Hello from Prognoz!

How to predict what will happen tomorrow

14.03.16 Prognoz Blog
2100 0
Êðèçèñ: âçëåò è ïàäåíèå

Today’s Prognoz Platform offers a full-featured set of modeling and forecasting tools to explore data, discover dependencies, and build real-life workflow models that help analysts and executives make informed decisions.

In this series of articles, I’ll talk about the modeling capabilities of the Prognoz Platform and provide some examples of real-world implementations.

Why Prognoz?

So, why should you use the Prognoz Platform for modeling tasks? Here are two very good reasons.

First, the platform’s modeling tool offers rich functionality, including a wide range of options and add-ins to construct composite chains of equations and run scenario-based multivariate calculations, including for multilevel forecasting and optimization problems.

Second, you get a rich library of methods and algorithms along with integration with other statistical packages.

Let’s discuss these reasons in greater detail.


In the figure below, you can see the architecture of the Prognoz Platform.



The platform’s modeler applies mostly to the Time Series Analysis and Modeling and Forecasting modules, which I’ve outlined in red in the architecture diagram. However, less sophisticated modeling capabilities are gradually becoming accessible in other modules as well. Let’s see what modeling tools are used for analysis in the Prognoz Platform.

First and foremost, we’re proud to say that the platform offers scenario-based forecasting, or the ability to answer the question, “What if…?” The analyst can add as many scenarios to the model as the number of ifs that he or she has. Scenarios may refer both to various calculation algorithms and to different inputs. Scenarios help you to compare forecasts visually and measure, say, the model’s sensitivity to inputs.


Second, the analyst can solve an objective function and answer the question, “What is needed for…?” The analyst can set the desired goal and then analyze factors to find the best combination for reaching this goal.



Models are visualized as charts with the support for nested structures. A model’s objects are placed in the workspace, where their links are specified. Once visualized, the model enables you to see at a glance what business problem is solved and what are its key entities.


Now, let’s move on to seamless integration with the data warehouse and other statistical packages. After dragging the desired indicator into the workspace, an invisible link is established between a variable and the database, which works in both directions and provides timely delivery of obtained results; I’ll provide more details about integration with other packages later on in this article. Built models and results are exportable to Microsoft Word or Excel in text or graphic format.

Designed to provide simplicity for analysts of different levels of expertise, the Prognoz Platform comes with intuitive and easy-to-use interfaces. Want to see the difference in modeling interfaces for different levels of proficiency? Let’s consider modeling interfaces for two embedded tools and one application powered by the methods and functionality of the Prognoz Platform.

Modeling tools

The first tool is called Time Series Analysis, which appears in the solid red box in the architecture diagram. This tool provides simple and easy analysis of changes in various indicators over time.

The modeler can find the desired series quickly, apply core statistical functions, and specify simple chains of equations. (It may not sound like much, but just wait until you see it in action!) Accessible from desktop and Web applications, Time Series Analysis is intended to make presentations and deliver analytics that don’t require complex multilevel analysis.


The Modeling and Forecasting tool, which appears as the solid yellow box in the architecture diagram, enables you to perform tasks of any complexity, from time series analysis to building models that are composed of thousands of equations. It’s also accessible in the Web and desktop applications.

The desktop application is designed for power users who can fully appreciate the modeling capabilities of the Prognoz Platform, where an equation is a separate object in the base.

The Web application is designed for novice modelers. It’s good for learning and gaining understanding of modeling fundamentals, running scenario calculations for executives, and making presentations. Nothing is easier than to explain the difference between a factor and a dependent variable by drawing a link from one indicator to another.

While supporting almost the same functionality as the desktop application, the Web application isn’t suitable for building complex models, since the output graph will be cluttered with objects.


Finally, let’s look at BIZone, or the Business Intelligence Zone application. BIZone is an add-in powered by the Prognoz Platform and used in multiple real-life implementations. It delivers data and models in an Excel-like interface. BIZone enables the user to build models of any complexity, parametrize them, and run models from different schemas concurrently. Here, an equation is not a separate object but an indicator. This helps to reduce the number of resultant entities.

Recently, we’ve enhanced BIZone with wizards that help the user to skip routines like selecting factors for multiple dependent indicators simultaneously and quality testing in bulk for a group of models based on specified criteria.



Methods library

What I’ve described above is the outer shell accessible to the user. But what controls the numbers inside the system and delivers results to the user?

Here we find our modeling core, or our library of mathematical and statistical methods and models. This library continuously expands as a result of new and complex implementations, competitive analysis, and the requirements of respected analytical agencies, in whose rankings Prognoz is included.

Currently, all methods can be divided into the following classes:

  • Analysis methods. These methods are for basic descriptive statistics and stationarity tests. We use them to analyze data structure, data distribution type, and scatter, as well as to perform other similar analyses at the data preparation stage.
  • Time series analysis. These methods are for the identification of trends, seasonality, and cycles, since they reveal regularities in the historical data of a modeled indicator. We use them to construct short-term forecasts.
  • Data mining. This is, so to say, a modern type of analysis applicable to any area, from the analysis of a consumer-goods shopping cart and the forecasting of customer outflow to the development of marketing campaigns, and so on.
  • Multivariate statistical analysis. This type of analysis is often used together with data mining. We apply it to cluster, segment, and reduce dimensionality of input data.
  • Correlation and regression analysis. This is the most popular forecasting method. An example of this method is impact analysis of macroeconomic indicators, world development indicators, and national policy measures for regional development.
  • Optimization and criterion problems. The most popular problem of these methods is the resource-constrained profit-maximization problem. A good example of our optimization projects is a solution for Danone designed to formulate dairy products.
  • Balance models. Such models include input and output balance.
  • And more. The Prognoz Platform supports general equilibrium models, simulation modeling, and high frequency modeling as well.

Model-building is an iterative process, where fitting a functional relationship is not the end of the process. To provide the required accuracy and adequacy of forecasts, we employ a multilevel quality assurance system that is comprised of:

  • Diagnostic tests for models, which test the factors included in a model for significance and the overall quality of the model
  • Diagnostic tests for forecasts, which verify forecasts, assess the stability of forecasts, and provide analysis of elasticities


In addition to methods from the Prognoz Platform library, users can benefit from the unlimited capabilities of the R modeling package. In our real-life implementations, we also provide integration with EViews.

For more details on methods, you can check out our platform documentation at http://help.prognoz.com/ru/.

Over the next couple articles, we’ll talk about our modeling tools and use cases in more detail. Bye for now!