Friday, March 16, 2007

Forecasting in Oxford

I went to a conference on forecasting in Oxford. Some researchers from the Lancaster Centre for Forecasting were presenting their work. Fildes and Goodwin were presenting evidence on sales forecasts by commercial companies.

It was startling to hear how forecasters in those companies were technophobe towards statistics and relied mostly on their own judgement, despite a poor track record. Goodwin explained some behavioural reasons for this fact. Apparently many sales forecasters would feel useless without meddling with the numbers. Also events were usually mis-anticiptated. Even confidence intervals would be seen as too advanced for them.

Some key lessons are:
• Don’t confuse your forecasts with your stock holding decision
􀀹 If you don’t know what your true forecast of demand is, you can never improve your performance.
• Forecasting accuracy is often not measured and monitored
􀀹 It should be!
• Benchmark your statistical forecast errors
􀀹 Some Statistical Systems forecasts are worse than useless!
• Understood why your forecasts are adjusted
􀀹 Too many small adjustments are made leading to no accuracy gains
􀀹 30% of adjustments are made in the wrong direction
• Consider developing your forecasting system to include
􀀹 Notes of why adjustments are made
􀀹 Summaries of the effects of similar adjustments, e.g. promotions

No comments: