At one point in the myth, Gretel (and Hansel) consume the sweet (eye) candy
house (architecture). Could this be an aplicable metaphor for Statistical
Learning Theory applied to Time Series Prediction? A chocolate covered
bicameralism? Happy Halloween and thanks for your help.
I intend to apply gretl (possibly R too) for the purpose of forecasting
models of the Foriegn Exchange. The platform from wich the Foriegn Exchange
time series data is observed and recorded is the MT4, Meta Trader 4 platform
from Metaqoutes. Scripts may be created in the metaqoutes language as well
as algorithmic trading rules. The menu path of the MT4 platform would be
under Tools then History (also F2), and from there various currency pairs
(EUR/USD) as well as time period (one hour chart, 4 hour chart, daily chart)
are available to export as .csv, .htm and .pm files. MT4 also offers tick by
tick data via a DDE (dynamic data exchange) link. The MT4 platform is one
that is used the most often in this industry and it is available with no
cost. Data is freely available from most retail Forex brokers when you sign
up for a practice account, in case you would like free data to play with. I
typically use Open Office calculator .ods files as my default spread sheet.
The gretl handbook talks about stacked time series and the commnad SETOBS on
pg.23-24 and then up to Ch.5 on pg.30 it mentions other useful commands for
stacked time series such as: UNITVAR, TIMEVAR, FREQ, STRCUTURE, STACK, etc.
It's my understanding that gretl expects data to be arranged by observation
and each row represents an observation.
The time series data I am working with is the exchange price of currency
pairs in multiple time frames (minute chart, hour chart, daily, weekly).
For example at 12PM noon on October, 31, 2008 the exchange price for a Euro
dollar in U.S. dollars is $1.27 (expressed as EUR/USD 1.2732). Further the
data from the MT4 is formated in rows as date, time, Opening price, High,
Low, Closing price, volume. This is the standard formatting and I am still
simply working on organizing the data from the MT4 platform to the gretl
platform. I was able to put the daily closing price all in a row and import
it into gretl but I still need to figure out the SETOBS command to get the
proper period. I also need to have a better understanding of how best to
stack this time series data (hour charts within daily charts within
My inital question then is for suggestions on how to stack these time
series. I'm seeking suggestions on strategies for nesting this data with
gretl for the purpose of forecasting models. The books that I am using as
guides for this project are "Time Series Prediction: Forecasting the Future
and Understanding the Past" by Andreas Weigend and Neil Gershenfeld and the
book "Statistical Learning Theory" by Vladimir Vapnik. There are several
other books that led me here such as "Chaos Under Control" by Peak and
I've noticed patterns on all scales from some spread sheet experiments with
multiple currency pairs and I want to investigate further with gretl. I have
not been formally educated past a college associate level and I am learning
on my own. Help and direction is greatly appreciated. The stacking question
I have is in reference to observing these self similar patterns from
multiple time frames. How may I investigate these self similar properties
within a stacking type of architecture in time series analysis is the best
way I may frame my question without going into further detail.
Time Series tricks 'n treats