Skip to content

Home

Jump to bottom
Liedo edited this page Nov 26, 2015 · 28 revisions

JDemetra+ Nowcasting

Nowcasting is often defined as the prediction of the present, the very near future and the very recent past. The plug-in developed at the National Bank of Belgium helps to operationalize the process of nowcasting. It can be used to specify and estimate dynamic factor models and visualize how the real-time dataflow updates expectations, as for instance in Banbura and Modugno (2010). The software can also be used to perform pseudo out-of-sample forecasting evaluations that consider the calendar of data releases, contributing to the formalization of the nowcasting problem originally proposed by Giannone, et al. (2008) or Evans (2005).

Examples constructed or replicated with JDemetra+:

Building a simple nowcasting model

###1. Create a new model inside your workspace### Here we show how to generate a model. Go to the menu STATISTICAL METHODS and select the option Nowcasting. Select the type of nowcasting model desired. For example, "dynamic factor models".

###2. Load the data###

Let's first load the data. Alternative formats are supported. If you work with data on excel, just make sure the dates are in the first column and the name of the series is in the first row:

  • Go to the Providers window, select Spreadsheets (right click) and "Open" your file.
  • Select the sheets that contains your data and copy each series or the whole sheet into the model space (use the drag and drop command).

###3. Model specification details ###

The so-called state-space representation of the factor model is written as follows:

Statespace

where the measurement equation links the N observables to the r underlying factors. Those factors, as shown in the second equation, follow a VAR of order p.

####3.1. Define the measurement equation####

In this example we have variables observed at the monthly and quarterly frequencies. There is the option to transform the series in multiple ways, including first differences or seasonal adjustment. The likelihood of the model which is will be important for the estimation, will be given by the transformed data. However, the forecasts will be calculated for the raw data.

The link between the transformed time series and the factors can be very sophisticated. Three options are possible for the moment:

  • Variables expressed in terms of monthly growth rates can be linked to a factor representing the underlying monthly growth rate of the economy if "M" is selected
![M](./images/M.png)
  • Monthly or quarterly variables that are correlated with the the underlying quarterly growth rate of the economy can be linked to a weighted average of the factors representing the underlying monthly growth rate of the economy. Such a weighted average is meant to represent quarterly growth rates, and it is implemented by selecting "Q":
![M](./images/Q.png)
  • The variables can also be linked to the cumulative sum of the last 12 monthly factors. If the model is designed in such a way that the monthly factors represent monthly growth rates, the resulting cumulative sum boils down to the year-on-year growth rate. Thus, variables expressed in terms of year-on-year growth rates or surveys that are correlated with the year-on-year growth rates of the reference series should be linked to the factors using this link:
![M](./images/YoY.png)

The factor loading structure can incorporate zero restrictions. Users should simply select which factors load on which variables. The following example helps to define a measurement equation for a very simple model for nowcasting German GDP:

####3.2. Define the transition equation#### The so-called transition equation is a representation of the r underlying factors in terms of a vector autoregressive model of order p. Both parameters can be determined by clicking on the tools M option of Model tab.

####3.3. Save your workspace and model specifications #### The workspace can be saved by clicking on the FILE menu and selecting "save as". Let's first save the workspace with the name "NowGermany_24_11_2015", in reference to today's date.

The name of our new model residing in our workspace ("Dfm1", by default) can be easily changed by applying the right click command. Select "Rename". Call it "Model r=2, p=3", in reference to the fact that the correlation among all variables is exclusively due to r=2 factors, which follow a VAR(p) of order p=3.

####3.4. In practice: Build your model in 3 minutes #### The following video shows how to execute all the steps mentioned in this section. We first load the data, then specify the model, and finally save the workspace and the model. It will only take 3 minutes because the input data does not require any transformation, thereby the column "Series transformations" is left blank. Notice, however, that the "Factors transformation" column needs to be manipulated to make sure the measurement equation is correctly specified.

M

###4. Modifying an existing model###

A workspace can contain a large number of models. In order to create a new model with a larger number of factors (r) or lags (p) than the one defined above, we can proceed as follows:

  • Go to the workspace window and right click on the model you want to modify. Select "Clone".

  • A new model will be added to the list of models that already exist inside your workspace. Proceed as before (right click) with this new model, and select the option "Rename". Call it "Model r=4, p=13", in reference to the fact that it has now four factors following a VAR(13).

###5. Estimation ### [to be completed]

JDemetra+ Nowcasting Home

Clone this wiki locally