Overview
Nowcasting is often defined as the prediction of the present, the very near future and the very recent past.
rjd3nowcasting provides helps to operationalize the process of nowcasting. It can be used to specify and estimate Dynamic Factor Models. Recent version of the package also includes news analysis. The way the package was conceived is inspired by the GUI add-in developed for JDemetra+ V2.
Installation
Running rjd3 packages requires Java 17 or higher. How to set up such a configuration in R is explained here
Latest release
To get the current stable version (from the latest release):
- From GitHub:
# install.packages("remotes")
remotes::install_github("rjdverse/rjd3toolkit@*release")
remotes::install_github("rjdverse/rjd3nowcasting@*release", build_vignettes = TRUE)- From r-universe:
install.packages("rjd3nowcasting", repos = c("https://rjdverse.r-universe.dev", "https://cloud.r-project.org"))Development version
You can install the development version of rjd3nowcasting from GitHub with:
# install.packages("remotes")
remotes::install_github("rjdverse/rjd3nowcasting")Usage
Once the package is loaded, there are four steps to follow:
- Import data
- Create or update the model
- Estimate the model
- Get results
Detailed information concerning each step can be found in the vignette.
2. Create or update the model
### Create a model from scratch
ns<-ncol(data0_ts)
types<-c("M","M","Q","YoY","YoY","M","M","YoY","YoY","YoY")
loadings1<-matrix(data=TRUE, ns, 2)
#loadings2<-matrix(data=c(rep(TRUE,ns+5),rep(FALSE,5)), ns, 2)
dfm0_init <- create_model(nfactors=2,
nlags=2,
factors_type = types,
factors_loading = loadings1,
var_init = "Unconditional")
### Update an existing model
est0 <- estimate_em(dfm0_init, data0_ts) # cfr. next step
dfm1_init <- est0$dfm # R object (list) to save from one time to another
# or, equivalently,
# dfm1_init <- create_model(nfactors=2,
# nlags=2,
# factors_type = types,
# factors_loading = loadings1,
# var_init = "Unconditional",
# var_coefficients = est0$dfm$var_coefficients,
# var_errors_variance = est0$dfm$var_errors_variance,
# measurement_coefficients = est0$dfm$measurement_coefficients,
# measurement_errors_variance = est0$dfm$measurement_errors_variance)3. Estimate the model
est0 <- estimate_ml(dfm0_init, data0_ts)
# or est0<-estimate_em(dfm0_init, data0_ts)
# or est0<-estimate_pca(dfm0_init, data0_ts) # to be avoided unless you only have monthly data4. Get results
Model and forecasts
rslt0 <- get_results(est0)
print(rslt0)
#> $loadings
#> Sample mean Stdev Coeff. F1 Coeff. F2 Idiosyncratic variance
#> FR_PVI -0.018 3.347 0.651 0.630 0.150
#> FR_TURN 0.178 3.915 0.621 0.603 0.219
#> FR_B1g_BCDE 0.110 4.219 0.145 0.141 0.257
#> FR_BS -5.962 6.726 0.212 0.006 0.126
#> FR_BS_prdexp 7.933 9.487 0.183 0.031 0.419
#> EA_PVI 0.027 2.695 0.646 0.627 0.162
#> EA_TURN 0.190 3.122 0.653 0.638 0.131
#> EA_BS -2.056 8.037 0.236 0.045 0.048
#> EA_BS_prdexp 8.349 9.070 0.235 0.100 0.173
#> EA_PMI_manuf 51.313 5.111 0.249 0.170 0.109
#>
#> $VAR_model
#> F1[-1] F2[-1] F1[-2] F2[-2]
#> F1 0.710 -0.253 -0.690 0.260
#> F2 -0.508 0.489 0.333 -0.624
#>
#> $Innovative_variance
#> F1 F2
#> F1 1.000 -0.151
