-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathG-news-v3-vs-v2.qmd
187 lines (100 loc) · 6.15 KB
/
G-news-v3-vs-v2.qmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
# New Features in v 3.x family {.unnumbered #v3-vs-v2}
## In this chapter
This chapter provides an overview of the new features in version 3.x as well as significant modification of display or content for features already available in v 2.2.4. Compared to its predecessor, version 3.x provides:
- Additional Algorithms for Seasonal Adjustment (SA), Benchmarking and Temporal Disaggregation (TD), Nowcasting, Revision Analysis
- More stand alone time series tools
- More "acceptable" frequencies in SA
- New SA (mass) production possibilities
<!-- ## New modular structure -->
## Seasonal Adjustment and Modelling
### Seasonal adjustment algorithms {#v3-vs-v2-sa}
#### Improvements on historical algorithms
Improvements X-13-Arima and Tramo-Seats (historical JD+)
- New acceptable data frequencies for seasonal adjustment and modelling of low frequency data:
- In v3.x Low frequency data: $p$ in ${2,3,4,6,12}$ is admissible in all
algorithms (historical and new)
- In version 2, only Tramo-Seats supported all these frequencies, whereas
X13-Arima was restricted to $p$ in ${2,4,12}$
- outlier correction taken into account when
selecting decomposition scheme
- ex-ante leap year correction added to Tramo-Seats (like in X13)
- automatic trading day regressors selection from pre-defined sets built, according to groups of days
- specification: split into two distinct concepts, which can be
directly manipulated by the user:
- reference (or domain) specification: a global set of constraints
inside of which estimation will be performed
- point (or estimation) specification: contains all parameter choices resulting from estimation
The user can transform a given "estimation specification" in a user defined specification.
#### New algorithms in v3.x
Tramo-Seats and X-13-Arima share a very similar and sophisticated
pre-adjustment process for the Arima model selection phase.
For new algorithms, the philosophy is to offer
- a simplified pre-adjustment on the arima modelling side, reduced to
airline model
- several enhanced decomposition options
- stl+ ("+" stands for airline based pre-adjustment)
- x12+: airline based pre-adjustment + new trend estimation
filters (Local Polynomials)
- seats+ (to come in the target v3 version): airline based
pre-adjustment + AMB decomposition
#### SA with Basic Structural Models (BSM) available in GUI
In version 3.x, SA with Basic Structural Models is a fully integrated
process with outlier detection, calendar correction and options on
external regressors.
Fundamentally it is a one-step estimation, performing pre-adjustment and
decomposition (with explicit components) in the same run
This makes regression variable selection more complicated:
- first step: a variable selection is performed with a Tramo like
airline model regression
- second: the entire structural model is estimated
In version 3.x this process is available from the graphical user interface.
#### SA algorithms extended for high-frequency data
All algorithms are available via an R package and will be available in GUI (in target v 3.x version)
- Extended Airline estimation, reg-Arima like (`rjd3highfreq` and GUI )
- Extended Airline Decomposition, Seats like (`rjd3highfreq` and GUI )
- MX12+ (`rjd3x11plus`, GUI upcoming)
- MSTL+ (`rjd3stl` and in GUI)
- MSTS (`rjd3sts`, GUI upcoming)
### Modelling Algorithms
| Algorithm | Access in GUI | Access in R (v2) | Access in R (v3) |
|------------------|---------------|---------------------|------------------|
| Reg-Arima | ✔️ | RJDemetra | rjd3x13 |
| Tramo | ✔️ | RJDemetra | rjd3tramoseats |
| Extended Airline | ✔️ (v3 only) | ✖ | rjd3highfreq |
| STS | ✔️ (v3 only) | rjdsts (deprecated) | rjd3sts |
## New SA (mass) production possibilities
<!-- (provide links) -->
### New R Tools for wrangling workspaces
With functions for
- changing raw data path
- customizing specifications
- merging workspaces by series names, as you would do with a data table
These functions are in `rjd3providers` and `rjd3workspace` packages, (already in a v 2.x stable precursor `rjdworkspace`)
### Production fully in R
without a workspace structure
- TS objects and full flexibility for customizing specifications
- new R functions enabling to apply revision policies (`rjd3x13::refresh` and `rjd3tramoseats::refresh`), with even more flexibility on data spans
Inherent shortcoming: data no readable by GUI, depriving of more sophisticated and visual feedback (compared to R) for manual fine tuning.
Solution : new R functions to create GUI readable dynamic workspaces on the fly (in aforementioned packages).
In the target 3.x, additional algorithms (X12+, STL+, BSM) will also be usable in production with a workspace and cruncher (on low frequency data)
## Time series general purpose tools
Version 3.x offers more stand alone tools (mainly in `rjd3toolkit`)
- Tests (seasonality, auto-correlation, normality, randomness...)
- (Fast) Arima Modelling
- Flexible Calendar regressors generation
- Auxiliary variables for pre-adjustment
- Spectral analysis (in GUI)
- Detection of multiple seasonal patterns (Canova-Hansen test)
- State space frame work as a toolbox (`rjd3sts`)
### Canova-Hansen test to identify multiple seasonal patterns
```{r, echo=TRUE, eval=FALSE}
rjd3toolkit::seasonality_canovahansen(data = df_daily$births,
p0 = min(ch.sp), p1 = max(ch.sp), np = max(ch.sp) - min(ch.sp) + 1)
```
{width=250px}
## Underway developments
- Moving Trading Days module integrated in all SA algorithms (for low and high frequency data), with two implementations one based on rolling windows and one on state space modelling
- Using Cubic Splines for smoother seasonal factors estimation of long periodicities ($p=365,25$)