|
PureBytes Links
Trading Reference Links
|
Andrew
The EL code for GARCH modeling as well as the code for other advanced time
series applications is currently being developed in a collaborative
arrangement between our firm and faculty at universities in the US and
Canada. The decision to market the code for Easy Language as open source or
using DLL calls will be made next week and the software should be available
by mid October. Negotiations with the publishers of the major statistical
software packages in order to link those programs with Omega products have
proven economically unfeasible leading us to write the code for use only
within TradeStation products. We believe that the result will be the most
powerful financial analysis software commercially available. The powerful
statistical functions combined with the moving window approach of
TradeStation should give traders an advantage not currently available in any
existing software package.
Some applications will include differencing for nonstationary means and
variances, Durbin-Watson statistics, ARIMA modeling, runs test, probability
testing using the geometric and other distributions, maximum likelihood
estimates of price direction and magnitude, least squares, non-linear least
squares, quadratic trend forecasting using regression, Kalman filtering,
goodness of fit testing, estimating polynomial distributed lag models,
Dickey-Fuller type methods to approximate lag, bandpass filters and other
methodologies. Traders may also look at traditional parameters and
statistics as variables which will provide a new dimension to forecasting.
GARCH stands for Generalized Autoregressive Conditional Heteroskedasticity.
Briefly, heteroskedasticity refers to variance in a population that varies
over time. The assumption of normality requires a constant variance. It is
easy to understand in cross-sectional data (spending may vary to a larger
degree at higher incomes than at low incomes) but is a little more confusing
in a time series. Expected variance may increase with time or have a higher
expectation at higher prices. Heteroskedasticity in part can help explain
fat tail distributions and volatility clusters. The result of
heteroskedasticity is that parameters computed through least squares are
still unbiased but are inefficient. Some solutions include transforming the
data by using logs or deflators. Hence, there are many variations and types
of GARCH modeling.
The presence of heteroskedasticity does not violate any Efficient Market
Theory assumptions. Dependent variances do not imply autocorrelated
residuals, the other part of GARCH modeling. Autocorrelation (serial
correlation) means that the error term from a time series forecast for
period (t) is correlated with the error term of the forecast for some period
(t-n), normally (t-1). The presence of autocorrelation does violate the
Efficient Market Hypothesis because now we are able to model price(t) as a
function of price(t-1), for example. It is my opinion that for the purposes
of building a better mouse trap, the construction of an appropriate AR or
ARMA model would be of more benefit than a GARCH approach. Entry and exit
timing would certainly benefit with a properly specified AR model, but I
doubt that GARCH would provide the same benefits.
I hope this has been of some help. If you have any questions don't
hesitate - on or off list.
Paul J. Pontillo
Technometal, Inc.
mailto:pontillo@xxxxxxxxxxxxxxxx
-----Original Message-----
From: Andrew Peskin [mailto:andrew@xxxxxxxxx]
Sent: Thursday, August 16, 2001 9:12 AM
To: omega-list@xxxxxxxxxx
Subject: Collaboration on a GARCH DLL Anyone?
I am interested in using/evaluating a measure of market volatility
derived from the GARCH model. I would be interested in working with
anyone who possesses a strong enough knowledge of mathematics and/or the
GARCH process, in order to code this routine and DLL. Once completed
the DLL could be used in any TradeStation ELA routine, VC, or VB
application. I have the coding skills necessary to code the DLL, I just
need someone to explain to me what exactly is going on in GARCH.
Anyone interested in pursuing this project, please contact me and let me
know.
Andrew
PS - If anyone has the full source code for GARCH, in either Basic (VB
or PB), C/C++, or Fortran, all the better.
|