|
PureBytes Links
Trading Reference Links
|
KyPlot, similar to TimeStat, 32-bit and on steroids (incredible
and free), long list of features
http://www.qualest.co.jp/Download/KyPlot/kyplot_e.htm
These are some of the Time Series Analysis functions, Kaufman has a section
on ARMA and Kalman filters in trading.
1) ARMA model: applies autoregressive moving average (ARMA) models to time
series data. AR and MA coefficients are calculated by maximum likelihood
method. A state space representation and Kalman filtering are used for
calculation.
2) Time Series Decomposition: applies a seasonal adjustment model given by
Kitagawa and Gersch (ref. 1) to data. A time series is decomposed into
trend, seasonal, autoregressive and noise components. Trading day factors
can be also incorporated. State space modeling and Kalman filter are used
and parameters are estimated by maximum likelihood method.
3) Time-Varying Variance Estimation: estimates the trend of the variance of
nonstationary time series data and normalizes the data by the estimated
variance.
4) Time-Varying Coefficient AR Model: applies the time-varying coefficient
AR model and estimates time-varying spectrum.
5) Locally Stationary AR Model: applies the locally stationary AR model by
decomposing the data into subintervals and estimates the spectra for the
subintervals.
6) Change Point Estimation: estimates the exact point of discontinuity in
statistical structures of time series data by applying the locally
stationary AR model and minimum AIC principle.
The programs are based on ref.2 and modified.
References
1. Kitagawa, G. and Gersch, W.: A smoothness priors-sate space modeling of
time series with trend and seasonality. J. Amer. Stat. Assoc., 79: 378-389,
1984.
2. Kitagawa, G.: Time Series Analysis Programming, Iwanami Shoten, Tokyp,
1993 (in Japanese)
|