Prozentor  >  Forecasting

Forecasting

Forecasting is founded upon a need to minimise uncertainty and risk, which can effectively be done if the forecasting is conducted with adequate precision.

We have been developing successful forecasting models for years using Prozentor forecasting software. These models are based on innovative and proven scientific methods, including particularly statistical time series analysis, which is primarily relevant for forecasting on the basis of historical price data.

Time series analysis modelling has been a recognised analysis technique at least since Clive W. J. Granger and Robert F. Engle received the2003 Nobel Prize in Economics , and is gaining in importance in commercial forecasting research.

SARIMAX models

SARIMAX models (Seasonal AutoRegressive Integrated Moving Average models with eXogeneous variables) are considered optimal for forecasting prices over time. In these models, the internal dependency structure of historical data is broken down into multiple components.

The auto-regressive (AR) coefficients describe short-term fluctuations. The associated lag order describes how many historical points in time still influence the next forecast. Moving average (MA) coefficients allow more economical parameterisation of short-term fluctuations. The degree of integration (I) factors in whether the absolute level of a time series or percentage changes in it are to be modelled.

Exogenous regressors represent the influence of events (such as football championships), calendar effects (holidays etc.) and other relevant variable factors (such as economic indicators). The error term describes all purely coincidental factors that consequently cannot be predicted, also referred to as ‘white noise’.

We generate short, medium and long-term forecasts, in contrast to many standard statistical forecasting applications.

Characteristics of our forecasting

• Point forecasting with confidence intervals
• Reliable forecasting even with small data volumes
• Utilisation of pre-transformations
• Model blending for enhanced forecasting quality
• Factoring in of major events, calendar effects and external influences (dates database)
• Consistent monthly and quarterly forecasting
• Simultaneous, consistent forecasting for hierarchical market structures
• High-quality multi-step forecasting
• Automated, objective model selection
• Robust forecasting even in extreme market phases and under special circumstances
• Automated identification and utilisation of relevant influence factors (e.g. interest rates, temperature, securities)

The forecast generation process

1. Determine values to be forecast
   (e.g. audience ratings on Saturday evening at 20:15)
2. Determine forecasting horizon (short-term, long-term)
3. Load time series data
4. Determine forecasting model using simulation study
5. Prepare forecast report, including forecast quality levels
6. Select time series well-suited for forecasting
7. Select forecast results export format (PDF file, Excel document, XML)
8. Automated forecast generation