[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Savannah-register-public] [task #5868] Submission of Special Interest G
From: |
Johann Rost |
Subject: |
[Savannah-register-public] [task #5868] Submission of Special Interest Group Time Series |
Date: |
Tue, 5 Sep 2006 11:10:12 +0000 |
User-agent: |
Mozilla/4.0 (compatible; MSIE 5.01; Windows 98) |
URL:
<http://savannah.gnu.org/task/?5868>
Summary: Submission of Special Interest Group Time Series
Project: Savannah Administration
Submitted by: johann0
Submitted on: Tuesday 09/05/2006 at 11:10
Should Start On: Tuesday 09/05/2006 at 00:00
Should be Finished on: Friday 09/15/2006 at 00:00
Category: Project Approval
Priority: 5 - Normal
Status: None
Privacy: Public
Assigned to: None
Percent Complete: 0%
Open/Closed: Open
Effort: 0.00
_______________________________________________________
Details:
A new project has been registered at Savannah
This project account will remain inactive until a site admin approves or
discards the registration.
== REGISTRATION ADMINISTRATION ==
While this item will be useful to track the registration process, approving
or discarding the registration must be done using the specific "Group
Administration" page, accessible only to site administrators, effectively
logged as site administrators (superuser):
<https://savannah.gnu.org/siteadmin/groupedit.php?group_id=8759>
== REGISTRATION DETAILS ==
Full Name:
----------
*Special Interest Group Time Series*
System Group Name:
-----------------
*sigtis*
Type:
-----
non-GNU software & documentation
License:
--------
GNU General Public License V2 or later
Other License:
--------------
I imagine to use a dual licensing model: one license is GPL the other
license could be a propriatary license (somthing similar like MySQL). What do
you think about it?
Description:
------------
Extensive library of algorithms for Time Series Analysis. (Time Series
Analysis is a part of Econometrics which belongs to Mathematical Statistics.
The library is written in C++. The Code is finished and tested. However it is
not online. For this reason I prefer to sent it by email. Please let me know
where I should send the code.
The following algoritms are implemented
Preprocessing: Eliminating Trend and Seasonality
Differencing
Logarithm
Additive Seasonality
- Parameter Estimation according Brockwell/Davis S1 – adapted to MSE, MAD,
MAPE, MdAPE
- Parameter Estimation according Brockwell/Davis S2 – adapted to MSE, MAD,
MAPE, MdAPE
- Optimum parameter estimations for MSE, MAD, MAPE, MdAPE
Multiplicative seasonality
- Parameter Estimation according Brockwell/Davis S1 – adapted to MSE, MAD,
MAPE, MdAPE
- Parameter Estimation according Brockwell/Davis S2 – adapted to MSE, MAD,
MAPE, MdAPE
- Optimum parameter estimations for MSE, MAD, MAPE, MdAPE
Extraction of a non-zero mean
- Optimum parameter estimations for MSE, MAD, MAPE, MdAPE
Linear Trend
- MSE parameter estimation according Abraham & Ledolter
- Optimum parameter estimations for MSE, MAD, MAPE, MdAPE
- Preliminary trend estimation heuristic for seasonal TS for MSE, MAD,
MAPE,
MdAPE
Logistic Trend
- Optimum Parameter estimation MSE – using LinMin optimization
- Optimum Parameter estimation MSE – using Simplex-Downhill optimization
- Optimum parameter estimations for MAD, MAPE, MdAPE
- Heuristic parameter estimation according Schlittgen
- Improved heuristic parameter estimations for MSE, MAD, MAPE, MdAPE
Exponential Trend
- Heuristic for Non-Seasonal TS for MSE, MAD, MAPE, MdAPE
- Heuristic for Seasonal TS for MSE, MAD, MAPE, MdAPE
- Optimum parameter estimation for MSE, MAD, MAPE, MdAPE
Damped Trend
- Heuristic parameter estimations for seasonal and for non-seasonal TS
using
MSE, MAD, MAPE, MdAPE
- Optimum parameter estimation for MSE, MAD, MAPE, MdAPE
Combined estimation of trend and seasonality
- Heuristic algorithms for MSE, MAD, MAPE, MdAPE
- Optimum algorithms for MSE, MAD, MAPE, MdAPE
Algorithms for extracting an identified mean, trend and/or seasonality from a
TS
3. ARMA Models
Model Identification
- Model identification according to Hannan/Rissanen
- Identification using Hold-Out Sample
- Model identification according to “Rectangle”-Heuristic
Preliminary Parameter Estimation
- Brockwell/Davis
- Hannan/Rissanen
- Burg’s Algorithm
- Durbin/Levinson
- Trench/Zohar
- “Hybrid” Heuristic
Final Parameter Estimation
- Conditioned Least Squares
- Unconditioned Least Squares
- Exact Likelihood (Luceno’s Algorithm)
- Exact Likelihood (Melard’s algorithm)
- Non-linear optimization algorithms for final parameter estimation:
DFP-MIN,
Powell’s Algorithm, FRPR-MIN, Simplex-Downhill
H-step ahead forecast and predicted values for observations
4. AR Models
“Model Identification”
- “Proximity” Algorithm
- “Rectangle” Heuristic
Exact Parameter Estimation
- ULS Algorithm
- CLS Algorithm
- Melard’s Algorithm for exact likelihood
- Luceno’s Algorithm for exact likelihood
Preliminary Parameter Estimation
- Burg’s Algorithm
- Durbin/Levinson
H-step ahead forecast and predicted values for observations
5. Exponential smoothing
Forecasting algorithms according Recurrence Form and Error-Correction Form
(where applicable)
Implemented algorithms for estimating the smoothing parameters
- Simplex-Downhill
- Numerical Method
- Grid-Min
- Grid-Min Heuristic
- Powell’s Algorithm
- FRPR-MIN
ES - Constant Level
Constant level, no seasonality - estimation of start-value
- First Value
- Local average
- Global average
- Backcast
- DLS (Discounted Least Squares)
- Combined estimation of start value and smoothing parameter.