#> F2 -0.151 1.000
fcsts0 <- get_forecasts(est0, n_fcst = 3)
print(fcsts0)
#> $forecasts_only
#> FR_PVI FR_TURN FR_B1g_BCDE FR_BS FR_BS_prdexp EA_PVI
#> Apr 2024 NA -0.26520376 -0.9066921 NA NA NA
#> May 2024 -0.5557794 -0.41682622 -1.0911187 NA NA -0.3988934
#> Jun 2024 0.1699863 0.36989280 -0.5998834 -9.138949 3.915219 0.1656424
#> Jul 2024 0.1685482 0.36225137 -0.5747306 -6.960827 6.381705 0.1606429
#> Aug 2024 -0.2416552 -0.07873533 -0.1738075 -6.907477 6.470485 -0.1560932
#> Sep 2024 -0.2809204 -0.09964879 -0.0680737 -8.495403 4.882437 -0.1728226
#> EA_TURN EA_BS EA_BS_prdexp EA_PMI_manuf
#> Apr 2024 -0.111319199 NA NA NA
#> May 2024 -0.291450017 NA NA NA
#> Jun 2024 0.301963860 -6.471711 3.067721 47.94640
#> Jul 2024 0.278853788 -3.807055 5.688887 49.26765
#> Aug 2024 -0.043318670 -3.706038 5.851589 49.39729
#> Sep 2024 0.002645719 -5.383872 4.670237 49.15457
plot(fcsts0, series_name = "FR_PVI")
# Note: impact of Covid-19 on the model estimate could be eliminated by treating affected indicators as missing data during this period. This has not been done in this example.News analysis
news1 <- get_news(est0, data1_ts, target_series = "FR_PVI", n_fcst = 3)
news1$impacts
#> series period expected_value observed_value
#> 1 FR_PVI 5-2024 -0.551604008710279 -2.10323226057412
#> 2 FR_TURN 4-2024 -0.158827203273009 1.5012792389185
#> 3 FR_BS 6-2024 -9.16421530477352 -7.9
#> 4 FR_BS_prdexp 6-2024 3.89240182586572 2.3
#> 5 EA_PVI 5-2024 -0.395537116403581 -0.615386557437869
#> 6 EA_TURN 4-2024 -0.0213558345482549 0.606324941105818
#> 7 EA_BS 6-2024 -6.49531828333652 -10.1
#> 8 EA_BS_prdexp 6-2024 3.05743620889786 0.5
#> 9 EA_PMI_manuf 6-2024 47.9509402506111 45.8
#> 10 TOTAL <NA> <NA> <NA>
#> news impacts(6-2024) impacts(7-2024)
#> 1 -1.55162825186384 0.0625426516614255 0.25062588561052
#> 2 1.6601064421915 -0.102937609850668 -0.0213553608864313
#> 3 1.26421530477352 -0.0324518053679323 -0.00408631517117787
#> 4 -1.59240182586572 -0.013283314097744 -0.00280694611263804
#> 5 -0.219849441034288 0.00987846987975813 0.0404545554456015
#> 6 0.627680775654073 -0.0853868086082372 -0.0178364405879784
#> 7 -3.60468171666348 -0.496517141653302 -0.101958959531745
#> 8 -2.55743620889786 -0.267633977600985 -0.0504429432184094
#> 9 -2.1509402506111 -1.17139914720656 -0.216781814558934
#> 10 <NA> -2.09718868284424 -0.124188339011193
#> impacts(8-2024)
#> 1 0.0349315424038756
#> 2 0.0315210711205507
#> 3 0.00797192122461071
#> 4 0.00468621804722846
#> 5 0.00571578862082551
#> 6 0.0254483699135738
#> 7 0.171450179562596
#> 8 0.0867506968615175
#> 9 0.374677629475995
#> 10 0.743153417230773
news1$forecasts
#> 6-2024 7-2024 8-2024
#> old_forecasts 0.1699863 0.16854824 -0.2416552
#> revised_forecasts 0.1376619 0.15925989 -0.2335544
#> new_forecasts -1.9595268 0.03507155 0.5095990
plot(news1)
Package Maintenance and contributing
Any contribution is welcome and should be done through pull requests and/or issues. pull requests should include updated tests and updated documentation. If functionality is changed, docstrings should be added or updated.
Licensing
The code of this project is licensed under the European Union Public Licence (EUPL).