Presentation on theme: "Introduction to Global and Local Models FORECASTING."— Presentation transcript:
Introduction to Global and Local Models FORECASTING
This presentation uses materials jointly developed by the Integrative Studies 1 teaching team
WHY IS FORECASTING USEFUL? Businesses need to plan for the future. This is true for all types of business. Requires some idea about what the future will be.
How do you do it? Method 1: ROULETTE Assume that the future will be the outcome of a pure chance process, and so is inherently unpredictable. So don’t forecast just guess and hope.
Method 2: OSTRICH Assume that the future will be identical to the present (nothing changes). Method 3: TRENDY The past contains patterns (trends) that will continue into the future. Future values are formed by identifying trends and projecting them forwards.
Method 4: MAD SCIENTIST Find a theoretical cause and effect model: use knowable changes in causes to predict changes in effects. Method 5: KALEIDOSCOPE Develop a set of scenarios that are plausible stories about the future, knowing all that you do right now. Keep all of these in mind. Evaluate your possible options against each of these.
Forecasting involves combining Statistics and our personal knowledge of the background of the data/problem Global Models use all the data Local Models use only a part of the data, often most recent
Singles: Sales Next Year = 0.8 * Sales this Year + 15.6 1997: Sales = 87 Million Forecast for 1998 is 0.8*87+15.6 = 85.2 Million OK for short-term - What about long-term forecasts?
Sales Next Year = 0.8 * Sales this Year + 15.6 Using this equation we can generate forecasts iteratively for any horizon, e.g. Forecast for 1999 = 0.8* Forecast for 1998 + 15.6 0.8*85.2 + 15.6 = 83.8
Proceeding in this way: Forecast for 2008 = 78.8 Million Which is much more credible than any of our Trend Forecasts Perhaps it is Pessimistic?
SALES NEXT YEAR = 0.8 * SALES THIS YEAR + 15.6 (Millions)
Use of a Forecasting method reflects our belief that the future behaviour of the data will be like it was in the past, and will not change over the forecast horizon. This is NOT a Statistical criterion and so cannot be objectively tested. Hence, choice of method requires much care & lots of reasoning as suggested, but also an Act of Faith.