Constant level, additive seasonality - estimation of start-value
- First Value
- Local average
- Global average
- Backcast
- Combined estimation of start value and smoothing parameter.
Constant level, multiplicative seasonality - estimation of start-value
- First Value
- Local average
- Global average
- Backcast
- Combined estimation of start value and smoothing parameter.
Constant level, multiplicative seasonality - estimation of start-value
- First Value
- Local average
- Global average
- Backcast
- Combined estimation of start value and smoothing parameter.
ES - Damped Trend
Damped Trend, no seasonality (Gardner’s “model 1”) - estimation of
start-values
- First Value
- Local average
- Global average
- Non-Seasonal “Fiber”- Heuristic
- Combined estimation of start values and smoothing parameters.
Damped Trend, no seasonality (Gardner’s “model 2”) - estimation of
start-values
- First Value
- Local average
- Global average
- Non-Seasonal “Bloc”- Heuristic
- Combined estimation of start values and smoothing parameters.
Damped Trend, additive seasonality - estimation of start-values
- Local average
- Global average
- Seasonal “Fiber”- Heuristic
- Combined estimation of start values and smoothing parameters.
Damped Trend, multiplicative seasonality - estimation of start-values
- Local average
- Global average
- Seasonal “Bloc”- Heuristic
- Combined estimation of start values and smoothing parameters.
ES - Exponential Trend
Exponential Trend, no seasonality - estimation of start-values
- First Value
- Local average
- Global average
- Combined estimation of start values and smoothing parameters.
Exponential Trend, additive seasonality - estimation of start-values
- Local average
- Global average
- Combined estimation of start values and smoothing parameters.
Exponential Trend, multiplicative seasonality - estimation of start-values
- Local average
- Global average
- Combined estimation of start values and smoothing parameters.
ES - Linear Trend
Double Exponential Smoothing
- First Value
- Local average
- Global average
- Heuristic
- Combined estimation of start value and smoothing parameter.
Holt-Winter’s non-seasonal Model - estimation of start-values
- First Value
- Local average
- Global average
- Combined estimation of start value and smoothing parameter.
Fast estimation of confidence limits according Yar & Chatfield
Holt-Winter’s Model with additive seasonality - estimation of start-values
- Local average
- Global average
- Combined estimation of start value and smoothing parameter.
Fast estimation of confidence limits according Yar & Chatfield
Holt-Winter’s Model with multiplicative seasonality - estimation of
start-values
- Local average
- Global average
- “Fiber”-Heuristic for linear trends
- Combined estimation of start value and smoothing parameter.
Fast estimation of confidence limits according Yar & Chatfield
Brown’s non-seasonal Model - estimation of start-values
- First Value
- Local average
- Global average
- Combined estimation of start value and smoothing parameter.
Brown’s Model with additive seasonality - estimation of start-values
- Local average
- Global average
- Combined estimation of start value and smoothing parameter.
Brown’s Model with additive seasonality - estimation of start-values
- Local average
- Global average
- Combined estimation of start value and smoothing parameter.
- “Bloc”-Heuristic for linear trends
6. Multivariate Linear Regression
Selection strategy for Variables
- Forward Selection
- Backward Elimination
- Stepwise Selection
- Complete Search
8. User defined Multivariate econometric models
Implementation of user defined models
7. Random Walk
Implementation of the Random-Walk model
8. Confidence Limits
Calculation of confidence limits for forecasts
9. Tests
- Has the TS a non-zero mean
- Has the TS a seasonality of given lengths
- Has the TS a trend
- Has the TS outliers
- TS has all-positive values
10. Model Evaluation
- Penalty function: BIC, AICC
- Evaluation via Hold-Out-Sample
- Error measure: MSE, MAD, MAPE, MdAPE
11. Transformations
Auto-Correlation
Auto-Covariance
- Algorithm for computing the Auto-Covariance directly
- Algorithm using Fast-Fourier-Transformation (FFT)
ACF
ACF according Tunicliffe & Wilson
Direct Computation of the ACF
Fast Fourier Transformation (FFT)
Periodogram
- Compute periodogram without filter
- Use Parzen-Window
Structural Math
- Selection of the optimum forecasting model - out of a given model set
- Selection of the necessary preprocessing (trend & season) – out of a
given “Preprocessing Set”
- Computation of optimum parameters for Models and Preprocessing
- Consistency Test for Multivariate Regression
- Generation of a report, why the suggested model has been preferred over
alternatives
_______________________________________________________
Reply to this item at:
<http://savannah.gnu.org/task/?5868>
_______________________________________________
Message sent via/by Savannah
http://savannah.gnu.org/
- [Savannah-register-public] [task #5868] Submission of Special Interest Group Time Series,
Johann Rost <=