: exe, msi, application, reg, php, js, htaccess, htpasswd, gitignore

...
Wait please