Predicted SSN Values

Posted by James Watson on 30 Oct, 2017

The SSN values provided by the SIDC are smoothed using a sliding window smoothing algorithm to minimise the effect of complex short-term, rapidly-varying components, but does not obscure the slowly-varying component. The window is centred on the current month and extends 6-months in either direction. This has the effect that final values are not fully determined until 6 months after the month in question. (See P.371, eq. 1 for the smoothing equation).

Proppy uses SSN values derived by combining historical/final values and predicted values. Historical SSN values are published by the SIDC at the following location.

The SIDC currently provide predictions produced by the following three algorithms (published at );

Podladchikova and Van der Linden (2012) advocate the use of a Kalman filter to further improve the accuracy of the predicted results by up to 30% for the McNish and Lincoln (ML) series.

Proppy has been modified to permit selection of predicted values from either the Standard Curves (SC) or McNish and Lincoln (ML) predictions with or without the Kalman filter. Although P.371-8 stipulates the use of the Standard Curves without the Kalman filter, users may wish to experiment with other values in a bid to improve the accuracy of predicted values.



 Share on: Twitter / Facebook / Google+ / Email