Project Objective
To develop advanced forecasting models for petrochemical prices to enable well-informed planning decisions and hedging against price volatility in a largely unregulated market.
Client
Industry business intelligence publication arm of one of the world's leading media publishing houses.
Approach
A long series of industry investigation and analysis and interviews with experts and other price forecasters informed the experimentation process. Experiments were undertaken with a large number of statistical modeling techniques to form the basis of the forecasting models. In order to maximize both short-term (1 month) forecasting accuracy and to develop a reliable view of long-term (10-12 months trends), a rigorous testing methodology was developed. In addition to the advanced models drawn from financial market practices, an experimental Market Sentiment Index was developed to attempt an increase in accuracy of predicting changes in direction of price evolution.
| Model Type |
Most suited Data |
Forecasting Horizon |
Model
Shelf-life |
Exponential
Smoothing |
No Trend, varying level |
Short |
Short |
| Holt's Method |
Varying trend & levels, No seasonality |
Short |
Short |
| Holt-Winter's Method |
Varying trend & levels, considers seasonality |
Short to Medium |
Medium |
| ARIMA |
Varying trend, levels, & seasonality |
Short to Medium |
Long |
| GARCH |
Data with Heteroskedasticity (unequal variance) |
Short to Medium |
Long |
Artificial Neural
Networks |
Large non-linear Datasets |
Short to Medium |
Medium |
|
Criteria for Selection of Forecasting Models |
Solution
A combination of sophisticated univariate and multivariate forecasting models that form the basis for publishing future prices and for editorial commentary in the published reports.
 |
Model’s Performance over a 12 Month Testing Period |
Benefits
Exceptionally accurate forecasts (greater than 95%) were generated for all products along with a price driver based understanding of price movements. Back to Top |
WHY DECISIONCRAFT