Introduction
Welcome! In this course we provide a basic introduction to forecasting using Microsoft Excel. The
course is practical in nature, with only a small amount of mathematical theory being introduced.
Forecasting is everywhere! In pretty much every situation where planning is required, forecasting (in
some form) is needed: to predict, and prepare for, changes in sales, profits, losses; set up next year’s
budget; scheduling staff in a call centre; building new power stations and other public utilities.
Forecasts may be short, or long term. Stock traders may need forecasts of only a few minutes ahead,
whereas city planners may need a ten year population forecast in order to give them enough time to
plan and implement infrastructure initiatives.
In this course we are going to focus on time series i.e., data that arrives at regular intervals (quarters,
months, days, minutes etc). The periods over which data is measured might change, but the
underlying ideas and approaches remain pretty much the same.
In order to forecast we need historical data. Our forecasting models (i.e., Excel) only have the
information contained in the historical data to help make their forecasts (an exception is Excel’s
What-If analysis which we explore later). Successful forecasting allows us to determine underlying
trends/patterns within that given historical data and use that information to make forecasts/predictions
for future time periods. This requires that we can distinguish between genuine trends/patterns and
random/noisy events within the data. This is not always easy.
A good question to ask yourself:
Is the period for which you are going to forecast, likely to be “typical” compared to your
historical data? For example, if the historical stock price has been rising consistently, then it
is extremely unlikely that any forecast model is going to be able to predict a massive drop
tomorrow. Typically forecast models can only predict versions of what they have “seen” in
the data. Forecasting in uncharted territory is dangerous!
Also, not everything can be forecast! Having the last five years winning lottery numbers will not help
you predict next week’s numbers. Some events are just random. Other events can be forecast with
great accuracy. Scientists can forecast the next solar eclipse with great precision. For the most part,
we will be concerned with data and events that fall between these two extremes.
I mentioned above that our forecasts rely purely on historical data. In certain circumstances you may
have extra information about the future (a new transformative product being released, increase in
marketing, staff layoffs etc). Excel’s What-If analysis tool allows forecasts to be adjusted to
incorporate this extra information.
The main focus points for the course are as follows:
- Forecasting models
- Forecast accuracy
- How to optimize forecasts using Solver
- What-If scenario analysis
Let’s get started!