Christoph123 schrieb:

Bei Forex ist es auch so, dass zu bestimmten Uhrzeiten eine Richtung dominater ist, als die andere. (Quelle: meine Untersuchungen)

Christoph123 schrieb:

BTW: Ich habe auch festgestellt, dass es z.B. bei AUD/USD zu stärkeren aber dafür selterenen Abwärtsbewegungen kommt. Entweder ist das nur eine Verzerrung, da mein gewählter Untersuchungszeitraum zu kurz gewählt war (aber was ist schon DIE richtige Länge?), oder es liegt ein Grundfehler in meiner Herangehensweise an die Untersuchung vor, oder ich liege richtig und es ist mal teilweise (zumindest für mich) bestägigt, dass es auch im FX Markt unterschiedliche Auspärgungen von den beiden Seiten gibt.

Christoph123 schrieb:

@ trash
Es ging um Cross Validation (goso's Post).
Also z.B.: IOS Periode=01.01. 2000-31.12.2007 und 01.01.2010-31.12.2011
OOS Periode= 01.01.2008 - 31.12.2009

Ist das möglich?


Ich habe übrigens nur die Begriffe OOS und IOS in meinem Post vertauscht, die Ergebnisse stimmen also.

Grüße
Christoph

Walk-forward testing carries the idea of ‘out-of-sample’ testing to the next level. Think of it as an ‘out-of-sample’ testing on steroids. It is a specific application of a technique known as Cross-validation. It means to take a segment of your data to optimize a system, and another segment of data to validate. Hence, here you optimize a window of data say past 1000 bars, and then test it on next 200 bars. Then roll the whole thing forward 200 bars and repeat the process. This gives you a large out of sample period and allows you to see how stable the system is over time.

Howard Bandy schrieb:

Cross validation is discussed in data mining books and it is appropriate for use in validating some categories of models, but it is not appropriate for systems trading financial time series.

Howard Bandy (Zitat aus Quantitative Trading Systems) schrieb:

Can out-of-sample periods precede in-sample periods?

No. A model that is fit to an in-sample period of 1985 to 1995 will prob-
ably show good results over the 1975 to 1985 period. But, the data we
will eventually trade is always more recent than the in-sample period,
never earlier. So the results over the earlier period have no predictive
value. Testing the model over a later time period, such as from 1995 to
the present, is meaningful. If the results over 1975 to 1985 are profitable
but the results from 1995 to the present are not, that is an indication that
the market has changed in such a way that the model being tested is no
longer profitable. If the results over 1975 to 1985 are profitable and the
results from 1995 to the present are also profitable, that is an indication
that the model may continue to trade profitably. The positive results
from 1975 to 1985 indicate that the underlying market is very stationary
with regard to that model.

For the same reasons, it is inappropriate to use the fold or jackknife
technique for dividing data into in-sample and out-of-sample periods.
The jackknife method divides all the data into several periods - say ten.
The modeling process makes ten passes; each pass treats nine of the pe-
riods as in-sample and reserves the other for out-of-sample. After the
ten passes, the results from the ten out-of-sample periods are gathered
together and analyzed as one out-of-sample period that includes all
the original data. Only one of the ten periods is truly out-of-sample.

Christoph123 schrieb:

Amibroker kann es leider nicht. Zumindest finde ich diese Funktion nicht.