Predicting the future, what formula is best?
They are all flawed in some way. But used together, the consensus model produces quite a realistic and reliable forecast.
Predicting the future is always a tricky thing to do, let alone do it well. So many factors influence what happens tomorrow, and many of these are beyond our control. The Sales Pipeline is a perfect example; you can never predict exactly when that next client will sign up.
While forecasting is hard, there are some very clever ways to create a sensible prediction of what may happen next. Typically, forecasting is performed using a mathematical formula applied to historical data. We’ve taken that idea one step further to use what we call the consensus method.
We run a whole range of mathematical formulas interdependently for predicting the future of each business metric. We then review all these results and discard any that don’t follow the consensus. Finally, we take the mean path of all predictions to get the final results.
It’s a little like asking 10 stockbrokers which stocks to buy. You’ll tend to go with the opinions of the majority rather than the few.
Interested in the maths?
Here are some of the mathematical processes we use to predict future business figures.
Naive forecast – “Sometimes, the best predictor for tomorrow is today.”
Autoregressive model – Uses weighted values from the past for future predictions
Moving Average – Takes into account long run seasonality over say, 3, 6 or 12 months
Weighted Moving Average – Evaluates long run seasonality where recent performance has more influence on predictions than the past
ARMA (Autoregressive Moving Average) – Uses a combination of Autoregressive and Moving average models
Exponential Smoothing – Takes into account local seasonality as well as longer term historic fluctuations
ARIMA (Autoregressive Integrated Moving Average) – Assesses historical patterns and places emphasis on more recent months
ARMAX (Autoregressive moving-average model with exogenous inputs model) – Autoregressive moving-average model that uses inputs other than the numbers being predicted
Neural Networks – A machine learning technique that tries to teach itself how to best predict the future based on historical data
You can use other forecast techniques. The trick is in finding the right one or ones that best predict based on the specific data you are forecasting.
Nerd alert: When forecasting time series data, there’s a lot more to consider than which formula to use. You should always run tests for stationarity, serial correlation, structural divergence, and so on to make sure the data is well understood and ‘behaves’ as expected.
Always remember to add your own experience and knowledge when assessing any forecasts as you are the best place to understand all the factors that could affect the future.